University of Central Florida STARS
Electronic Theses and Dissertations, 2004-2019
2017
Factors Influencing Hypoglycemia Care Utilization and Outcomes Among Adult Diabetic Patients Admitted to Hospitals: A Predictive Model
Waleed Kattan University of Central Florida
Part of the Health Services Research Commons, and the Public Affairs Commons Find similar works at: https://stars.library.ucf.edu/etd University of Central Florida Libraries http://library.ucf.edu
This Doctoral Dissertation (Open Access) is brought to you for free and open access by STARS. It has been accepted for inclusion in Electronic Theses and Dissertations, 2004-2019 by an authorized administrator of STARS. For more information, please contact [email protected].
STARS Citation Kattan, Waleed, "Factors Influencing Hypoglycemia Care Utilization and Outcomes Among Adult Diabetic Patients Admitted to Hospitals: A Predictive Model" (2017). Electronic Theses and Dissertations, 2004-2019. 5416. https://stars.library.ucf.edu/etd/5416
FACTORS INFLUENCING HYPOGLYCEMIA CARE UTILIZATION AND OUTCOMES AMONG ADULT DIABETIC PATIENTS ADMITTED TO HOSPITALS: A PREDICTIVE MODEL
by
WALEED M KATTAN M.H.A. The George Washington University, 2010 MBBS, King Abdul-Aziz University, 2002
A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Public Affairs in the College of Health and Public Affairs at the University of Central Florida Orlando, Florida
Spring Term 2017
Major Professor: Thomas T.H. Wan
Waleed M. Kattan 2017
ii ABSTRACT
Diabetes Miletus (DM) is one of the major health problems in the United States. Despite all efforts made to combat this disease, its incidence and prevalence are steadily increasing. One of the common and serious side effects of treatment among people with diabetes is hypoglycemia
(HG), where the level of blood glucose falls below the optimum level. Episodes of HG vary in their severity. Nevertheless, many require medical assistance and are usually associated with higher utilization of healthcare resources such as frequent emergency department visits and physician visits. Additionally, patients who experience HG frequently have poor outcomes such as higher rates for morbidities and mortality.
Although many studies have been conducted to explore the risk factors associated with
HG as well as others that looked into the level of healthcare utilization and outcomes among patients with HG, most of these studies failed to establish a theoretical foundation and integrate a comprehensive list of personal risk factors. Therefore, this study aimed to employ Andersen’s health Behavior Model of health care utilization (BM) as a framework to examine the problems of HG. This holistic approach facilitates enumerating predictors and examining differential risks of the predisposing (P), enabling (E) and need-for-care (N) factors influencing HG and their effects on utilization (U) and outcomes (O).
The population derived from the national inpatient sample of the Healthcare Cost and
Utilization Project (HCUP) database and included all non-pregnant adult diabetic patients admitted to hospitals’ Emergency Departments (EDs) with a diagnosis of HG from 2012-2014.
Based on the BM framework, different factors influencing HG utilization and outcome were grouped under the P, E, or N component. Utilization was measured by patients’ length of stay
(LoS) in the hospital and the total charges incurred for the stay. Outcome was assessed based on
iii the severity ranging from mortality (the worst), severe complications, mild complications, to no complications (the best).
Structural Equation Modeling (SEM) followed by Decision Tree Regression (DTREG) were performed. SEM helped in testing multiple hypotheses developed in the study as well as exploring the direct and indirect impact of different risk factors on utilization and outcome. The results of the analysis show that N is the most influential component of predictors of U and O.
This is parallel to what was repeatedly found in different studies that employed the BM.
Regarding the other two components, P was found to have some effect on O, while E influences the total charge. Interaction effects of predictors were noted between some components, which indicate the indirect effect of these components on U and O. Subsequently, DTREG analysis was conducted to further explore the probability of the different predictor variables on LoS, total charge, and outcome. Results of this study revealed that the presence of renal disease and DM complications among HG patients play a key role in predicting U and O. Furthermore, age, socio-economic status (SES), and the geographical location of the patients were also found to be vital factors in determining the variability in U and O among HG patients.
In conclusion, findings of this study lend support to the use of the BM approach to health services use and outcomes and provide some practical applications for healthcare providers in terms of using the predictive model for targeting patient subgroups (HG patients) for interventions among diabetic patients. Moreover, policy implications, particularly related to the
Central Florida area, for decision makers regarding how to approach the growing problem of DM can be drawn from the study results.
iv ACKNOWLEDGMENTS
At the end of this long trip I would like to emphasize my deep appreciation to all who helped me in various aspects of my academic journey and made this dissertation possible. First and above all, I praise God, the almighty for providing me this opportunity and granting me the capability to proceed successfully. I thank Him for answering my prayers for giving me the strength and blessings to plod on during each and every phase of my life.
There are many people who helped me during the journey of preparing this dissertation.
So, I will attempt to give them their due here, and I sincerely apologize for any omissions. First and foremost, I am extremely grateful to my parents, the ones who can never ever be thanked enough, for the overwhelming love, prayers, care, wisdom, and sacrifices for educating and preparing me for my future. Without their support and proper guidance, it would have been impossible for me to complete my higher education.
My special, profound and affectionate thanks, love, warm gratitude, and deep indebtedness are due to my wife, who has been struggling with me, hand by hand, to secure and shape brighter future. Her love, trust, understanding, sacrifices, prayers, support, commitment and looking after my children during my study all stand behind my achievements.
I would also like to express my love and thanks to ‘the beats of my heart,’ my children for fulfilling my life with everlastingly joyful and happy moment, and it is their love and innocent smiles that have made the hardship of this task bearable. Also I express my deep thanks to my sisters, brothers, father in law and mother in law for their support, inspiration, and valuable prayers.
I immensely thankful to the chair of my dissertation committee and my academic advisor,
Dr. Thomas T. H. Wan for his perfect guidance, mentorship, patience and support throughout my
v doctoral study. I also would like to thank all my dissertation committee members; Dr. Bernardo
Ramirez, Dr. Robyne Stevenson, Dr. Varadraj Gurupur, and Dr. Richard Pratley for their invaluable comments and guidance. It was really honor to work under the guidance of these distinguished doctors.
I would like to offer my sincere thanks to all of the professors who taught me during my graduate studies. They did not hesitate to share their knowledge with me and to inspire my interest of learning. I am also grateful to my all friends and classmates for making my study journey and campus life enjoyable.
vi TABLE OF CONTENTS
LIST OF FIGURES ...... xiii LIST OF TABLES ...... xv LIST OF ABBREVIATIONS ...... xvii CHAPTER 1 : INTRODUCTION ...... 1 1.1 Organization of the chapters: ...... 2
1.2 Background ...... 4
1.2.1 Diabetes background:...... 4
1.2.2 Hypoglycemia background ...... 5
1.2.3 The problem of hypoglycemia ...... 5
1.3 The Study problem ...... 7
1.4 The Study Purpose ...... 9
1.5 Significance of the Study ...... 9
1.5.1 Policy Relevance ...... 9
1.5.2 Practical Application ...... 10
1.6 Theoretical framework: Andersen’s Behavioral Model of Health Services Use (BM) 11
1.7 Research Hypotheses ...... 13
1.8 Research Questions ...... 14
CHAPTER 2 : LITERATURE REVIEW AND THEORETICAL FRAMEWORK ...... 15 2.1 Review of Conceptual/Theoretical Perspectives ...... 15
2.1.1 Previous work on Andersen’s Model ...... 18
2.1.2 Studies that employed the BM using SEM ...... 19
2.2 Prior studies on HG risks factors, utilization, and outcome ...... 20
2.2.1 Studies examined risk factors to hypoglycemia...... 20
vii 2.3 The predisposing component (P) ...... 21
2.3.1 Age ...... 22
2.3.2 Ethnicity ...... 23
2.3.3 Gender ...... 24
2.3.4 Dementia ...... 25
2.3.5 Depression...... 27
2.3.6 Tobacco use ...... 29
2.3.7 Alcohol use ...... 30
2.3.8 Drug abuse ...... 31
2.4 The Enabling component (E) ...... 31
2.4.1 Socio-economic Status (SES) ...... 31
2.4.2 Insurance coverage...... 33
2.4.3 Geographical location of the patient and the hospital ...... 34
2.5 The Need-for-Care component (N) ...... 35
2.5.1 Type of DM...... 35
2.5.2 Diabetes Complications ...... 36
2.5.3 Uncontrolled Diabetes ...... 36
2.5.4 Chronic Kidney Disease (CKD) ...... 37
2.5.5 Liver disease ...... 38
2.5.6 Hypertension ...... 39
2.5.7 Malignancy ...... 39
2.5.8 Hyperlipidemia ...... 39
2.5.9 Low Body Mass Index (BMI) ...... 40
viii 2.5.10 Charlson Comorbidities Index ...... 41
2.6 Studies examined the utilization and/or outcomes among HG patients ...... 42
2.6.1 HG Utilization (U) ...... 43
2.6.2 HG Outcomes (O) ...... 46
2.7 Summary of Gaps in the Literature ...... 50
CHAPTER 3 : METHODOLOGY ...... 51 3.1 Introduction ...... 51
3.2 Research Questions ...... 51
3.3 Research Design ...... 52
3.4 Data source and sample ...... 52
3.5 Subjects for the Study ...... 53
3.6 Study Variables ...... 54
3.6.1 Predisposing component (P): ...... 56
3.6.2 Enabling component: ...... 56
3.6.3 Need-for-Care (N):...... 57
3.6.4 Utilization (U) ...... 57
3.6.5 Outcome (O) ...... 58
3.7 Power Analysis and Sample Size Justification ...... 58
3.8 Statistical Analysis ...... 59
3.8.1 Structural Equation Modeling (SEM) ...... 59
3.8.2 Confirmatory Factor Analysis (CFA) ...... 61
3.9 Measurement Models ...... 62
3.10 Structural Equation Model with the Measurement Models ...... 64
ix 3.11 Validation of the Models ...... 66
3.11.1 Factor loadings (standardized regression coefficient) ...... 66
3.11.2 Goodness-of-fit statistics ...... 66
3.11.3 Modification indices between errors ...... 68
3.12 Decision Tree Regression (DTREG) ...... 68
CHAPTER 4 : RESULTS ...... 69 4.1 Introduction ...... 69
4.2 Descriptive Statistics ...... 69
4.3 Correlation Analysis ...... 71
4.4 Confirmatory Factor Analysis (CFA) ...... 72
4.4.1 The Predisposing component (P) or Risk Propensity ...... 72
4.4.2 The Enabling component (E) ...... 74
4.4.3 The Need-for-Care Component (N) ...... 77
4.5 Structural Equation Modeling with the Measurement Models ...... 79
4.5.1 LoS Model ...... 79
4.5.2 Total Charge Model ...... 82
4.6 Decision Tree Regression (DTREG) Results ...... 84
4.6.1 DTREG for LoS ...... 84
4.6.2 DTREG for Total Charge ...... 86
4.6.3 DTREG for Outcome ...... 88
4.7 Hypothesis Testing ...... 90
4.7.1 Hypothesis One ...... 90
4.7.2 Hypothesis Two ...... 91
x 4.7.3 Hypothesis Three ...... 91
4.7.4 Hypothesis Four ...... 91
4.7.5 Hypothesis Five ...... 92
4.8 Summary of the Chapter ...... 93
CHAPTER 5 : DISCUSSION, IMPLICATIONS, LIMITATIONS AND CONCLUSIONS ...... 94 5.1 Summary ...... 94
5.2 Discussion of the impact of different components on length of stay ...... 94
5.3 Discussion of the impact of different components on total charge ...... 97
5.4 Discussion of the impact of different components on outcome ...... 98
5.5 Implications ...... 99
5.5.1 Policy Implications ...... 99
5.5.2 Theoretical Implications ...... 102
5.5.3 Practical Implications...... 103
5.6 Limitations ...... 104
5.7 Recommendations for Future Research ...... 105
5.8 Conclusions ...... 106
APPENDIX A: PREVIOUS STUDIES RELATED TO HG RISK FACTORS, UTILIZATION, AND OUTCOMES ...... 108 APPENDIX B: PREVIOUS STUDIES THAT EMPLOYED THE BM...... 112 APPENDIX C: FREQUENCY TABLE FOR THE CATEGORIZED VARIABLES ...... 115 APPENDIX D: HISTOGRAMS FOR THE VARIABLES ...... 118 APPENDIX E: SPEARMAN'S RHO CORRELATIONS...... 124 APPENDIX F: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR LoS ...... 129
xi APPENDIX G: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR TOTAL CHARGE ...... 132 APPENDIX H: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR OUTCOME ...... 135 APPENDIX I: INTERNATIONAL CLASSIFICATION OF DISEASES NINTH REVISION (ICD-9) DIAGNOSTIC CODES USED TO EXTRACT VARIABLES ...... 138 APPENDIX J: ORANGE, OSCEOLA, AND SEMINOLE COUNTY STATISTICS ...... 143 APPENDIX K: UCF IRB APPROVAL LETTER ...... 145 REFERENCES ...... 147
xii LIST OF FIGURES
Figure 1. The Initial Health Behavior Model (Andersen, 1968) ...... 15
Figure 2. The revised Health Behavior Model (Andersen, 1995)...... 16
Figure 3. Common Configurations of the Health Behavior Model (Graham et al., 2017)...... 17
Figure 4. Measurement Model for the Predisposing Component (Risk Propensity as an
Exogenous Latent Variable)...... 62
Figure 5. Measurement Model for the Enabling Component (Exogenous Latent Variable) ...... 63
Figure 6. Measurement Model of the Need-for-Care Component (Need-for-care as an
Exogenous Latent Variable)...... 63
Figure 7. Andersen Behavior Model of Healthcare Utilization: Predictors of Hypoglycemia LoS
and Outcome - Simplified ...... 64
Figure 8. Andersen Behavior Model of Healthcare Utilization: Predictors of Hypoglycemia Total
Charges and Outcome - Simplified ...... 65
Figure 9. Measurement Model for the Predisposing Component (An Exogenous Latent Variable)
- Generic ...... 72
Figure 10. Measurement Model for the Predisposing Component (An Exogenous Latent
Variable) - Modified ...... 73
Figure 11. Measurement Model for the Enabling Component (An Exogenous Latent Variable) -
Generic ...... 75
Figure 12. Measurement Model for the Enabling Component (An Exogenous Latent Variable) -
Modified ...... 75
Figure 13. Measurement Model for the Need-for-Care Component (An Exogenous Latent
Variable) - Generic...... 77
xiii Figure 14. Measurement Model for the Need-for-Care Component (An Exogenous Latent
Variable) – Modified ...... 78
Figure 15. Structural Equation Model with the Measurement Models for LoS and Outcome ..... 80
Figure 16. Structural Equation Model with the Measurement Models for Total Charge and
Outcome ...... 82
Figure 17. DTREG Tree of the Predictors for LoS ...... 85
Figure 18. DTREG Tree of the Predictors for Total Charge ...... 87
Figure 19. DTREG Tree of the Predictors for Outcome ...... 89
Figure 20. Relative Importance of the Variables in (P) ...... 130
Figure 21. Relative Importance of the Variables in (E) ...... 130
Figure 22. Relative Importance of the Variables in (N) ...... 130
Figure 23. Relative Importance of the Variables in the overall LoS model ...... 131
Figure 24. Relative Importance of the Variables in (P) ...... 133
Figure 25. Relative Importance of the Variables in (E) ...... 133
Figure 26. Relative Importance of the Variables in (N) ...... 133
Figure 27. Relative Importance of the Variables in the overall Total Charge model ...... 134
Figure 28. Relative Importance of the Variables in (P) ...... 136
Figure 29. Relative Importance of the Variables in (E) ...... 136
Figure 30. Relative Importance of the Variables in (N) ...... 136
Figure 31. Relative Importance of the Variables in the overall Outcome model ...... 137
xiv LIST OF TABLES
Table 1. List of study variables ...... 12
Table 2. Operational Definition and Measurement Instruments for Study Variables ...... 55
Table 3. Goodness-of-Fit Indices and Criteria for Model Validation ...... 67
Table 4. Descriptive Statistics...... 70
Table 5. Parameter Estimates for the Predisposing Component Measurement Models ...... 74
Table 6. Goodness-of-Fit Statistics of Generic and Revised Model for the Predisposing
Component ...... 74
Table 7. Parameter Estimates for the Enabling Component Measurement Models ...... 76
Table 8. Goodness-of-Fit Statistics of Generic and Revised Model for the Predisposing
Component ...... 76
Table 9. Parameter Estimates for the Need-for-Care Component Measurement Models ...... 78
Table 10. Goodness-of-Fit Statistics of Generic and Revised Need-for-Care Component ...... 79
Table 11. Goodness-of-Fit Statistics of Generic SEM for LoS ...... 81
Table 12. Parameter Estimates for the Generic LoS SEM...... 81
Table 13. Goodness-of-Fit Statistics of Generic and Revised SEM for Total Charge ...... 83
Table 14. Parameter Estimates for the Generic Total Charge SEM ...... 83
Table 15. Summary of DTREG Analysis of the Predictors for LoS – ranked from highest to
lowest ...... 86
Table 16. Summary of DTREG Analysis of the Predictors for Total Charge – ranked from
highest to lowest ...... 88
Table 17. Summary of DTREG Analysis of the Predictors of Outcome – ranked from worst to
best ...... 90
xv Table 18. Summary of Hypothesis Testing Results ...... 93
Table 19. Previous studies related to HG risk factors, utilization, and outcomes ...... 109
Table 20. Previous studies that employed the BM ...... 113
Table 21. Frequency Table for the Categorized Variables ...... 116
Table 22. Spearman's rho Correlations for (P) ...... 125
Table 23. Spearman's rho Correlations for (E) ...... 126
Table 24. Spearman's rho Correlations for (N) ...... 127
Table 25. Spearman's rho Correlations for LoS, Total Charge, and Outcome ...... 128
Table 26. ICD-9 Codes Used to Extract Variables from Original HCUP NIS Data ...... 139
Table 27. Orange, Osceola, and Seminole County Statistics Table...... 144
xvi LIST OF ABBREVIATIONS
AA African American AACE American Association of Clinical Endocrinologists ADA American Diabetes Association AGFI Adjusted Goodness-of-Fit Index AHRQ Agency for Healthcare Research and Quality AMOS Analysis of Moment Structures BG Blood Glucose BGL Blood Glucose Level BM Andersen’s Health Behavior Model BMI Body Mass Index CAD Coronary Artery Disease CDC Center for Disease Control and Prevention CFI Comparative Fit Index CI Confidence Interval CKD Chronic Kidney Disease CN Hoelter’s Critical N CR Critical Ratio CV Cardiovascular df Degree of Freedom ED Emergency Department eGFR estimated Glomerular Filtration Rate GFI Goodness-of-Fit Index GLM Glucose Lowering Medication HCUP Healthcare Cost and Utilization Project HF Heart Failure HG Hypoglycemia HR Hazard Ratio
xvii ICU Intensive Care Unit IDeg Insulin Degludec IGlar Insulin Glargine IRB Institutional Review Board ICD9 International Classification of Diseases Ninth Revision JAMA Journal of American Medical Association LoS Length of Stay MI Myocardial Infarction NHIS National Health Interview Survey
NIDDK National Institute of Diabetes and Digestive and Kidney Diseases
NIS National Inpatient
NPH Neutral Protamine Hagedorn OR Odds Ratio QoL Quality of Life PCP Primary Care Physician
RF Renal Failure
RMSEA Root Mean Square Error Approximate RR Relative Risk SEM Structural Equation Modeling SES Socio-Economic Status SH Severe Hypoglycemia SPSS Statistical Package for the Social Sciences SRW Standardized Regression Weight T1DM Type 1 Diabetes Mellitus T2DM Type 2 Diabetes Mellitus TLI Tucker Lewis Index UCF University of Central Florida UK United Kingdom
xviii URW Unstandardized Regression Weight U.S. United States VA Veteran Affairs VBP Value-Based Payment
xix CHAPTER 1 : INTRODUCTION
Diabetes is one of the major health problems in the United States (Diabetes.org, 2014).
Despite all efforts made to combat this disease, its incidence and prevalence are steadily increasing (CDC, 2014). When diagnosed, doctors advise their patients to change their lifestyle, improve their eating behaviors, and maintain regular physical exercise (NIDDK, 2014). All these efforts aim to achieve better control on the level of glucose in the patients’ blood. However, only a few patients succeed in doing so, whereas most require supplementation with medications called “glucose lowering medications” (GLM), which help in reducing the amount of glucose circulating in their blood to achieve what is known as a state of “glycemic control” (Briscoe &
Davis, 2006).
Unfortunately, sometimes the blood glucose level (BGL) falls below the optimum level and patients develop an event called “hypoglycemia” (HG) (ADA, 2015). Episodes of HG can be mild, moderate, or severe hypoglycemia (SH), depending on the amount of drop in blood glucose
(BG) below the optimum BGL (Barendse, Singh, Frier, & Speight, 2012). It is well-known that
HG is usually associated with higher utilization of healthcare resources such as emergency department (ED) visits, physician visits, Intensive Care Unit (ICU) transfers, and medications
(Curkendall et al., 2009). Additionally, patients who experience HG also have poor outcomes such as more readmissions, higher morbidity, and higher odds of mortality (Bloomfield et al.,
2012).
HG is not uncommon among diabetic patients. It is estimated that 25% of hospital admissions for diabetes are because of SH (Greco & Angileri, 2004). Therefore, there have been many studies conducted to explore the risk factors associated with HG. In addition, many studies have looked at healthcare utilization among these patients as well as their clinical outcomes.
1 However, most of these studies lacked a solid theoretical framework and therefore produced inconsistent knowledge that did not help to resolve the overall problem. Additionally, previous approaches focused on either the risk factors or the utilization and outcomes, without integrating different factors into a coherent and comprehensive profile associated with health services utilization and outcomes.
Therefore, this study aims to employ Andersen’s health Behavior Model of health care utilization (BM) (Babitsch, Gohl, & von Lengerke, 2012) as a framework to examine the problem of HG. This holistic approach helps in determining different risk factors and their implications for improving patients’ utilization and outcomes. To achieve this, the study used
Structural Equation Modeling (SEM) (Weston & Gore, 2006), which is a powerful analytical technique that helps in establishing causal sequence, examining multiple research hypotheses, and determining the overall goodness of fit of the data to the BM. Plenty of studies used SEM to simplify complex relationships between variables as well as to detect any indirect relationships between them (Guo, Perron, & Gillespie, 2009). Hence, this study also provides rich information to healthcare providers and policy-makers and can assist them in better understanding the problem and implementing new approaches to reducing the occurrence of such a serious health disorder.
1.1 Organization of the chapters:
Chapter One begins with a brief background on diabetes and HG, including the problem of HG on the individual level as well as the overall healthcare system. Then, the study problem, purpose, and significance are introduced. Afterward, a brief discussion on the theoretical
2 framework of the study is presented, followed by a summary of the research hypotheses and questions.
Chapter Two reviews the relevant literature. It outlines a brief history of the BM and previous work done that employed the BM. Next, HG risk factors documented in the literature, including P, E, and N components, are discussed in detail. Last, previous studies that focused on
HG utilization and outcomes are presented.
Chapter Three presents the methodology applied to this study. The research questions, design, subjects, and data collection methodology are introduced, followed by a comprehensive discussion about variables and their operationalization. Then, it presents power analysis and sample size justification. Last, the methodology section presents SEM, measurement models, validation of the models, and decision tree regression (DTREG) analysis.
Chapter Four presents the results of the analysis including descriptive statistics, correlation analysis, Confirmatory Factor Analysis (CFA), Structural equation model, and the results from the DTREG analysis. At the end of the chapter, test results of each of the five hypotheses of the study are revealed.
Chapter Five provides a discussion on the study findings, with a special focus on the three main endogenous variables: LoS, total charge, and outcome. Afterward, theoretical, policy, and practical implications of the study are discussed. Finally, major limitations for the study are noted and followed by pertinent recommendations for future research. A conclusion is drawn from this theory-based investigation of HG and its impact on utilization of health services and outcomes of diabetes care.
3 1.2 Background
1.2.1 Diabetes background:
Diabetes is a national health problem in the U.S. that affects about one in every ten
Americans, which is about 30 million people (Diabetes.org, 2014). That number is expanding as there are about 1.7 million new cases of diabetes diagnosed annually (CDC, 2014). From an economic perspective, diabetes imposes a substantial burden on society. In 2012, the estimated total economic cost of diabetes care in the U.S. was $245 billion (CDC, 2014). Medical expenditures incurred by diabetic patients are estimated to be $13,700 per year, which is approximately 2.3 times higher than non-diabetic patients (ADA, 2013), with an average patient’s out-of-pocket expenditure of $1,373 per year (Cunningham & Carrier, 2014). Hospital inpatient care for diabetic patients is considered to be the largest component of their medical expenditures, which is about 43% of the total medical costs related to diabetes (ADA, 2013).
According to the American Association of Clinical Endocrinologists (AACE) and American
Diabetes Association (ADA), more than 20% of all hospital inpatient days were related to the care of diabetic patients (Moghissi et al., 2009).
The two primary types of diabetes are type I diabetes mellitus (T1DM) and type II diabetes mellitus (T2DM). T1DM used to be called “juvenile diabetes” or “insulin-dependent diabetes” because it mostly occurs in younger people who require insulin. It only represents 5-
10% of all diagnosed diabetes (NIDDK, 2014). On the other hand, T2DM, also referred to
“adult-onset diabetes” or “non-insulin-dependent diabetes,” is the most common type that usually affects middle-aged or older people. About 90-95% of people with diabetes are affected by T2DM (NIDDK, 2014).
4 1.2.2 Hypoglycemia background
The primary aim of long-term management of diabetes is to lower the amount of glucose circulating in the blood, or BGL (Briscoe & Davis, 2006). However, when BGL falls below the optimum range, the person is considered to be having an episode of HG (Barendse et al., 2012).
The ADA defines hypoglycemia as “a condition characterized by abnormally low BGL, usually less than 70 mg/dl” (ADA, 2015). HG can be mild, moderate, or severe, based on the amount of drop in BGL below the 70 mg/dl. There are no clear demarcations between each of those categories, but according to AACE and ADA, many clinicians consider a drop in BGL below 50 severe hypoglycemia (SH) (Moghissi et al., 2009). Acute symptoms that are usually associated with hypoglycemia are tachycardia, sweating, anxiety, shakiness, dry mouth, and hunger (ADA,
2005; Ahrén, 2013; Barnett et al., 2010). However, the number of symptoms and their severity are correlated with the acuity of the HG episode.
1.2.3 The problem of hypoglycemia
One of the considerably prevalent complications among diabetic patients is HG. A 2015 systematic review and meta-analysis of population-based studies revealed that the incidence of
HG episodes among diabetic per person, per year is about 20 times (Edridge et al., 2015). While not all HG events are considered severe and many can be self-managed, often, patients will seek medical assistance by visiting the ED, and they will sometimes end up being admitted into the hospital. The Centers for Disease Control and Prevention (CDC) estimated that nearly 300,000
ED visits for adults in the U.S. occurred in 2011, with HG being the first-listed diagnosis and diabetes as another diagnosis (CDC, 2014). Analysis of data from the 1993–2005 National
Hospital Ambulatory Medical Care Survey showed that there were about five million ED visits 5 because of HG, which means that the ED visit rate was 34 (95% CI 30–37) per 1,000 diabetic patients with HG (Ginde, Espinola, & Camargo, 2008).
Hospitalization of ED-treated HG cases is also common. A study published in the Journal of the American Medical Association (JAMA) found that almost one-third (29.3%, CI, 21.8%–
36.8%) of the cases presented to ED with HG resulted in hospitalization (Geller et al., 2014).
Ginde et al. also found that one-fourth of ED-treated HG cases were admitted to the hospital
(Ginde et al., 2008). There are other studies showing that diabetics with HG have higher odds of being hospitalized (Hsu et al., 2013; Jakubczyk & Rdzanek, 2015). Analysis of 10 years’ worth of records from 1998-2009 from the National Health Insurance Research Database of Taiwan with over 77,600 T2DM patients showed that individuals with HG had 3.45 times higher risk of being hospitalized (Hsu et al., 2013). The utilization of healthcare services seemed to be higher among those with more severe HG. A recently published systematic review and meta-analysis estimated that about 10% of SH required hospital treatment in the ED or admission (Jakubczyk
& Rdzanek, 2015). Davis, et al. (2010) conducted a longitudinal observational cohort study with data from 1999-2006 focusing on diabetic patients and found that 24.2% of SH cases requested ambulance attendances only, 43.9% visited the ED, and about 31.8% needed to be admitted to the hospital. Another Canadian study with data from 2004-2009 found that patients who experienced previous episodes of SH had higher risks for hospitalizations than those who never experienced SH episodes (HR 2.80, 95% CI 1.55–5.06) (Majumdar et al., 2013).
In addition to the increasing odds of ED visits and hospital admissions among diabetic patients with HG, those who were admitted to the hospital tended to stay an average of three days longer than others who did not have HG (Curkendall et al., 2009; Gómez et al., 2015;
Signorovitch et al., 2013; Turchin et al., 2009). Moreover, they were at a higher risk of being
6 transferred to the intensive care unit (Blosch, Chernoff, & Nijjar, 2010; Farrokhi et al., 2012;
Garg, Hurwitz, Turchin, & Trivedi, 2013). Readmission rates among HG patients tended to be higher among diabetics who experienced HG (Quilliam, Simeone, & Ozbay, 2011; Zapatero et al., 2014). Because of all of the above-mentioned utilization channels related to HG, the cost associated with the treatment of HG cases was found to be tremendously higher than those with no HG (Curkendall et al., 2009; Quilliam, Simeone, Ozbay, & Kogut, 2011; Zhang et al., 2010).
Another problem of HG identified by Bloomfield, et al. (2012) was that those who experienced HG usually developed unpleasant medical complications and poor outcomes. For example, HG was consistently found to be associated with cardiovascular (CV) complications such as MI, stroke, and arrhythmias studies (Goto, Arah, Goto, Terauchi, & Noda, 2013; Yeh et al., 2015). Other common consequences of HG included falls and fall-related fractures, sepsis, and seizures. Additionally, inpatient mortality rates among HG patients were 7-12% higher than the rates of diabetics without HG (Curkendall et al., 2009; Gómez et al., 2015). There are also many other long-term consequences of HG that are not included in this study. These include issues that affect the quality of life (QoL) such as absence from work and decreased productivity at work, dementia, and impairment of cognitive functions (Bloomfield et al., 2012; Lopez,
Annunziata, Bailey, Rupnow, & Morisky, 2014).
1.3 The Study problem
There are numerous well-conducted studies on the risk factors for HG as well as the utilization and outcomes of HG events (see Appendix A). Studies that focused on risk factors usually included national data and concentrated on the incidence and prevalence of the HG events. In addition, many of these studies were very specific and focused only on the HG risks
7 among certain medications, specific age groups such as the elderly, or specific comorbidities such as kidney diseases. Moreover, several studies looked only into one type, either T1DM or
T2DM. All of these studies provided useful results about the most relevant risk factors of HG.
However, the review showed that these studies lack a holistic approach, which integrates risk factors into a coherent predictive model with the level of utilization and/or outcomes.
Furthermore, most of these studies did not include a control group. The analyses of the risk factors were descriptive in nature or included minimal use of regression analysis or predictive modeling.
On the other hand, studies that examined the level of utilization, such as LoS or cost of care, or the outcomes, provided details about these elements. However, they typically did not include any risk factors, or they focused only on some specific ones such as specific GLMs used and their correlation with LoS or specific age group and their level of utilization and outcomes.
Moreover, neither type of study applied the BM, which integrates the predictors with utilization and outcome in a full and comprehensive model.
The burden that HG imposes on diabetic patients and the overall healthcare system is enormous. At the patient level, diabetics with HG usually have higher out-of-pocket expenditures, greater risk for short-term and long-term complications, and higher mortality rates than other diabetics who do not have HG. In addition, HG episodes usually require hospitalizations of those patients and also are linked to longer hospital stays and higher mortality risks (Jakubczyk & Rdzanek, 2015). On the other hand, HG has a massive impact on the economy as well as resource utilizations of the healthcare system. Therefore, there is a pressing need for studies that assist in providing the bigger picture of this problem. According to the
8 AACE and ADA consensus statement on inpatient glycemic control, there are several topics related to HG that need further exploration (Moghissi et al., 2009). These include:
(1) What is the profile of inpatients at greatest risk for severe hypoglycemia?
(2) What are the short-term and long-term outcomes of patients experiencing severe
hypoglycemia?
(3) What are the true costs of inpatient hypoglycemia?
1.4 The Study Purpose
The literature is loaded with multiple risk factors that contribute to the development of
HG events that require hospitalization. These studies also show that different factors play different roles in the severity of the attack as well as the prognosis and outcomes (Berkowitz et al., 2013). However, there is very limited knowledge about the correlation between different risk factors of HG and their impact on utilization of healthcare services or their correlation with the overall outcome of hospitalization of diabetic patients with HG. Therefore, the primary goal of the study is to identify the relative importance of the determinants in predicting utilization and outcome among those patients and to examine the mechanisms and interactions between those factors. Furthermore, the study seeks to explore the magnitudes of the main effects of each of the predictor variables and their interaction effects on health services use and outcomes.
1.5 Significance of the Study
1.5.1 Policy Relevance
Diabetes is a huge problem in Florida with over 13% of adults having been diagnosed with diabetes, and about 39% having prediabetes (ADA, 2016; Floridacharts.com, 2016). The
9 number of ED visits with a diagnosis of diabetes has increased by 47.2% from 2009 to 2014 and the number of hospitalizations increased by about 16% (Floridahealth.gov, 2017). According to the ADA, total annual diabetes medical expenses in Florida are estimated to be $24.3 billion
(ADA, 2016). HG is the second most common complication among DM patients regarding the percent of discharges in Florida (22.36% of discharges) and the costliest in terms of total charges
(Floridahealth.gov, 2017). Therefore, this study will be useful to health planners and policy makers in identifying diabetes patients who are at higher risks for developing HG events and those who are susceptible to more utilizations and worse outcomes. Consequently, special education and awareness programs can be tailored to target those at higher risks. In addition, community-specific initiatives can be implemented based on the patient characteristics in each community affected by HG and other related complications.
1.5.2 Practical Application
The primary contribution of this study is to operationalize and test the multi-level or nested characters of risk factors as well as the associated resources utilizations and health outcomes among patients who develop HG. Therefore, the study could aid understanding of the level of hospital care utilization as well as the overall treatment outcomes among diabetic patients who might develop HG. In addition, the study shows the magnitude of the effect of each of the factors as well as the presence of any interaction among the factors. This will help healthcare providers, who see patients during clinic visits or in the hospital at the time of admission, to pay particular attention to diabetic patients who have any of these factors and potentially prevent some of the unwanted outcomes. Consequently, those patients might consume fewer healthcare services and achieve better outcomes.
10 1.6 Theoretical framework: Andersen’s Behavioral Model of Health Services Use (BM)
One of the most widely acknowledged models that is used as an explanatory framework in identifying predictors of health care utilization is the BM, which was developed in the 1960s by Ronald M. Andersen (Aday & Andersen, 1974; Andersen, 1995; Andersen & Aday, 1978;
Andersen & Newman, 1973; Chern, Wan, & Begun, 2002; Wan & Soifer, 1974). The BM is a multivariate model that integrates three different individual determinants of health services utilization: Predisposing (P), enabling (E), and need-for-care (N) component.
The later revised model incorporated a critical component for researchers and policy makers, which is outcome (Andersen, 1995). Outcome is an important dimension that provides a broader understanding of utilization behavior since it is considered a subsequent predisposing component to the individual’s future healthcare utilization.
Predisposing component (P): Ps are “individual characteristics which exist prior to the onset of specific episodes of illness” (Andersen & Newman, 1973). The propensity to utilize medical services differs from one patient to another, and it is subject to a variety of social and demographic factors (Wan & Soifer, 1974). Based on the previous studies that examined risk factors of HG, this study includes the following most commonly used risk factors under (P): age, gender, and ethnicity (Bloomfield et al., 2012) (Table 1). In addition, dementia, depression, tobacco use, alcohol use, and drugs abuse will be included as they were repeatedly considered in many studies that employed the BM (Babitsch et al., 2012).
11 Table 1. List of study variables Predisposing Enabling Need-for-Care Utilization Outcome 1. Age 1. Insurance status 1. DM Complications 1. Hospital LoS 1. Severity of 2. Sex 2. Socioeconomic 2. Type of DM 2. Cost the adverse 3. Ethnicity Status 3. BMI underweight outcome 4. Dementia 3. Hospital Location 4. Hypertension 5. Depression 4. Patient Location 5. Hyperlipidemia 6. Tobacco use 6. Liver Failure 7. Alcohol use 7. Renal Disease 8. Drugs Abuse 8. Uncontrolled DM 9. Cancer 10. Charlson Comorbidity Index (CCI)
Enabling component (E): For people who have the predisposing component, there are means that enhance or inhibit their use of services (Aday & Andersen, 1974; Wan & Yates,
1974). This study considers the most relevant factors related to HG for (E): medical insurance type, socio-economic status (Bloomfield et al., 2012). In addition, two other factors are considered: the geographical location of the patient, and the geographical location of the hospital, as they both repeatedly showed a strong influence on utilization (U) and outcome (O) among HG patients (Table 1).
Need-for-care component (N): This refers to the level of illness that the individual has
(Chern et al., 2002). Parallel to the presence of predisposing and enabling conditions, illness is required to be perceived by the patient or his family in order for the patient to proceed and use health services, such as a clinic or hospital visit (Andersen & Newman, 1973; Wan & Soifer,
1975). Among all other factors, (N) is considered “the most immediate cause of health service use” (Andersen & Newman, 1973). While there are plenty of elements that can be included under (N), this study includes a group of factors that were commonly found to impact (U) and
(O) among diabetic patients who develop HG: DM complications, type of DM, BMI
12 underweight, hypertension, liver failure, kidney disease, uncontrolled DM, and the Charlson
Comorbidity Index (CCI) (Table 1).
1.7 Research Hypotheses
Based on the literature summarized in Appendix A, all three components (P, E, and N) play a role in determining the variability in utilization (U) and outcomes (O). For example, old age, low income level, and the presence comorbidities might increase the individual’s length of stay, health care costs, readmissions, patient falls, and mortality (Berkowitz et al., 2013;
Johnston, Conner, Aagren, Ruiz, & Bouchard, 2012; Nirantharakumar et al., 2012). However, the level of importance of each of these factors might vary. Another issue is that LoS does not always correlate with the total cost. For example, some patients might have a shorter hospital stay while the total costs associated with the intensity of care provided to them are very high. In contrast, some patients might stay longer times in the hospital with some reasonable total costs associated.
Therefore, the first three hypotheses for this study include:
Ha1: The magnitude of different components varies with regard to determining the LoS
among patients who develop HG.
Ha2: The magnitude of different components varies with regard to determining the total
cost among patients who develop HG.
Ha3: The magnitude of different components varies with regard to determining the
outcome among patients who develop HG.
As mentioned earlier, (N), which represented mainly as the level of illness, is the most important element in determining the level utilization. Therefore, plenty of studies focused only
13 on risk factors related to (N), such as the presence of kidney disease and its effects on HG and showed that kidney disease is directly linked to higher costs and LoS (Yu, Lin, Chang, Sung, &
Kao, 2014). So, the fourth hypothesis for this study is:
Ha4: The need-for-care component is the most important component of the predictors of
use, while the effects of other predictors are being simultaneously considered.
Last, certain studies showed that interaction effects among different predictor variables existed. For example, older patients who use Sulfonylurea have longer LoS and have a higher risk for adverse events such as falls or fractures (Deusenberry, Coley, Korytkowski, & Donihi,
2012). By considering this, the fifth hypothesis for this study is:
Ha5: There are statistically significant interaction effects (e.g., P*N, N*E, and E*N) of
predictor variables on utilization and outcomes.
1.8 Research Questions
RQ1. Which component is most important in determining the LoS among patients who
develop HG, when other factors are held constant?
RQ2. Which component is most important in determining the total cost among patients who
develop HG, when other factors are held constant?
RQ3. Which component is most important in determining the outcome among patients who
develop HG, when other factors are held constant?
RQ4. Is the need-for-care the most important predictor of use, while the effects of other
predictors are simultaneously considered?
RQ5. Do the predictor variables (P, E, and N) show any significant interaction effects on
utilization and outcomes?
14 CHAPTER 2 : LITERATURE REVIEW AND THEORETICAL FRAMEWORK
2.1 Review of Conceptual/Theoretical Perspectives
The behavioral model of health services utilization, which is often referred to as the
“health behavior model” (BM) (Babitsch, Gohl, & von Lengerke, 2012), was developed by
Ronald Andersen in the late 1960s as a framework to help understand which factors influence families to use health services (Figure 1) (Andersen, 1995). Since then, the BM has undergone multiple revisions and modifications until the last revised model, which was published in 1995
(Aday & Andersen, 1974; Andersen, 1995; Andersen & Aday, 1978; Andersen & Newman,
1973).
Figure 1. The Initial Health Behavior Model (Andersen, 1968)
The most recent model embraced two major aspects: the focus on individuals rather than families and the addition of outcome. Because of the heterogeneity among family members, it was more appropriate to employ the model at the individual level. Furthermore, the revised model helps in providing more accurate reflections about the different factors. On the other hand, including outcome in the model represents a dynamic correlation between the after-effects of
15 utilizing health services. It also denotes a feedback loop of possible future use of health care services (Figure 2) (Shepherd, Locke, Zhang, & Maihafer, 2014).
Figure 2. The revised Health Behavior Model (Andersen, 1995).
The fundamental principles of the BM rely on categorizing an individual’s determinants of health service utilization into three categories: (1) predisposing component; (2) enabling component; (3) need-for-care (Andersen & Newman, 1973). These factors individually and collectively influence the individual’s level of health services use (Conwell & Boult, 2008).
Individual characteristics that pre-exist before the onset of illness are considered predisposing.
Age, sex, and ethnicity are typical examples of factors under the predisposing component. On the other hand, the enabling component refers to the means and resources available that enable the individual to use the services (Gelberg, Andersen, & Leake, 2000). For example, income and insurance coverage facilitate the use of health services for patients. Last is the need-for-care, which is considered the most immediate cause for utilization (Aday & Andersen, 1974). Need- for-care refers to the individual’s illness level and may include elements such as comorbidities and diabetes type.
16 The BM provides a reliable framework to examine health services utilization because it can be applied for multiple purposes (Gochman, 1997). One of the major advantages of the BM is that it works as an integrated theoretical framework that is practical and applicable to a broad range of studies. Another advantage is that it arrays different factors that influence utilization into three categories. This helps researchers in conceptualizing relevant predictors of utilization rather than just listing all factors without any order. Lastly, it facilitates testing multiple research hypotheses generated from a wide-array of predictors and utilization variables (Gochman, 1997).
Recently, various configurations have been developed among different studies that employed the Andersen’s BM (Graham, Hasking, Brooker, Clarke, & Meadows, 2017). While they all provide the same concept regarding the three components: predisposing, enabling, and need-for care, the difference is about how they impact utilizations (Stiffman et al., 2001). The figure below shows the most common configurations for the BM:
Figure 3. Common Configurations of the Health Behavior Model (Graham et al., 2017).
Reviewing different studies and how they employed the BM model reveals that two major elements usually define the most appropriate model to use: study context and sample
17 characteristics (Babitsch et al., 2012; Graham et al., 2017). Thus, reviewing the literature is key to determining how different components are related to each other and which configuration is relevant to the study.
2.1.1 Previous work on Andersen’s Model
Since the BM was introduced, it has frequently been used in different studies that examined the level of health care utilizations (Babitsch et al., 2012). These studies investigated a broad spectrum of diseases such as diabetes, hypertension, depression, heart diseases, cancer, and many others, with the level of patients’ utilizations of healthcare services (Appendix B).
While the BM has been used to examine wide varieties of illnesses including diabetes, it was not applied to examining factors influencing HG care utilization and outcomes.
With diabetes, the BM was employed in plenty of studies that examined the utilization of care among diabetic patients. Most of these studies examined either T2DM alone or T1DM and
T2DM. Some studies examined general utilization among diabetes patients including the frequency of ED visits, the number of hospitalizations during a certain amount of time, and the number of visits to doctors’ clinics (Gamble & Chang, 2014; Gucciardi, DeMelo, Booth,
Tomlinson, & Stewart, 2009; Jayawant, 2008; Ou et al., 2012). A few others focused on the utilization of specific services such as the eye care exam (Baumeister et al., 2015; DeLawnia, Lu,
& Zo, 2014). Utilization among diabetics was also studied to examine the presence of any racial differences (Chandler & Monnat, 2015; Shenolikar, 2006; Yeboah-Korang, Kleppinger, &
Fortinsky, 2011). Diabetics’ adherence to medication and self-care and their impact on utilization was also examined in other studies (Patel et al., 2015; Yamashita, Kart, & Noe, 2012). Last, the
18 level of healthcare costs associated with diabetes care was investigated in several studies (Hu,
Shi, Pierre, Zhu, & Lee, 2015; Jayawant, 2008; Shenolikar, 2006).
As mentioned earlier, the BM was employed in a broad range of issues other than diabetes. Two of the most frequently examined types of diseases using BM are mental and psychological illnesses. Several studies examined the influence of mental diseases and psychological issues on healthcare utilization (Blaskowitz, 2014; Kilany, 2014; Ko, 2015;
Rhoades et al., 2014; Springer, 2015). Another common issue was examining the utilization level among older populations (Clement, 2015; Manski et al., 2013; Scheetz, 2010; Snih et al., 2006;
Wolinsky, Stump, & Johnson, 1995) or racial differences in utilization (Bazargan, Bazargan, &
Baker, 1998; Wilkinson-Lee, 2008). There were some other studies that specifically examined the impact of having insurance coverage on utilization (Akobirshoev, 2015; Gelberg et al., 2000;
Mallya, 2006; Roberge, 2008; Shepherd et al., 2014).
Similar to diabetes studies, most of the studies that examined mental and psychological illnesses considered hospitalizations, LoS, ED visits, and clinic visits as measures of utilization.
However, not all of the above-mentioned studies used SEM and some employed only the traditional multivariate techniques, which lack the ability to demonstrate the presence of any interaction effects among different predictors. Moreover, the use of traditional multivariate techniques does not help detect any interaction effects among different the predictors.
2.1.2 Studies that employed the BM using SEM
In contrast to the studies mentioned above, there are some others that employed BM to examine the level of utilization and applied SEM as the primary method of analysis. Although these studies examined different arrays of health conditions, two predominant factors were found
19 to influence the level of utilization. The first predictor of utilization was the level of illness, which was usually the most significant component in most of these studies (Brook, Lee, Balka,
Finch, & Brook, 2014; Chern et al., 2002; Stein, Andersen, Robertson, & Gelberg, 2012; Tan,
2009; Wan & Soifer, 1974). The other common predictor was the presence of health insurance coverage, which is usually included in the enabling component (LaHousse, 2009; Stein,
Andersen, & Gelberg, 2007; Tan, 2009; Wan & Soifer, 1974).
2.2 Prior studies on HG risks factors, utilization, and outcome
Numerous studies examined HG risk factors and provided rich information about the relevant factors that lead to hypoglycemia (see Appendix A). Moreover, many others examined the level of utilization and/or the health outcomes among patients with HG. However, these studies either looked at the risk factors and/or the utilization and outcomes, which means that they did not employ a holistic approach that integrates the risk factors with the level of utilization and outcomes among those patients. Also, these studies did not use SEM.
The following sections will discuss the two primary different types of studies: studies that examined HG risk factors only and studies that examined HG utilizations and outcomes.
Relevant findings from these studies will be discussed in a later section (sections 2.3-2.6) when reviewing the different factors.
2.2.1 Studies examined risk factors to hypoglycemia
Plenty of studies examined risk factors to HG among diabetic patients and provided comprehensive knowledge about the most relevant risk factors. Some included large databases with multiple years to determine the incidence of HG and the most significant risk factors
20 (Bruderer et al., 2014; Edridge et al., 2015; Ginde et al., 2008; Giorda et al., 2015; Sämann et al.,
2013). Other studies looked at drug-related risks of HG (Czech et al., 2015; Murad et al., 2012;
Murad et al., 2009), while others either examined patients on insulin regimens or those on oral
GLMs (Gold, Frier, MacLeod, & Deary, 1997; Kostev, Dippel, & Rathmann, 2014; Kostev,
Dippel, & Rathmann, 2015; Sonoda et al., 2015; Weinstock et al., 2015), and some others focused on patients with continuous glucose monitoring (Cox, Gonder-Frederick, Ritterband,
Clarke, & Kovatchev, 2007; Fiallo-Scharer et al., 2011). There are also some other studies that focused on older populations only (Bruderer et al., 2014). Although these studies provided useful information about the risk factors related to HG (sections 2.3-2.6), they focused only on HG events that occurred in the outpatient settings.
There are other studies that investigated risk factors of HG among hospitalized diabetic patients (Abdelhafiz, Bailey, & Sinclair, 2012; D'Netto, Murphy, Mitchell, & Dungan, 2015;
Dendy et al., 2014; Deusenberry et al., 2012; Elliott, Schafers, McGill, & Tobin, 2012; Farrokhi et al., 2012; Gómez et al., 2015; Lin et al., 2010; Pathak et al., 2015; Schloot et al., 2016).
Similar to the studies of outpatients, they did not examine the level of utilization or patient care outcomes (sections 2.4-2.6).
2.3 The predisposing component (P)
A systematic review of studies using BM from 1998–2011 showed that there was a broad range of variables utilized in these studies. However, few “key variables” were found to be repeatedly used (Babitsch et al., 2012). For example, age, sex, and ethnicity were used under (P), whereas indicators such as income and health insurance were used under (E). Because this study is the first to examine HG and because of the nature of secondary data, it will focus only on the
21 key variables to be able to have a solid understanding about HG. Therefore, this study will include the following risk factors under (P): age, ethnicity, gender, depression, dementia, tobacco use, alcohol use, and drug abuse.
2.3.1 Age
Age is consistently reported to increase the risk of HG in most studies. A recent study published by the ADA, which included an observation of over 900,000 patients with diabetes from 2005 to 2011, found that the rate of HG significantly increases with age (Pathak et al.,
2015). Another population-based cohort study in Canada, which included more than 85,000 patients from 2004 to 2009, also found that age substantially increased the risk of HG among diabetic patients (Majumdar et al., 2013). A sensitivity analysis of 8,767 type 2 diabetic patients in China who were recruited between 1995 and 2007 found that old age is an independent predictor for SH (Kong et al., 2014). The hazard ratio per 10 years was found to be 1.50 (1.24–
1.81). Another study that included documents from eight hospitals regarding HG at their EDs found that for every five-year increase in age, the odds of SH among T2DM patients increased by 30% (OR: 1.3, 95% CI 1.20-1.45) (Liatis et al., 2015).
Older people usually experience alterations in their metabolism, become more vulnerable to malnutrition, and frequently have other comorbidities (Lin et al., 2010). Therefore, the risk of
HG among older diabetics increased with GLMs. For example, sulfonylureas, an oral GLM, was repeatedly found to increase the risk of HG among older diabetics. A 2012 nested case-control study in a tertiary care academic medical center that included hospitalized patients aged 65 years and older found that those elderly diabetics were at increased risk for sulfonylurea-related HG
(OR 3.07, 95% CI 1.67–5.63) (Deusenberry et al., 2012). Insulin also is a strong GLM that is
22 usually associated with a high risk for developing HG events among older diabetics (OR 1.04,
95% CI 1.02–1.08) (Farrokhi et al., 2012). A recent study by Geller, et al., (2014) that investigated a national database from 2007-2011 found that older diabetics on insulin have twice the risk for ED visits and five times the risk of hospitalization due to hypoglycemia.
2.3.2 Ethnicity
Ethnicity refers to “a concept that incorporates social, religious, linguistic, dietary, and other variables to identify individual persons and populations” (Witzig, 1996). While race is usually used to classify people by biological features such as skin color, ethnicity is typically more concerned with cultural characteristics such as language and food habits (Caprio et al.,
2008). Therefore, it is more appropriate to examine ethnicity differences when examining chronic illnesses such as DM (Schwartz, 2001) than it is to study race.
Regarding DM, a recent report by the CDC shows that there are differences among ethnicities for diabetes prevalence in the U.S. (CDC, 2014). The highest group was found to be
Native Americans with 15.9%, followed by African Americans 13.2%, Hispanics 12.8%, Asian
Americans 9%, and last are the non-Hispanic whites with only 7.6%. However, a national study on 2012 data that included over 70,000 diabetic patients found that Hispanics have the highest odds of HG episodes when compared to whites and African Americans (Lopez et al., 2014).
Another cross-sectional analysis of a random sample of adults in the Kaiser Permanente
Northern California diabetes registry using data from 2005–2006 showed that the odds of developing HG are 35% and 33% higher in Latinos and African Americans, respectively, than
Whites (Berkowitz et al., 2014). Regarding hospitalization among DM patients, a recent study on
23 eight Southeastern States showed the existence of racial disparities in DM hospitalization among rural Medicare beneficiaries (Wan, Lin, & Ortiz, 2016).
2.3.3 Gender
A systematic review of predictors and consequences of SH showed mixed findings about gender (Bloomfield et al., 2012). However, when considering utilization, women were repeatedly found to use more health services than men used across studies that employed BM on different kinds of diseases (Babitsch et al., 2012). Likewise in diabetes, men usually consumed fewer healthcare services than women did (OR 0.96, 95% CI 0.86-1.91) (Gamble & Chang, 2014).
Analysis of data from 1993–2005 from the National Hospital Ambulatory Medical Care Survey shows that ED visit rates are higher among diabetic females (4.0 per 1,000 ED visits, 95% CI
3.5-4.5) than males (3.4 per 1,000 ED visits, 95% CI 3.0-3.8) (Ginde et al., 2008). Similarly, a study using the 2004-2008 MarketScan® database found that the odds of ED visits among
T2DM males are lower than they are for females (OR 0.82, 95% CI 0.78-0.85) (Simeone &
Quilliam, 2012). Correspondingly, hospital admissions related to HG were lower for males than for females (OR 0.84, 95% CI 0.73–0.96) (Quilliam et al., 2011).
In addition to the gender differences among diabetics’ healthcare utilization, females who develop SH usually had higher utilization as well. It has been found that symptomatic HG in
T2DM (OR 1.08, 95% CI 0.97–1.20) (Berkowitz et al., 2014) as well as in T1DM (IRR 1.52,
95% CI 1.33–1.74) (Giorda et al., 2015) was more common among women . Surprisingly, the overall in-hospital mortality was higher among males. Additionally, a retrospective study using a national database of hospitalized Japanese diabetic patients from 2008–2012 that reviewed 22.7
24 million discharge records showed that in-hospital mortality among females is lower than it is for males (OR 0.637, 95% CI 0.551 -0.736) (Sako et al., 2015).
2.3.4 Dementia
Dementia is defined as “a general term for a decline in mental ability severe enough to interfere with daily life” (alz.org, 2017). It has been a measure considered under (P) in multiple studies that employed the BM for diseases other than HG (Babitsch, Gohl, & von Lengerke,
2012). Concerning its relationship with HG, several studies confirmed that dementia has a positive correlation with HG among diabetic patients (Bruce et al., 2009; Feil et al., 2011;
Feinkohl et al., 2014; Yaffe et al., 2013). A recent systematic review and meta-analysis examined the interaction between HG and cognitive impairment in elderly patients treated with
GLMs among studies conducted between 2005-2015 and found that patients with dementia have a higher risk of developing HG with a pooled odds ratio (95% CI 1.25, 2.06) (Mattishent &
Loke, 2016). Another recent systematic review on predictors and consequences of severe HG in adults with diabetes concluded that dementia is a risk factor for severe HG (Bloomfield et al.,
2012). Similarly, a recent review article on diabetes, dementia, and HG also indicated that multiple studies showed that dementia increases the odds of developing severe HG (Meneilly &
Tessier, 2016).
Plenty of other well-recognized studies also showed a strong correlation between dementia and HG. In the Action in Diabetes and Vascular Disease: Preterax and Diamicron
Modified Release Controlled Evaluation (ADVANCE) study, which included more than 11,000
T2DM patients, higher cognitive function was significantly associated with a lower risk for developing HG (HR 0.93, 95% CI 0.87 to 0.99) (Zoungas et al., 2010). In contrast, severe
25 cognitive dysfunction was found to significantly increase the risk of severe HG (HR 2.10, 95%
CI 1.14–3.87; p = 0.018) (de Galan et al., 2009). In the Action to Control Cardiovascular Risk in
Diabetes (ACCORD) study, which was done at 52 clinical sites in North America with about
3,000 T2DM patients, lower baseline cognitive scores were found to be a predictor of an increased likelihood of clinically significant HG over the following years (Launer et al., 2011;
Meneilly & Tessier, 2016; Punthakee et al., 2012). Similar findings were reported in the
Outcome Reduction with an Initial Glargine Intervention (ORIGIN) trial, which included more than 12,500 patients (Meneilly & Tessier, 2016; Riddle, 2015). Moreover, a large longitudinal cohort study that included 16,667 T2DM patients who were followed up for 27 years from 1980-
2007 found that the fully adjusted dementia HR among patients with one HG episode was 1.42
(95% CI, 1.12-1.78), and among those with 2 or more HG episodes, the HR was 2.36 (HR 2.36,
95% CI, 1.57-3.55) when compared to those with 0 episodes (Whitmer, Quesenberry, & Selby,
2009). In Germany, a large retrospective study of 32,545 patients from 1,072 practices found that dementia increases the adjusted odds for developing HG among T2DM (1.49; 1.31-1.69)
(Kostev, Dippel, & Rathmann, 2014).
Some studies tried to explore why patients with dementia tend to have higher odds of
HG, especially among older diabetic patients. The Fremantle diabetes study examined more than
300 older DM patients and concluded that the inability to self-manage medications among patients with dementia might be the main reason for the development of HG (HR 4.17, 95% CI
1.43–12.13) (Bruce et al., 2009). Another retrospective cohort study of national Veterans Affairs
(VA) administrative/clinical data and Medicare claims that included about 16,000 older T2DM veterans with dementia between 2008-2009 found that in addition to the inability to self-manage medications, those patients had difficulties following prescribed regimens, showed significant
26 weight loss that is usually related to changes in their eating habits such as low appetite, and demonstrated lower capacity to recognize and respond properly to symptoms (Thorpe et al.,
2015).
2.3.5 Depression
It is also referred to as “major depressive disorder” or “clinical depression”, which is “a mood disorder that causes a persistent feeling of sadness and loss of interest” (Mayoclinic.org,
2017). Similar to dementia, depression also has been used in different studies that employed the
BM under (P) (Babitsch et al., 2012). Regarding its relationship to HG, a 2012 systematic review
(Bloomfield et al., 2012) as well as other reviews confirmed that depression significantly increases the odds of developing HG among diabetic patients (Barendse et al., 2012; Jafari &
Britton, 2015). Another large 2015 study was conducted to examine the association of HG with fall-related outcomes among older T2DM patients from 2008-2011 (Kachroo et al., 2015). The
Kachroo et al., study included more than 1.1 million T2DM patients’ records and found that the percentage of dementia among patients who had HG was twice what it was among those without
HG. Very similar results were shown in another observational cohort study that was conducted from 2005 to 2011 and included more than 900,000 diabetic patients (Pathak et al., 2015). That study found that the risk of severe HG was 50% higher among depressed patients when compared to others with no depression (P < 0.001). Last, the ORIGIN trial that included over
12,500 patients found that the HR for HG among depressed patients was 1.28 (1.11-1.48,
P<0.001) (Meneilly & Tessier, 2016; Riddle, 2015).
Studies with smaller samples also provided comparable results (Green, Fox, & Grandy,
2012; Williams, Pollack, & DiBonaventura, 2011). A longitudinal cohort study of 4,117 diabetic
27 patients found that the HR for HG is higher among depressed patients when compared to the non-depressed (HR 1.42, 95% CI, 1.03–1.96) (Katon et al., 2013). The study also found that the number of HG episodes was higher among those depressed patients (OR 1.34, 95% CI, 1.03–
1.74). Another study that included over 3,800 T2DM patients found that depression is three times greater among those who have a history of HG (Tschope et al., 2011).
All the findings mentioned above match the ones that were conducted in other European countries. In Germany, a study included more than 32,500 T2DM patients found that the odds of
HG among depression patients were 1.24 (95% CI 1.13-1.35) (Kostev et al., 2014). Another smaller study in Germany of 420 patients also found that depression was associated with Severe
HG (Hermanns, Kulzer, Krichbaum, Kubiak, & Haak, 2005). In Belgium, a 2016 longitudinal study of over 43,000 patients from 20% of all Belgian hospital beds confirmed that depression was a significant risk for HG (Chevalier, Vandebrouck, De Keyzer, Mertens, & Lamotte, 2016).
Interestingly, numerous studies reported possible correlations between HG in diabetes, depression, low income, and lower educational attainment (Berkowitz et al., 2014). This relationship explains why diabetic patients with depression usually have suboptimal self-care, including being less adherent to recommended diet, exercise, and foot care as well as the pervasive lack of regular meals (Katon et al., 2013; Kostev et al., 2014; Zhang et al., 2015). In addition, many other studies found that those patients are more likely to have issues with their medication-taking behavior (Katon et al., 2013; Krass, Schieback, & Dhippayom, 2015; Zhang et al., 2015) as well as poor adherence to blood glucose monitoring (Katon et al., 2013; Kostev et al., 2014). A large study of more than 38,000 veterans with diabetes who had at least three outpatient visits in the previous year found that the rates for secondary prevention of diabetes such as retina examination and foot sensory examination were significantly lower among those
28 with depression (Conwell & Boult, 2008; Desai, Druss, & Perlin, 2002). All of the above findings point toward a positive correlation between HG and depression in diabetes patients.
2.3.6 Tobacco use
Smoking significantly increased the risks of developing T2DM, according to a 2015 systematic review on the relationship of active, passive, and quitting smoking with T2DM (Pan,
Wang, Talaei, Hu, & Wu, 2015). For those who were already diagnosed with T2DM, it was found that the RR for developing repeated HG events among smokers is 1.34 (1.02-2.02, P<0.05)
(Akram, Pedersen-Bjergaard, Carstensen, Borch-Johnsen, & Thorsteinsson, 2006). Another study found that history of smoking, regardless of current smoking status, increased the odds of developing severe HG episodes among T2DM patients (adjusted HR 1.43, 95% CI 1.09-1.788,
P<0.01) (Zoungas et al., 2010). Among T1DM, smokers also had higher odds of developing HG when compared to non-smokers (OR 2.40, 95% CI 1.30-4.40) (Hirai, Moss, Klein, & Klein,
2007).
On top of the risk of developing HG, smoking adversely affected diabetic patients by increasing the risks for other comorbidities as well as mortality. A 2013 systematic review and meta-analysis with bias analysis on the association between severe HG and risk of cardiovascular disease among T2DM patients showed that smokers had a higher RR for HG and CV diseases
(2.37, 95% CI 1.61-3.47) than others who were non-smokers (RR 1.91, 95% CI 1.59-2.29) (Goto et al., 2013). Regarding other comorbidities, both T1DM and T2DM smoker patients significantly differed from non-smokers in terms of the presence of microalbuminuria as well as microvascular and macrovascular complications (Nilsson, Gudbjörnsdottir, Eliasson, Cederholm,
& Steering Committee of the Swedish National Diabetes, 2004). In the Translating Research Into
29 Action for Diabetes (TRIAD) study, examining records of 8,334 patients revealed that smoking increased the all-cause mortality among T2DM (HR 1.59, 95% CI 0.94-1.63) (McEwen et al.,
2012).
2.3.7 Alcohol use
The use of alcohol has been included under (P) in many previous studies (Babitsch et al.,
2012). Alcohol has also been examined as a risk factor for HG, and the findings affirm that relationship (Barendse et al., 2012). A 2012 narrative review on the impact of HG on the quality- of-life and related patient-reported outcomes in T2DM found that alcohol consumption increased the risk of developing HG among T2DM patients (Barendse et al., 2012; Braak et al., 2000). A more recent systematic review and meta-analysis of population-based studies that included 46 studies of a total of 532,542 T2DM patients identified alcohol as a risk factor for developing HG
(Edridge et al., 2015). In the Fremantle diabetes study, daily alcohol consumption was found to increase the odds for HG among T2DM patients (HR 1.38, 95% CI 0.55-3.46) (Davis et al.,
2010). In another study, the annual incidence of HG was found to be positively correlated with the amount of alcoholic drinks per week (Miller et al., 2010).
There are many answers as to how alcohol consumption might lead to HG among diabetic patients. Alcohol reduces the endogenous production of glucose, and it inhibits the process of gluconeogenesis (Cryer, Davis, & Shamoon, 2003). Moreover, inebriation itself affects the level of self-care and adherence to recommended diet or medications (Cryer et al.,
2003). Hence, the consumption of alcohol facilitates the occurrence of HG episodes among diabetic patients.
30 2.3.8 Drug abuse
While there are plenty of studies that have considered drug abuse as an element in (P) for different diseases (Babitsch et al., 2012), only a few studies have examined its relationship with
HG among diabetic patients. One large nested case-control analysis study in the U.K. between
1998 and 2012 found that tramadol use by patients increased the risk of HG requiring hospitalization (Fournier, Azoulay, Yin, Montastruc, & Suissa, 2015). Another study found a significant dose-response relationship between HG and the use of methadone (Flory, Wiesenthal,
Thaler, Koranteng, & Moryl, 2016). Desai et al. (2002) found that there was a significant difference in the use of retina examination, foot sensory examination, and other secondary prevention care of diabetes among diabetic patients with mental disorders who abused drugs and others who did not. These findings might be one way to explain the relationship between drug abuse and HG (Conwell & Boult, 2008).
2.4 The Enabling component (E)
As mentioned earlier, key variables related to HG are examined in this study, including socio-economic status (SES), insurance coverage, and the geographical location of the patient and the hospital.
2.4.1 Socio-economic Status (SES)
Lower SES has been regularly found to be correlated with diabetes. Results from the
National Health Examination and Nutrition Examination Survey showed that patients who were at a lower SES were more likely to have diabetes than those with higher SES (OR 2.1, 95% CI
1.1–4.0) (Seligman, Bindman, Vittinghoff, Kanaya, & Kushel, 2007). A 2011 systematic review
31 and meta-analysis also revealed the same findings (RR 1.40, 95% CI 1.04-1.88) (Agardh,
Allebeck, Hallqvist, Moradi, & Sidorchuk, 2011). Children who were in lower SES families were also found at a higher risk of developing diabetes (HR 1.3; 95% CI=0.7, 2.6) (Maty, James,
& Kaplan, 2010). The incidence of diabetes among unemployed adults was also found to be higher than it was for those who were employed (OR for men 2.9; 95% CI 1.9-4.4; OR for women 1.7, 95% CI 0.8-3.7) (Kumari, Head, & Marmot, 2004). The TRIAD study showed that all-cause mortality was also higher among low SES T2DM patients (fully adjusted HR 1.80, 95%
CI 1.36-2.38) vs. others with a higher SES (fully adjusted HR 1.20, 95% CI 0.91-1.58) (McEwen et al., 2012).
In addition to the risk of developing diabetes, the odds of HG also increased among the low SES group because of low health literacy, food insecurity and prolonged fasting, lack of access to quality health care, and inability to afford BG monitoring devices (Berkowitz et al.,
2013; Duran-Nah, Rodriguez-Morales, Smitheram, & Correa-Medina, 2008; Nelson, Brown, &
Lurie, 1998; Seligman et al., 2011). A 2013 systematic review published in the Journal of Health
Service Research found that lower SES independently increased the risk of HG among adults with diabetes (Berkowitz et al., 2013). The Canadian Journal of Diabetes also published a systematic review on T1DM that also found a significant correlation between low SES and HG
(Sawka et al., 2007). A recent study examined the association between HG and low SES and found that diabetics with low SES have about twice the risk of developing HG when compared to others with a higher SES (16% vs. 8.8%) (Berkowitz et al., 2014). Hospital admissions related to
HG were also found to be higher among people with lower SES (OR 1.26, 95% CI 0.99 to 1.61)
(Wild et al., 2010). One reason for this might be because those with a lower SES were consistently found to have lower rates of regular doctors’ follow-ups, screenings, and
32 preventions (Babitsch et al., 2012; Brechner, Howie, Herman, Will, & Harris, 1993; Brown et al., 2004; Broyles, McAuley, & Baird-Holmes, 1999; Chen, Kazanjian, & Wong, 2008; Conwell
& Boult, 2008; Thode, Bergmann, Kamtsiuris, & Kurth, 2005).
2.4.2 Insurance coverage
Most studies that employed the BM found that the presence of insurance coverage significantly increased the likelihood of using health services (Babitsch et al., 2012) and people with healthcare coverage were more likely to participate in diabetic management programs (OR
1.08, 95% CI 0.76-1.54) (Gamble & Chang, 2014). Rates of ED visits due to HG varied among different insurance coverage types. Analysis of National Hospital Ambulatory Medical Care
Survey data from 1993–2005 showed that the ED visit rate per 1,000 people with diabetes is higher in public insurance vs. private and self-pay (5.8 vs. 2.6 and 2.2) (Ginde et al., 2008).
There were also differences in the characteristics of patients and their levels of utilizations of health care services among different insurance providers. Medicaid patients usually had more severe chronic medical conditions and more comorbid conditions (Adelmann,
2003). Therefore, these coexisting medical conditions lead to higher costs and service usage rates among Medicaid patients when compared to others covered by other health insurance providers
(Bailey, Wan, Brunt, & Somes, 2001; Conwell & Boult, 2008; Ou et al., 2012). A systematic literature review that was conducted to determine the relationship between low SES and severe
HG among T1DM patients found that people under Medicaid had twice the risk of HG-requiring medical assistance than others who were covered by other insurance providers (OR 10.2, 95%
CI) (Sawka et al., 2007).
33 2.4.3 Geographical location of the patient and the hospital
This variable has been used consistently in many studies that employed the Andersen BM
(Aday & Andersen, 1974; Babitsch et al., 2012). Likewise, it has been used in many studies that focused on HG among urban vs. rural diabetic patients from different aspects such as their level of ED utilizations, regular physician follow-ups, regular diabetes-related screening and prevention measures, and the degree of diabetes goal-attainments.
Generally, urban areas have lower odds of diabetes HG events when compared to rural areas. Kostev et al. (2014) found the odds ratio is 0.74 (0.68–0.80), even after controlling for demographics and comorbidities. Data from over five million ED visits from 1993-2005 analyzed by Ginde et al. (2008) showed that ED visits with HG are higher in non-urban areas
(4.5 per 1,000 ED visit, 95% CI 3.0-6.1) vs. 3.5 per 1,000 (95% CI 3.1-4.0) in urban areas.
Another recent study in Taiwan also found that the level of severe HG requiring medical assistance was higher among those not living in urban areas (Chen, Yang, Huang, Chen, & Hwu,
2015).
Regarding the differences in the frequency of regular doctor visits, follow-ups, screenings, and preventions, the systematic review conducted by Babitsch et al., (2012) of studies from 1998-2011 that utilized the BM found that residents of urban areas were more likely to be adherent to these practices than others who live in the countryside. Other studies also showed that usually there were fewer healthcare services in the countryside than there were in urban areas (Broyles et al., 1999; Thode et al., 2005). Another large study with 30,589 Medicare diabetic patients also showed that patients living in rural areas received almost half as much outpatient care and were less likely to receive the recommended tests, screenings, and prevention
34 measures when compared with others who live in larger urban areas (Rosenblatt et al., 2001).
Many other studies on diabetic patients found similar results, and therefore, patients in smaller towns were usually less likely to attain their recommended goals (Andrus, Kelley, Murphey, &
Herndon, 2004; Conwell & Boult, 2008; Unger, Warren, Canway, Manderson, & Grigg, 2011;
Weiner, Garnick, Fowles, Lawthers, & Palmer, 1995).
2.5 The Need-for-Care component (N)
Based on the review of previous studies’ findings, this study includes the following key variables, which are usually found to be correlated to HG: DM complications, type of DM, underweight, hypertension, liver Failure, Kidney disease, uncontrolled DM, and the Charlson
Comorbidity Index (CCI).
2.5.1 Type of DM
The odds of developing HG episodes among diabetic patients with T1DM are known to be higher than those with T2DM (Cariou et al., 2015). Nevertheless, most patients seen in the ED with HG are T2DM (Leese et al., 2003). This is because the prevalence of T2DM is much higher than T1DM (Cryer, 2008; Jafari & Britton, 2015). A recently published systematic review and meta-analysis shows that there is not a significant difference in the overall resources’ utilization among patients with T1DM and T2DM who develop SH (Jakubczyk & Rdzanek, 2015).
However, the same study found that patients with T1DM tend to have higher estimates of professional medical treatment such as doctor consultation or calling an ambulance than T2DM tend to have (23.61% vs. 15.54%). On the other hand, the overall hospital admissions because of
HG seemed to be higher among T2DM than T1DM (15.28% vs. 8.6%). Another study on 5,026
35 diabetic patients admitted in 2010 focusing only on SH found more admissions among T1DM than T2DM (21% vs. 5.1%, P = .000) (Dendy et al., 2014). In that study, the OR for T1DM was found to be 3.43 (95% CI 1.81-6.49).
2.5.2 Diabetes Complications
Diabetes is a chronic disease that can lead to several complications if it is not properly managed and controlled (Mayoclinic.org, 2017). The Mayo Clinic study (2017) showed that common long-term complications included cardiovascular diseases including micro and macro- vascular complications, nerve damage (neuropathy), kidney damage (nephropathy), eye damage
(retinopathy), foot damage, and skin problems. Several studies, including the 2012 systematic review by Bloomfield et al. reported that diabetes complications such as nephropathy, retinopathy, and neuropathy were considered independent risk factors for severe HG (Bloomfield et al., 2012; Kachroo et al., 2015; Malabu, Vangaveti, & Kennedy, 2014; McEwen et al., 2012).
Williams et al. (2012) found that confirmed hypoglycemia was three-to-four times more likely to occur among those diabetic patients who reported neuropathy than others.
2.5.3 Uncontrolled Diabetes
Uncontrolled diabetes happens when a diabetic patient shows a consistently high blood glucose level that is above the recommended target (ADA, 2009). Usually, long-term uncontrolled diabetes leads to several complications that affect the heart, blood vessels, eyes, kidneys, and the nerves (Amiel et al., 2015). In contrast, tight glycemic control among particular groups of patients usually contributes to HG (Bloomfield et al., 2012). Elderly, underweight, or
36 patients who take certain kinds of GLMs are at a higher risk for developing HG if they are under intensive blood glucose control (Edridge et al., 2015; Jafari & Britton, 2015).
2.5.4 Chronic Kidney Disease (CKD)
Renal insufficiency is usually caused by a disease in the kidney and recognized with an elevation in serum creatinine level or decrease in “estimated glomerular filtration rate” (eGFR)
(Bloomfield et al., 2012). It is considered one of the major risk factors for hypoglycemia in patients with or without diabetes (Abdelhafiz et al., 2012; Moen et al., 2009). Diabetic patients with CKD, who develop HG are usually found to have higher mean serum creatinine levels
(Kilpatrick et al., 2014; Lee et al., 2015) and lower eGFR (Brian & Simon, 2012; Schloot et al.,
2016). Insulin also increases the risk of HG among diabetics with CKD. Therefore, appropriate adjusting of the dose of insulin is recommended in those high-risk patients (Baldwin et al., 2012;
Farrokhi et al., 2012).
Several studies showed that the odds of developing HG episodes increase among diabetic patients with CKD (Duran-Nah et al., 2008; Kostev et al., 2014; Pathak et al., 2015). A retrospective cohort study with data from 70 hospitals found that diabetics who developed HG had higher odds of comorbidity with CKD (OR 28.4, 95% CI 27.4-29.4) than had those with no
HG (OR 11.5, 95% CI 11.3-11.7) (Curkendall et al., 2009). Another study on 5,026 diabetic patients admitted in 2010 showed that the odds of having CKD with HG were statistically higher
(OR 69.1% vs. 46.9%, P<.001) than they were for those with diabetics with no HG (Dendy et al.,
2014). Hsu et al. (2013) also found that CKD was independently associated with HG (OR 3.26,
95% CI 2.76–3.86). Therefore, the rate of HG among patients with CKD was higher than it was for those with no CKD (10.72 vs. 5.33 per 100 patient-months) (Moen et al., 2009).
37 The severity of CKD highly impacted the severity and frequency of HG episodes (Moen et al., 2009). Renal failure (RF), a more advanced stage of CKD (Kidneyfund.org, 2016), was found to cause 26% higher HG episodes (OR 1.26, 95% CI 1.16-1.37) (Kostev et al., 2014).
Curkendall et al. (2009) also found that diabetics with RF were more likely to have HG than those without RF (OR 3.5, 95% CI 3.0-4.1) vs. (OR 1.2, 95% CI 1.1-1.3). Similarly, Quilliam et al. (2011) found the odds ratios of T2DM admissions for HG and RF to be 3.10 (95% CI 2.05-
4.67).
2.5.5 Liver disease
Liver diseases have been commonly found as a comorbidity among diabetic patients with
HG. A retrospective cohort study with data from 70 hospitals found that diabetics who developed HG had higher odds of comorbidity with liver diseases (OR 2.2, 95% CI 1.8-2.7) compared to (OR 2.0, 95% CI 1.9-2.1) those with no HG (Curkendall et al., 2009). Another nested case-control study with a large database from 2004-2008 reported that liver disease was associated with HG-related ED visits (OR 1.55, 95% CI 1.30-1.84) (Simeone & Quilliam, 2012).
Liver cirrhosis, a late stage of fibrosis of the liver, was also found to be a common comorbidity. A study derived from the National Health Insurance Research Database between
1998-2009 with over 77,000 T2DM patients in Taiwan found that liver cirrhosis was independently associated with HG (OR 1.71, 95% CI 1.17–2.48) (Hsu et al., 2013). Another study on 5,026 diabetic patients admitted in 2010 found statistically higher proportions of cirrhosis among patients with SH (14.8% vs. 7.2%) compared to the non-SH group (Dendy et al.,
2014).
38 2.5.6 Hypertension
Hypertension is a very common comorbidity found among diabetic patients. It is estimated that 71% of adults with diabetes have blood pressure over 140/90 (CDC, 2014). A positive correlation was usually found between HG and the presence of hypertension (Duran-
Nah et al., 2008; Moore, Li, Hung, Downs, & Nebeker, 2009). The high blood pressure might have been caused by the kidney damage associated with diabetes. Hsu et al. (2013) also found that hypertension was independently associated with HG among patients with T2DM (HR 1.75,
95% CI 1.57–1.96).
2.5.7 Malignancy
Many studies reported that malignancies were usually found in diabetic patients who presented with HG. A consecutive cohort study of 8,767 T2DM patients in China between 1995-
2007 found that the incidence of all-site cancer in patients with SH was higher than it was for those without SH (13.4 vs. 6.4%) (Kong et al., 2014). Liatis et al. (2015) also reported that cancer was independently associated with SH among T2DM patients (OR 3.21, 95% CI 1.64–
6.29). Likewise, Hsu et al. (2013) found that cancer was independently associated with HG in
T2D patients (OR 2.73, 95% CI 2.12-3.50). Abdelhafiz et al. (2012) found that HG patients were more likely to be diagnosed with cancer than were others with no HG (10% vs. 3.5%, P = 0.04).
2.5.8 Hyperlipidemia
Hyperlipidemia is a condition that is usually found among diabetic patients (Rowe et al.,
2015), and the relationship between hyperlipidemia and HG have been documented in several studies (Lopez, Bailey, & Rupnow, 2015). A 2013 systematic review and meta-analysis with
39 bias analysis study found that T2DM patients with dyslipidemia have a higher RR for developing
HG and cardiovascular diseases (RR 1.93, 95%CI 1.70-2.18) (Goto et al., 2013). Kachroo et al.
(2015) found in their 2015 study of over 1.1 million T2DM patients that dyslipidemia is the second most common comorbidity (39.4%) found in HG patients. In the TRIAD study, it was found that diabetic patients with a history of dyslipidemia had an increased risk of all-cause mortality when compared to their counterparts (McEwen et al., 2012). A large 2015 study was conducted to examine the association of HG with fall-related outcomes among older T2DM patients from 2008-2011(Kachroo et al., 2015). The study included more than 1.1 million T2DM patients’ records and found that the percentage of dyslipidemia among patients who have HG was twice the rate shown among those without HG.
2.5.9 Low Body Mass Index (BMI)
BMI is a more appropriate technique than measuring the weight alone because it considers the height of the person as well as the weight. It is calculated by dividing the individual’s weight in kilograms by the square of height in meters. Based on the value, the person is categorized as underweight (BMI<18.5), normal weight (BMI 18.5–24.9), overweight
(BMI 25–29.9), or obese (BMI>30) (nhlbi.nih.gov, 2016). BMI is inversely related to HG.
Higher levels of BMI have been found to be associated with lower numbers of HG episodes among diabetics, while lower BMI levels have been consistently found to be a risk factor for HG among diabetic patients. Therefore, dose adjustments for GLMs is recommended based on the
BMI level of each diabetic patient (Elliott et al., 2012; Kilpatrick et al., 2014; Olveira et al.,
2015; Sako et al., 2015; Schafers, Naunheim, Vijayan, & Tobin, 2012; Schloot et al., 2016).
40 A large epidemiological analysis of more than 10,000 diabetic patients showed that patients with a BMI of 30 or more had a 35% lower incidence of HG than those with less than 25
BMI (HR 0.65, 95% CI 0.5 to 0.85). Another study that included more than 11,000 patients recruited from 215 centers in 20 different countries between 2001-2003 found that each unit increase in the BMI was associated with a 5% decrease in risk of HG (HR 0.95, 95% CI 0.93-
0.98) (Zoungas et al., 2010). Giorda et al. (2015) examined T1DM patients only and found similar odds associated with HG (IRR 0.95, 95% CI 0.94–0.97). In contrast, Kong et al. (2014) also conducted a study on a consecutive cohort of 8,767 T2DM patients in China between 1995-
2007 and found that 1 kg/m2 increase in the BMI was associated with 5% reduction in SH risk.
In-hospital HG-related mortality also appeared to be higher among those with low BMI. Analysis of the National Inpatient Database in Japan from 2008–2012 with 22.7 million inpatients’ records focusing on HG among diabetic patients revealed that the highest mortality was among underweight (BMI<18.5) diabetics with HG (OR 2.345, 95% CI 1.918-2.868) (Sako et al., 2015).
2.5.10 Charlson Comorbidities Index
The Charlson Comorbidity Index (CCI) is the most commonly used index, and it was developed and validated as a measure of predicting in-hospital mortality (Charlson, Pompei,
Ales, & MacKenzie, 1987; Huang et al., 2014; Quan et al., 2011). The most recent revised CCI consists of 17 comorbidities and, depending on the mortality risk associated with each condition, each is assigned a score from one to six, and the total score for all comorbidities reflects the CCI score (Charlson et al., 1987; Roffman, Buchanan, & Allison, 2016).
Numerous studies concluded that the higher the CCI score among diabetic patients, the higher the odds of developing HG (Curkendall et al., 2009; Kostev et al., 2014; Kostev et al.,
41 2015; Lopez et al., 2014; Moore et al., 2009; Pathak et al., 2015; Signorovitch et al., 2013).
Bruderer et al. (2014) found that patients with a CCI score between four and six had higher odds for experiencing HG (OR 3.44, 95% CI 2.87-4.13), and the odds were even higher among those with a CCI of seven or more (OR 8.18, 95% CI 5.71-11.72) when compared to others with a one to three CCI score.
Higher CCI scores also aggravate the adverse outcomes that follow HG (Abdelhafiz et al., 2012). Additionally, Yeh et al. (2015) found that those with HG were at higher risks for developing CV complications when their CCI were higher. In another study, which focused in falls related to HG, Malabu et al. (2014) reported that the odds for falls following HG increased among diabetic patients with higher CCI. Several other studies found that HG-related mortality was higher among those with higher CCI (McEwen et al., 2012; Monami et al., 2007; Sako et al.,
2015; Zapatero et al., 2014).
Besides poor outcomes associated with higher CCI, utilization for healthcare services among those patients were significantly higher than their counterparts. This includes the number of hospitalizations, ED visits, OPD visits, readmissions, LoS, total medical costs, and diabetes care-related costs (Ou et al., 2012; Quilliam et al., 2011; Zapatero et al., 2014). Ou et al. (2012) also found that those with higher CCI scores were usually less adherent to their antidiabetic medications.
2.6 Studies examined the utilization and/or outcomes among HG patients
In contrast to the above studies, many studies focused on the level of utilization and/or outcomes among diabetic patients without much attention to the risk factors associated with it.
These studies were focused either on patients who developed HG in outpatient settings or on
42 those who developed HG in the hospital. Studies that examined HG episodes in outpatient settings usually focused on the level of healthcare utilization alone (Davis et al., 2010; Duran-
Nah et al., 2008; Geller et al., 2014; Gu et al., 2015; Heller, Frier, Herslov, Gundgaard, &
Gough, 2016; Home, Fritsche, Schinzel, & Massi-Benedetti, 2010; Liatis et al., 2015; Majumdar et al., 2013; Mier et al., 2012; Quilliam et al., 2011; Quilliam et al., 2011; Simeone & Quilliam,
2012; Solomon et al., 2013; Williams et al., 2012) or outcomes only (Kong et al., 2014; Simon et al., 2015). Few others investigated utilization and outcomes together (Cariou et al., 2015; Sako et al., 2015). As anticipated, clinic visits, the number of ED visits, the frequency of hospitalizations, and LoS were the most common measures used in these studies to gauge the level of utilization in diabetic patients who developed HG as an outpatient. On the other hand, mortality and quality of life were usually used to measure outcomes.
Regarding studies that focused on HG among hospitalized patients, most either included both utilization and outcomes (Brodovicz et al., 2013; Chevalier et al., 2016; Curkendall et al.,
2009; Turchin et al., 2009; Zapatero et al., 2014) or outcomes only (Olveira et al., 2015;
Vriesendorp, DeVries, Rosendaal, & Hoekstra, 2008). Among these studies, LoS and cost were the two measures that usually were used for utilization, while mortality was the most common measure used for outcomes. There were also other measures for outcomes used such as readmission, septic shocks, falls, and fall-related fractures.
2.6.1 HG Utilization (U)
Based on the literature related to HG and other studies that employed the BM on other illnesses, the two most commonly used indicators for (U) are: hospital LoS and cost. This study includes those two indicators as measures for (U) among hospitalized HG patients.
43 2.6.1.1 Hospital LoS
Diabetic patients were usually found to stay longer in the hospital than were non- diabetics (Moghissi et al., 2009). Among diabetic patients, those who had an HG event while hospitalized were consistently found to have had longer LoS than others without HG have had
(Brodovicz et al., 2013; Geller et al., 2014; Gómez et al., 2015; Jakubczyk & Rdzanek, 2015;
Nirantharakumar et al., 2012; Turchin et al., 2009; Zapatero et al., 2014). A retrospective cohort study using a large database from Spain from 2005-2010 found that HG among hospitalized patients was significantly associated with a 24% increase in the odds of longer stays (OR 1.24,
95% CI 1.15-1.35). (Zapatero et al., 2014). Gómez-Huelgas et al. (2015) also found a strong association between HG and LoS (HR 2.036, 95% CI 1.862-2.227).
Numerous studies sought to find the average difference in LoS among HG and non-HG diabetic patients. They found that the average LoS among diabetic patients who developed HG was three days longer than it was for others who did not. Curkendall et al. (2009) reported 3.0 days more among HG patients (95% CI 2.8-3.2). Gómez-Huelgas et al. (2015) found that the average LoS among HG patients was 12 days vs. nine days among non-HG patients.
Signorovitch et al. (2013) also found that T2DM patients with HG have a mean LoS of 10.95 vs.
8.31 days with no HG. Turchin et al. (2009) also reported that the LoS of diabetics who developed HG was 2.5 higher than those without HG (P < 0.0001). The duration of stay was found to be correlated with the severity of HG. Nirantharakumar et al. (2012) analyzed data from
6,374 admissions between 2007–2010 in the UK and found that the adjusted LoS increased by
2.33 times (95% CI 1.91–2.84) among the group with the most severe HG compared to 1.51
(95% CI 1.35–1.68) among less severe HG. Therefore, a strong correlation was found between
44 LoS and mortality. Finally, the Turchin et al. (2009) study showed that each additional day with
HG was associated with 85.3% more risk of inpatient mortality.
2.6.1.2 Cost
Research indicates that HG imposes a substantial economic burden on the healthcare system. However, estimating the exact magnitude of the expenses associated with HG care is a complex issue. This complexity is due to the wide variety of indirect costs related to each HG episode, such as comorbidities, adverse events, complications, and the long-term negative impact on quality of life and work productivity (Ahrén, 2013). Additionally, estimating the direct costs associated with HG is also not an easy task because each episode varies in its severity and the causes that lead to it (Liu, Zhao, Hempe, Fonseca, & Shi, 2012).
Plenty of studies examined the economic impact of HG in diabetic patients in the U.S.
Most of these studies focused on the costs associated with home medications, physician visits, and quality of life, according to a 2010 systematic review (Zhang et al., 2010). Curkendall et al.
(2009) conducted a study on 70 hospitals and measured the costs associated with inpatient HG for T1DM and T2DM patients. They found that patients who develop HG incurred average charges of about $86,000 compared to $54,000 for those who did not have HG. Their analysis concluded that HG was associated with 38.9% higher total charges (95% CI, 35.6%- 42.1%).
Quilliam et al. (2011) investigated a large database from MarketScan® Research Databases from
2004-2008 and found that HG represented 1% of all inpatient costs and 2.7% of ED costs. The average cost of HG admission was about $17,500, which is $4,000 higher than other diabetes- related admissions, while ED visit average costs for HG were about $1,400 compared to $320 for other diabetes-related ED visits.
45 While there were big differences in the estimated costs found in the above studies, it is evident that the cost of HG-related admissions or ED visits were substantially higher among diabetics with HG than they were for those without. There were also some other studies in other countries such as Germany, U.K., and Sweden that estimated these costs (Zhang et al., 2010).
Cost estimates were different from the U.S. because of the dissimilarities in the payment systems, care delivery, and extent of treatment (Liu et al., 2012). All in all, there is comprehensive agreement that hospital-related costs are substantially higher among HG patients.
2.6.2 HG Outcomes (O)
This study includes five outcome measures under the umbrella of (O). The first measure is mortality, which is the most commonly used measure for (O) among HG studies as well as studies that employed the BM for other diseases. Second is CV complications; they are found to be significantly associated with HG in numerous studies. Third is sepsis, which is a slightly less common outcome, but it is far-reaching and usually associated with higher levels of (U). The last two, patient falls and seizures, are less severe but more commonly found among HG patients.
2.6.2.1 Mortality
Diabetic patients who present to the ED with HG or develop in-hospital HG are considered at a higher risk of death. A recently published study examined the hospital discharge records of more than 5.4 million diabetic patients from 1997-2010 and found that those who developed HG in the hospital were 12% more likely to die in the hospital (OR 1.12, 95% CI
1.09-1.15) (Gómez et al., 2015). Another study examined extensive data from 70 hospitals from
2000-2006 and found the likelihood of dying among HG patients to be 7% higher (OR 1.07, 95%
46 CI 1.2-1.11) (Curkendall et al., 2009). However, the risk of mortality varied based on the severity of HG. Those with more severe HG were usually at a higher risk of mortality than were others with less severe HG. A retrospective cohort study analyzed 4,368 admissions between
2003-2004 and found that the odds of inpatient death increased three times for every 10 mg/dl decrease in BG (Turchin et al., 2009).
The relationship between in-hospital HG and mortality was also consistent with other studies conducted in other countries. In the UK, one study found that inpatient mortality was 5-
15% among patients with HG (Nirantharakumar et al., 2012) and another UK study found it to be
7.6% (Brodovicz et al., 2013). In Spain, discharged patients’ records between 2005-2010 showed that 10.2% of those who develop inpatient HG died in the hospital compared to 9.5% without
HG (OR 1.08, 95% CI 1.05-1.11) (Zapatero et al., 2014).
2.6.2.2 Cardiovascular (CV) Complications
HG steadily has been found to be a risk factor for developing a broad range of adverse
CV complications in many systematic review studies (Goto et al., 2013; Yeh et al., 2015). These complications include MI, stroke, and arrhythmias (Johnston et al., 2011; Khunti et al., 2015).
Another systematic review was conducted in 2013 to summarize the available data on the pathophysiology behind HG and CV complications (Hanefeld, Duetting, & Bramlage, 2013). It found that HG lead to CV complications by elevating the thrombotic tendency, affecting cardiac repolarization, provoking inflammation, and accelerating the process of development of atherosclerosis.
A 2013 systematic review and meta-analysis found that SH was strongly associated with higher odds of developing CV diseases (RR 2.05, 95% CI 1.74 to 2.42) (Goto et al., 2013). Hsu
47 et al. found that stroke was a common comorbidity that independently associated with HG (OR
2.84, 95% CI 2.31–3.48) followed by heart failure (HF) (OR 2.04, 95% CI 1.65–2.51) (Hsu et al., 2013). Simeone and Quilliam (2012) found that ED-treated HG cases usually had HF (OR
1.70, 95% CI 1.49-1.93), peripheral vascular disease (OR 1.80, 95% CI 1.60-2.02), arrhythmia
(OR 1.22, 95% CI 1.04-1.44), and stroke (OR 1.81, 95% CI 1.41-2.32). In another study,
Quilliam et al. (2011) found various CV complications that were independently associated with inpatient HG. For example, theu study found that arrhythmia is higher among HG patients (OR
1.69, 95%CI 1.17–2.44), coronary artery disease (CAD) (OR 1.48, 95%CI 1.21–1.81), HF (OR
2.33, 95%CI 1.72–3.15), and stroke (OR 2.78, 95%CI 1.62–4.77) (Quilliam et al., 2011).
Hospital admissions were common among HG patients with CV complications as well. When comparing patients without HG with those having HG, Hsu et al. (2013) found a higher hospitalization rate per 1,000 person-years for stroke (27.97 vs. 69.05), HF (27.96 vs. 65.51) and other CV diseases (106.00 vs. 323.36).
2.6.2.3 Sepsis
Sepsis is one of the most common unpleasant in-hospital complications among HG patients. A retrospective chart review of adult patients showed that the prevalence of sepsis was higher among those who developed HG than it was for those without HG (Blosch et al., 2010).
Another retrospective cohort study of data from 70 hospitals also showed that diabetics who present to the hospital with HG had higher odds of developing sepsis (OR 6.9, 95% CI 6.3-7.4) than other diabetics without HG had(OR 2.2, 95% CI 2.1-2.3) (Curkendall et al., 2009). Besides sepsis at presentation, a study on over 5,000 diabetic patients admitted in 2010 found that the odds ratio of developing in-hospital sepsis among HG patients was 2.64 (95% CI 1.6-4.35)
48 (Dendy et al., 2014). Dendy et al. (2014) noted that the risk of developing septicemias among
HG patients increased with the severity of HG, and found that the SH group had statistically higher proportions of sepsis (49.4%) compared to non-SH (12.5%). Abdelhafiz et al. (2012) found that sepsis was more likely to be found among HG patients than others with no HG (18% vs. 1.8%, P < 0.001).
2.6.2.4 Patient falls
In general, falls are undesired because they usually lead to fractures, hospital admissions, longer LoS, more utilization of healthcare resources, higher costs, and sometimes deaths
(Chevalier et al., 2016). HG is one of the major factors that leads to patient falls at their homes. It has been estimated that HG events increase the odds of falls-related fractures by 70% (OR 1.7,
CI 95% 1.58–1.83) (Johnston et al., 2012). Among hospitalized patients, falls and fall-related fractures are not uncommon among HG patients. One study in Australia included over 11,000 patients between 1996-1998 and showed that patient falls occurred at a rate of 291.2 per 100,000 people per year (Kennedy, Chapman, Nayar, Grant, & Morris, 2002; Malabu et al., 2014).
Another Belgian study estimated that accidental falls were reported in 11.2% of the HG-related hospitalizations (Chevalier et al., 2016). A study included a large U.S. database from 1998–2010 of non-insulin treated T2DM and found that HG increased the hazards of falls by 17% among hospitalized patients younger than 65 years (HR 1.17, 95% CI 0.88, 1.57), and 36% for those older than 65 years (HR 1.36, 95% CI 1.13–1.6) (Signorovitch et al., 2013). Kachroo et al.
(2015) also found that elderly patients were usually found to have twice the risk for falls than younger patients were.
49 2.6.2.5 Seizures
Seizures that are due to hypoglycemia are one of the unpleasant neurological complications of HG because they sometimes lead to permanent neuronal damage or even death
(Stafstrom, 2008). A retrospective cohort study of data from 70 hospitals found that diabetics who presented to the hospital with HG had higher odds of developing seizures (OR 3.0, 95% CI
2.7-3.4) than did other diabetics without HG (OR 2.2, 95% CI 2.1-2.2) (Curkendall et al., 2009).
Holstein, Hammer, and Egberts (2003) estimated that 5% of HG cases presented to the ED had seizures. Another prospective, population-based study over four years in Germany revealed that about 7% of hospitalized diabetic patients with HG had seizures (Holstein & Egberts, 2001).
2.7 Summary of Gaps in the Literature
While plenty of studies examined different aspects of the HG problem such as risk factors, utilizations, or outcomes, this literature review revealed that most of these studies were not based on a logically-constructed theoretical framework. This resulted in segregated findings of some risk factors and some other information about the level of utilization or outcomes. In addition, SEM was not employed; hence, these studies did not demonstrate the presence of any interaction effects among different predictors. Moreover, the presence or absence of indirect effects among the factors has never been examined without SEM. Therefore, there is a need to conduct rigorous studies that include multiple risk factors of HG and examine their impact on hospital utilization and outcomes. One way is to employ the BM with SEM to be able to delineate differential effects of risk factors on utilization and outcome, coupled with the examination of the presence of any interaction effects among predictor variables.
50 CHAPTER 3 : METHODOLOGY
3.1 Introduction
This chapter presents the methodology applied to this study. The research questions are introduced, followed by the research design, data sources, sample, and subject of the study. Next, all study variables are described with their definitions and measurements. Afterward, sample size justification and power analysis are described. Last, details about the process of multivariate statistical analysis methodology are explained followed by the DTREG analysis.
3.2 Research Questions
As mentioned earlier, the main goal of this study is to identify the relative importance of factors related to HG utilization and outcomes among diabetic patients. In addition, the study explores the magnitude of the effects of each of the independent variables and whether there are any interaction effects among them. Therefore, the research questions for this study will be:
RQ1. Which component is most important in determining the LoS among patients who
develop HG, when other factors are held constant?
RQ2. Which component is most important in determining the total cost among patients who
develop HG, when other factors are held constant?
RQ3. Which component is most important in determining the outcome among patients who
develop HG, when other factors are held constant?
RQ4. Is the need-for-care the most important predictor of use, while the effects of other
predictors are simultaneously considered? 51 RQ5. Do the predictor variables (P, E, and N) show any significant interaction effects on
utilization and outcomes?
3.3 Research Design
A quantitative, non-experimental and retrospective design was employed in this study by utilizing data from a national database for the years of 2012, 2013, and 2014. Although this observational study is non-experimental, it is more feasible, especially because of the availability of the comprehensive national database. Another issue is that the lack of previous studies examined HG under the guidance of the BM stimulate the need for such a study to help other researchers when formulating any future studies related to HG risk factors, utilization, and outcome.
3.4 Data source and sample
The retrospective design of the study was based on the selection of the study subjects from the Healthcare Cost and Utilization Project (HCUP) National database. HCUP is a group of healthcare databases sponsored by the Agency for Healthcare Research and Quality (AHRQ).
HCUP databases aggregates all different data collection efforts of State data organizations, hospital associations, private data organizations, and the Federal government to establish a unified national information resource of encounter-level health care data (HCUP-us.ahrq.gov,
2016). Since 1988, data is gathered every year from 47 States and the District of Columbia, representing 97% of all inpatient hospital discharges. AHRQ converts administrative health care data that were acquired from HCUP Partners into uniform databases that are easy to use by researchers.
52 The data for this study was extracted from the National Inpatient Sample (NIS), which is part of a family of databases drawn from all different State inpatient databases in HCUP. A 20% stratified sample of all discharges from community hospitals, not including rehabilitation or long-term acute care hospitals, is approximated by the NIS (AHRQ.gov, 2016). The data includes all patients, regardless of payer, including individuals covered by Medicaid, Medicare, private insurance, or uninsured. Nis is considered the largest all-payer inpatient care database in the U.S. as it contains data from more than seven million hospital stays each year (AHRQ.gov,
2016).
This study included all cases presented to the ED in the years 2012, 2013, and 2104.
Based on the International Classification of Diseases Ninth Revision (ICD-9) diagnostic codes, only adults 18 and older, who were diagnosed with T1DM or T2DM and were diagnosed with
HG were included in the study.
3.5 Subjects for the Study
The unit of analysis of this study is individual patients. Data on patients who presented to the ED between January 1, 2012 and December 31, 2014, with the diagnosis of T1DM or T2DM, were included. Therefore, inclusion criteria for patients was:
C1: Age > or = 18
C2: Presented to ED between January 1, 2012 and December 31, 2014
C3: Diagnosed with diabetes
C4: Admitted with or developed HG during hospitalization
53 Because of the nature of the study, non-diabetics and diabetics without HG were excluded from the study. Patients who were younger than 18 years were also excluded because genetic diseases are the predominant factors leading to HG among children (Wang, Shandro,
Sohoni, & Fassl, 2011). In addition, factors that lead to HG are different between newborns, infants, toddlers, and older children (Brook, Clayton, & Brown, 2010). Last, when compared to non-pregnant adults, HG during pregnancy has different maternal risk factors, different pregnancy outcomes, and a wide-range of fetal-related complications (Naik et al., 2016;
Ringholm, Pedersen-Bjergaard, Thorsteinsson, Damm, & Mathiesen, 2012; Wallis, 2008).
Therefore, all pregnant women were excluded from this study.
3.6 Study Variables
The BM is the principal framework for this study. Different indicators or variables were considered under each of the main three constructs/components: P, E, and N. In addition, the study also included other indicators under the utilization and outcome constructs. The following table summarizes the study variables:
54 Table 2. Operational Definition and Measurement Instruments for Study Variables Variable Variable name Definition Scale type Predisposing component (P) Age of the patient. All adults 1 Age Exogenous Continuous in years 18 years and above 2 Gender Exogenous Male or Female Categorical (Dichotomous) Male=0, Female=1 The ethnicity of the patient. Whether patient’s ethnicity is Categorical (Dichotomous). 1= African American or 3 AA_Hisp Exogenous African American or Hispanic Hispanic, 0=Others or not 4 Dementia Exogenous Patient has dementia or not Categorical (Dichotomous) No dementia=0, Yes=1 Categorical (Dichotomous) No b depression=1, 5 No Depression Exogenous Patient has no depression Depression=0 Refer to the healthy lifestyle Categorical 2=Patient has a healthy lifestyle with no Healthy lifestyle for the patient, which is 6 Exogenous smoking, no alcohol, and no drugs. 1= Patient does one of (HLS) tobacco-free, alcohol-free, the above, 0= Patient does 2 or more of the above. and no drug abuse Enabling component (E) Whether the patient is covered Categorical (Dichotomous): Medicaid=1, Others (Medicare, 1 Medicaid Exogenous by Medicaid or not Private, no insurance)=0 The socio-economic status of Categorical: 1= 0-25th percentile, 2= 26th to 50th the patient based on the 2 SES Exogenous percentile, 3= 51st to 75th percentile, 4= 76th to 100th median household income for percentile patient Hospital located in an Urban 3 Urban_hosp Exogenous Categorical (Dichotomous). Urban=1, Rural=0 or rural area Categorical: 1= Not metropolitan or micropolitan county, 2= Patient home location. Based Micropolitan county, 3= Counties in metro areas of 50,000- 4 Patient Location Exogenous on the location's county 249,999 population, 4= Counties in metro areas of 250,000- population. 999,999 population, 5= Counties of metro areas of >=1 million population Need-for-Care component (N) Patient has any DM specific DM complication (eye, 1 Exogenous Categorical (Dichotomous). No complications=0, Yes=1 Complications neurological, cardiac, renal, others) Patient has DM that is 2 Uncontrolled DM Exogenous Categorical (Dichotomous). No=0, Yes=1 uncontrolled 3 DMII Exogenous Patient is T2DM Categorical (Dichotomous). T2DM=1, T1DM=0 BMI Patient BMI category is 4 Exogenous Categorical (Dichotomous). No=0, Yes=1 underweight underweight 5 Hypertension Exogenous Patient has hypertension Categorical (Dichotomous). No=0, Yes=1 6 Hyperlipidemia Exogenous Patient has hyperlipidemia Categorical (Dichotomous). No=0, Yes=1 Patient has moderate to severe 7 Liver_dis Exogenous Categorical (Dichotomous). No=0, Yes=1 liver disease 8 Renal_disease Exogenous Patient has Renal disease Categorical (Dichotomous). No=0, Yes=1 9 Cancer Exogenous Patient has malignancy Categorical (Dichotomous). No=0, Yes=1 Charlson 10 Comorbidity Control Score of the CCI Categorical. Scores 1-25 Index (CCI) Utilization (U) 1 Hospital LoS Endogenous Patient days in the hospital Continuous: in days Total charges in USD for the 2 Cost Endogenous Continuous: in USD admission Outcome (O) Categorical: 3= patient died in hospital, 2= patient The severity level of the developed CV complications and/or sepsis, 1= patient had 1 Outcome Endogenous outcome ranging from no seizures and/or fall, 0= none of the above adverse outcomes to death
55 3.6.1 Predisposing component (P):
As mentioned earlier, the study includes the main key variables used in previous studies that were found to be relevant risk factors related to HG among diabetic patients. Regarding the predisposing component or risk propensity profile, age, ethnicity, gender, dementia, depression, and healthy life status were considered as indicators in this study. The age of the patients was collected in years. Gender was either male or female. Ethnicity was either African American,
Hispanic, or others. These categorizations were based on evidence from the literature review that showed that African Americans and Hispanics had higher odds of developing HG when compared to others. Regarding healthy life status, patients were categorized into three groups: those who do not drink, smoke, or abuse drugs; those who do one of these; and those who do two or more.
3.6.2 Enabling component:
The study included the four major indicators for the enabling construct, which were frequently used in previous studies to facilitate or impede the use of health services: medical insurance type, SES level, the location of the patient, and the hospital location. Health insurance was categorized into two groups: those who were on Medicaid and those who had either
Medicaid, private insurance, or no insurance. This categorization is relevant to this study as
Medicaid patients found to be the sickest group when compared to the other groups (Adelmann,
2003). Regarding SES, the categorization was based on the median household income for the patient. Of the four categories that represent income percentile level, category one represented the poorest from the 0-25th percentile, while category four represented the richest from the 75th to
100th percentile. Hospital locations were categorized into urban or rural. Lastly, patient locations 56 were categorized into five groups based on the population density of the county related to patients’ homes: the smaller the category number, the smaller the county population size.
3.6.3 Need-for-Care (N):
Based on the review of the literature, the indicators for N were the DMII, the presence of diabetes-related complications (DM_complications), uncontrolled diabetes (uncontrolled_DM), underweight (BMI underweight), hypertension, hyperlipidemia, liver disease (liver_dis), renal disease (renal_disease), and cancer. All of these variables are dichotomous in nature, No=0 while
Yes=1. These variables were extracted from the ICD-9 codes associated with each case in the database (Appendix I, Table 26).
As discussed earlier, CCI was found to have a broad range of consequences in addition to
HG, which included CV complications, falls, higher costs, longer LoS, and more risks to mortality. Therefore, it was more appropriate for this study to consider CCI as a control variable.
It is also worth mentioning that this approach of controlling CCI was employed among numerous previous studies that examined risk factors for HG among diabetic patients (Fayfman et al.,
2016; Khunti et al., 2015; Moen et al., 2009; Nirantharakumar et al., 2012; Patel et al., 2015;
Zapatero et al., 2014).
3.6.4 Utilization (U)
As discussed earlier, the two indicators examined in this study for U are LoS and the total cost. LoS is the duration patients spent in the hospital in days. Total cost refers to the total cost of care in U.S. dollars. The number of hospital days does not necessarily correlate with the total cost of the hospital stay. In addition, some variables might influence LoS or the total charges
57 without affecting the other variables. Therefore, the study treated the two utilization variables separately.
3.6.5 Outcome (O)
To facilitate data analysis and interpretations, the outcome variable was coded as a categorical variable from 0-3 depending on the severity. Category 3 is the most adverse outcome that refers to a patient’s death in the hospital. Category 2 is when the patient did not die but experienced severe adverse outcomes including sepsis and/or CV complications such as MI, HF, stroke, or arrhythmias. Category 1 includes minor adverse outcomes: seizures and/or falls.
Lastly, category 0 represents patients discharged from the hospital without any of the previous complications. Thus, the higher the outcome index, the poorer the health outcome observed.
3.7 Power Analysis and Sample Size Justification
Statistical power indicates the odds of observing a treatment effect when it occurs. Power analysis helped in determining the appropriate sample size by taking the alpha size, beta size, standard deviation, and mean. Power analysis was conducted before the study to help determine the sample size needed, and again at the end of the study to make sure that the included sample size was appropriate. The higher the power, the more trustworthy the results and applicable to the population. However, it is necessary to note that statistical significance in hypothesis testing is associated with the sample size. Power analysis calculation is as follows:
1. Power =1-beta (type II error)
2. Effect size = signal (program effect) / noise (error)
58 Effect size (ES) is a name given to a family of indices that measure the magnitude of a treatment effect: 0.1 is minimal, 0.3 is optimistic.
Power analysis also helped avoid making a Type I or Type II error. Type I error occurs when incorrectly rejecting a true null hypothesis and concluding that a statistically significant relationship exists when it does not. Therefore, the level of significance (alpha) is set at the 0.05 level to minimize the chance of Type I error. Type II error occurs when failing to reject a false null hypothesis and concludes that no significant relationship exists when it does. To minimize this risk, the beta value was set at 0.8 or higher.
3.8 Statistical Analysis
Initially, descriptive statistics of the data were determined using the SPSS software.
General characteristics of the data were inspected as well as the presence of any missing results.
Then, SPSS was used to look for multicollinearity or correlations among different variables for each construct. Highly correlated variables (>0.9) were not accepted in the model because of the multi-collinearity problem.
3.8.1 Structural Equation Modeling (SEM)
SEM was used as the main multivariate statistical method for the analysis. This is because SEM is a very versatile method that allows for the examination of latent constructs with multiple indicators; accounts for measurement errors; and conducts Confirmatory Factor
Analysis (CFA). Moreover, because this study employs the BM, multiple hypotheses needed to be examined, and SEM effectively served this purpose. Another advantage of SEM is that it can establish causal sequence and confirm the stability of the measurement model over time.
59 Path analysis differs from the traditional regression analysis in certain ways. Regression analysis, or covariance analysis, is a multivariate analysis methodology that is used to capture multiple relationships while simultaneously providing a simple and fast estimation result (Ahn,
2002). Path analysis, on the other hand, is a complementary methodology to regression analysis that uses a structural equation model to specify relationships among a set of variables, through path diagrams showing a system of simultaneous equations (Land, 1969; Petraitis, Dunham, &
Niewiarowski, 1996).
While path analysis is an extension of regression, its principal advantage is the simplification of complex relationships among the study variables through the visualization of relationships between study variables using a path diagram (Lleras, 2005; Wan & Cooper, 1988).
Moreover, path analysis does not show only direct relationships, as regression analysis does, but also, path analysis traces out indirect relationships, a step which is a key feature of the path analysis (Wan, 2002). Also, it is not possible to specify mediating variables in one regression analysis, but it is possible with path analysis (Streiner, 2005). However, it should be understood that path analysis cannot confirm causation between variables (Lleras, 2005; Petraitis et al.,
1996; Wan, 2002).
For this study, the IBM® SPSS® Amos Graphics 22 software was used to construct the measurement models. Then, CFA was conducted to test the construct validity of the measurement models and make any modifications based on the results. Lastly, a structural equation model was performed, based on the BM, and then analyzed by the goodness of fit statistics of the model.
60 3.8.2 Confirmatory Factor Analysis (CFA)
Confirmatory Factor Analysis (CFA) is a statistical methodology that was developed to facilitate the evaluation of the validity of the latent constructs’ measurement models (Byrne,
2001). Thus, it was used to examine whether the number of factors and loading of the measured
(indicator) variables on them complies with the expected, which is originally based on a theoretical background (Albright & Park, 2009). Hence, CFA helped in examining whether the measures of a construct were consistent with the model proposed.
One of the strengths of CFA is that when it is paired with a strong theoretical foundation prior to analyzing data, it can provide a power causal inference model by its ability to test the hypothesis indicated in the study. In addition, CFA allows the researcher to impose constraints on the model to determine specific correlations among different variables (Wan, 2002).
However, if the assumptions of the model are violated, the results have a tendency to be biased or can lead to drawing a wrong conclusion. Therefore, the validity and reliability of the measurement model need to be carefully examined before conducting SEM.
The three assumptions of CFA are: latent and observable variables are measured as deviations from their means, observable variables in the indicators are more prevalent than the unobservable ones, and the different factors are not correlated (Wan, 2002). Therefore, correlation analysis was also performed prior to CFA. Afterward, CFA was used to validate each of the three measurement models, (P, E, and N) by examining the relationships between the latent constructs and their indicators. This sequence identified whether the number of factors and loadings of the measured variables conformed to what was expected by the pre-established theory of this study.
61 3.9 Measurement Models
Measurement models illustrate the relationships between latent constructs and their indicators. The first component is the predisposing component or risk propensity profile (a latent exogenous variable). As a latent construct, risk indicators including age, gender, ethnicity, depression, dementia, and healthy lifestyle (HLS) were coded consistently in the same direction.
Factors with higher scores were considered to have an increased in the risk propensity.
Figure 4. Measurement Model for the Predisposing Component (Risk Propensity as an Exogenous Latent Variable)
Next is the enabling latent construct, which includes four indicators as shown in the following illustration:
62
Figure 5. Measurement Model for the Enabling Component (Exogenous Latent Variable)
N is assumed to be affected by P or E, and therefore, N was categorized as an endogenous variable. The following diagram shows the measurement model for N:
Figure 6. Measurement Model of the Need-for-Care Component (Need-for-care as an Exogenous Latent Variable)
63 3.10 Structural Equation Model with the Measurement Models
SEM is a statistical analysis process utilizing a linear variance-covariance matrix to analyze latent variables (both exogenous and endogenous) using factor loading, and thereby making it a subtype of CFA (Preacher, Cai, & MacCallum, 2007). The model overcomes weaknesses of factor analysis and structural equation modeling by merging them into a single model that estimates both latent variables and structural relations among the latent variable
(Wan, 2002). Another strength of SEM is that it allows for a direct test of the theory in terms of how well the model determines data (Byrne, 2001). The path diagram produced by SEM simplifies the relationship among all different variables (Bentler & Bonett, 1980; Schoenberg,
1989). Therefore, SEM is useful for this study as it contained a large number of the study variables with hypothesized causal links.
Based on the BM, two models were proposed. In the first model, U, an endogenous variable, is represented by LoS:
Figure 7. Andersen Behavior Model of Healthcare Utilization: Predictors of Hypoglycemia LoS and Outcome - Simplified
64 As is shown in the above figure, P and E are correlated with N. Also, all three components (P, E, and N) have a direct impact on LoS as well as on O. In addition, LoS is considered to have a direct impact on O. CCI is a control variable that is correlated with N. This configuration was constructed based on the findings of previous studies of HG that show how different factors that were inter-related with each other impacted the level of utilization and outcome among HG patients. For example, different studies that were discussed in the previous chapter show the relationship between age, severity of HG, LoS, total cost, and outcome. Also, some other studies showed the inter-relation between SES, medication adherence, severity and frequency of HG, LoS, total cost, and outcome. Therefore, the above model was formulated to examine the inter-relations between P and E on N.
When considering that total charges represent U, a similar model was constructed:
Figure 8. Andersen Behavior Model of Healthcare Utilization: Predictors of Hypoglycemia Total Charges and Outcome - Simplified
65 3.11 Validation of the Models
The above models were validated through the following statistics:
1. Factor loadings (standardized regression coefficient)
2. Goodness-of-fit statistics
3. Modification indices between errors
The model validation process included validating each measurement model developed, followed by a causal model in the SEM to determine the integrity of the proposed causal model.
3.11.1 Factor loadings (standardized regression coefficient)
Factor loadings with critical ratio value (equal to +1.96 or higher, and -1.96 or lower) were considered statistically significant at 0.05 level (Byrne, 2001). Indicators with insignificant values were excluded. Additionally, indicators with very low standardized regression coefficients might also be excluded from the original model to generate a more fit model.
3.11.2 Goodness-of-fit statistics
Goodness-of-fit statistics were produced by the AMOS software to help identify whether the measurement models fit the data (Bentler & Bonett, 1980). Therefore, the overall model fit was evaluated to understand how well the models fit the data. Consequently, goodness-of-fit statistics facilitated the decision to accept the SEM or not. The total number of known variances and covariances were compared to the estimated elements or parameters in the available sample
(Bentler, 1990). The target was to have an over-identified model (with a surplus of information or degrees of freedom). Chi-square value was used to show whether the model was reasonably
66 fitted to the data. In the case of a just-identified model, df should be equal to 0. If the model did not fit well with the data based on the selected goodness-of-fit statistics, the model was revised.
In order to improve the new revised model, first, insignificant factor loadings were eliminated. If removing insignificant factor loadings did not prove the researcher’s acceptable goodness-of-fit statistics, then measurement errors of factor loadings would be correlated with each other to get a better fit looking at the modification index (MI).
The following indicators (Table 3) were considered in the study to determine the construct validity of each measurement model:
Table 3. Goodness-of-Fit Indices and Criteria for Model Validation
Index Criterion Definition
The discrepancy should evaluates the degree of discrepancy between the sample and Chi-square (x²) be minimal (<5) fitted covariances (Hooper, Coughlan, & Mullen, 2008; Hu &
Bentler, 1999) The difference between the number of parameters to be estimated Degrees of Freedom (df) greater than or equal to 0 from the number of known elements (correlations) in the
correlation matrix (Weston & Gore, 2006) Tests the null hypothesis that the sample is obtained from a smaller than 5.0 suggests Likelihood Ratio (x²/df) population with characteristics of the covariance matrix (Wan, a good fit 2002)
0–1 (the larger, the Summarize the variance explained for the entire model (Weston GFI (goodness-of-fit index) better) & Gore, 2006) >0.9
0–1 (the larger, the AGFI (adjusted goodness- Measures GFI while considering the degrees of freedom better) of-fit index) available >0.9
Root Mean Squared Error smaller than 0.09 Assesses the degree of model adequacy on the basis of of Approximation suggests a good fit population discrepancy (Wan, 2002) (RMSEA)
Evaluates the model by comparing the χ2 the model to the null Comparative Fit Index >0.90 suggest a good fit model, while taking into account the sample size (Hooper et al., (CFI) 2008)
also known as NNFI, compares the discrepancy between the χ2 Tucker Lewis Index (TLI) >0.90 suggest a good fit of the hypothesized model to the one in the null model (Hu &
Bentler, 1999)
Hoelter's Critical N (CN) = or > 200 Specifies the largest sample size to be accepted (Hoelter, 1983)
67
3.11.3 Modification indices between errors
To improve the model fit, modification indices were examined. The pairs of error terms with high modification indices were correlated to each other in an effort to reduce the chi-square and to improve the model fit.
3.12 Decision Tree Regression (DTREG)
Decision Tree Regression is a robust application that is usually used for data mining, which is a process of extracting relevant information from the database, to generate user-friendly and easy-to-interpret models such as the decision tree model (dtreg.com, 2017). The software generates a series of branched “nodes” that describe the relationship between variables in the data. Examining the values in terminal nodes, which enclose the value of the predictor
(independent) variables, allowed predicting the value of the target (dependent) variable (Sherrod,
2014). Therefore, DTREG modeling was conducted following path analysis to further examine the probability of the different target variables included in this study.
68 CHAPTER 4 : RESULTS
4.1 Introduction
This chapter includes the results of the data analysis. Results of descriptive statistics and correlation analysis of all variables used in this study, generated by the SPSS statistical software program, are shown first. Then, CFA results for all the three latent constructs: (P, E, and N), are presented. It includes the results of the generic models followed by the modified models. Next, the two generated structural equation models with the measurement models for LoS and total charge are presented. Afterward, results of the DTREG analysis for the three target variables:
LoS, total charge, and O are shown, including a decision tree model for each of them. Last, each of the five hypotheses of the study were examined.
4.2 Descriptive Statistics
After merging the databases for the three years, cases involving patients that were younger than 18 years were excluded. Afterwards, all cases that included pregnant women were also excluded. Then, only cases that were admitted through ED, diagnosed with T1DM or
T2DM, and presence of HG were selected. When examining the selected cases, only a few variables (sex, AA_Hisp, patient_loc, Medicaid, and SES) had missing data, and the missing data for each of the variables constituted less than 5%. Therefore, missing data was replaced by the calculated mean. When checking for normality, there were few outliers among the two variables:
LoS and total charge, and these outliers were less than 0.3% of the data. Therefore, these outliers were excluded from the analysis in order to have normally distributed data. The following table shows the descriptive statistics of the variables after cleaning the data:
69 Table 4. Descriptive Statistics Std. Vali Miss Skew Kurto Mean Deviat Range Min. Max. Sum d ing ness sis ion
Age 4,822 0 58.42 17.08 -0.22 -0.64 72 18 90 281,700 Sex 4,822 0 0.48 0.50 0.08 -1.99 1 0 1 2,317 AA_Hisp 4,822 0 0.32 0.47 0.78 -1.40 1 0 1 1,540 Dementia 4,822 0 0.08 0.27 2.94 7.88 1 0 1 377 No_ 4,822 0 0.61 0.49 -0.44 -1.81 2,928 Depression 1 0 1 HLS 4,822 0 1.57 0.68 -1.31 0.34 2 0 2 7,592 Urban_ 4,822 0 0.77 0.42 -1.30 -0.31 3,726 hosp 1 0 1 Patient_ 4,822 0 4.18 1.16 -1.30 0.59 20,140 Loc 4 1 5 SES 4,822 0 2.21 1.07 0.36 -1.14 3 1 4 10,678 Medicaid 4,822 0 0.19 0.39 1.57 0.46 1 0 1 924 CCI 4,822 0 2.43 2.43 1.36 2.22 16 0 16 11,715 DM_ comp 4,822 0 0.23 0.42 1.28 -0.37 1 0 1 1,113 Uncont DM 4,822 0 0.24 0.42 1.24 -0.45 1 0 1 1,138 DMII 4,822 0 0.85 0.35 -2.01 2.04 1 0 1 4,120 Under_ Wt 4,822 0 0.14 0.34 3.84 8.53 1 0 1 476 Hyper- 4,822 0 0.40 0.49 0.39 -1.85 1,947 lipidemia 1 0 1 Renal_ 4,822 0 0.29 0.45 0.95 -1.10 1,380 Disease 1 0 1 Liver_ dis 4,822 0 0.06 0.23 3.82 12.61 1 0 1 275 Hyper- 4,822 0 0.76 0.43 -1.24 -0.47 3,679 tension 1 0 1 Cancer 4,822 0 0.08 0.27 3.13 7.82 1 0 1 379 LoS 4,822 0 3.59 3.56 2.71 9.97 27 0 27 17,287 27,30 205,78 207,49 131,662,57 4,822 0 27,381 2.75 9.86 TOTCHG 5 6 1713 9 9 Outcome 4,822 0 0.90 0.98 0.30 -1.70 3 0 3 4,336 YEAR 4,822 0 2,013 0.82 0.00 -1.50 2 2012 2014 9,706,689
As is shown in the above table, the total sample size was 4,822 with no missing data. A sample size of 4,822 participants seems sufficient as the general guidelines for SEM sufficient sample indicate 10 to 20 participants per estimated parameter (Weston & Gore, 2006). To examine normality, the skewness and kurtosis indices for each variable were inspected to
70 determine its normality (Weston & Gore, 2006). Extreme values were considered when the absolute values for skewness are >3.0, and >10.0 for kurtosis (Kline, 2005; Weston & Gore,
2006). In examining the results, it seems that the skewness and kurtosis indices for all variables were within the recommended range, except for the three variables: under_wt, liver_dis, and cancer, where the values were above the suggested range. Appendix D also shows histograms for all of the variables, and Appendix C demonstrates the frequency of the data. Again, frequency analysis showed that those three variables: under_wt, liver_dis, and cancer had very skewed distributions, about 10% with a value =1 vs. 90% =0. Therefore, these variables were excluded from the models.
4.3 Correlation Analysis
Since most of the variables in the data were categorical in nature, the Spearman's rank- order correlation was used to examine the correlation among the variables (Hauke & Kossowski,
2011). The relationship between two variables was expressed as a range between -1 to +1.
Correlation coefficients that are >0.85 are considered to be strong positive relationships and <-
0.8 are regarded as a strong negative correlation between two variables and may indicate potential problems (Kline, 2005; Weston & Gore, 2006). As is shown in the correlation tables in
Appendix E, none of the variables was highly correlated with the other, and therefore all variables were considered in the models.
71 4.4 Confirmatory Factor Analysis (CFA)
This study has three latent variables: P, E, and N and all are exogenous variables. A separate measurement model was developed for each of the latent variables and then independently validated by CFA.
4.4.1 The Predisposing component (P) or Risk Propensity
As mentioned earlier, the P construct was composed of six indicators: age, sex, AA_Hisp, dementia, depression, and HLS. Using AMOS Graphics 23, a CFA model was formulated:
Figure 9. Measurement Model for the Predisposing Component (An Exogenous Latent Variable) - Generic
As depicted in Figure 9, all indicators had a strong positive correlation with the latent variable P except for sex and AA_Hisp where the correlation was very weak. However, all
72 indicators for the construct were statistically significant at <.05 with critical values >1.96 except for sex, where P=0.81 and CR =0.24 (Table 5). Therefore, those two variables were excluded from the original model in the modified/nested model:
Figure 10. Measurement Model for the Predisposing Component (An Exogenous Latent Variable) - Modified
As is shown in the above-modified model for P, all indicators had a strong positive correlation with the latent variable P. In addition, there was a correlation between HLS and no_depression, which indicated an indirect effect on the latent variable P. As shown in the following table (Table 5), the modified model had all critical ratios >1.96 and they were statistically significant at p ≤ .05.
73 Table 5. Parameter Estimates for the Predisposing Component Measurement Models Generic Model Revised Model Indicator U.R.W S.R.W S.E C.R. P U.R.W S.R.W S.E C.R. P * HLS < ---P 0.023 0.455 0.001 19.121 *** 0.015 0.368 0.001 11.42 ** No_Depressio 11.44 * 0.016 0.457 0.001 19.159 *** 0.011 0.372 0.001 n < ---P 2 ** Dementia < --- 11.22 * 0.007 0.367 0 17.204 *** 0.005 0.329 0 P 3 ** Age < ---P 1 0.798 1 0.965 AA_Hisp < --- 0.002 0.052 0.001 3.05 0.002 - - - - - P Sex < ---P 0 0.004 0.001 0.24 0.81 - - - - - * d4 < -- > d3 0.041 0.143 0.006 7.439 ** *** . Correlation is significant at the 0.05 level. Note: U. R.W. = Unstandardized Regression Weight; S. R. W. = Standardized Regression Weight; S. E. = Standard Error; C. R. = Critical Ratio
The goodness-of-fit statistics were also significantly improved in the revised model as shown in the following table:
Table 6. Goodness-of-Fit Statistics of Generic and Revised Model for the Predisposing Component Fit Indices Criterion Generic Model Revised Model Chi-square (x²) < 5 341.5 1.45 Degrees of freedom (df) ≥ 0 9 1 Likelihood ratio (x²/df) <4 37.9 1.45 Probability (p or p-close) ≥ .05 0 0.228 Goodness-of-Fit Index (GFI) >.90 0.98 1 Adjusted GFI (AGFI) >.90 0.95 0.99 Tucker Lewis Index (TLI) >.90 0.75 0.99 Comparative Fit Index (CFI) >.90 0.85 1 Root Mean Square Error of Approximation (RMSEA) ≤.09 0.088 0.01 Hoelter’s Critical N (CN) > 200 239 12747
4.4.2 The Enabling component (E)
The E construct was composed of 4 indicators: urban_hosp, patient_loc, SES, and
Medicaid. Using AMOS Graphics 23, a CFA model was formulated:
74
Figure 11. Measurement Model for the Enabling Component (An Exogenous Latent Variable) - Generic
As is shown, all indicators had a strong positive correlation with the latent variable E except for Medicaid, where the correlation was very weak and adverse. In addition, all indicators for the construct were statistically significant at <.05 with critical values >1.96 except for
Medicaid, where P=0.171 (Table 7). Therefore, the variable “Medicaid” was excluded in the modified model:
Figure 12. Measurement Model for the Enabling Component (An Exogenous Latent Variable) - Modified
75 The above modified model for E shows that all indicators had a strong positive correlation with the latent variable P. In addition, the modified model had all critical ratios >1.96 and they were statistically significant at p ≤ .05.
Table 7. Parameter Estimates for the Enabling Component Measurement Models Generic Model Revised/Nested Model Indicator U.R.W S.R.W S.E C.R. P U.R.W S.R.W S.E C.R. P SES 0.251 0.244 0.058 4.349 *** 0.265 0.252 0.058 4.541 *** < --- E Patient_Loc 1 0.903 1 0.88 < --- E Urban_hosp 0.099 0.247 0.023 4.352 *** 0.104 0.254 0.023 4.543 *** < --- E Medicaid -0.014 -0.029 0.008 -1.711 0.171 - - - - - < --- E *** . Correlation is significant at the 0.05 level. Note: U. R.W. = Unstandardized Regression Weight; S. R. W. = Standardized Regression Weight; S. E. = Standard Error; C. R. = Critical Ratio
Because the model has only three variables, it was considered a “just identified model”
(Table 8):
Table 8. Goodness-of-Fit Statistics of Generic and Revised Model for the Predisposing Component Fit Indices Criterion Generic Model Revised Model
Chi-square (x²) < 5 1.75 - Degrees of freedom (df) ≥ 0 2 0 Likelihood ratio (x²/df) <4 0.877 - Probability (p or p-close) ≥ .05 0.4 - Goodness-of-Fit Index (GFI) >.90 1 1 Adjusted GFI (AGFI) >.90 0.99 - Tucker Lewis Index (TLI) >.90 1 -
Comparative Fit Index (CFI) >.90 1 1 Root Mean Square Error of Approximation ≤.09 0 0.08 (RMSEA) Hoelter’s Critical N (CN) > 200 16,477 -
As is shown above, the RMSEA was 0.08, an acceptable model fit, for the revised model.
76 4.4.3 The Need-for-Care Component (N)
As mentioned earlier, the N construct was composed of six indicators: DM_comp, DMII, renal_disease, hyperlipidemia, hypertension, and uncontrolled_DM. using AMOS Graphics 23, a
CFA model was formulated:
Figure 13. Measurement Model for the Need-for-Care Component (An Exogenous Latent Variable) - Generic
As shown in Figure 13, all indicators had a strong positive correlation with the latent variable N except for uncontrolled_DM, where the correlation was negative. In addition, all indicators for the construct were statistically significant at <.05 with critical values >1.96 except for uncontrolled_DM, where P=0.09 (Table 9). Therefore, uncontrolled_DM was excluded in the modified model:
77
Figure 14. Measurement Model for the Need-for-Care Component (An Exogenous Latent Variable) – Modified As is shown, the above-modified model for N shows that all indicators had a strong positive correlation with the latent variable (N). In addition, some correlations between the indicators had been established to improve the model fit. As shown in the following table (Table
9), the modified model had all critical ratios >1.96 or <-1.96 and they were statistically significant at p ≤ .05.
Table 9. Parameter Estimates for the Need-for-Care Component Measurement Models Generic Model Revised Model
Indicator U.R.W S.R.W S.E C.R. P U.R.W S.R.W S.E C.R. P 15 ** DMII < --- N 0.453 0.317 0.029 0.813 0.518 0.061 13.281 *** .6 * 8. ** DM_comp < --- N 0.265 0.155 0.033 0.298 0.159 0.053 5.657 *** 1 * 21 ** Hyperlipidemia < --- N 1.051 0.529 0.048 0.485 0.222 0.041 11.959 *** .7 * Renal_Disease < --- N 1 0.514 1 0.497
Hypertension < --- N 0.95 0.58 0.95 0.502 - - - 0. UncontDM < --- N 0.032 - - - - - 0.054 0.031 1.6 09 n14 <--> n15 0.072 0.412 0.003 21.607 ***
n11 <--> n12 -0.024 -0.19 0.003 -8.971 ***
n11<--> n13 0.028 0.17 0.003 8.25 ***
n12<--> n13 -0.023 -0.197 0.003 -8.262 *** *** . Correlation is significant at the 0.05 level. Note: U. R.W. = Unstandardized Regression Weight; S. R. W. = Standardized Regression Weight; S. E. = Standard Error; C. R. = Critical Ratio
78
The goodness-of-fit statistics were also significantly improved in the revised model as shown in Table 10:
Table 10. Goodness-of-Fit Statistics of Generic and Revised Need-for-Care Component Fit Indices Criterion Generic Model Revised Model Chi-square (x²) < 5 978 7.14 Degrees of freedom (df) ≥ 0 10 2 Likelihood ratio (x²/df) <4 97.8 3.57 Probability (p or p-close) ≥ .05 0 0.02 Goodness-of-Fit Index (GFI) >.90 0.9 0.99 Adjusted GFI (AGFI) >.90 0.86 0.99 Tucker Lewis Index (TLI) >.90 0.374 0.98 Comparative Fit Index (CFI) >.90 0.58 0.99 Root Mean Square Error of Approximation (RMSEA) ≤.09 0.14 0.02 Hoelter’s Critical N (CN) > 200 91 4,043
4.5 Structural Equation Modeling with the Measurement Models
This last step was the formulation of the SEM with the measurement models. The statistical process in the SEM allowed for examining how well the data fit the theoretically- formulated model. As mentioned earlier, two separate models were developed and examined, one using LoS and another one for total charge.
4.5.1 LoS Model
After revising the models for P, E, and N, a structural equation model with the measurement models for LoS was constructed based on the BM:
79
Figure 15. Structural Equation Model with the Measurement Models for LoS and Outcome
As is shown, N had the strongest impact on LoS, P had a very weak one, while E had an insignificant relationship with LoS. Regarding O, N was also the strongest component, followed by P, and E has no impact on O. There was also an inter-relation between P and N, which indicated an indirect effect on LoS and O. CCI, as a control variable, also strongly correlated with N.
Regarding the model’s goodness-of-fit, it seems that the model was not optimally fit because the likelihood ratio was 39.6 (Table 11). However, some indicators such as RMSEA and
80 GFI showed that the model was reasonably fit. In addition, additional modifications can be made in the model that might improve its fit to the data.
Table 11. Goodness-of-Fit Statistics of Generic SEM for LoS Fit Indices Criterion Generic Model Chi-square (x²) < 5 3128 Degrees of freedom (df) ≥ 0 79 Likelihood ratio (x²/df) <4 39.6 Probability (p or p-close) ≥ .05 0 Goodness-of-Fit Index (GFI) >.90 0.92 Adjusted GFI (AGFI) >.90 0.6 Tucker Lewis Index (TLI) >.90 0.6 Comparative Fit Index (CFI) >.90 0.7 Root Mean Square Error of Approximation (RMSEA) ≤.09 0.08 Hoelter’s Critical N (CN) > 200 156
Table 12. Parameter Estimates for the Generic LoS SEM Generic Model Indicator U.R.W S.R.W S.E C.R. P LoS <--- N 2.961 0.159 0.281 10.523 *** LoS<--- P -0.012 -0.053 0.003 -3.342 *** LoS<--- E 0.017 0.005 0.056 0.3 0.764 Renal_Disease<--- N 1 0.462 DMII<--- N 0.24 0.13 0.027 8.976 *** No_Depression<--- P 0.011 0.379 0.001 17.053 *** Dementia<--- P 0.005 0.33 0 16.073 *** Hypertension<--- N 0.95 0.396 HLS<--- P 0.016 0.374 0.001 16.958 *** SES<--- E 0.26 0.248 0.058 4.454 *** Patient_Loc<--- E 1 0.887 Urban_hosp<--- E 0.103 0.252 0.023 4.457 *** Hyperlipidemia<--- N 0.473 0.182 0.033 14.374 *** DM_comp<--- N 1.115 0.511 0.038 29.506 *** Severity_of_Adv_outcome <--- P 0.013 0.224 0.001 12.474 *** Severity_of_Adv_outcome <--- N 1.175 0.234 0.075 15.718 *** Severity_of_Adv_outcome <--- E 0.005 0.005 0.015 0.316 0.752 Severity_of_Adv_outcome <--- LoS 0.027 0.099 0.004 7.294 *** AGE<--- P 1 0.953 *** . Correlation is significant at the 0.05 level. Note: U. R.W. = Unstandardized Regression Weight; S. R. W. = Standardized Regression Weight; S. E. = Standard Error; C. R. = Critical Ratio
81 4.5.2 Total Charge Model
After revising the models for P, E, and N, a structural equation model with the measurement models for total charge was constructed based on the BM:
Figure 16. Structural Equation Model with the Measurement Models for Total Charge and Outcome
Similar to the LoS model, N in this model had the strongest impact on total charge.
However, E appeared to have a moderate impact on total charge, while P was the weakest.
Regarding O, N remained the strongest predictive component, followed by P, and E had no
82 impact on O. The inter-relation between P and N was also present as well as the correlation between CCI and N. The model was also reasonably fit, as some of the indices were at the target level while others were not.
Table 13. Goodness-of-Fit Statistics of Generic and Revised SEM for Total Charge Fit Indices Criterion Generic Model Chi-square (x²) < 5 3072 Degrees of freedom (df) ≥ 0 79 Likelihood ratio (x²/df) <4 38.8 Probability (p or p-close) ≥ .05 0 Goodness-of-Fit Index (GFI) >.90 0.93 Adjusted GFI (AGFI) >.90 0.89 Tucker Lewis Index (TLI) >.90 0.66 Comparative Fit Index (CFI) >.90 0.74 Root Mean Square Error of Approximation (RMSEA) ≤.09 0.08 Hoelter’s Critical N (CN) > 200 159
Table 14. Parameter Estimates for the Generic Total Charge SEM Indicator U.R.W S.R.W S.E C.R. P TOTCHG<--- N 23615 0.168 2138 11.04 *** TOTCHG<--- P -88.353 -0.053 26.129 -3.38 *** TOTCHG<--- E 4140.641 0.148 779.62 5.31 *** Renal_Disease<--- N 1 0.468 DMII<--- N 0.244 0.135 0.027 9.16 *** No_Depression<--- P 0.011 0.378 0.001 17.04 *** Dementia<--- P 0.005 0.33 0 16.07 *** Hypertension<--- N 0.95 0.402 HLS<--- P 0.016 0.373 0.001 16.95 *** SES<--- E 0.282 0.258 0.047 6.02 *** Patient_Loc<--- E 1 0.847 Urban_hosp<--- E 0.113 0.265 0.019 6.03 *** Hyperlipidemia<--- N 0.472 0.184 0.033 14.40 *** DM_comp<--- N 1.112 0.517 0.038 29.56 *** Severity_of_Adv_outcome <--- P 0.013 0.228 0.001 12.74 *** Severity_of_Adv_outcome <--- N 1.103 0.261 0.074 15.00 *** Severity_of_Adv_outcome <--- E -0.018 -0.018 0.016 -1.10 0.27 Severity_of_Adv_outcome <--- TOTCHG 0 0.162 0 11.85 *** AGE<--- P 1 0.955 *** . Correlation is significant at the 0.05 level. Note: U. R.W. = Unstandardized Regression Weight; S. R. W. = Standardized Regression Weight; S. E. = Standard Error; C. R. = Critical Ratio 83
4.6 Decision Tree Regression (DTREG) Results
As mentioned earlier, DTREG modeling was conducted on each of the three target variables: LoS, total charge, and O to further explore the probability of the different predictor
(independent) variables on the target (dependent) variable. Initially, only those predictor variables under (P, E, and N) that were found significant in path analysis were analyzed separately against the target variable. Afterward, an overall decision tree analysis was conducted, which included high-impact variables on the target variable. Results of the analysis, as well as the decision tree models, are shown following the analysis of each of the target variable.
4.6.1 DTREG for LoS
As described above, a separate DTREG analysis was conducted for each of the groups of predictor variables influencing LoS. The results of the analysis for the first group (P), including:
Age, dementia, no_depression, and HLS, showed that age (relative importance =100) was by far the most important variable that influenced LoS followed by HLS, no_depression, and dementia
(relative importance = 18.6, 12.4, 9.5 respectively) (Appendix F, Figure 20). For the (E) group,
SES was of highest importance followed by patient_loc and then urban_hosp (relative importance = 100, 72.1, 63.3 respectively) (Appendix F, Figure 21). Regarding the (N) group, the top two predictor variables were DM_comp and renal_disease (relative importance = 100,
62.5 respectively), followed by hyperlipidemia, DMII, and hypertension (relative importance =
49.4, 47.15, 45.29 respectively) (Appendix F, Figure 22).
The results of the path analysis showed a very weak impact of P and E on LoS while N had the largest impact (Table 12). Therefore, the overall DTREG analysis for LoS included the
84 following predictor variables: age, SES, DM_comp, renal_disease hyperlipidemia, DMII, and hypertension. The results of the analysis revealed that among all factors, age had the highest relative impact on LoS, followed by SES, DM_comp, renal_disease, and hypertension
(Appendix F, Figure 23). The proportion of variance explained by the model (R^2) was 7.3%, and RMSE was 3.4. A DTREG tree of the predictors for LoS was generated as well as a summary table for the results as shown below:
Figure 17. DTREG Tree of the Predictors for LoS
85
Table 15. Summary of DTREG Analysis of the Predictors for LoS – ranked from highest to lowest
LoS Score (days) Characteristics
9.80 Age <22.5, with DM complications, SES 1,2
6.80 Age >82.5, with DM complications, no renal disease, SES 3,4
4.90 Age <67.5, with DM complications and Renal disease, SES 3,4
4.24 Age >22.5, with DM complications, SES 1,2
3.64 Age >57.5 with no DM complication
3.24 Age <57.5 with no DM complication
3.16 Age >67.5, with DM complications and Renal disease, SES 3,4
3.12 Age <82.5, with DM complications, no renal disease, SES 3,4
The results indicated that those young DM patients who were younger than 22.5 years, who have one or more DM complications and who have a lower SES, have the longest LoS.
Second to them were those elderly DM patients who were older than 82.5 years and at a higher
SES, with DM complications and no renal disease. On the other hand, shorter LoS was shown among those younger than 82.5 years with DM complications.
4.6.2 DTREG for Total Charge
Similar steps were conducted for the DTREG analysis with total charge. Again, the results of the analysis for the (P) group showed that age (relative importance =100) is by far the most important variable that influenced total charge followed by HLS, no_depression, and dementia (relative importance = 15.2, 11.2, 7.1 respectively) (Appendix G, Figure 24). For the
(E) group, patient_loc was the highest importance followed by SES and then urban_hosp
86 (relative importance = 100, 23.7, 13.4 respectively) (Appendix G, Figure 25). Regarding the (N) group, the top two predictor variables were renal_disease and DM_comp (relative importance =
100, 49.3 respectively), followed by DMII, hyperlipidemia, and hypertension (relative importance = 27.5, 19.9, 18.6, respectively) (Appendix G, Figure 26).
Since the path analysis showed that P has a very weak impact on total charge (Table 14), the overall DTREG analysis included the following predictor variables: age, patient_loc, SES, urban_hosp, renal_disease, DM_comp, and DMII. The results of the analysis indicated that age had the highest relative impact on total charge, followed by patient_loc, SES, renal_disease,
DM_comp, and DMII (Appendix G, Figure 27). The proportion of variance explained by the model (R^2) was 11.97%, and RMSE was 25,687. A DTREG tree of the predictors for LoS was generated as well as a summary table for the results as shown in Figure 18:
Figure 18. DTREG Tree of the Predictors for Total Charge
87 Table 16. Summary of DTREG Analysis of the Predictors for Total Charge – ranked from highest to lowest
Total Charge Score Characteristics
$33,548 Age <81.5 with renal disease, pt. loc 4,5 $30,329 Age 70-80, pt loc 1,2,3 $28,426 Pt loc 5 and no Renal disease $24,933 Age >81.5 with renal disease, pt. loc 4,5 $24,074 Pt loc 4 and no Renal disease $22,734 Age <70, with DM comp, Pt loc 1,2,3 $17,829 Age >80, pt loc 1,2,3 $17,643 Age <70, no DM comp, pt loc 1,2,3
As shown in Table 16, highest charges were associated with those younger than 81.5 years who are located in larger metropolitan areas and who had renal disease, followed by those who were between 70-80 years old and who lived in smaller towns. In contrast, those who were younger than 70, had no DM complication and lived in smaller towns had the lowest charges, followed by those elderly DM patients who were older than 80 and lived in smaller towns.
4.6.3 DTREG for Outcome
The last DTREG analysis was for O. The results of the analysis for the P group showed that age (relative importance =100) was the most important variable that influenced O followed by HLS, no_depression, and dementia (relative importance = 15.8, 11.2, 6.04 respectively)
(Appendix H, Figure 28). For the E group, patient_loc had the highest importance followed by
SES and then urban_hosp (relative importance = 100, 30.5, 17.6 respectively) (Appendix H,
Figure 29). Regarding the N group, the top two predictor variables were renal_disease and
88 DM_comp (relative importance = 100, 49.3 respectively), followed by DMII, hyperlipidemia, and hypertension (relative importance = 27.5, 19.9, 18.6 respectively) (Appendix H, Figure 30).
Both path analyses showed that E had a minimal impact on O (Table 12 and 14). Thus, the overall DTREG analysis for O included the following predictor variables: age, HLS, patient_loc, renal_disease, and DM_comp. The results of the analysis indicated that age had the highest relative impact on O, followed by renal_disease, patient_loc, HLS, and DM_comp
(Appendix H, Figure 31). The proportion of variance explained by the model (R^2) was 18.6%, and RMSE was 0.8. A DTREG tree of the predictors for LoS was generated as well as a summary table for the results as shown in Figure 19:
Figure 19. DTREG Tree of the Predictors for Outcome
89 Table 17. Summary of DTREG Analysis of the Predictors of Outcome – ranked from worst to best Outcome Score Characteristics
1.7 Age > 87.5 with renal disease
1.36 Age 61.5-87.5 with renal disease
1.34 Age > 75.5 with no renal disease
1.08 Age 51.5-61.5 with renal disease
1.007 Age 61.5-75.5 with no renal disease
0.97 Age <51.5 with renal disease
0.7 Age 52.5-61.5 with no renal disease
0.47 Age <52.5 with no renal disease
The results indicated that the worst outcome was among diabetics who were older than
61.5 years with a renal disease. Better outcomes were found among those who were younger than 61.5 with no renal disease.
4.7 Hypothesis Testing
This last section is concerned with reviewing the five hypotheses for this study and evaluating whether the results of the study support each one of them.
4.7.1 Hypothesis One
The first two hypotheses for this study were about the magnitude of different components on utilization. The first one was for LoS:
Ha1: The magnitude of different components varies with regard to determining the LoS among patients who develop HG.
90 The SEM analysis clearly revealed that all the three factors: P, E, and N had different impacts on LoS (Table 12). N was found to have the strongest impact on LoS, while P and E had a very weak direct impact on LoS. Therefore, hypothesis 1 was supported.
4.7.2 Hypothesis Two
The second hypothesis for the study was about the other utilization measure, total charge:
Ha2: The magnitude of different components varies with regard to determining the total
cost among patients who develop HG.
The results of the analysis indicated that N had the strongest impact on total charge, followed by E. On the other hand, P seemed to have a very week direct impact on total charge
(Table 14). Thus, the study results also supported hypothesis 2.
4.7.3 Hypothesis Three
The third hypothesis was related to the magnitude of different components on outcome:
Ha3: The magnitude of different components varies with regard to determining the
outcome among patients who develop HG.
The analysis showed that N and P had a stronger direct impact on O when compared to E
(Table 12 and 14). In addition, both measures for U: LoS and total charge showed a moderate impact on the O. Therefore, these findings supported the third hypothesis.
4.7.4 Hypothesis Four
The fourth hypothesis is about whether N was the most important predictor for utilization:
91 Ha4: The need-for-care component is the most important component of the predictors of
use, while the effects of other predictors are being simultaneously considered.
When considering LoS, N showed the strongest significant direct effect on LoS, while P had a very weak but significant effect on LoS, and E had a non-significant effect. Regarding total charge, all factors showed a statistically significant impact. However, N still had the strongest effect (Table 12 and 14). Thus, that hypothesis was supported.
4.7.5 Hypothesis Five
The last hypothesis was related to the interaction effects among predictor variables on utilization
and outcome:
Ha5: There are statistically significant interaction effects (e.g., P*N, N*E, and E*N) of
predictor variables on utilization and outcomes.
When examining the models, we noted that there were no inter-relations between E and
N. However, there was a strong inter-relation between P and N in the LoS model as well as the total charge model (SRW= 0.25 for both models). These findings indicated indirect effects on
LoS and O. Using SEM, interaction effects can be accurately determined. However, a closer look at the DTREG analysis results showed that certain factors had high impact on target variables.
For example, age, SES, and renal disease each had a high impact. However, the results indicated that their effect varied in severity based on the presence or absence of other factors. Hence, an interaction effect is present among these factors and therefore the fifth hypothesis was supported.
92 Table 18. Summary of Hypothesis Testing Results Hypotheses Test Result
Ha1: the magnitude of different components varies with regard to determining the Supported (Positive) LoS among patients who develop HG.
Ha2: the magnitude of different components varies with regard to determining the Supported (Positive) total cost among patients who develop HG.
Ha3: the magnitude of different components varies with regard to determining the Supported (Positive) outcome among patients who develop HG.
Ha4: The need-for-care component is the most important component of the Supported (Positive) predictors of use, while the effects of other predictors are being simultaneously considered.
Ha5: There are statistically significant interaction effects (e.g., P*N, N*E, and Supported (Positive) E*N) of predictor variables on utilization and outcomes.
4.8 Summary of the Chapter
This chapter presented all the results of the data analysis. Starting with descriptive analysis, few variables were excluded from the analysis as they breached the normality assumptions. Afterward, correlation analysis showed that none of the remaining variables was highly correlated with the other. Then, all measurement models were tested and modified.
Subsequently, the SEM for LoS and total charge was constructed, and the results were revealed.
Afterward, DTREG analysis results were discussed for each of the target variables followed with a summary of high-impact variables for each target variable. Last, all the five hypotheses were examined based on the results of the analysis. Through the various analyses of the data, all five hypotheses were supported. More discussion about these results will be in the following chapter.
93 CHAPTER 5 : DISCUSSION, IMPLICATIONS, LIMITATIONS AND CONCLUSIONS
5.1 Summary
This study had five research questions and five accompanying hypotheses. The study findings provided strong support for all of them. The results indicated that different elements in the behavior model vary in terms of their impact on the level of utilization and outcome among
HG patients. The results of the study also revealed that the need-for-care component was the most important predictor for length of stay, total charge, and outcome. Moreover, important inter-relations were found between some components such as the predisposing and need-for-care components, which indicated that these components had indirect impacts on utilization and outcome along with their direct ones.
In this section, the two different elements for utilization: length of stay and total charge will be discussed separately, followed by outcome. Afterward, the various implications of the study including policy, practical, and theoretical implications were discussed. Then, major limitations of the study were presented, followed by recommendations for future research and conclusion.
5.2 Discussion of the impact of different components on length of stay
Results of the SEM for length of stay did go along with the theoretical background for this study; the results showed that the three different components (predisposing, enabling, and need-for-care) had different impacts on utilization. Moreover, the results indicated that the need- for-care had the strongest effect on the duration of stay with a weight of 0.16 (Table 12). This mimics the classical findings of the original developer of the BM, Ronald Andersen (Aday &
Andersen, 1974; Andersen & Newman, 1973). The predisposing component had a smaller effect,
94 but statistically significant impact on patients’ length of stay. This means that patients with some risk factors under the predisposing component or the enabling component will likely be expected to stay longer in the hospital. For example, older patients with diabetes complications who are located in larger metropolitan areas are expected to spend more time in the hospital. Last, the enabling component shows no effect on the duration of stay among diabetic patients with hypoglycemia. This means that the socio-economic status and geographical location do not play a significant role in determining the duration of stay in hospitals among those patients. Thus, when trying to tackle the issue of length of stay among HG patients, this study suggests a focus on issues related to factors under the predisposing and need-for-care components rather than the enabling one.
However, this study showed that the main influence on patients’ length of stay is when there are factors related to illness, which was clearly demonstrated by the inter-relation between the predisposing component and need-for-care. In turn, this finding goes along with the BM, where the predisposing component had a weak direct impact on utilization but a stronger indirect impact through need-for-care. Thus, patients who have some factors under the predisposing component do not necessarily have longer stays in hospital unless they have other factors under the need for care component. This means that patients who have renal disease and diabetes complications are more likely to stay longer in hospitals if they are older and have dementia. In contrast, older patients with dementia are not likely to stay longer in hospitals if they do not show any diabetes complication, renal diseases, or other factors under the need-for-care component.
These findings also support the findings from other studies in which certain factors such as dementia, tobacco use, alcohol use, or drug use played an important role in patients’ behavior,
95 including adherence to medication, regular follow-ups for screening and prevention, and self- care measures. Consequently, those patients demonstrated more comorbidities and illnesses, which lead to longer stays in the hospital.
Another finding is related to comorbidities, demonstrated as the control variable CCI.
Since these comorbidities have a great impact on utilization and outcome, the CCI variable was applied as a controlled variable in an attempt to hold it constant during the statistical analysis in order to test the relative relationship among the main component in the model. The strong correlation between comorbidities and the need-for-care component shown in the results was expected because of the great number of previous studies that reported the impact of different comorbidities on the severity of illness among HG patients. This relationship provides great knowledge to healthcare providers when considering the issue of length of stay among HG patients. This is because comorbidities seem to have a major impact on the severity of illness as well as the health consequences among HG patients. Therefore, investing more effort and time in treating those comorbidities might help to reduce the severity of HG episodes as well as the duration of stay in hospitals.
Along with the SEM results, the DTREG results of this study showed that age was the most important predictor variable among all the other indicators under the predisposing component. Elderly diabetic patients have been consistently shown to have higher risks for developing HG as well as developing complications, staying longer in the hospital, and acquiring higher total costs (Geller et al., 2014; Kong et al., 2014; Majumdar et al., 2013; Pathak et al.,
2015). on the other hand, patient’s socio-economic status was found to have the strongest impact on length of stay among all different factors under the enabling component. Again, based on the theoretical background of this study, the presence of factors under the need-for-care component
96 is crucial to impact utilization. The results of this study showed that the presence of DM complications and renal disease, which are factors under the need-for-care component, played a vital role in impacting length of stay (Table 15). These findings in the DTREG analysis explain the indirect impact of predisposing and enabling components on length of stay that was revealed in the SEM.
5.3 Discussion of the impact of different components on total charge
When exploring the results of the SEM for the second element of utilization, total charge, this study found once more that the three different components had different impacts on utilization, with the need-for-care being the strongest with a S.R.W of 0.17 (Table 14). Thus, patients who have some factors under the need-for-care component are more likely to incur higher charges, regardless the presence or absence of any other factors under the predisposing or enabling component. However, the results also show inter-relations between the predisposing and the need-for-care components. This means that some factors such as patients’ age and dementia can have an impact on the total charge if there are other factors under the need-for-care component such as renal disease and hypertension present in the patient. Likewise, the DTREG results showed that the presence of renal disease and DM complication, as factors under the need-for-care component, were the key factors in defining the total charges associated with the admission. Yet, higher chargers were found to be among older patients, a factor under the predisposing component, who have renal diseases and DM complications.
Regarding comorbidities, the findings in the total charge model are similar to those in the length of stay model. The results show a great relation between comorbidities and the need-for- care component. This means that patients with a higher comorbidity index are more likely to
97 incur higher charges for their admission. Thus, when considering reducing the charges among
HG patients, it is important to invest more time and effort in treating other comorbidities associated with those patients.
Another finding is related to the enabling component in the total charges model. While there was no impact of the enabling component on length of stay, there is a relatively strong and statistically significant impact of the enabling component on total chargers. These findings indicated that patients with higher socio-economic status, living in larger metropolitan areas, and visiting urban hospitals incurred higher charges than their counterparts in rural areas. Parallel findings were found in the results of DTREG analysis as patients living in larger metropolitan areas with a higher SES incurred higher charges than the others. These findings were contrary to what was reported in the literature, which was that patients in smaller towns usually had higher
ED visits as well more severe HG episodes (Andrus et al., 2004; Babitsch et al., 2012; Chen et al., 2015; Conwell & Boult, 2008; Ginde et al., 2008; Rosenblatt et al., 2001; Unger et al., 2011;
Weiner et al., 1995). One explanation for the discrepancy might be that smaller hospitals in rural areas provide only essential services with no follow-up services (Broyles et al., 1999; Thode et al., 2005), therefore resulting in fewer charges. Another possibility for the difference is that because patients with lower SES who live in smaller towns are sicker, some of them die in the hospital and hence incur fewer charges than their counterparts who survive.
5.4 Discussion of the impact of different components on outcome
Results regarding outcome are very similar between the models of length of stay and total charges. The ranking is as follows: need-for-care component had the highest impact on outcome, followed by the predisposing component, then utilization, while the enabling component had the
98 weakest impact on outcome. This means that patients with more severe illness status were more likely to have worse outcomes, regardless the presence of any other factors under all other components. However, the impact of the other two components, predisposing and enabling, can be demonstrated indirectly through the need-for-care component. Also, comorbidities are shown to indirectly impact outcomes. Thus, the outcome could be expected to get worse among patients who have some factors under the need-for-care components as well as other factors under the two other component or other comorbidities. For example, older diabetic patients who develop
HG usually have more comorbidities and thus were expected to have very poor outcomes when compared to younger patients. The DTREG results also revealed the same findings (Table 16), where older patients with renal disease demonstrated more adverse outcomes, while younger patients with no renal disease had the best outcomes. In conclusion, in order to improve the outcomes among HG patients, efforts should focus on factors under the need-for-care component such as treating underlying renal disease and DM complications. Also, other high impact factors under the other components such as age, dementia, and comorbidities can impact the outcome and they need to be considered as well when trying to improve the outcome among HG patients.
5.5 Implications
5.5.1 Policy Implications
Florida is one of the largest states in terms of the percentage of adults with diabetes, and the three Central Florida counties of Orange, Osceola, and Seminole are close to Florida’s average of over 10% of adults diagnosed with diabetes in 2013 (Floridacharts.com, 2016) and more than 26% of adults being obese. In 2014 alone, there were more than 36,000 hospitalizations in Orange county related to diabetes (Floridacharts.com, 2016). According to a 99 2013 Central Florida community health needs assessment report, diabetes is the most prevalent chronic disease in the region, is a leading cause of death, and is therefore considered one of the top five health priorities in the region (Floridahospital.com, 2013).
Results of this study also showed that the geographical location of diabetic patients with
HG impacts the level of utilization as well as outcome. Recent data about the geographical distribution of diabetes in Florida shows that 14% of adults in urban areas have diabetes, compared to 15.5% in suburban areas and 18.7% in rural areas (UnitedHealthFoundation, 2016).
A recent study conducted in eight southeastern states also showed racial disparities in diabetes hospitalization of rural Medicare beneficiaries (Wan, Lin, & Ortiz, 2016). Therefore, this study suggests that more effort is needed toward residents in rural areas. These measures might include conducting community needs assessments to evaluate the availability of healthcare services in patients’ neighborhoods as well as examining what other challenges are present among those residents that might affect their diabetes-related care.
Another major finding of this study is the great impact of age on utilization and outcome among diabetic patients with HG. In Florida, most of the state’s counties have older populations than do counties in the U.S. overall (Reynolds, Gunderson, & Bamford, 2015), and the number of Floridians aged 85 and older increased by about 4% from 2000 to 2010 (Reynolds et al.,
2015). Additionally, the rate of ED visits for diabetics who are 75 and older increased 44% from
2009 to 2014 (Floridahospital.com, 2013). At the county level, it is estimated that by 2025,
Orange county will have the highest number of people aging 65 and older with over 215,000 and accounting for 14% of the county’s population (Appendix J, Table 27). In contrast, Seminole county have the highest percentage of elderly people in Central Florida and it is estimated that this percentage will reach 18% in 2025 (Appendix J, Table 27). When specifically examining
100 certain races such as Blacks and Hispanics, Osceola county shows the highest percentage of elderly people among those races (Appendix J, Table 27). Thus, this study indicates a recommendation for decision makers in Central Florida to develop certain strategies targeting older diabetic patients in Orange and Seminole counties to improve their health literacy, enhance their adherence to medications, and be more encouraged to perform regular prevention and screening visits. In addition, certain strategies can be formulated to target elderly Blacks and
Hispanics living in Osceola county.
Another issue is that the findings of this study show relationships between old age, comorbidities, HG severity, Socio-economic status, and living in rural areas with the level of utilization and the severity outcomes. Recent statistics show that Osceola county has the highest percentage increase of disabled people younger than 65 years from 2011-2015 in Central Florida
(Appendix J, Table 27). Moreover, when compared to the other two counties in Central Florida,
Osceola county has the highest percentage of uninsured people younger than 65 years, the lowest median household income, and the highest number of persons in poverty (Appendix J, Table 27).
Orange county, on the other hand, has the highest increase in the number and percentage of veterans from 2011-2015 (Appendix J, Table 27). Therefore, with a transdisciplinary approach, policy makers might need to consider those issues to implement better strategies to reduce the growing negative economic and social impact of the HG problem among diabetic Floridians. To approach such a huge problem, a collaboration among different sectors and stakeholders, including health care providers, social workers, policy makers, financial experts, health educators, community leaders, and other experts is called for in order to understand the size of the problem, discuss the challenges from all perspectives, examine successful strategies implemented in other areas, propose some policy options, and then choose the best feasible
101 option. This approach will ensure the implementation of better strategies that are economically and practically feasible. In addition, including experts from different disciples will provide a larger pool of knowledge, which could reduce potential mistakes and increase success.
Last, there are some difference between the three counties in Central Florida regarding their racial profile. The 2017 estimates show that Osceola county has the highest percentage of
Hispanic population, Orange county has the highest percentage of African Americans, while
Seminole county has the highest percentage of White people (Appendix J, Table 27). From 2010-
2014, The percentage of Hispanics increased the highest in Osceola county while the number of
Hispanics increased the highest in Orange county (Appendix J, Table 27). Although this study showed that racial difference did not play a significant role in impacting the utilization and outcome among HG patients, it is important for policy makers to consider the racial differences between the three counties especially when developing certain educational or behavioral modification policies and strategies.
5.5.2 Theoretical Implications
The BM offers a framework for examining the impact of different risk factors for HG utilization and outcome. Employing SEM to analyze the magnitude of effects of different factors provided a conceptual understanding of the relationship between all the components of the BM.
Moreover, the BM emphasizes the direct and indirect impact of different components on utilization and outcome. Thus, the use of advanced statistical methodologies such as SEM was crucial to examining these relationships and showing the difference between the direct and indirect impacts of each component on utilization and outcome. In addition, each of the components was conceptualized and examined separately as a latent construct. This provided in-
102 depth knowledge regarding each of the different risk factors under each component and the presence of any inter-relations among them.
The BM theoretical base of the study also played a vital role when applying DTREG analysis. Rather than including all factors into one single inquiry, certain factors were grouped together based on the component they are related to. This approach provided a comprehensiveness to each component in terms of the most important factors in each group that influenced the target variable. Consequently, only high impact factors under each component were included in the final analysis. This approach provided more accurate and interpretable results that are also more practical.
5.5.3 Practical Implications
With recent transition toward the value-based payments (VBP), healthcare providers are being encouraged to deliver more high-quality services at lower cost (CMS.gov, 2017). This is because their reimbursement is now based on quality rather than quantity of visits or the number of services provided (Wagner, 2015). In addition, stakeholders impose continuous pressure on healthcare organizations to provide sustainable, high quality, and cost-effective services
(Ramirez, West, & Costell, 2013). Therefore, understanding which groups are at high risk for developing complications or using more services for a common disease such as DM is crucial for healthcare providers (Henkel & Maryland, 2015).
This study revealed that certain groups are at higher risk of longer stays in hospitals, higher total charges, and worse outcomes. Hospitals can employ a digital predictive model to help identify those high-risk patients and proactively provide the best care for them. Also, healthcare providers can target this group with educational interventions by employing the
103 Knowledge, Attitude, Preventive Practice, and Outcomes (KAP-O) framework to improve their health status, prevent severe adverse outcomes, and reduce their level of healthcare utilization
(Marathe, Wan, & Marathe, 2016; Wan, 2014). Last, collaborations among different levels of providers: clinics, hospitals, long-term, and post-acute care organizations might also facilitate monitoring wearable devices among high-risk DM patients to prevent HG episodes (Clarka ,
Elswickb, Gabrielb, Gurupur, & Wisniewski, 2016).
5.6 Limitations
Although this study is robust and provided rich knowledge about different HG risk factors and their impact on utilization and outcome, there were certain limitations. First, the study used secondary data, which is on one hand the largest national database available. On the other hand, the nature of secondary data usually limits the number of variables to the ones that are already accessible in the database. So, some variables, such as patients’ educational attainment, the presence of a primary care physician, duration of DM, history of previous HG, medications used, and whether the patient was transferred to ICU or not, would be important when examining HG risk factors, but they were not recorded in the database. Also, some variables that were already included in the database need further clarification to be useful. For instance, the data did not differentiate between those who were admitted with HG from those who developed HG after admission. Another issue is that because the database included only inpatient data, only short-term in-hospital complications were captured. Thus, long-term complications, such as readmissions and mortality after discharge could not be captured.
Moreover, other HG patients who sought medical assistance through clinics or other avenues also were not captured.
104 Last, there were some assumptions related to the data used in this study. The first assumption was that any diagnosis that was not recorded in the patients’ ICD codes was considered to be absent. However, because of the nature of the ICD coding and its sole purpose for medical billing, there were some occasions in which some problems happened to patients and might have been recorded in the doctors’ notes, but they were not recorded under the patients’
ICD codes because the hospital would not have been reimbursed for them. These missing pieces of information lead to a second assumption about the accuracy of the data: since there was no way to check the accuracy of data entry in the database, the study assumed that all data are accurate. The last assumption relates to some variables. For example, the study assumed all patients had one source of insurance coverage, as the data recorded only one type of insurance provider for each patient. However, there are some patients with dual insurance, such as
Medicaid and Medicare. Unfortunately, information about patients with dual insurance could not be attained, so the assumption made for this study was that all patients had only one single insurance provider.
5.7 Recommendations for Future Research
It is recommended that future studies be designed in a prospective format, which could allow for the inclusion of some other key variables. Therefore, it is suggested that variables such as the duration of DM (Amiel, Dixon, Mann, & Jameson, 2008; Karges et al., 2015), type of glucose-lowering medications (Chan & Colagiuri, 2015; Handelsman, Bode, Endahl, Mersebach,
& King, 2015; Heller, Frier, Herslov, Gundgaard, & Gough, 2015; Monami et al., 2014;
Thompson, 2015), type of antibiotics (Dzygalo, Kowalska, & Szypowska, 2015; Kendall &
105 Wooltorton, 2006) and transfer to ICU (Gangji, Gerstein, Goldsmith, & Clase, 2007; Hermanides et al., 2010) be included in further studies that would explore their impact on HG U and O.
In addition, it is recommended that further research be conducted concerning the timing and frequency of HG episodes among admitted patients as well as separating those who come to
ED with HG from others who develop in-hospital HG. Another area that needs further exploration is related to the pattern of HG episodes across the month. HG events usually occur when there are disturbances in patients’ diet or their intake, and end-of-month exhaustion of food budgets among low SES patients contribute to the increase of HG occurrence during that time of the month (Seligman, Bolger, Guzman, López, & Bibbins-Domingo, 2014).
Another area of possible future research would be to explore the long-term HG effects such as readmissions (Majumdar et al., 2013), history of HG, and annual frequency of HG
(Hirsch et al., 2012; Kaur, Markley, Schlauch, & Izuora, 2015), as well as other complications that are treated in outpatient settings. Last, HG can occur among diseases other than diabetes. So, it might also be important to conduct studies that examine HG among non-diabetic patients and their risk factors, utilization, and outcomes to compare them to patients with diabetes.
5.8 Conclusions
To the best of our knowledge, the present study is the first study aimed at examining the multilevel character of a variety of factors that contribute to HG and their effects on utilization and outcome while employing Andersen’s BM. This study’s use of the largest national inpatient database combined with the robust analytical methods used provided rich information about the matter. On one hand, SEM helped in testing multiple hypotheses developed for the study as well as exploring the direct and indirect impact of different factors on utilization and outcome. On the
106 other hand, by conducting DTREG analysis, the study further explored the probability of the various predictor variables on length of stay, total charge, and outcome, which provided a better picture about HG risk factors and their individual impacts on each target variable.
Thus, especially in light of the current shift toward VBP, these results provide practical applications for healthcare providers as they could be used as predictors of HG among diabetic patients. Based on the study findings, healthcare providers can more easily proactively deal with the high risk groups who are elderly patients with comorbidities, who have diabetes-related complications, and who present with renal disease. Targeting those groups could help reduce the frequency and severity of HG episodes. Consequently, better results could be achieved, such as shorter hospital stays, less cost, and better health outcomes.
The study results also have great policy implications for decision makers regarding how to approach the growing problem of DM in the area of Central Florida. One major finding is the pronounced impact of age on all different target variables in this study: length of stay, total charges, and outcome. With the increasing number of older people and diabetes in Central
Florida, policy makers need to integrate innovative policies with collaboration among different stakeholders and experts to enhance patients’ levels of self-care and adherence to recommended care. Another finding is related to the geographical locations of the DM patients. In Florida, the percentage of diabetes is higher in the rural areas of the state. Thus, in-depth assessments of these areas regarding access to care, level of healthcare provided, and barriers to care among DM patients residing there is critical. These assessments could help reduce the unpleasant medical, social, and economic consequences of HG among those groups.
107 APPENDIX A: PREVIOUS STUDIES RELATED TO HG RISK FACTORS, UTILIZATION, AND OUTCOMES
108 Table 19. Previous studies related to HG risk factors, utilization, and outcomes Authors, year Diabetes types Factors examined Utilization Outcomes
Studies on HG risk factors among primary care-related HG
(Sämann et al., 2013) T1DM, T2DM incidence No utilization No outcomes
(Kostev et al., 2014) T2DM insulin-treated No utilization No outcomes
(Fiallo-Scharer et al., 2011) T1DM CGM No utilization No outcomes
(Bordier et al., 2015) T2DM elderly 70+ No utilization No outcomes
(Cox et al., 2007) T1DM, T2DM CGM No utilization No outcomes
(Bruderer et al., 2014) T2DM incidence No utilization No outcomes
(Giorda et al., 2015) T1DM incidence No utilization No outcomes
(Sonoda et al., 2015) T2DM insulin-treated No utilization No outcomes
(Weinstock et al., 2015) T1DM insulin-treated among No utilization No outcomes elderly (Kostev et al., 2015) T2DM insulin-treated, No utilization No outcomes compare regimen (Edridge et al., 2015) T2DM incidence No utilization No outcomes
(Ginde et al., 2008) T1DM, T2DM incidence No utilization No outcomes
(UKHSG, 2007) T1DM, T2DM HG agent No utilization No outcomes
(Gold et al., 1997) T1DM Insulin-dependent No utilization No outcomes DM, fear, awareness
Studies on HG utilization among primary care-related HG
(Mier et al., 2012) T2DM no factors physician No outcomes visits, eye exam, ED (Simeone & Quilliam, 2012) T2DM no factors Ed/OPD visit No outcomes
(Gu et al., 2015) T2DM no factors hospitalization No outcomes
(Liatis et al., 2015) T1DM, T2DM no factors ED visits No outcomes
(Davis et al., 2010) T2DM no factors ED/ No outcomes Ambulance/ hospitalization s (Williams et al., 2012) T2DM no factors office visits, No outcomes ED, hospitalization s LoS and cost (Quilliam et al., 2011) T2DM no factors office visits, No outcomes ED,
109 hospitalization s (Majumdar et al., 2013) T2DM no factors hospitalization No outcomes s (Fayfman et al., 2016) T1DM, T2DM no factors ED, No outcomes hospitalization s (Quilliam et al., 2011) T2DM no factors hospitalization No outcomes
(Duran-Nah et al., 2008) T2DM no factors hospitalization No outcomes
Studies on HG outcomes among primary care-related HG
(Kong et al., 2014) T2DM no factors clinic FU death and cancer (Simon et al., 2015) T2DM no factors No utilization QoL, Fear
(Sako et al., 2015) T1DM, T2DM no factors hospitalization mortality
Studies on hospital-related HG
(Farrokhi et al., 2012) T2DM non-critically ill No utilization No outcomes diabetics (Schloot et al., 2016) T2DM sulfonylurea No utilization No outcomes
(Lin et al., 2010) T2DM Recurrent HG No utilization No outcomes
(Pathak et al., 2015) T2DM ED and inpatient No utilization No outcomes
(Elliott et al., 2012) T2DM Treatment and HG No utilization No outcomes
(D'Netto et al., 2015) T1DM, T2DM Recurrent HG No utilization No outcomes
(Abdelhafiz et al., 2012) NA non-diabetics No utilization No outcomes
(Gómez et al., 2015) T2DM descriptive statistics No utilization No outcomes
(Dendy et al., 2014) T1DM, T2DM hospitalized with DM No utilization No outcomes developed SH (Deusenberry et al., 2012) T2DM hospitalized with DM No utilization No outcomes developed HG (Vriesendorp et al., 2008) T2DM ICU No utilization septic shock, ventilator, etc. (Olveira et al., 2015) T2DM TPN patients insulin No utilization Mortality
(Brodovicz et al., 2013) T2DM Chronic LoS Mortality
(Chevalier et al., 2016) T1DM, T2DM no factors cost, LoS Mortality, fractures, falls (Curkendall et al., 2009) T1DM, T2DM no factors LoS Mortality
(Zapatero et al., 2014) T2DM age sex only LoS Mortality, readmission
110 (Turchin et al., 2009) T2DM descriptive stat LoS Mortality
Studies on drugs and HG
(Czech et al., 2015) T1DM, T2DM systematic review No utilization No outcomes and meta-analysis (Schopman et al., 2014) T2DM sulfonylurea, No utilization No outcomes systematic review and meta-analysis (Murad et al., 2009) T2DM systematic review No utilization death, HG, and meta-analysis infection, stroke, MI (Home et al., 2010) T2DM insulin-treated, ED/ No outcomes Systemic meta hospitalization (Geller et al., 2014) T2DM Insulin-treated DM ED/ No outcomes hospitalization (Solomon et al., 2013) T2DM insulin type ED/ No outcomes hospitalization (Cariou et al., 2015) T1DM, T2DM insulin-treated hospitalization trauma
111 APPENDIX B: PREVIOUS STUDIES THAT EMPLOYED THE BM
112 Tab le 20. Previous studies that employed the BM Authors, year Issue examined Utilization
General non-HG studies that employed BM
(Rhoades et al., 2014) Homeless mental health utilization OPD, hospitalization, counseling
(Wan & Yates, 1975) Dental services utilization
(Springer, 2015) factors mental health utilization OPD, hospitalization
(Manski et al., 2013) older American HC utilization OPD, Hospitalization, Surgery, Home health (Matthews, 1990) Predictors to access to home service (Ko, 2015) Mental help seeking behavior
(Akobirshoev, 2015) Parental insurance and health on child utilization (Clement, 2015) Psychological factors in older rural OPD, ED health care use (Scheetz, 2010) Utilization in old OPD, Hospitalization, Surgery, Home health (Wilkinson-Lee, 2008) Ethnicity and Utilization Testing and screening
(Mallya, 2006) Asthma insurance and utilization medication use, OPD, hospitalization, ED
(Blaskowitz, 2014) Adults with developmental Ed, hospitalization disability utilization (Bazargan et al., 1998) AA Utilization Ed, hospitalization
(Roberge, 2008) utilization Children with public ED insurance (Wolinsky et al., 1995) older adults utilization OPD, hospitalization
(Kilany, 2014) Mental illness utilization OPD
(Snih et al., 2006) HC utilization older Mexican LoS, OPD American
Diabetes-related studies that employed BM
(Patel et al., 2015) T2DM Medical adherence
(Yeboah-Korang et al., T2DM Racial differences Healthcare use 2011) (Ou et al., 2012) T2DM Prediction of HC utilization and outcomes. OPD, hospitalization, ED (Yamashita et al., 2012) T2DM Prediction of adherence and self-care
(Gucciardi et al., 2009) T2DM use of educational programs
(Hu et al., 2015) T1DM and T2DM Medical expenditure in DM
113 (Baumeister et al., 2015) T1DM and T2DM Eye care exam use
(Chandler & Monnat, T1DM and T2DM Racial differences 2015) (Gamble & Chang, T2DM OPD 2014) (DeLawnia et al., 2014) T1DM and T2DM Eye care exam use
(Shenolikar, 2006) T1DM and T2DM Racial differences related to medication adherence and utilization: hospitalization ED, cost (Jayawant, 2008) T2DM Medicaid patients hospitalization ED, cost
BM studies using SEM
(Chern et al., 2002) expenditure cost
(Stein et al., 2012) Vulnerable population OPD, hospitalization, ED
(Wan & Soifer, 1974) physician utilization physician utilization
(Stein et al., 2007) Vulnerable population HC use in OPD, hospitalization homeless women (Tan, 2009) Utilization of older Asians in the LoS, hospitalization, OPD USA (LaHousse, 2009) Factors utilization of screening mammography screening (Stein, Andersen, Utilization with cerebral palsy OPD, hospitalization Koegel, & Gelberg, adults 2000) (Brook et al., 2014) Utilization of mental services by OPD minority adults
114 APPENDIX C: FREQUENCY TABLE FOR THE CATEGORIZED VARIABLES
115 Table 21. Frequency Table for the Categorized Variables Variable Name Value Frequency Percent Valid Percent Cumulative Percent Sex 0 2,505 51.9 51.9 51.9
1 2,317 48.1 48.1 100.0 Total 4,822 100.0 100.0
AA_Hisp 0 3,282 68.1 68.1 68.1
1 1,540 31.9 31.9 100.0 Total 4,822 100.0 100.0
Dementia 0 4,445 92.2 92.2 92.2
1 377 7.8 7.8 100.0 Total 4,822 100.0 100.0 No_Depression 0 1,894 39.3 39.3 39.3
1 2,928 60.7 60.7 100.0 Total 4,822 100.0 100.0 HLS 0 525 10.9 10.9 10.9
1 1,002 20.8 20.8 31.7
2 3,295 68.3 68.3 100.0 Total 4,822 100.0 100.0 Urban_hosp 0 1,096 22.7 22.7 22.7
1 3,726 77.3 77.3 100.0 Total 4,822 100.0 100.0 Patient_Loc 1 187 3.9 3.9 3.9
2 441 9.1 9.1 13.0
3 433 9.0 9.0 22.0
4 1,033 21.4 21.4 43.4
5 2,728 56.6 56.6 100.0 Total 4,822 100.0 100.0 SES 1.00 1,580 32.8 32.8 32.8
2.00 1,418 29.4 29.4 62.2
3.00 1,034 21.4 21.4 83.6
4.00 790 16.4 16.4 100.0 Total 4,822 100.0 100.0 Medicaid 0 3,898 80.8 80.8 80.8
1 924 19.2 19.2 100.0 Total 4,822 100.0 100.0 CCI 0 1,209 25.1 25.1 25.1
1 894 18.5 18.5 43.6
2 835 17.3 17.3 60.9
3 609 12.6 12.6 73.6
4 445 9.2 9.2 82.8
5 297 6.2 6.2 88.9
6 179 3.7 3.7 92.7
7 126 2.6 2.6 95.3
8 113 2.3 2.3 97.6
9 45 .9 .9 98.5
116
10 30 .6 .6 99.2
11 15 .3 .3 99.5
12 11 .2 .2 99.7
13 5 .1 .1 99.8
14 4 .1 .1 99.9
15 4 .1 .1 100.0
16 1 .0 .0 100.0 Total 4,822 100.0 100.0 DM_comp 0 3,709 76.9 76.9 76.9
1 1,113 23.1 23.1 100.0 Total 4,822 100.0 100.0 UncontDM 0 3,684 76.4 76.4 76.4
1 1,138 23.6 23.6 100.0 Total 4,822 100.0 100.0 DMII 0 702 14.6 14.6 14.6
1 4,120 85.4 85.4 100.0 Total 4,822 100.0 100.0 Under_wt 0 4,346 90.1 90.1 90.1 1 476 9.9 9.9 100.0 Total 4,822 100.0 100.0 Hyperlipidemia 0 2,875 59.6 59.6 59.6
1 1,947 40.4 40.4 100.0 Total 4,822 100.0 100.0 Renal_Disease 0 3,442 71.4 71.4 71.4
1 1,380 28.6 28.6 100.0 Total 4,822 100.0 100.0 Liver_Dis 0 4,547 94.3 94.3 94.3
1 275 5.7 5.7 100.0 Total 4,822 100.0 100.0 Hypertension 0 1,143 23.7 23.7 23.7
1 3,679 76.3 76.3 100.0 Total 4,822 100.0 100.0 Cancer 0 4,443 92.1 92.1 92.1
1 379 7.9 7.9 100.0 Total 4,822 100.0 100.0 Outcome 0 2,524 52.3 52.3 52.3
1 332 6.9 6.9 59.2
2 1,894 39.3 39.3 98.5
3 72 1.5 1.5 100.0 Total 4,822 100.0 100.0 Year 2012 1,605 33.3 33.3 33.3
2013 1,609 33.4 33.4 66.7
2014 1,608 33.3 33.3 100.0
Total 4,822 100.0 100.0
117 APPENDIX D: HISTOGRAMS FOR THE VARIABLES
118
119
120
121
122
123 APPENDIX E: SPEARMAN'S RHO CORRELATIONS
124 Table 22. Spearman's rho Correlations for (P)
No_ Deme TOT Out Ye Age Sex AA_ Hisp Depre HLS CCI LoS ntia CHG come ar ssion
Age Correlation 1.00 Coefficient p-value
Sex Correlation 0.00 1.00 Coefficient p-value 0.87
AA Correlation 0.01 .069* 1.00 _ Coefficient * His p-value 0.70 0.00 p De Correlation .324** .029* -0.01 1.00 men Coefficient tia p-value 0.00 0.04 0.39
No_ Correlation .370** - .165** .112** 1.00 Dep Coefficient .069* ress * ion p-value 0.00 0.00 0.00 0.00
HL Correlation .385** .101* .070** .135** .263** 1.00 S Coefficient * p-value 0.00 0.00 0.00 0.00 0.00
CCI Correlation .341** - .082** .164** .228** .190** 1.00 * Coefficient .034 p-value 0.00 0.02 0.00 0.00 0.00 0.00
LoS Correlation .083** 0.01 -0.03 .061** -.061** 0.01 .202** 1.00 Coefficient p-value 0.00 0.54 0.07 0.00 0.00 0.53 0.00
TO Correlation .069** - 0.03 .035* 0.00 0.00 .205** .654** 1.00 T Coefficient .031* CH p-value 0.00 0.03 0.06 0.02 0.79 0.91 0.00 0.00 G Out Correlation .338** - 0.02 .124** .194** .133** .343** .165** .205** 1.00 com Coefficient .051* e * p-value 0.00 0.00 0.17 0.00 0.00 0.00 0.00 0.00 0.00 YE Correlation 0.00 0.01 0.00 0.00 -0.02 0.00 0.01 -0.01 .034* 0.00 1.0 AR Coefficient 0 p-value 0.95 0.58 0.82 0.78 0.22 0.84 0.61 0.60 0.02 0.99 **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
125 Table 23. Spearman's rho Correlations for (E)
Urban_ Patient TOT YE SES Medicaid CCI LoS Outcome hosp _Loc CHG AR
Urban_ Correlation 1.00 hosp Coefficient p-value
Patient_ Correlation .186** 1.00 Loc Coefficient p-value 0.00
SES Correlation .057** .211** 1.00 Coefficient p-value 0.00 0.00
Medi Correlation .050** .042** - 1.00 caid Coefficient .127** p-value 0.00 0.00 0.00
CCI Correlation .035* .046** 0.01 -.121** 1.00 Coefficient p-value 0.02 0.00 0.51 0.00
LoS Correlation .032* 0.00 - -.031* .202** 1.00 ** Coefficient .050 p-value 0.02 0.89 0.00 0.03 0.00
TOT Correlation .092** .175** .033* -0.02 .205** .654** 1.00 CHG Coefficient p-value 0.00 0.00 0.02 0.28 0.00 0.00
Out Correlation 0.02 .039** .029* -.108** .343** .165** .205** 1.00 come Coefficient p-value 0.29 0.01 0.04 0.00 0.00 0.00 0.00
YEAR Correlation .197** .031* -0.02 .031* 0.01 -0.01 .034* 0.00 1.00 Coefficient p-value 0.00 0.03 0.22 0.03 0.61 0.60 0.02 0.99
**. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
126 Table 24. Spearman's rho Correlations for (N) Renal Y Hyper- DM_ Uncon _ Hyper- TOT Out- E CCI DMII lipi LoS comp t DM Diseas tension CHG come A demia e R
CCI Correlation 1.00 Coefficient p-value
DM_ Correlation .559 1.00 comp Coefficient ** p-value 0.00
Uncont Correlation .256 .116** 1.00 DM Coefficient ** p-value 0.00 0.00
DMII Correlation .065 -.078** -.176** 1.00 Coefficient ** p-value 0.00 0.00 0.00
Hyper- Correlation .071 .031* 0.00 .137** 1.00 lipidemia Coefficient ** p-value 0.00 0.03 0.81 0.00
Renal_ Correlation .613 .225** 0.02 .112** .087** 1.00 Disease Coefficient ** p-value 0.00 0.00 0.22 0.00 0.00
Hyper- Correlation .203 .080** -0.01 .261** .459** .249** 1.00 tension Coefficient ** p-value 0.00 0.00 0.51 0.00 0.00 0.00
LoS Correlation .202 .086** .114** 0.02 -.051** .077** 0.02 1.00 ** Coefficient p-value 0.00 0.00 0.00 0.14 0.00 0.00 0.28
TOT Correlation .205 .105** .080** 0.02 -0.02 .098** .034* .654** 1.00 CHG Coefficient ** p-value 0.00 0.00 0.00 0.22 0.26 0.00 0.02 0.00
Out Correlation .343 .056** 0.00 .136** .066** .218** .117** .165** .205** 1.00 come Coefficient ** p-value 0.00 0.00 0.96 0.00 0.00 0.00 0.00 0.00 0.00 YEAR Correlation 0.01 .029* -0.02 0.00 0.02 0.01 0.01 -0.01 .034* 0.00 1. Coefficient 00 p-value 0.61 0.05 0.27 0.90 0.11 0.66 0.72 0.60 0.02 0.99 **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
127 Table 25. Spearman's rho Correlations for LoS, Total Charge, and Outcome
CCI LoS TOTCHG Outcome YEAR
CCI Correlation Coefficient 1.00
p-value
LoS Correlation Coefficient .202** 1.00
p-value 0.00
TOTCHG Correlation Coefficient .205** .654** 1.00
p-value 0.00 0.00
Outcome Correlation Coefficient .343** .165** .205** 1.00
p-value 0.00 0.00 0.00 YEAR Correlation Coefficient 0.01 -0.01 .034* 0.00 1.00
p-value 0.61 0.60 0.02 0.99
**. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
128 APPENDIX F: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR LoS
129
Variable Importance ------AGE 100.00 HLS 18.6 No_Depression 12.4 Dementia 9.5
Figure 20. Relative Importance of the Variables in (P)
Variable Importance ------SES 100.000 Patient_Loc 72.124 Urban_hosp 63.397
Figure 21. Relative Importance of the Variables in (E) Variable Importance ------DM_comp 100.000 Renal_Disease 62.5 Hyperlipidemia 49.4 DMII 47.1 Hypertension 45.3
Figure 22. Relative Importance of the Variables in (N)
130 Variable Importance ------AGE 100.000 SES 27.959 DM_comp 13.252 Renal_Disease 10.975 Hypertension 10.559 Hyperlipidemia 10.552 DMII 9.536
Figure 23. Relative Importance of the Variables in the overall LoS model
131 APPENDIX G: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR TOTAL CHARGE
132
Variable Importance ------AGE 100.000 HLS 15.2 No_Depression 11.558 Dementia 7.1
Figure 24. Relative Importance of the Variables in (P)
Variable Importance ------Patient_Loc 100.000 SES 23.709 Urban_hosp 13.408
Figure 25. Relative Importance of the Variables in (E) Variable Importance ------Renal_Disease 100.000 DM_comp 49.319 DMII 27.577 Hyperlipidemia 19.890 Hypertension 18.629
Figure 26. Relative Importance of the Variables in (N)
133
Variable Importance ------AGE 100.000 Patient_Loc 40.982 SES 29.889 Renal_Disease 13.962 DM_comp 10.746 DMII 10.292
Figure 27. Relative Importance of the Variables in the overall Total Charge model
134 APPENDIX H: RELATIVE IMPORTANCE OF THE VARIABLES IN DTREG ANALYSIS FOR OUTCOME
135
Variable Importance ------AGE 100.000 HLS 15.802 No_Depression 11.219 Dementia 6.048
Figure 28. Relative Importance of the Variables in (P)
Variable Importance ------Patient_Loc 100.000 SES 30.571 Urban_hosp 17.680
Figure 29. Relative Importance of the Variables in (E) Variable Importance ------Hyperlipidemia 100.000 Hypertension 60.302 Renal_Disease 54.957 DM_comp 34.036 DMII 14.997
Figure 30. Relative Importance of the Variables in (N)
136 Variable Importance ------AGE 100.000 Renal_Disease 18.193 Patient_Loc 16.820 HLS 11.316 DM_comp 5.811
Figure 31. Relative Importance of the Variables in the overall Outcome model
137 APPENDIX I: INTERNATIONAL CLASSIFICATION OF DISEASES NINTH REVISION (ICD-9) DIAGNOSTIC CODES USED TO EXTRACT VARIABLES
138 Table 26. ICD-9 Codes Used to Extract Variables from Original HCUP NIS Data Alcohol abuse 29181 30303 30501 53530 7903 2915 30301 30393 4255 5712 9803 2911 29182 30390 30502 53531 9800 2918 30302 30500 5353 5713 9808 2912 29189 30391 30503 5710 9801 9809 V113 9468 9469 9463 9467 2913 30300 30392 3575 5711 9802 9446 9453 9461 9462 2914
Arrhythmia 42689 42731 42761 7850 V4501 42613 4271 42742 42789 V450 V533 4260 4269 42732 42769 99601 V4502 4267 4272 42760 4279 V4500 V5331 42610 4270 42741 42781 99604 V4509 V5339 V5332 42612
BMI Underweight V850
Cerebrovascular disease 43330 43411 43819 43881 43400 43301 4339 4373 43831 4333 43410 36234 43331 4349 43820 43882 43811 4331 43390 438 43832 43812 43850 430 4338 43490 43821 4389 43842 43310 43391 4380 43840 43813 43851 431 43380 43491 43822 43320 43401 43311 4340 43810 43841 43814 43852 4330 43381 436 43830 43321 4341 4332 43300
Congestive Heart Failure 40403 4254 4280 42823 42840 40291 40491 4258 42821 42832 42843 39891 40411 4255 4281 42830 42841 40401 40493 4259 42822 42833 4289 40201 40413 4257 42820 42831 42842 40211
Dementia 29012 29021 29041 2908 29410 29011 29020 29040 29043 2941 29420 2900 29013 2903 29042 2909 29411 29421 3312 29010
Depression 29625 29633 29641 29646 29654 29662 29689 96902 2967 9690 96904 29626 29634 29642 29650 29655 29663 2980 96903 29680 96900 96905 29630 29635 29643 29651 29656 29664 3004 30112 29682 96901 96909 29631 29636 29644 29652 29660 29665 29666 311 29645 29653 29661 29632 29640
DM I 25013 25033 25053 25073 25093 25011 25023 25043 25063 25083 25071 25001 25021 25041 25061 25081 25091 25012 25031 25051 25003
DM II 25022 25042 25062 25082 25080 25010 25032 25052 25072 25092 25040 25000 25030 25050 25070 25090 25060 25020 25002
DM with CVD
25071 25072 25073 25070
DM with Eye dis. 25051 25052 25053 36201 36202 36203 36204 36205 36206 36207 25050
DM with Neuropathy 25061 25062 25063 3572 25060
DM with other compl 25091 25092 25093 25090
DM with Renal Dis. 25041 25042 25043 25040
139 Drug Abuse 2920 29211 29212 2922 29281 30472 30473 30480 30481 30482 30483 2919 29283 29284 29285 29289 2929 30490 30491 30492 30493 30520 30521 29282 30401 30402 30403 30410 30411 30522 30523 30530 30531 30532 30533 30400 30413 30420 30421 30422 30423 30540 30541 30542 30543 30550 30551 30412 30431 30432 30433 30440 30441 30552 30553 30560 30561 30562 30563 30430 30443 30450 30451 30452 30453 30570 30571 30572 30573 30580 30581 30442 30461 30462 30463 30470 30471 30582 30583 30590 30591 30592 30593 30460 9454 9464 9465 9466 9467 3090 3091 9899 V6542 9468 9469 9445
HG 2512 2510
HIV 420 421 422 429 430 7953 79571 7958 V08 449 440 42
432 433 439 431
Hyperlipidemia 2724 2720
Hypertension 40210 40300 4039 40401 40411 40201 4030 40311 40400 40410 40490 4010 40211 40301 40390 40402 40412 40491 40493 40509 40519 40599 40591 4011 40290 4031 40391 40403 40413 40492 40501 40511 4040 4041 4049 4019 40291 40310 40200
Malignancy 1639 1954 20147 20420 1960 1451 1722 1990 20192 2059 V1091 1400 1640 1955 20148 20421 1961 1452 1723 1991 20193 20590 1613 1401 1641 1958 20150 20422 1962 1453 1724 1992 20194 20591 1618 1403 1642 1960 20151 2048 1963 1454 1725 20000 20195 20592 1619 1404 1643 1961 20152 20480 1965 1455 1726 20001 20196 2060 1620 1405 1648 1962 20153 20481 1966 1456 1727 20002 20197 20600 1622 1406 1649 1963 20154 20482 1968 1458 1728 20003 20198 20601 1623 1408 1650 1965 20155 2049 1969 1459 1729 20004 20200 20602 1624 1409 1658 1966 20156 20490 1970 1460 1740 20005 20201 2061 1625 1410 1659 1968 20157 20491 1971 1461 1741 20006 20202 20610 1628 1411 1700 1969 20158 20492 1972 1462 1742 20007 20203 20611 1629 1412 1701 1970 20160 2050 1973 1463 1743 20008 20204 20612 1630 1413 1702 1971 20161 20500 1974 1464 1744 20010 20205 2062 1631 1414 1703 1972 20162 20501 1975 1465 1745 20011 20206 20620 1638 1415 1704 1973 20163 20502 1976 1466 1746 20012 20207 20621 193 1416 1705 1974 20164 2051 1977 1467 1748 20013 20208 20622 1940 1418 1706 1975 20165 20510 1978 1468 1749 20014 20210 2068 1941 1419 1707 1976 20166 20511 1980 1469 1750 20015 20211 20680 1943 1420 1708 1977 20167 20512 1981 1470 1759 20016 20212 20681 1944 1421 1709 1978 20168 2052 1982 1471 1760 20017 20213 20682 1945 1422 1710 1980 20170 20520 1983 1472 1761 20018 20214 2069 1946 1428 1712 1981 20171 20521 1984 1473 1762 20020 20215 20690 1948 1429 1713 1982 20172 20522 1985 1478 1763 20021 20216 20691 1949 1430 1714 1983 20173 2053 1986 1479 1764 20022 20217 20692 1950 1431 1715 1984 20174 20530 1987 1480 1765 20023 20218 2070 1951 1438 1716 1985 20175 20531 19881 1481 1768 20024 20220 20700 1952 1439 1717 1986 20176 20532 19882 1482 1769 20025 20221 20701 1953 1440 1718 1987 20177 2058 19889 1483 179 20026 20222 20702 20145 1441 1719 19881 20178 20580 1990 1488 1800 20027 20223 2071 20146 1448 1720 19882 20190 20581 1991 1489 1801 20028 20224 20710 20262 1449 1721 19889 20191 20582 1992 1490 1808 20030 20225 20711 20263 1450 1886 20072 20267 V1000 V1072 1491 1809 20031 20226 20712 20264 1541 1887 20073 20268 V1001 V1079 1498 181 20032 20227 2072 20265 1542
140 1888 20074 20270 V1002 V1081 1499 1820 20033 20228 20720 20266 1543 1889 20075 20271 V1003 V1082 1500 1821 20034 20230 20721 2383 1548 1890 20076 20272 V1004 V1083 1501 1828 20035 20231 20722 2384 1550 1891 20077 20273 V1005 V1084 1502 1830 20036 20232 2078 2385 1551 1892 20078 20274 V1006 V1085 1503 1832 20037 20233 20780 2386 1552 1893 20080 20275 V1007 V1086 1504 1833 20038 20234 20781 3573 1560 1894 20081 20276 V1009 V1087 1505 1834 20040 20235 20782 20066 1561 1898 20082 20277 V1011 V1088 1508 1835 20041 20236 2080 20067 1562 1899 20083 20278 V1012 V1089 1509 1838 20042 20237 20800 20068 1568 1900 20084 20280 V1020 V109 1510 1839 20043 20238 20801 20070 1569 1901 20085 20281 V1021 V1090 1511 1840 20044 20240 20802 20071 1570 1902 20086 20282 V1022 2038 1512 1841 20045 20241 2081 1881 1571 1903 20087 20283 V1029 20380 1513 1842 20046 20242 20810 1882 1572 1904 20088 20284 V103 20381 1514 1843 20047 20243 20811 1883 1573 1905 20100 20285 V1040 20382 1515 1844 20048 20244 20812 1884 1574 1906 20101 20286 V1041 2040 1516 1848 20050 20245 2082 1885 1578 1907 20102 20287 V1042 20400 1518 1849 20051 20246 20820 20140 1579 1908 20103 20288 V1043 20401 1519 185 20052 20247 20821 20141 1580 1909 20104 20290 V1044 20402 1520 1860 20053 20248 20822 20142 1588 1910 20105 20291 V1045 2041 1521 1869 20054 20250 2088 20143 1589 1911 20106 20292 V1046 20410 1522 1871 20055 20251 20880 20144 1590 1912 20107 20293 V1047 20411 1523 1872 20056 20252 20881 1536 1591 1913 20108 20294 V1048 20412 1528 1873 20057 20253 20882 1537 1598 1914 20110 20295 V1049 2042 1529 1874 20058 20254 2089 1538 1599 1915 20111 20296 V1050 20123 1530 1875 20060 20255 20890 1539 1600 1916 20112 20297 V1051 20124 1531 1876 20061 20256 20891 1540 1601 1917 20113 20298 V1052 20125 1532 1877 20062 20257 20892 2031 1602 1918 20114 2030 V1053 20126 1533 1878 20063 20258 2380 20310 1603 1919 20115 20300 V1059 20127 1534 1879 20064 20260 2381 20311 1604 1920 20116 20301 V1060 20128 1535 1880 20065 20261 2382 20312 1605 1921 20117 20302 V1061 V1062 V1063 20121 1928 1611 1923 20120 1608 1922 20118 1610 V1071 V1069 20122 1929 1612 1609
Moderate or severe liver disease 4561 45620 45621 5722 4560 5724 5728 570 5715 5723
Myocardial infarction 41012 41031 41050 4107 4108 4101 41022 41041 41060 41062 4105 4100 4102 41032 41051 41070 41080 41010 4103 41042 41061 41091 41082 41000 41020 4104 41052 41071 41081 41011 41030 4106 41072 4109 41092 41001 41021 41040 41090 412 41002
Patient fall E8843 E8844 E8845 E888 E8880 E8881 E8888 E8889 E8842
Peripheral vascular disease 43814 43840 4402 44031 43810 43822 43851 44023 4409 44381 44322 930 43819 43841 44020 44032 43811 43830 43852 44024 4410 44382 44323 4373 43820 43842 44021 4404 43812 43831 4400 44029 44100 44389 44324 438 43821 43850 44022 4408 43813 43832 4401 44030 44101 4439 44329 4380 4412 44102 4417 4414 4411 4431 5579 5571 V434 4471 44321 4415 4413 44103 4419 4416
Renal disease 40402 40501 5836 5881 40310 40411 5824 5852 5889 40391 40490 4030 40403 5820 5837 5888 40311 40412 58281 5853 V420 4040 40491 40300 4041 5821 585 58881 4039 40413 58289 5854 V451 40400 40492 40301 40410 5822 5851 58889 40390 4049 5829 5855 V4511 40401 40493 4031 586 V561 5831 5859 V560 V562 5830 5856 V4512 5834 5880 5832
141
40493 V4512 V560 5859 40301 5852 58889 V5631 5881 40402 5855 V5631 585 5888 V561 586 40311 5853 5889 V5632 40412 40403 5856 V5632 5851 58881 V562 5880 40391 5854 V420 V568 40413 V4511 40492 V568 V451
Rheumatic disease 7100 7101 7102 7103 7104 7140 7141 7142 71481 71489 725 4465 seizures 34570 34571 3458 34580 78031 78032 78033 78039 34590 34591 7803 3457 3459 34581
Septicemia
3812 3840 3844 78552 3810 382 3842 388 99591 380 3819 3841 3849 7907 3811 383 3843 389 99592 381
Stroke 43311 4338 43401 V1254 43300 43321 4339 43411 436 43310 43331 430 4332 43380 4341 V171 43301 4333 43390 4349 43491 43490 4340 431 43320 43381 43410 43400 4331 43330 43391 4330 TOBACCO USE 30510 30511 30512 30513 3051 98984
Uncont DM 25020 25032 25062 25092 25010 25022 25042 25072 25012 25030 25052 25002 25021 25033 25063 25093 25011 25023 25043 25073 25013 25031 25053 25003 25083 25082
142 APPENDIX J: ORANGE, OSCEOLA, AND SEMINOLE COUNTY STATISTICS
143 Table 27. Orange, Osceola, and Seminole County Statistics Table
Osceola Orange Seminole Population % of Total Population % of Total Population % of Total Total Population, 2017 estimates 268,685 100% 1,145,956 100% 422,718 100% White 190,641 70% 728,795 63% 330,664 78% Hispanic or Latino 122,146 45% 308,244 26% 72,457 17% Black or African American 30,369 11% 238,241 20% 47,107 11% Some Other Race 27,623 10% 77,216 6% 15,692 3% Two or More Races 10,900 4% 56,581 4% 15,421 3% Asian 7,406 2% 39,325 3% 12,190 2% American Indian 1,452 Below 1% 4,532 Below 1% 1,386 Below 1% Three or more races 817 Below 1% 2,852 Below 1% 939 Below 1% Native Hawaiian Pacific Islander 294 Below 1% 1,266 Below 1% 258 Below 1%
Population growth estimates April 1, 2015 total population estimates 308,327 1,252,396 442,903 2025 total population estimates 461,900 1,669,000 536,800 Total population increase 2015-2025 153,573 50% 416,604 33% 93,897 21% Hispanic population 2010-2014 growth 32,041 26.2% 57,178 18.5% 12,610 17.4%
Gender statistics 276 Below 1% Total Population, 2016 estimates 265,423 1,112,252 419,203 Male Population: 129,442 48.8% 546,392 49.1% 202,981 48.4% Female Population: 135,981 51.2% 565,860 50.9% 216,222 51.6%
Age-related statistics 2015 65 and older 38,251 12.50% 136,704 11% 61,981 14% 2025 65 and older 66,391 15.50% 215,629 14% 90,084 18%
2015 Non-Hispanic White 65 and older 19,930 17.6% 79,018 14.5% 48,238 16.8%
2025 Non-Hispanic White 65 and older 28,778 23.2% 109,729 19.4% 67,952 22.5% 2015 Non-Hispanic Black 65 and older 3,312 10.5% 21,376 8.1% 4,828 9.5%
2025 Non-Hispanic Black 65 and older 6,832 14.2% 38,894 11.0% 7,574 12.0% 2015 Hispanic 65 and older 13,892 9.1% 29,712 8.1% 7,111 8.5% 2025 Hispanic 65 and older 28,426 11.9% 55,636 10.5% 11,596 10.4%
Other social statistics Veterans, 2011-2015 16,150 19.8% 63,185 20.0% 29,945 12.0% Persons with a disability, under age 65 years, 2011-2015 10.3% 7.2% 6.8% Persons without health insurance, under age 65 years 17.6% 15.7% 11.9% Median household income (in 2015 dollars), 2011-2015 44,254 47,943 57,010 Persons in poverty, percent 18.5% 15.6% 11.5%
144 APPENDIX K: UCF IRB APPROVAL LETTER
145
146 REFERENCES
Abdelhafiz, A. H., Bailey, C., & Sinclair, A. (2012). Hypoglycaemia in hospitalized non-diabetic older people. Review. European Geriatric Medicine Journal, 3(3), 174-178. doi:http://dx.doi.org/10.1016/j.eurger.2012.02.008
ADA. (2005). Defining and reporting hypoglycemia in diabetes: a report from the American Diabetes Association Workgroup on Hypoglycemia. Diabetes Care Journal(0149-5992 (Print)).
ADA. (2009). Diagnosis and Classification of Diabetes Mellitus. Diabetes Care Journal, 32(Suppl 1), S62-S67. doi:10.2337/dc09-S062
ADA. (2013). Economic costs of diabetes in the U.S. in 2012. Diabetes Care Journal, 36(4), 1033-1046. doi:10.2337/dc12-2625
ADA. (2015). Hypoglycemia (Low Blood Glucose). Retrieved from http://www.diabetes.org/living-with-diabetes/treatment-and-care/blood-glucose- control/hypoglycemia-low-blood.html
ADA. (2016). The Burden of Diabetes in Florida. Retrieved from American Diabetes Association: http://main.diabetes.org/dorg/PDFs/Advocacy/burden-of- diabetes/florida.pdf
Aday, L. A., & Andersen, R. (1974). A framework for the study of access to medical care. Health Services Research Journal, 9(3), 208-220.
Adelmann, P. K. (2003). Mental and substance use disorders among Medicaid recipients: prevalence estimates from two national surveys. Adm Policy Ment Health, 31(2), 111- 129.
Adelmann, P. K. (2003). Mental and substance use disorders among Medicaid recipients: prevalence estimates from two national surveys. Administration and Policy in Mental Health Journal, 31(2), 111-129.
Agardh, E., Allebeck, P., Hallqvist, J., Moradi, T., & Sidorchuk, A. (2011). Type 2 diabetes incidence and socio-economic position: a systematic review and meta-analysis. The International Journal of Epidemiology, 40(3), 804-818. doi:10.1093/ije/dyr029
Ahn, J. (2002). Beyond Single Equation Regression Analysis: Path Analysis and Multi-Stage Regression Analysis. American Journal of Pharmaceutical Education, 66(1). doi:aj660105.pdf
Ahrén, B. (2013). Avoiding hypoglycemia: a key to success for glucose-lowering therapy in type 2 diabetes. Vascular Health and Risk Management Journal, 9, 155-163. doi:10.2147/VHRM.S33934
147 AHRQ.gov. (2016, December 2016). Overview of the National (Nationwide) Inpatient Sample (NIS). HCUP Databases. Retrieved from https://www.hcup-us.ahrq.gov/nisoverview.jsp
Akobirshoev, I. (2015). The Impact of Parental Health Insurance on Health Outcomes and Health Care Utilization For Insured Children. (3723584 Ph.D.), Brandeis University, The Heller School for Social Policy and Management, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Akram, K., Pedersen-Bjergaard, U., Carstensen, B., Borch-Johnsen, K., & Thorsteinsson, B. (2006). Frequency and risk factors of severe hypoglycaemia in insulin-treated Type 2 diabetes: a cross-sectional survey. Diabetic Medicine Journal, 23(7), 750-756. doi:10.1111/j.1464-5491.2006.01880.x
Albright, J. J., & Park, H. M. (2009). Confirmatory factor analysis using Amos, LISREL, Mplus, and SAS/STAT CALIS. Working Paper. Indiana University. The University Information Technology Services (UITS) Center for Statistical and Mathematical Computing, Indiana University. Retrieved from www.iu.edu/~statmath/stat/all/cfa/cfa2008.pdf alz.org. (2017). What Is Dementia? Retrieved from http://www.alz.org/what-is-dementia.asp
Amiel, S. A., Aschner, P., Childs, B., Cryer, P. E., de Galan, B., Heller, S. R., . . . Grp, I. H. S. (2015). Minimizing Hypoglycemia in Diabetes. Diabetes Care Journal, 38(8), 1583- 1591. doi:10.2337/dc15-0279
Amiel, S. A., Dixon, T., Mann, R., & Jameson, K. (2008). Hypoglycaemia in Type 2 diabetes. Diabetic Medicine Journal, 25(3), 245-254. doi:10.1111/j.1464-5491.2007.02341.x
Andersen, R. (1968). A behavioral model of families' use of health services. [Chicago]: Center for Health Administration Studies, University of Chicago.
Andersen, R. (1995). Revisiting the behavioral model and access to medical care: does it matter? J Health Soc Behav, 36(1), 1-10.
Andersen, R., & Aday, L. A. (1978). Access to medical care in the U.S.: realized and potential. Medical Care Journal, 16(7), 533-546.
Andersen, R., & Newman, J. F. (1973). Societal and individual determinants of medical care utilization in the United States. The Milbank Memorial Fund quarterly Health and society, 51(1), 95-124.
Andrus, M. R., Kelley, K. W., Murphey, L. M., & Herndon, K. C. (2004). A comparison of diabetes care in rural and urban medical clinics in Alabama. Journal of Community Health, 29(1), 29-44.
Babitsch, B., Gohl, D., & von Lengerke, T. (2012). Re-revisiting Andersen's Behavioral Model of Health Services Use: a systematic review of studies from 1998-2011. Psychosocial Medicine Journal, 9, Doc11. doi:10.3205/psm000089
148 Babitsch, B., Gohl, D., & von Lengerke, T. (2012). Re-revisiting Andersen’s Behavioral Model of Health Services Use: a systematic review of studies from 1998–2011. GMS Psycho- Social-Medicine, 9, Doc11. doi:10.3205/psm000089
Bailey, J. E., Wan, J. Y., Brunt, D. L., & Somes, G. W. (2001). Improvements in diabetes management and outcomes in a Medicaid managed care system. Tenn Med(1088-6222 (Print)).
Baldwin, D., Zander, J., Munoz, C., Raghu, P., Delange-Hudec, S., Hong, L., . . . Molitch, M. (2012). A Randomized Trial of Two Weight-Based Doses of Insulin Glargine and Glulisine in Hospitalized Subjects With Type 2 Diabetes and Renal Insufficiency? Diabetes Care Journal, 35(10), 1970-1974. doi:10.2337/dc12-0578
Barendse, S., Singh, H., Frier, B. M., & Speight, J. (2012). The impact of hypoglycaemia on quality of life and related patient-reported outcomes in Type 2 diabetes: a narrative review. Diabet Medicine, 29(3), 293-302. doi:10.1111/j.1464-5491.2011.03416.x
Barnett, A. H., Cradock, S., Fisher, M., Hall, G., Hughes, E., & Middleton, A. (2010). Key considerations around the risks and consequences of hypoglycaemia in people with type 2 diabetes. Int J Clin Pract, 64(8), 1121-1129. doi:10.1111/j.1742-1241.2009.02332.x
Baumeister, S. E., Schomerus, G., Andersen, R. M., Tost, F., Markus, M. R., Volzke, H., & Jurgens, C. (2015). Trends of barriers to eye care among adults with diagnosed diabetes in Germany, 1997-2012. Nutr Metab Cardiovasc Dis, 25(10), 906-915. doi:10.1016/j.numecd.2015.07.003
Bazargan, M., Bazargan, S., & Baker, R. S. (1998). Emergency department utilization, hospital admissions, and physician visits among elderly African American persons. The Gerontologist, 38(1), 25-36.
Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin Journal, 107(2), 238-246.
Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin Journal, 88(3), 588-606. doi:10.1037/0033- 2909.88.3.588
Berkowitz, S. A., Aragon, K., Hines, J., Seligman, H., Lee, S., & Sarkar, U. (2013). Do clinical standards for diabetes care address excess risk for hypoglycemia in vulnerable patients? A systematic review. Health Serv Res, 48(4), 1299-1310. doi:10.1111/1475-6773.12048
Berkowitz, S. A., Karter, A. J., Lyles, C. R., Liu, J. Y., Schillinger, D., Adler, N. E., . . . Sarkar, U. (2014). Low Socioeconomic Status is Associated with Increased Risk for Hypoglycemia in Diabetes Patients: the Diabetes Study of Northern California (DISTANCE). Journal of Health Care for the Poor & Underserved, 25(2), 478-490 413p. doi:10.1353/hpu.2014.0106
149 Blaskowitz, M. G. (2014). Predictors of Emergency Room and Hospital Utilization Among Adults with Intellectual and Developmental Disabilities (IDD) in New York City: University at Albany. Department of Health Policy, Management and Behavior.
Bloomfield, H. E., Greer, N., Newman, D., MacDonald, R., Carlyle, M., Fitzgerald, P., . . . Wilt, T. J. (2012). Predictors and Consequences of Severe Hypoglycemia in Adults with Diabetes - A Systematic Review of the Evidence. Retrieved from Evidence-based Synthesis Program: https://login.ezproxy.net.ucf.edu/login?auth=shibb&url=http://search.ebscohost.com/logi n.aspx?direct=true&db=cmedm&AN=23256228&site=ehost-live
Blosch, A. K., Chernoff, A., & Nijjar, P. S. (2010). Can Hypoglycemia be Used as a Marker of Outcome in Hospitalized Non-Diabetics∗. Journal of Clinical Lipidology, 4(3), 228. doi:http://dx.doi.org/10.1016/j.jacl.2010.03.068
Bordier, L., Buysschaert, M., Bauduceau, B., Doucet, J., Verny, C., Lassmann Vague, V., & Le Floch, J. P. (2015). Predicting factors of hypoglycaemia in elderly type 2 diabetes patients: Contributions of the GERODIAB study. Diabetes & Metabolism, 41(4), 301- 303. doi:http://dx.doi.org/10.1016/j.diabet.2015.03.001
Braak, E. W., Valk, H. W., Bije, Y. F., Laak, M. F., Haeften, T. W., & Erkelens, D. W. (2000). Response to training in blood glucose awareness is related to absence of previous hypoglycemic coma. Diabetes Care Journal, 23(8), 1199-1200.
Brechner, R. J., Howie, L. J., Herman, W. H., Will, J. C., & Harris, M. I. (1993). Ophthalmic examination among adults with diagnosed diabetes mellitus. Jama(0098-7484 (Print)).
Brian, M. F., & Simon, R. H. (2012). 1. Epidemiology and Impact of Hypoglycemia on Patients with Diabetes Translational Endocrinology & Metabolism: Hypoglycemia in Diabetes Update (Vol. 3 (4), pp. 15-46): The Endocrine Society.
Briscoe, V. J., & Davis, S. N. (2006). Hypoglycemia in type 1 and type 2 diabetes: physiology, pathophysiology, and management. Clinical Diabetes(3), 115.
Brodovicz, K. G., Mehta, V., Zhang, Q., Zhao, C., Davies, M. J., Chen, J., . . . Engel, S. S. (2013). Association between hypoglycemia and inpatient mortality and length of hospital stay in hospitalized, insulin-treated patients. Current Medical Research & Opinion, 29(2), 101-107 107p. doi:10.1185/03007995.2012.754744
Brook, C. G. D., Clayton, P., & Brown, R. (2010). Brook's clinical pediatric endocrinology, 6th ed: Book News, Inc.
Brook, J. S., Lee, J. Y., Balka, E. B., Finch, S. J., & Brook, D. W. (2014). Utilization of Mental Health Services by Minority Urban Adults: Psychosocial Predictors. Journal of Urban Health, 91(4), 720-735. doi:http://dx.doi.org/10.1007/s11524-014-9878-3
150 Brown, E. R., Davidson, P. L., Yu, H., Wyn, R., Andersen, R. M., Becerra, L., & Razack, N. (2004). Effects of community factors on access to ambulatory care for lower-income adults in large urban communities. Inquiry, 41(1), 39-56.
Broyles, R. W., McAuley, W. J., & Baird-Holmes, D. (1999). The medically vulnerable: their health risks, health status, and use of physician care. Journal of Health Care Poor Underserved, 10(2), 186-200.
Bruce, D. G., Davis, W. A., Casey, G. P., Clarnette, R. M., Brown, S. G. A., Jacobs, I. G., . . . Davis, T. M. E. (2009). Severe hypoglycaemia and cognitive impairment in older patients with diabetes: the Fremantle Diabetes Study. Diabetologia, 52(9), 1808. doi:10.1007/s00125-009-1437-1
Bruderer, S. G., Bodmer, M., Jick, S. S., Bader, G., Schlienger, R. G., & Meier, C. R. (2014). Incidence of and risk factors for severe hypoglycaemia in treated type 2 diabetes mellitus patients in the UK--a nested case-control analysis. Diabetes, Obesity and Metabolism, 16(9), 801-811. doi:10.1111/dom.12282
Byrne, B. M. (2001). Structural equation modeling with AMOS: Basic concepts, applications, and programming. Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Caprio, S., Daniels, S. R., Drewnowski, A., Kaufman, F. R., Palinkas, L. A., Rosenbloom, A. L., & Schwimmer, J. B. (2008). Influence of Race, Ethnicity, and Culture on Childhood Obesity: Implications for Prevention and Treatment. A consensus statement of Shaping America's Health and the Obesity Society, 31(11), 2211-2221. doi:10.2337/dc08-9024
Cariou, B., Fontaine, P., Eschwege, E., Lievre, M., Gouet, D., Huet, D., . . . Charbonnel, B. (2015). Frequency and predictors of confirmed hypoglycaemia in type 1 and insulin- treated type 2 diabetes mellitus patients in a real-life setting: results from the DIALOG study. Diabetes Metabolism, 41(2), 116-125. doi:10.1016/j.diabet.2014.10.007
CDC. (2014). National Diabetes Statistics Report: Estimates of Diabetes and Its Burden in the United States.
Chan, S. P., & Colagiuri, S. (2015). Systematic review and meta-analysis of the efficacy and hypoglycemic safety of gliclazide versus other insulinotropic agents. Diabetes Research & Clinical Practice(1872-8227 (Electronic)).
Chandler, R. F., & Monnat, S. M. (2015). Racial/Ethnic Differences in Use of Health Care Services for Diabetes Management. Health Education & Behavior, 42(6), 783-792. doi:10.1177/1090198115579416
Charlson, M. E., Pompei, P., Ales, K. L., & MacKenzie, C. R. (1987). A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis, 40(5), 373-383.
151 Chen, A. W., Kazanjian, A., & Wong, H. (2008). Determinants of mental health consultations among recent Chinese immigrants in British Columbia, Canada: implications for mental health risk and access to services. The Journal of Immigrant and Minority Health, 10(6), 529-540. doi:10.1007/s10903-008-9143-5
Chen, Y.-J., Yang, C.-C., Huang, L.-C., Chen, L., & Hwu, C.-M. (2015). Increasing trend in emergency department visits for hypoglycemia from patients with type 2 diabetes mellitus in Taiwan. Primary Care Diabetes, 9(6), 490-496. doi:http://dx.doi.org/10.1016/j.pcd.2015.04.002
Chern, J. Y., Wan, T. T., & Begun, J. W. (2002). A structural equation modeling approach to examining the predictive power of determinants of individuals' health expenditures. The Journal of Medical Systems, 26(4), 323-336.
Chevalier, P., Vandebrouck, T., De Keyzer, D., Mertens, A., & Lamotte, M. (2016). Cost and co- morbidities associated with hypoglycemic inpatients in Belgium. J Med Econ, 19(1), 44- 52. doi:10.3111/13696998.2015.1086775
Clarka , S. E., Elswickb, S., Gabrielb, M., Gurupur, V., & Wisniewski, P. (2016). Transitions of Care: A Patient-Centered Perspective of Health Information Systems That Support Post- Acute Care. Journal of Integrated Design and Process Science, 20(1), 95-110. doi:10.3233/jid-2016-0008
Clement, L. (2015). The influence of psychological determinants on rural health care utilization among older Hispanic and non-Hispanic whites: The San Luis Valley Health and Aging Study. (1605082 M.A.), University of Colorado at Denver, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
CMS.gov. (2017). Hospital Value-Based Purchasing. Retrieved from https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment- Instruments/hospital-value-based-purchasing/index.html
Conwell, L. J., & Boult, C. (2008). The effects of complications and comorbidities on the quality of preventive diabetes care: a literature review. Population Health Management Journal, 11(4), 217-228. doi:10.1089/pop.2007.0017
Cox, D. J., Gonder-Frederick, L., Ritterband, L., Clarke, W., & Kovatchev, B. P. (2007). Prediction of Severe Hypoglycemia. Diabetes Care Journal, 30(6), 1370-1373. doi:10.2337/dc06-1386
Cryer, P. E. (2008). The Barrier of Hypoglycemia in Diabetes. Diabetes, 57(12), 3169-3176. doi:10.2337/db08-1084
Cryer, P. E., Davis, S. N., & Shamoon, H. (2003). Hypoglycemia in Diabetes. Diabetes Care Journal, 26(6), 1902-1912. doi:10.2337/diacare.26.6.1902
152 Cunningham, P., & Carrier, E. (2014). Trends in the financial burden of medical care for nonelderly adults with diabetes, 2001 to 2009. Am J Manag Care, 20(2), 135-142.
Curkendall, S. M., Natoli, J. L., Alexander, C. M., Nathanson, B. H., Haidar, T., & Dubois, R. W. (2009). Economic and clinical impact of inpatient diabetic hypoglycemia. Endocrine Practice, 15(4), 302-312. doi:10.4158/EP08343.OR
Czech, M., Rdzanek, E., Paweska, J., Adamowicz-Sidor, O., Niewada, M., & Jakubczyk, M. (2015). Drug-related risk of severe hypoglycaemia in observational studies: a systematic review and meta-analysis. BMC Endocr Disord, 15, 57. doi:10.1186/s12902-015-0052-z
D'Netto, M., Murphy, C. V., Mitchell, A., & Dungan, K. (2015). Predictors of recurrent hypoglycemia following a severe hypoglycemic event among hospitalized patients. Hospital Practice, 1-8. doi:10.1080/21548331.2016.1130584
Davis, T. M., Brown, S. G., Jacobs, I. G., Bulsara, M., Bruce, D. G., & Davis, W. A. (2010). Determinants of severe hypoglycemia complicating type 2 diabetes: the Fremantle diabetes study. The Journal of Clinical Endocrinology & Metabolism, 95(5), 2240-2247. doi:10.1210/jc.2009-2828 de Galan, B. E., Zoungas, S., Chalmers, J., Anderson, C., Dufouil, C., Pillai, A., . . . Woodward, M. (2009). Cognitive function and risks of cardiovascular disease and hypoglycaemia in patients with type 2 diabetes: the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) trial. Diabetologia, 52(11), 2328-2336. doi:10.1007/s00125-009-1484-7
DeLawnia, C.-H., Lu, N. L., & Zo, R. (2014). UTILIZATION OF DILATED EYE EXAMS AMONG ADULTS WITH DIABETES. European Scientific Journal.
Dendy, J. A., Chockalingam, V., Tirumalasetty, N. N., Dornelles, A., Blonde, L., Bolton, P. M., . . . Andrews, S. S. (2014). Identifying risk factors for severe hypoglycemia in hospitalized patients with diabetes. Endocrine Practice, 20(10), 1051-1056. doi:10.4158/EP13467.OR
Desai, M. M., Druss, B. G., & Perlin, J. B. (2002). Mental disorders and quality of diabetes care in the veterans health administration. Am J Psychiatry(0002-953X (Print)).
Deusenberry, C. M., Coley, K. C., Korytkowski, M. T., & Donihi, A. C. (2012). Hypoglycemia in Hospitalized Patients Treated with Sulfonylureas. Pharmacotherapy, 32(7), 613-617. doi:10.1002/j.1875-9114.2011.01088.x
Diabetes.org. (2014). Statistics About Diabetes. Retrieved from http://www.diabetes.org/diabetes-basics/statistics/?referrer=https://www.google.com/ dtreg.com. (2017). DTREG Software For Predictive Modeling and Forecasting. Retrieved from https://www.dtreg.com/methodology
153 Duran-Nah, J. J., Rodriguez-Morales, A., Smitheram, J., & Correa-Medina, C. (2008). Risk factors associated with symptomatic hypoglycemia in type 2 diabetes mellitus patients. Revista de Investigación Clínica Journal, 60(6), 451-458.
Dzygalo, K., Kowalska, A., & Szypowska, A. (2015). The beneficial effect of insulin degludec on nocturnal hypoglycaemia and insulin dose in type 1 diabetic patients: a systematic review and meta-analysis of randomised trials. Acta Diabetol(1432-5233 (Electronic)).
Edridge, C. L., Dunkley, A. J., Bodicoat, D. H., Rose, T. C., Gray, L. J., Davies, M. J., & Khunti, K. (2015). Prevalence and Incidence of Hypoglycaemia in 532,542 People with Type 2 Diabetes on Oral Therapies and Insulin: A Systematic Review and Meta-Analysis of Population Based Studies. PLoS ONE, 10(6), e0126427. doi:10.1371/journal.pone.0126427
Elliott, M. B., Schafers, S. J., McGill, J. B., & Tobin, G. S. (2012). Prediction and Prevention of Treatment-Related Inpatient Hypoglycemia. J Diabetes Sci Technol, 6(2), 302-309. doi:10.1177/193229681200600213
Farrokhi, F., Klindukhova, O., Chandra, P., Peng, L., Smiley, D., Newton, C., . . . Umpierrez, G. (2012). Risk factors for inpatient hypoglycemia during subcutaneous insulin therapy in non-critically ill patients with type 2 diabetes. The Journal of Diabetes Science and Technology, 6(5), 1022-1029.
Fayfman, M., Vellanki, P., Alexopoulos, A.-S., Buehler, L., Zhao, L., Smiley, D., . . . Umpierrez, G. E. (2016). Report on Racial Disparities in Hospitalized Patients with Hyperglycemia and Diabetes. The Journal of Clinical Endocrinology & Metabolism. doi:10.1210/jc.2015-3220
Feil, D. G., Rajan, M., Soroka, O., Tseng, C.-L., Miller, D. R., & Pogach, L. M. (2011). Risk of Hypoglycemia in Older Veterans with Dementia and Cognitive Impairment: Implications for Practice and Policy. J Am Geriatr Soc, 59(12), 2263-2272 2210p. doi:10.1111/j.1532- 5415.2011.03726.x
Feinkohl, I., Aung, P. P., Keller, M., Robertson, C. M., Morling, J. R., McLachlan, S., . . . Price, J. F. (2014). Severe Hypoglycemia and Cognitive Decline in Older People With Type 2 Diabetes: The Edinburgh Type 2 Diabetes Study. Diabetes Care Journal, 37(2), 507-515. doi:10.2337/dc13-1384
Fiallo-Scharer, R., Cheng, J., Beck, R. W., Buckingham, B. A., Chase, H. P., Kollman, C., . . . Wolpert, H. (2011). Factors predictive of severe hypoglycemia in type 1 diabetes: analysis from the Juvenile Diabetes Research Foundation continuous glucose monitoring randomized control trial dataset. Diabetes Care Journal, 34(3), 586-590. doi:10.2337/dc10-1111
154 Floridacharts.com. (2016). FLORIDA DEPARTMENT OF HEALTH Division of Public Health Statistics & Performance Management- Chronic Diseases. Retrieved from http://www.floridacharts.com/charts/ChronicDiseases/default.aspx
Floridahealth.gov. (2017). Florida Diabetes Report - 2017. Retrieved from http://www.floridahealth.gov/:
Floridahospital.com. (2013). Community Health Needs Assessment Retrieved from https://www.floridahospital.com/sites/default/files/pdf/chna_with_interactive_toc.pdf
Flory, J. H., Wiesenthal, A. C., Thaler, H. T., Koranteng, L., & Moryl, N. (2016). Methadone Use and the Risk of Hypoglycemia for Inpatients With Cancer Pain. Journal of Pain and Symptom Management, 51(1), 79-87.e71. doi:http://dx.doi.org/10.1016/j.jpainsymman.2015.08.003
Fournier, J. P., Azoulay, L., Yin, H., Montastruc, J. L., & Suissa, S. (2015). Tramadol use and the risk of hospitalization for hypoglycemia in patients with noncancer pain. JAMA Intern Med, 175(2), 186-193. doi:10.1001/jamainternmed.2014.6512
Gamble, T. L., & Chang, L. (2014). Healthcare Utilization and Diabetes Management Programs: Indiana 2006-2010. American Journal of Managed Care, 20(10), e461-468 461p.
Gangji, A. S., Gerstein, H. C., Goldsmith, C. H., & Clase, C. M. (2007). A systematic review and meta-analysis of hypoglycemia and cardiovascular events: a comparison of glyburide with other secretagogues and with insulin. Diabetes Care Journal(0149-5992 (Print)).
Garg, R., Hurwitz, S., Turchin, A., & Trivedi, A. (2013). Hypoglycemia, with or without insulin therapy, is associated with increased mortality among hospitalized patients. Diabetes Care Journal, 36(5), 1107-1110. doi:10.2337/dc12-1296
Gelberg, L., Andersen, R. M., & Leake, B. D. (2000). The Behavioral Model for Vulnerable Populations: application to medical care use and outcomes for homeless people. Health Serv Res, 34(6), 1273-1302.
Geller, A. I., Shehab, N., Lovegrove, M. C., Kegler, S. R., Weidenbach, K. N., Ryan, G. J., & Budnitz, D. S. (2014). National estimates of insulin-related hypoglycemia and errors leading to emergency department visits and hospitalizations. JAMA Intern Med, 174(5), 678-686. doi:10.1001/jamainternmed.2014.136
Ginde, A. A., Espinola, J. A., & Camargo, C. A., Jr. (2008). Trends and disparities in U.S. emergency department visits for hypoglycemia, 1993-2005. Diabetes Care Journal, 31(3), 511-513. doi:10.2337/dc07-1790
Giorda, C. B., Ozzello, A., Gentile, S., Aglialoro, A., Chiambretti, A., Baccetti, F., . . . Rossi, M. C. (2015). Incidence and risk factors for severe and symptomatic hypoglycemia in type 1 diabetes. Results of the HYPOS-1 study. Acta Diabetol, 52(5), 845-853. doi:10.1007/s00592-015-0713-4 155 Gochman, D. S. (1997). Handbook of health behavior research 1: Personal and social determinants. New York: Plenum Press.
Gold, A. E., Frier, B. M., MacLeod, K. M., & Deary, I. J. (1997). A structural equation model for predictors of severe hypoglycaemia in patients with insulin-dependent diabetes mellitus. Diabet Medicine, 14(4), 309-315. doi:10.1002/(sici)1096-9136(199704)14:4<309::aid- dia345>3.0.co;2-#
Gómez, R., Guijarro, R., Zapatero, A., Barba, R., Guijarro, A., Tinahones, F., & Bernal, R. (2015). The frequency and impact of hypoglycemia among hospitalized patients with diabetes: A population-based study. Journal of Diabetes and its Complications, 29(8), 1050-1055. doi:http://dx.doi.org/10.1016/j.jdiacomp.2015.07.018
Goto, A., Arah, O. A., Goto, M., Terauchi, Y., & Noda, M. (2013). Severe hypoglycaemia and cardiovascular disease: systematic review and meta-analysis with bias analysis. BMJ, 347, f4533. doi:10.1136/bmj.f4533
Graham, A., Hasking, P., Brooker, J., Clarke, D., & Meadows, G. (2017). Mental health service use among those with depression: an exploration using Andersen's Behavioral Model of Health Service Use. Journal of Affective Disorders, 208, 170-176. doi:http://dx.doi.org/10.1016/j.jad.2016.08.074
Greco, D., & Angileri, G. (2004). Drug-induced severe hypoglycaemia in Type 2 diabetic patients aged 80 years or older. Diabetes Nutr Metab, 17(1), 23-26.
Green, A. J., Fox, K. M., & Grandy, S. (2012). Self-reported hypoglycemia and impact on quality of life and depression among adults with type 2 diabetes mellitus. Diabetes Research & Clinical Practice, 96(3), 313-318. doi:10.1016/j.diabres.2012.01.002
Gu, W., Ren, Y., Ji, L., Hong, T., Mu, Y., Guo, L., . . . Yang, X. (2015). Non-linear associations of risk factors with mild hypoglycemia among Chinese patients with type 2 diabetes. Journal of Diabetes and its Complications. doi:10.1016/j.jdiacomp.2015.12.024
Gucciardi, E., DeMelo, M., Booth, G., Tomlinson, G., & Stewart, D. E. (2009). Individual and contextual factors associated with follow-up use of diabetes self-management education programmes: a multisite prospective analysis. Diabet Medicine, 26(5), 510-517. doi:10.1111/j.1464-5491.2009.02713.x
Guo, B., Perron, B., & Gillespie, D. (2009). A Systematic Review of Structural Equation Modelling in Social Work Research. The British Journal of Social Work, 39(8), 1556- 1574.
Handelsman, Y., Bode, B. W., Endahl, L. A., Mersebach, H., & King, A. B. (2015). Patients Achieving Good Glycemic COntrol (Hba1c <7%) Experience A Lower Rate Of Hypoglycemia With Insulin Degludec Than With Insulin Glargine: A Meta- Analysis Of Phse 3a Trials. Endocrine Practice(1934-2403 (Electronic)).
156 Hanefeld, M., Duetting, E., & Bramlage, P. (2013). Cardiac implications of hypoglycaemia in patients with diabetes - a systematic review. Cardiovasc Diabetol, 12, 135. doi:10.1186/1475-2840-12-135
Hauke, J., & Kossowski, T. (2011). Comparison of Values of Pearson's and Spearman's Correlation Coefficients on the Same Sets of Data. Quaestiones Geographicae Journal, 30(2), 87.
HCUP-us.ahrq.gov. (2016). Overview of HCUP. Retrieved from https://www.hcup- us.ahrq.gov/overview.jsp
Heller, S. R., Frier, B. M., Herslov, M. L., Gundgaard, J., & Gough, S. C. (2015). Severe hypoglycaemia in adults with insulin-treated diabetes: impact on healthcare resources. Diabet Medicine(1464-5491 (Electronic)).
Heller, S. R., Frier, B. M., Herslov, M. L., Gundgaard, J., & Gough, S. C. (2016). Severe hypoglycaemia in adults with insulin-treated diabetes: impact on healthcare resources. Diabet Medicine, 33(4), 471-477. doi:10.1111/dme.12844
Henkel, R. J., & Maryland, P. A. (2015). The Risks and Rewards of Value-Based Reimbursement. Frontiers of Health Services Management, 32(2), 3-16.
Hermanides, J., Vriesendorp, T. M., Dotsch, R., Rosendaal, F. R., Zandstra, D. F., Hoekstra, J. B. L., & DeVries, J. H. (2010). Hypoglycemia is associated with intensive care unit mortality. Crit Care Med(1530-0293 (Electronic)).
Hermanns, N., Kulzer, B., Krichbaum, M., Kubiak, T., & Haak, T. (2005). Affective and anxiety disorders in a German sample of diabetic patients: prevalence, comorbidity and risk factors. Diabet Medicine, 22(3), 293-300. doi:10.1111/j.1464-5491.2005.01414.x
Hirai, F. E., Moss, S. E., Klein, B. E., & Klein, R. (2007). Severe hypoglycemia and smoking in a long-term type 1 diabetic population: Wisconsin Epidemiologic Study of Diabetic Retinopathy. Diabetes Care Journal, 30(6), 1437-1441. doi:10.2337/dc06-2264
Hirsch, I. B., Courreges, J.-P., Dykiel, P., Franek, E., Hermansen, K., King, A., . . . Davies, M. (2012). Insulin degludec/insulin aspart administered once daily at any meal, with insulin aspart at other meals versus a standard basal-bolus regimen in patients with type 1 diabetes: a 26-week, phase 3, randomized, open-label, treat-to-target trial. Diabetes Care Journal(1935-5548 (Electronic)). doi:D - NLM: PMC3476910 EDAT- 2012/08/31 06:00 MHDA- 2013/04/18 06:00 CRDT- 2012/08/31 06:00 PHST- 2012/08/28 [aheadofprint] AID - dc11-2503 [pii] AID - 10.2337/dc11-2503 [doi] PST - ppublish
Hoelter, J. W. (1983). The Analysis of Covariance Structures: Goodness-of-Fit Indices. Sociological Methods & Research Journal, 11(3), 325.
157 Holstein, A., & Egberts, E. H. (2001). Lower incidence of severe hypoglycaemia in patients with type 2 diabetes treated with glimepiride versus glibenclamide. Diabetes Metab Res Rev(1520-7552 (Print)).
Holstein, A., Hammer, C., & Egberts, E. H. (2003). Characteristics and time course of severe glimepiride- versus glibenclamide-induced hypoglycaemia. Eur J Clin Pharmacol(0031- 6970 (Print)).
Home, P. D., Fritsche, A., Schinzel, S., & Massi-Benedetti, M. (2010). Meta-analysis of individual patient data to assess the risk of hypoglycaemia in people with type 2 diabetes using NPH insulin or insulin glargine. Diabetes, Obesity and Metabolism, 12(9), 772- 779. doi:10.1111/j.1463-1326.2010.01232.x
Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural Equation Modelling: Guidelines for Determining Model Fit. Electronic Journal of Business Research Methods, 6(1), 53- 59.
Hsu, P.-F., Sung, S.-H., Cheng, H.-M., Yeh, J.-S., Liu, W.-L., Chan, W.-L., . . . Chuang, S.-Y. (2013). Association of Clinical Symptomatic Hypoglycemia With Cardiovascular Events and Total Mortality in Type 2 Diabetes: A nationwide population-based study. Diabetes Care Journal, 36(4), 894-900 897p. doi:10.2337/dc12-0916
Hu, L.-t., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling A Multidisciplinary Journal, 6(1), 1-55. doi:10.1080/10705519909540118
Hu, R., Shi, L., Pierre, G., Zhu, J., & Lee, D. C. (2015). Diabetes and medical expenditures among non-institutionalized U.S. adults. Diabetes Research & Clinical Practice, 108(2), 223-234. doi:10.1016/j.diabres.2015.02.016
Huang, Y.-q., Gou, R., Diao, Y.-s., Yin, Q.-h., Fan, W.-x., Liang, Y.-p., . . . Liu, F. (2014). Charlson comorbidity index helps predict the risk of mortality for patients with type 2 diabetic nephropathy. Journal Of Zhejiang University. Science. B, 15(1), 58-66. doi:10.1631/jzus.B1300109
Jafari, B., & Britton, M. E. (2015). Hypoglycaemia in elderly patients with type 2 diabetes mellitus: a review of risk factors, consequences and prevention. Journal of Pharmacy Practice & Research, 45(4), 459.
Jakubczyk, M., & Rdzanek, E. (2015). Economic resources consumption structure in severe hypoglycemia episodes: a systematic review and meta-analysis. Expert Rev Pharmacoecon Outcomes Res, 15(5), 813-822 810p. doi:10.1586/14737167.2015.1076338
Jayawant, S. S. (2008). Effect of dosing regimens on medication use, healthcare resource utilization, and costs in Medicaid enrolled type 2 diabetes mellitus patients. (3300107
158 Ph.D.), The Ohio State University, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Johnston, S. S., Conner, C., Aagren, M., Ruiz, K., & Bouchard, J. (2012). Association between hypoglycaemic events and fall-related fractures in Medicare-covered patients with type 2 diabetes. Diabetes, Obesity and Metabolism, 14(7), 634-643. doi:10.1111/j.1463- 1326.2012.01583.x
Johnston, S. S., Conner, C., Aagren, M., Smith, D. M., Bouchard, J., & Brett, J. (2011). Evidence linking hypoglycemic events to an increased risk of acute cardiovascular events in patients with type 2 diabetes. Diabetes Care Journal, 34(5), 1164-1170. doi:10.2337/dc10-1915
Kachroo, S., Kawabata, H., Colilla, S., Shi, L., Zhao, Y., Mukherjee, J., . . . Fonseca, V. (2015). Association between hypoglycemia and fall-related events in type 2 diabetes mellitus: analysis of a U.S. commercial database. Journal of Managed Care Pharmacy, 21(3), 243- 253.
Karges, B., Rosenbauer, J., Holterhus, P. M., Beyer, P., Seithe, H., Vogel, C., . . . Initiative, D. P. V. (2015). Hospital admission for diabetic ketoacidosis or severe hypoglycemia in 31,330 young patients with type 1 diabetes. European Journal of Endocrinology, 173(3), 341- 350. doi:10.1530/EJE-15-0129
Katon, W. J., Young, B. A., Russo, J., Lin, E. H. B., Ciechanowski, P., Ludman, E. J., & Von Korff, M. R. (2013). Association of Depression With Increased Risk of Severe Hypoglycemic Episodes in Patients With Diabetes. Ann Fam Med, 11(3), 245-250. doi:10.1370/afm.1501
Kaur, G., Markley, B., Schlauch, K., & Izuora, K. E. (2015). Outcomes of Less Intensive Glycemic Target for a Subcutaneous Insulin Protocol in Hospitalized Patients. The American Journal of the Medical Sciences, 350(6), 442-446. doi:10.1097/MAJ.0000000000000577
Kendall, C., & Wooltorton, E. (2006). People with diabetes should avoid antibiotic gatifloxacin. CMAJ : Canadian Medical Association journal, 174(8), 1089-1090. doi:10.1503/cmaj.060356
Kennedy, R. L., Chapman, A. J., Nayar, R., Grant, P., & Morris, A. D. (2002). Accidents in patients with insulin-treated diabetes: increased risk of low-impact falls but not motor vehicle crashes--a prospective register-based study. Journal of Trauma(0022-5282 (Print)).
Khunti, K., Davies, M., Majeed, A., Thorsted, B. L., Wolden, M. L., & Paul, S. K. (2015). Hypoglycemia and risk of cardiovascular disease and all-cause mortality in insulin- treated people with type 1 and type 2 diabetes: a cohort study. Diabetes Care Journal, 38(2), 316-322. doi:10.2337/dc14-0920
159 Kidneyfund.org. (2016). Kidney failure/ESRD. Retrieved from http://www.kidneyfund.org/kidney-disease/kidney-failure/
Kilany, M. (2014). Patient-centered medical homes for patients with severe mental illness: Utilization and quality of care variation in rural and urban areas. (3622390 Ph.D.), The University of North Carolina at Chapel Hill, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Kilpatrick, C. R., Elliott, M. B., Pratt, E., Schafers, S. J., Blackburn, M. C., Heard, K., . . . Tobin, G. S. (2014). Prevention of inpatient hypoglycemia with a real-time informatics alert. J Hosp Med, 9(10), 621-626. doi:10.1002/jhm.2221
Kline, R. B. (2005). Principles and Practice of Structural Equation Modeling: Guilford Press.
Ko, J. (2015). Help-seeking Behavior among Working-age Adults with Suicidal Ideation: An integration of the Andersen Model of Health Care Utilization and the Three-stage Model of Mental Health Help-seeking. (3704587 Ph.D.), University of Maryland, Baltimore, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Kong, A. P., Yang, X., Luk, A., Ma, R. C., So, W. Y., Ozaki, R., . . . Chan, J. C. (2014). Severe hypoglycemia identifies vulnerable patients with type 2 diabetes at risk for premature death and all-site cancer: the Hong Kong diabetes registry. Diabetes Care Journal, 37(4), 1024-1031. doi:10.2337/dc13-2507
Kostev, K., Dippel, F.-W., & Rathmann, W. (2014). Predictors of hypoglycaemia in insulin- treated type 2 diabetes patients in primary care: A retrospective database analysis. Primary Care Diabetes Journal, 8(2), 127-131. doi:http://dx.doi.org/10.1016/j.pcd.2013.10.001
Kostev, K., Dippel, F. W., & Rathmann, W. (2014). Predictors of hypoglycaemia in insulin- treated type 2 diabetes patients in primary care: a retrospective database analysis. Prim Care Diabetes, 8(2), 127-131. doi:10.1016/j.pcd.2013.10.001
Kostev, K., Dippel, F. W., & Rathmann, W. (2015). Risk of hypoglycaemia in type 2 diabetes patients under different insulin regimens: a primary care database analysis. German Medical Science, 13, Doc01. doi:10.3205/000205
Krass, I., Schieback, P., & Dhippayom, T. (2015). Adherence to diabetes medication: a systematic review. Diabet Medicine, 32(6), 725-737. doi:10.1111/dme.12651
Kumari, M., Head, J., & Marmot, M. (2004). Prospective study of social and other risk factors for incidence of type 2 diabetes in the Whitehall II study. Arch Intern Med, 164(17), 1873-1880. doi:10.1001/archinte.164.17.1873
LaHousse, S. F. (2009). Factors associated with mammography screening utilization among Latinas: A revision of the behavioral model of health services use. (3364000 Ph.D.), 160 Michigan State University, Ann Arbor. Ethnic NewsWatch; ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Land, K. C. (1969). Principles of Path Analysis. Sociological Methodology, 3.
Launer, L. J., Miller, M. E., Williamson, J. D., Lazar, R. M., Gerstein, H. C., Murray, A. M., . . . Bryan, R. N. (2011). Effects of intensive glucose lowering on brain structure and function in people with type 2 diabetes (ACCORD MIND): a randomised open-label substudy. The Lancet Neurology, 10(11), 969-977. doi:http://dx.doi.org/10.1016/S1474- 4422(11)70188-0
Lee, Y.-L., Chen, B. K., Lin, K.-D., Su, R.-W., Lee, M.-Y., Hsiao, P.-J., & Shin, S.-J. (2015). The impact of severe hypoglycemia on renal impairment in type 2 diabetes. Diabetes Research and Clinical Practice, 108(3), 448-455. doi:http://dx.doi.org/10.1016/j.diabres.2015.02.028
Leese, G. P., Wang, J., Broomhall, J., Kelly, P., Marsden, A., Morrison, W., . . . Morris, A. D. (2003). Frequency of severe hypoglycemia requiring emergency treatment in type 1 and type 2 diabetes: a population-based study of health service resource use. Diabetes Care Journal, 26(4), 1176-1180.
Liatis, S., Mylona, M., Kalopita, S., Papazafiropoulou, A., Karamagkiolis, S., Melidonis, A., . . . Kokkinos, A. (2015). Hypoglycaemia requiring medical assistance in patients with diabetes: A prospective multicentre survey in tertiary hospitals. Diabetes & Metabolism, 41(2), 126-131. doi:http://dx.doi.org/10.1016/j.diabet.2014.10.006
Lin, Y. Y., Hsu, C. W., Sheu, W. H. H., Chu, S. J., Wu, C. P., & Tsai, S. H. (2010). Risk Factors for Recurrent Hypoglycemia in Hospitalized Diabetic Patients Admitted for Severe Hypoglycemia.
Liu, S., Zhao, Y., Hempe, J. M., Fonseca, V., & Shi, L. (2012). Economic burden of hypoglycemia in patients with Type 2 diabetes. Expert Rev Pharmacoecon Outcomes Res, 12(1), 47-51. doi:10.1586/erp.11.87
Lleras, C. (2005). Path Analysis: Elsevier, Inc.
Lopez, J. M., Annunziata, K., Bailey, R. A., Rupnow, M. F., & Morisky, D. E. (2014). Impact of hypoglycemia on patients with type 2 diabetes mellitus and their quality of life, work productivity, and medication adherence. Patient Prefer Adherence, 8, 683-692. doi:10.2147/PPA.S58813
Lopez, J. M. S., Bailey, R. A., & Rupnow, M. F. T. (2015). Demographic Disparities Among Medicare Beneficiaries with Type 2 Diabetes Mellitus in 2011: Diabetes Prevalence, Comorbidities, and Hypoglycemia Events. Population Health Management Journal, 18(4), 283-289. doi:10.1089/pop.2014.0115
161 Majumdar, S. R., Hemmelgarn, B. R., Lin, M., McBrien, K., Manns, B. J., & Tonelli, M. (2013). Hypoglycemia associated with hospitalization and adverse events in older people: population-based cohort study. Diabetes Care Journal, 36(11), 3585-3590. doi:10.2337/dc13-0523
Malabu, U. H., Vangaveti, V. N., & Kennedy, R. L. (2014). Disease burden evaluation of fall- related events in the elderly due to hypoglycemia and other diabetic complications: a clinical review. Journal of Clinical Epidemiology, 6, 287-294. doi:10.2147/CLEP.S66821
Mallya, U. G. (2006). National perspective on the effect of health insurance on medication and resource utilization among adults with asthma. (3238028 Ph.D.), University of Michigan, Ann Arbor. ABI/INFORM Complete; ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Manski, R. J., Moeller, J. F., Chen, H., Schimmel, J., Clair, P. A. S., & Pepper, J. V. (2013). Patterns of Older Americans’ Health Care Utilization Over Time. American Journal of Public Health, 103(7), 1314-1324. doi:10.2105/AJPH.2012.301124
Marathe, K. R., Wan, T. T. H., & Marathe, S. (2016). A SYSTEMATIC REVIEW ON THE KAP-O FRAMEWORK FOR DIABETES EDUCATION AND RESEARCH. Medical Research Archives, 4(1).
Matthews, J. A. (1990). Predictors of hospitalized patients' access to postdischarge home health services. (PH.D.), UNIVERSITY OF MARYLAND AT BALTIMORE. Retrieved from https://login.ezproxy.net.ucf.edu/login?auth=shibb&url=http://search.ebscohost.com/logi n.aspx?direct=true&db=rzh&AN=109870030&site=ehost-live Available from EBSCOhost rzh database.
Mattishent, K., & Loke, Y. K. (2016). Bi-directional interaction between hypoglycaemia and cognitive impairment in elderly patients treated with glucose-lowering agents: a systematic review and meta-analysis. Diabetes, Obesity and Metabolism, 18(2), 135-141. doi:10.1111/dom.12587
Maty, S. C., James, S. A., & Kaplan, G. A. (2010). Life-Course Socioeconomic Position and Incidence of Diabetes Mellitus Among Blacks and Whites: The Alameda County Study, 1965-1999. American Journal of Public Health, 100(1), 137-145. doi:10.2105/Ajph.2008.133892
Mayoclinic.org. (2017). Depression (major depressive disorder). Diseases and Conditions. Retrieved from http://www.mayoclinic.org/diseases- conditions/depression/basics/definition/con-20032977
Mayoclinic.org. (2017). Diabetes Complications. Diseases and Conditions. Retrieved from http://www.mayoclinic.org/diseases-conditions/diabetes/basics/complications/con- 20033091
162 McEwen, L. N., Waitzfelder, B. E., Crosson, J. C., Marrero, D. G., Mangione, C. M., & Herman, W. H. (2012). Predictors of mortality over 8 years in type 2 diabetic patients: Translating Research Into Action for Diabetes (TRIAD). Diabetes Care(1935-5548 (Electronic)). doi:D - NLM: PMC3357242 EDAT- 2012/03/21 06:00 MHDA- 2012/09/19 06:00 CRDT- 2012/03/21 06:00 AID - dc11-2281 [pii] AID - 10.2337/dc11-2281 [doi] PST - ppublish
Meneilly, G. S., & Tessier, D. M. (2016). Diabetes, Dementia and Hypoglycemia. Canadian Journal of Diabetes, 40(1), 73-76. doi:10.1016/j.jcjd.2015.09.006
Meneilly, G. S., & Tessier, D. M. (2016). Review: Diabetes, Dementia and Hypoglycemia. Canadian Journal of Diabetes, 40, 73-76. doi:10.1016/j.jcjd.2015.09.006
Mier, N., Wang, X., Smith, M. L., Irizarry, D., Treviño, L., Alen, M., & Ory, M. G. (2012). Factors Influencing Health Care Utilization in Older Hispanics with Diabetes along the Texas-Mexico Border. Population Health Management Journal, 15(3), 149-156 148p. doi:10.1089/pop.2011.0044
Miller, M. E., Bonds, D. E., Gerstein, H. C., Seaquist, E. R., Bergenstal, R. M., Calles-Escandon, J., . . . Genuth, S. (2010). The effects of baseline characteristics, glycaemia treatment approach, and glycated haemoglobin concentration on the risk of severe hypoglycaemia: post hoc epidemiological analysis of the ACCORD study. BMJ, 340. doi:10.1136/bmj.b5444
Moen, M. F., Zhan, M., Hsu, V. D., Walker, L. D., Einhorn, L. M., Seliger, S. L., & Fink, J. C. (2009). Frequency of hypoglycemia and its significance in chronic kidney disease. Clinical Journal of the American Society of Nephrology, 4(6), 1121-1127. doi:10.2215/CJN.00800209
Moghissi, E. S., Korytkowski, M. T., DiNardo, M., Einhorn, D., Hellman, R., Hirsch, I. B., . . . Umpierrez, G. E. (2009). American Association of Clinical Endocrinologists and American Diabetes Association Consensus Statement on Inpatient Glycemic Control. Diabetes Care Journal, 32(6), 1119-1131. doi:10.2337/dc09-9029
Monami, M., Dicembrini, I., Kundisova, L., Zannoni, S., Nreu, B., & Mannucci, E. (2014). A meta‐analysis of the hypoglycemic risk in randomized controlled trials with sulphonylureas in patients with type 2 diabetes (Provisional abstract). In M. Monami, I. Dicembrini, L. Kundisova, S. Zannoni, B. Nreu, & E. Mannucci (Eds.). BMC Endocrine Disorders: BMC Endocrine Disorders.
Monami, M., Lamanna, C., Lotti, E., Marsili, A., Masotti, G., Marchionni, N., & Mannucci, E. (2007). Are comorbidity indices useful in predicting all-cause mortality in Type 2 diabetic patients? Comparison between Charlson index and disease count. Aging Clin Exp Res(1594-0667 (Print)).
163 Moore, C., Li, J., Hung, C., Downs, J., & Nebeker, J. R. (2009). Predictive value of alert triggers for identification of developing adverse drug events. Journal of Patient Safety, 5(4), 223- 228 226p.
Murad, M. H., Coburn, J. A., Coto-Yglesias, F., Dzyubak, S., Hazem, A., Lane, M. A., . . . Montori, V. M. (2012). Glycemic control in non-critically ill hospitalized patients: a systematic review and meta-analysis. The Journal of Clinical Endocrinology & Metabolism, 97(1), 49-58. doi:10.1210/jc.2011-2100
Murad, M. H., Coto-Yglesias, F., Wang, A. T., Sheidaee, N., Mullan, R. J., Elamin, M. B., . . . Montori, V. M. (2009). Drug-Induced Hypoglycemia: A Systematic Review. The Journal of Clinical Endocrinology & Metabolism, 94(3), 741-745. doi:10.1210/jc.2008-1416
Naik, D., Shyamasunder, A. H., Mruthyunjaya, M. D., Gupta Patil, R., Paul, T. V., Christina, F., . . . Thomas, N. (2016). Masked hypoglycemia in pregnancy. J Diabetes. doi:10.1111/1753-0407.12485
Nelson, K., Brown, M. E., & Lurie, N. (1998). Hunger in an adult patient population. Jama, 279(15), 1211-1214. nhlbi.nih.gov. (2016). Calculate Your Body Mass Index. Retrieved from http://www.nhlbi.nih.gov/health/educational/lose_wt/BMI/bmicalc.htm
NIDDK. (2014, Feb 2014). Your Guide to Diabetes: Type 1 and Type 2. Retrieved from http://www.niddk.nih.gov/health-information/health-topics/Diabetes/your-guide- diabetes/Pages/index.aspx
Nilsson, P. M., Gudbjörnsdottir, S., Eliasson, B., Cederholm, J., & Steering Committee of the Swedish National Diabetes, R. (2004). Smoking is associated with increased HbA1c values and microalbuminuria in patients with diabetes--data from the National Diabetes Register in Sweden. Diabetes & Metabolism, 30(3), 261-268.
Nirantharakumar, K., Marshall, T., Kennedy, A., Narendran, P., Hemming, K., & Coleman, J. J. (2012). Hypoglycaemia is associated with increased length of stay and mortality in people with diabetes who are hospitalized. Diabet Medicine, 29(12), e445-448. doi:10.1111/dme.12002
Olveira, G., Tapia, M. J., Ocon, J., Cabrejas-Gomez, C., Ballesteros-Pomar, M. D., Vidal- Casariego, A., . . . Tous, M. C. (2015). Hypoglycemia in noncritically ill patients receiving total parenteral nutrition: a multicenter study. (Study group on the problem of hyperglycemia in parenteral nutrition; Nutrition area of the Spanish Society of Endocrinology and Nutrition). Nutr J, 31(1), 58-63. doi:10.1016/j.nut.2014.04.023
Ou, H. T., Mukherjee, B., Erickson, S. R., Piette, J. D., Bagozzi, R. P., & Balkrishnan, R. (2012). Comparative performance of comorbidity indices in predicting health care-related
164 behaviors and outcomes among Medicaid enrollees with type 2 diabetes. Population Health Management Journal, 15(4), 220-229. doi:10.1089/pop.2011.0037
Pan, A., Wang, Y., Talaei, M., Hu, F. B., & Wu, T. (2015). Relation of active, passive, and quitting smoking with incident type 2 diabetes: a systematic review and meta-analysis. Lancet Diabetes Endocrinol, 3(12), 958-967. doi:10.1016/s2213-8587(15)00316-2
Patel, I., Erickson, S. R., Caldwell, C. H., Woolford, S. J., Bagozzi, R. P., Chang, J., & Balkrishnan, R. (2015). Predictors of medication adherence and persistence in Medicaid enrollees with developmental disabilities and type 2 diabetes. Research in Social and Administrative Pharmacy Journal. doi:10.1016/j.sapharm.2015.09.008
Pathak, R. D., Schroeder, E. B., Seaquist, E. R., Zeng, C., Lafata, J. E., Thomas, A., . . . O'Connor, P. J. (2015). Severe Hypoglycemia Requiring Medical Intervention in a Large Cohort of Adults With Diabetes Receiving Care in U.S. Integrated Health Care Delivery Systems: 2005-2011. Diabetes Care Journal, 39(3), 363-370. doi:10.2337/dc15-0858
Petraitis, P. S., Dunham, A. E., & Niewiarowski, P. H. (1996). Inferring Multiple Causality: The Limitations of Path Analysis. Functional Ecology, 421.
Preacher, K. J., Cai, L., & MacCallum, R. C. (2007). Alternatives to traditional model comparison strategies for covariance structure models. In T. D. Little, J. A. Bovaird, N. A. Card, T. D. Little, J. A. Bovaird, & N. A. Card (Eds.), Modeling contextual effects in longitudinal studies. (pp. 33-62). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.
Punthakee, Z., Miller, M. E., Launer, L. J., Williamson, J. D., Lazar, R. M., Cukierman-Yaffee, T., . . . Gerstein, H. C. (2012). Poor Cognitive Function and Risk of Severe Hypoglycemia in Type 2 Diabetes. Diabetes Care Journal, 35(4), 787-793. doi:10.2337/dc11-1855
Quan, H., Li, B., Couris, C. M., Fushimi, K., Graham, P., Hider, P., . . . Sundararajan, V. (2011). Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol, 173(6), 676- 682. doi:10.1093/aje/kwq433
Quilliam, B. J., Simeone, J. C., & Ozbay, A. B. (2011). Risk Factors for Hypoglycemia-Related Hospitalization in Patients With Type 2 Diabetes: A Nested Case–Control Study. Clinical Therapeutics, 33(11), 1781-1791 1711p. doi:10.1016/j.clinthera.2011.09.020
Quilliam, B. J., Simeone, J. C., Ozbay, A. B., & Kogut, S. J. (2011). The incidence and costs of hypoglycemia in type 2 diabetes. Am J Manag Care, 17(10), 673-680.
Ramirez, B., West, D. J., & Costell, M. M. (2013). Development of a culture of sustainability in health care organizations. J Health Organ Manag, 27(5), 665-672. doi:10.1108/jhom-11- 2012-0226
165 Reynolds, J., Gunderson, J., & Bamford, M. (2015). FLORIDA’S AGING POPULATION: Critical Issues for Florida’s Future | 2015. Retrieved from Pepper Institute on Aging and Public Policy: http://pepperinstitute.fsu.edu/sites/g/files/imported/storage/original/application/22487bf2 3ab907d4c17b39227dccf000.pdf
Rhoades, H., Wenzel, S. L., Golinelli, D., Tucker, J. S., Kennedy, D. P., & Ewing, B. (2014). Predisposing, enabling and need correlates of mental health treatment utilization among homeless men. Community Ment Health J, 50(8), 943-952. doi:10.1007/s10597-014- 9718-7
Riddle, M. C. (2015). Predictors of Nonsevere and Severe Hypoglycemia During Glucose- Lowering Treatment With Insulin Glargine or Standard Drugs in the ORIGIN Trial. Diabetes Care Journal, 38(1), 22-28. doi:10.2337/dc14-1329
Ringholm, L., Pedersen-Bjergaard, U., Thorsteinsson, B., Damm, P., & Mathiesen, E. R. (2012). Hypoglycaemia during pregnancy in women with Type 1 diabetes. Diabet Med, 29(5), 558-566. doi:10.1111/j.1464-5491.2012.03604.x
Roberge, B. L. (2008). Health Care Utilization of the Emergency Department of the Hospital by Children with Healthy Families, a Publicly Funded Health Care Insurance: University of La Verne.
Roffman, C. E., Buchanan, J., & Allison, G. T. (2016). Charlson Comorbidities Index. Journal of Physiotherapy, 62(3), 171. doi:http://dx.doi.org/10.1016/j.jphys.2016.05.008
Rosenblatt, R. A., Baldwin Lm Fau - Chan, L., Chan L Fau - Fordyce, M. A., Fordyce Ma Fau - Hirsch, I. B., Hirsch Ib Fau - Palmer, J. P., Palmer Jp Fau - Wright, G. E., . . . Hart, L. G. (2001). Improving the quality of outpatient care for older patients with diabetes: lessons from a comparison of rural and urban communities. Journal of Family Practice(0094- 3509 (Print)).
Rowe, B. H., Singh, M., Villa-Roel, C., Leiter, L. A., Hramiak, I., Edmonds, M. L., . . . Campbell, S. (2015). Acute Management and Outcomes of Patients with Diabetes Mellitus Presenting to Canadian Emergency Departments with Hypoglycemia. Canadian Journal of Diabetes, 39, Supplement 4, 9-18. doi:http://dx.doi.org/10.1016/j.jcjd.2015.09.084
Sako, A., Yasunaga, H., Matsui, H., Fushimi, K., Hamasaki, H., Katsuyama, H., . . . Yanai, H. (2015). Hospitalization for Hypoglycemia in Japanese Diabetic Patients: A Retrospective Study Using a National Inpatient Database, 2008-2012. Medicine Journal, 94(25), e1029- e1029 1021p. doi:10.1097/MD.0000000000001029
Sämann, A., Lehmann, T., Heller, T., Müller, N., Hartmann, P., Wolf, G. B., & Müller, U. A. (2013). A retrospective study on the incidence and risk factors of severe hypoglycemia in primary care. Fam Pract, 30(3), 290-293. doi:10.1093/fampra/cms071
166 Sawka, A. M., Boulos, P., Talib, A. S., Gafni, A., Thabane, L., Papaioannou, A., . . . Gerstein, H. C. (2007). Low Socioeconomic Status and Increased Risk of Severe Hypoglycemia in Type 1 Diabetes: A Systematic Literature Review. Canadian Journal of Diabetes, 31(3), 233-241. doi:http://dx.doi.org/10.1016/S1499-2671(07)13010-0
Schafers, S., Naunheim, R., Vijayan, A., & Tobin, G. (2012). Incidence of hypoglycemia following insulin-based acute stabilization of hyperkalemia treatment. J Hosp Med, 7(3), 239-242.
Scheetz, L. C. T. (2010). Utilization of care services among the oldest old. (1488174 M.S.), Iowa State University, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Schloot, N. C., Haupt, A., Schutt, M., Badenhoop, K., Laimer, M., Nicolay, C., . . . Holl, R. W. (2016). Risk of severe hypoglycemia in sulfonylurea-treated patients from diabetes centers in Germany/Austria: How big is the problem? Which patients are at risk? Diabetes Metab Res Rev, 32(3), 316-324. doi:10.1002/dmrr.2722
Schoenberg, R. (1989). Covariance Structure Models. Annual Review of Sociology, 425.
Schopman, J. E., Simon, A. C., Hoefnagel, S. J., Hoekstra, J. B., Scholten, R. J., & Holleman, F. (2014). The incidence of mild and severe hypoglycaemia in patients with type 2 diabetes mellitus treated with sulfonylureas: a systematic review and meta-analysis. Diabetes Metab Res Rev, 30(1), 11-22. doi:10.1002/dmrr.2470
Schwartz, R. S. (2001). Racial Profiling in Medical Research. New England Journal of Medicine, 344(18), 1392-1393. doi:doi:10.1056/NEJM200105033441810
Seligman, H., Jacobs, E. A., Lopez, A., Sarkar, U., Tschann, J., & Fernandez, A. (2011). Food insecurity and hypoglycemia among safety net patients with diabetes. Jama(1538-3679 (Electronic)). doi:D - NLM: NIHMS640268 D - NLM: PMC4230711 EDAT- 2011/07/13 06:00 MHDA- 2011/09/02 06:00 CRDT- 2011/07/13 06:00 AID - 171/13/1204 [pii] AID - 10.1001/archinternmed.2011.287 [doi] PST - ppublish
Seligman, H. K., Bindman, A. B., Vittinghoff, E., Kanaya, A. M., & Kushel, M. B. (2007). Food insecurity is associated with diabetes mellitus: Results from the national health examination and nutrition examination survey (NHANES) 1999-2002. J Gen Intern Med, 22(7), 1018-1023. doi:10.1007/s11606-007-0192-6
Seligman, H. K., Bolger, A. F., Guzman, D., López, A., & Bibbins-Domingo, K. (2014). Exhaustion Of Food Budgets At Month's End And Hospital Admissions For Hypoglycemia. Health Affairs, 33(1), 116-123. doi:10.1377/hlthaff.2013.0096
Shenolikar, R. A. (2006). Racial differences in oral antidiabetic medication adherence, healthcare utilization and costs associated with type 2 diabetes in Medicaid enrollees.
167 (3230882 Ph.D.), The Ohio State University, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Shepherd, J. G., Locke, E., Zhang, Q., & Maihafer, G. (2014). Health services use and prescription access among uninsured patients managing chronic diseases. The Journal of Community Health, 39(3), 572-583. doi:10.1007/s10900-013-9799-1
Sherrod, P. (2014). DTREG: Predictive Modeling Software. Retrieved from https://www.dtreg.com/
Signorovitch, J. E., Macaulay, D., Diener, M., Yan, Y., Wu, E. Q., Gruenberger, J. B., & Frier, B. M. (2013). Hypoglycaemia and accident risk in people with type 2 diabetes mellitus treated with non-insulin antidiabetes drugs. Diabetes, Obesity and Metabolism, 15(4), 335-341. doi:10.1111/dom.12031
Simeone, J. C., & Quilliam, B. J. (2012). Predictors of Emergency Department and Outpatient Visits for Hypoglycemia in Type 2 Diabetes: An Analysis of a Large US Administrative Claims Database. Annals of Pharmacotherapy, 46(2), 157-168 112p.
Simon, D., Velasco, P. d. P., Parhofer, K. G., Frederick, L. G., Lomon, I. D., Vandenberghe, H., . . . Bradley, C. (2015). Hypoglycaemic episodes in patients with type 2 diabetes – risk factors and associations with patient-reported outcomes: The PANORAMA Study. Diabetes & Metabolism, 41(6), 470-479. doi:http://dx.doi.org/10.1016/j.diabet.2015.08.007
Snih, S. A., Markides, K. S., Ray, L. A., Freeman, J. L., Ostir, G. V., & Goodwin, J. S. (2006). Predictors of healthcare utilization among older Mexican Americans. Ethnicity & Disease, 16(3), 640-646.
Solomon, M. D., Vijan, S., Forma, F. M., Conrad, R. M., Summers, N. T., & Lakdawalla, D. N. (2013). The impact of insulin type on severe hypoglycaemia events requiring inpatient and emergency department care in patients with type 2 diabetes. Diabetes Research and Clinical Practice, 102(3), 175-182. doi:http://dx.doi.org/10.1016/j.diabres.2013.09.013
Sonoda, N., Morimoto, A., Ugi, S., Morino, K., Sekine, O., Nemoto, K., . . . Miyamatsu, N. (2015). Predictors for Mild and Severe Hypoglycemia in Insulin-Treated Japanese Diabetic Patients. PLoS ONE, 10(6), e0130584. doi:10.1371/journal.pone.0130584
Springer, E. L. (2015). Factors impacting mental health service utilization among adults with mental illness. (1588145 M.S.W.), San Diego State University, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Stafstrom, C. E. (2008). Not All Sweetness and Light: The Role of Glycogen in Hypoglycemic Seizures. Epilepsy Currents, 8(4), 105-107. doi:10.1111/j.1535-7511.2008.00258.x
168 Stein, J. A., Andersen, R., & Gelberg, L. (2007). Applying the Gelberg-Andersen Behavioral Model for Vulnerable Populations to Health Services Utilization in Homeless Women. J Health Psychol, 12(5), 791.
Stein, J. A., Andersen, R. M., Koegel, P., & Gelberg, L. (2000). Predicting health services utilization among homeless adults: A prospective analysis. J Health Care Poor Underserved, 11(2), 212-230.
Stein, J. A., Andersen, R. M., Robertson, M., & Gelberg, L. (2012). Impact of Hepatitis B and C Infection on Health Services Utilization in Homeless Adults: A Test of the Gelberg- Andersen Behavioral Model for Vulnerable Populations. Health Psychology, 31(1), 20- 30. doi:10.1037/a0023643
Stiffman, A. R., Striley, C., Horvath, V. E., Hadley-Ives, E., Polgar, M., Elze, D., & Pescarino, R. (2001). Organizational context and provider perception as determinants of mental health service use. J Behav Health Serv Res, 28(2), 188-204.
Streiner, D. L. (2005). Finding our way: An introduction to path analysis. Canadian Journal of Psychiatry, 50(2), 115-122.
Tan, J. (2009). Health service utilization among Chinese, Filipino, and Asian Indian older adults in the United States. (3359719 Ph.D.), Washington University in St. Louis, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Thode, N., Bergmann, E., Kamtsiuris, P., & Kurth, B. M. (2005). [Predictors for ambulatory medical care utilization in Germany]. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz, 48(3), 296-306. doi:10.1007/s00103-004-1004-3
Thompson, A. E. (2015). Hypoglycemia. Jama, 313(12), 1284-1284. doi:10.1001/jama.2015.0876
Thorpe, C. T., Gellad, W. F., Good, C. B., Zhang, S., Zhao, X., Mor, M., & Fine, M. J. (2015). Tight glycemic control and use of hypoglycemic medications in older veterans with type 2 diabetes and comorbid dementia. Diabetes Care Journal, 38(4), 588-595. doi:10.2337/dc14-0599
Tschope, D., Binz, C., Krekler, M., Plate, T., Deeg, E., & Gitt, A. K. (2011). Antidiabetic pharmacotherapy and anamnestic hypoglycemia in a large cohort of type 2 diabetic patients--an analysis of the DiaRegis registry. Cardiovasc Diabetol(1475-2840 (Electronic)). doi:D - NLM: PMC3162488 EDAT- 2011/07/16 06:00 MHDA- 2012/01/27 06:00 CRDT- 2011/07/16 06:00 PHST- 2011/06/02 [received] PHST- 2011/07/14 [accepted] AID - 1475-2840-10-66 [pii] AID - 10.1186/1475-2840-10-66 [doi] PST - epublish
169 Turchin, A., Matheny, M. E., Shubina, M., Scanlon, J. V., Greenwood, B., & Pendergrass, M. L. (2009). Hypoglycemia and clinical outcomes in patients with diabetes hospitalized in the general ward. Diabetes Care Journal, 32(7), 1153-1157 1155p. doi:10.2337/dc08-2127
UKHSG, U. H. S. G. (2007). Risk of hypoglycaemia in types 1 and 2 diabetes: effects of treatment modalities and their duration. Diabetologia(0012-186X (Print)).
Unger, C. C., Warren, N., Canway, R., Manderson, L., & Grigg, K. (2011). Type 2 diabetes, cardiovascular disease and the utilisation of primary care in urban and regional settings. Rural Remote Health Journal, 11(4), 1795.
UnitedHealthFoundation. (2016). Florida Annual Report. Retrieved from http://www.americashealthrankings.org/explore/2015-annual- report/measure/Diabetes/state/FL
Vriesendorp, T. M., DeVries, J. H., Rosendaal, F., & Hoekstra, J. B. (2008). Severe hypoglycemia in critically ill: risk and outcomes. Crit Care Med, 36(4), 1390; author reply 1390-1391. doi:10.1097/CCM.0b013e31816a1788
Wagner, K. (2015). Shifting Gears. Healthcare Executive Journal, 30(3), 10-18.
Wallis, J. (2008). Hypoglycaemia in pregnancy. Midwives Magazine, 18.
Wan, T. (2002). Evidence-based health care management : multivariate modeling approaches: Boston : Kluwer Academic Publishers, c2002.
Wan, T. (2014). A TRANSDISCIPLINARY APPROACH TO HEALTH POLICY RESEARCH AND EVALUATION. International Journal of Public Policy, 10(4-5), 161-177. doi:10.1504/IJPP.2014.063094
Wan, T., & Cooper, P. L. (1988). A structural equation model of physician productivity. J Health Care Mark, 8(4), 37-45.
Wan, T., Lin, Y.-L., & Ortiz, J. (2016). Racial Disparities in Diabetes Hospitalization of Rural Medicare Beneficiaries in 8 Southeastern States. Health Services Research & Managerial Epidemiology, 3. doi:10.1177/2333392816671638
Wan, T., Lin, Y.-L., & Ortiz, J. (2016). Racial Disparities in Diabetes Hospitalization of Rural Medicare Beneficiaries in 8 Southeastern States. Health Services Research and Managerial Epidemiology, 3, 2333392816671638. doi:doi:10.1177/2333392816671638
Wan, T., & Soifer, S. (1974). Determinants of physician utilization: a causal analysis. J Health Soc Behav, 15(2), 100-108.
Wan, T., & Soifer, S. (1975). A multivariate analysis of the determinants of physician utilization. Socio-Economic Planning Sciences, 9(5), 229-237. doi:http://dx.doi.org/10.1016/0038- 0121(75)90007-5
170 Wan, T., & Yates, A. (1974). Prediction of dental services utilization: a multivariate approach. Inquiry, 12(2), 143-156.
Wan, T. T., & Yates, A. S. (1975). Prediction of dental services utilization: a multivariate approach. Inquiry, 12(2), 143-156.
Wang, N. E. A., Shandro, J., Sohoni, A., & Fassl, B. (2011). A practical guide to pediatric emergency medicine; caring for children in the emergency department. : Cambridge U. Press
Weiner, J. P., Garnick, D. W., Fowles, J., Lawthers, A. G., & Palmer, R. H. (1995). Variation in office-based quality. A claims-based profile of care provided to Medicare patients with diabetes. Jama(0098-7484 (Print)).
Weinstock, R. S., DuBose, S. N., Bergenstal, R. M., Chaytor, N. S., Peterson, C., Olson, B. A., . . . Hirsch, I. B. (2015). Risk Factors Associated With Severe Hypoglycemia in Older Adults With Type 1 Diabetes. Diabetes Care Journal. doi:10.2337/dc15-1426
Weston, R., & Gore, P. A. (2006). A Brief Guide to Structural Equation Modeling. The Counseling Psychologist, 34(5), 719-751. doi:doi:10.1177/0011000006286345
Whitmer, R. A., Quesenberry, C. P., & Selby, J. V. (2009). Hypoglycemic episodes and risk of dementia in older patients with type 2 diabetes mellitus. Jama(1538-3598 (Electronic)). doi:D - NLM: NIHMS126351 D - NLM: PMC2782622 EDAT- 2009/04/16 09:00 MHDA- 2009/04/21 09:00 CRDT- 2009/04/16 09:00 AID - 301/15/1565 [pii] AID - 10.1001/jama.2009.460 [doi] PST - ppublish
Wild, S. H., McKnight, J. A., McConnachie, A., Lindsay, R. S., Glasgow, & Lothian Diabetes Register Data, G. (2010). Socioeconomic status and diabetes-related hospital admissions: a cross-sectional study of people with diagnosed diabetes. Journal of Epidemiol Community Health, 64(11), 1022-1024. doi:10.1136/jech.2009.094664
Wilkinson-Lee, A. M. (2008). Health care utilization among Mexican-, Cuban-, and Puerto Rican-American adolescents: Examining Andersen's behavioral model of health services use. (3315864 Ph.D.), The University of Arizona, Ann Arbor. ProQuest Dissertations & Theses A&I; ProQuest Dissertations & Theses Global database.
Williams, S. A., Pollack, M. F., & DiBonaventura, M. (2011). Effects of hypoglycemia on health-related quality of life, treatment satisfaction and healthcare resource utilization in patients with type 2 diabetes mellitus. Diabetes Research and Clinical Practice, 91(3), 363-370. doi:http://dx.doi.org/10.1016/j.diabres.2010.12.027
Williams, S. A., Shi, L., Brenneman, S. K., Johnson, J. C., Wegner, J. C., & Fonseca, V. (2012). The burden of hypoglycemia on healthcare utilization, costs, and quality of life among
171 type 2 diabetes mellitus patients. Journal of Diabetes and its Complications, 26(5), 399- 406. doi:http://dx.doi.org/10.1016/j.jdiacomp.2012.05.002
Witzig, R. (1996). The medicalization of race: scientific legitimization of a flawed social construct. Ann Intern Med, 125(8), 675-679.
Wolinsky, F. D., Stump, T. E., & Johnson, R. J. (1995). Hospital Utilization Profiles among Older Adults over Time - Consistency and Volume among Survivors and Decedents. Journals of Gerontology Series B-Psychological Sciences and Social Sciences, 50(2), S88-S100.
Yaffe, K., Falvey, C. M., Hamilton, N., Harris, T. B., Simonsick, E. M., Strotmeyer, E. S., . . . Schwartz, A. V. (2013). Association between hypoglycemia and dementia in a biracial cohort of older adults with diabetes mellitus. JAMA Intern Med, 173(14), 1300-1306. doi:10.1001/jamainternmed.2013.6176
Yamashita, T., Kart, C., & Noe, D. (2012). Predictors of adherence with self-care guidelines among persons with type 2 diabetes: results from a logistic regression tree analysis. J Behav Med, 35(6), 603-615. doi:10.1007/s10865-011-9392-y
Yeboah-Korang, A., Kleppinger, A., & Fortinsky, R. H. (2011). Racial and Ethnic Group Variations in Service Use in a National Sample of Medicare Home Health Care Patients with Type 2 Diabetes Mellitus. J Am Geriatr Soc, 59(6), 1123-1129. doi:10.1111/j.1532- 5415.2011.03424.x
Yeh, J. S., Sung, S.-H., Huang, H.-M., Yang, H.-L., You, L.-K., Chuang, S.-Y., . . . Chen, C.-H. (2015). Hypoglycemia and risk of vascular events and mortality: a systematic review and meta-analysis. Acta Diabetol.
Yu, T. M., Lin, C. L., Chang, S. N., Sung, F. C., & Kao, C. H. (2014). Increased risk of stroke in patients with chronic kidney disease after recurrent hypoglycemia. Neurology, 83(8), 686-694. doi:10.1212/WNL.0000000000000711
Zapatero, A., Gomez-Huelgas, R., Gonzalez, N., Canora, J., Asenjo, A., Hinojosa, J., . . . Barba, R. (2014). Frequency of hypoglycemia and its impact on length of stay, mortality, and short-term readmission in patients with diabetes hospitalized in internal medicine wards. Endocrine Practice, 20(9), 870-875. doi:10.4158/EP14006.OR
Zhang, Y., Ting, R. Z. W., Yang, W., Jia, W., Li, W., Ji, L., . . . China Depression in Chinese Patients with Type 2 Diabetes Study, G. (2015). Depression in Chinese patients with type 2 diabetes: associations with hyperglycemia, hypoglycemia, and poor treatment adherence. J Diabetes, 7(6), 800-808. doi:10.1111/1753-0407.12238
Zhang, Y., Wieffer, H., Modha, R., Balar, B., Pollack, M., & Krishnarajah, G. (2010). The burden of hypoglycemia in type 2 diabetes: a systematic review of patient and economic perspectives. Journal of Clinical Outcomes Management, 17(12), 547-557 511p.
172 Zoungas, S., Patel, A., Chalmers, J., de Galan, B. E., Li, Q., Billot, L., . . . Heller, S. (2010). Severe Hypoglycemia and Risks of Vascular Events and Death. New England Journal of Medicine, 363(15), 1410-1418. doi:doi:10.1056/NEJMoa1003795
173