Intra-Hospital Transfers and the Associated Risk of Hospital-Onset

Clostridium Difficile Infection

Thesis

Presented in Partial Fulfillment of the requirements for the Degree Master of Arts

in the Graduate School of The Ohio State University

Megan McHaney, B.A.

Graduate Program in Geography

The Ohio State University

2018

Thesis Committee:

Elisabeth Root, Adviser

Courtney Hebert

Harvey Miller

Megan McHaney

2018

Abstract

One of the greatest challenges faced by modern hospitals is the mitigation of nosocomial infections. Part of what makes their elimination so elusive is that hospitals have yet to fully understand the many facets that allow them to spread so successfully. The synergism of the hospital environment, medical practices, and the patients themselves create an ever-shifting landscape, making the infections difficult to pinpoint.

Patient level factors that contribute to the acquisition of illnesses are generally well understood, as are the environmental. However, commonplace medical practices, such as intra-hospital patient transfers, could be propagating nosocomial infections.

This research explores if the number of intra-hospital transfers a patient experiences could be associated with the likelihood of them being diagnosed with hospital-onset Clostridium difficile infection, a common nosocomial illness. This is accomplished through mixed methods including logistic regression, social network analysis and spatial analysis utilizing geographic information systems. These indicate that there is a positive association between the number of intra-hospital transfers and diagnosis with hospital-onset

Clostridium difficile. Further, local spatial dependency was found when examining the location of rooms associated with the diagnosis of Clostridium difficile. The results provide several potential avenues of intervention including reduction of unnecessary intra-hospital transfers and better targeted sanitization regimes.

Acknowledgments

This research could not have been completed without the support and expertise of my advisor, Dr.

Elisabeth Root. Additional thanks to my committee members, Dr. Hebert and Dr. Miller, whose help was equally indispensable and the Institute for the Design of Environments Aligned for Patient Safety

(IDEA4PS) at The Ohio State University which is sponsored by the Agency for Healthcare Research &

Quality (P30HS024379).

iii

Vita

2012-2016 Undergraduate Research Assistant, Department of Geography, Kansas

State University

2013-2016 Undergraduate Teaching Assistant, Department of Anthropology,

Kansas State University

2016 BA, Geography, Kansas State University

2016-Present Graduate Research Assistant, Department of Geography, The Ohio

State University

Publications

2018 McHaney, R., Reychav, I., Zhu, L., & McHaney-Lindstrom, M.

Iterative conceptual modeling: A case study in cardiac patient survival

simulation. Operations Research for Health Care.

Fields of Study

Major Field: Geography

Table of Contents

ABSTRACT ...... ii

ACKNOWLEDGMENTS ...... iii

VITA ...... iv

LIST OF TABLES ...... vi

LIST OF FIGURES ...... vii

CHAPTER ONE: INTRODUCTION ...... 1

RESEARCH OBJECTIVES ...... 1

BACKGROUND ...... 2

Hospital Infection Control ...... 2

Nosocomial Infections ...... 4

Previous Research...... 6

Research Approach ...... 7

Institutional Review Board Approval of Research ...... 12

CHAPTER TWO: ANALYSIS OF INTRA-HOSPITAL TRANSFERS AND CLOSTRIDIUM DIFFICILE

...... 13

BRIDGE ...... 19

CHAPTER THREE: NETWORK ANALYSIS OF INTRA-HOSPITAL TRANSFERS AND

CLOSTRIDIUM DIFFICILE ...... 20

CHAPTER FOUR: CONCLUSION...... 40

REFERENCES ...... 42

List of Tables

Table 1. Definition of Variables ...... 14

Table 2. Case and Control Descriptive Statistics ...... 15

Table 3. Logistic Regression Results ...... 16

Table 4. SNA Overall Metrics ...... 29

List of Figures

Figure 1. Disease Ecology Framework adapted from Meade 1977 ...... 8

Figure 2. Matching Propensity Jitter Graph ...... 16

Figure 3. Case and Control SNA Visualization ...... 29

Figure 4. RPOS and RPROS ...... 31

Figure 5. RPOS 'Onset' Risk Rings ...... 32

Figure 6. RPROS Eigenvector and LISA Map ...... 34

vii

Chapter One: Introduction

Research Objectives

Hospital environments affect patient life in a multitude of ways: from overall wellbeing to reduction of infections (Huisman, Morales, van Hoof, & Kort, 2012). Research indicates that well-designed hospitals have important positive outcomes for those moving within their interior, including reduction of staff and patient stress, improved healthcare quality, and ultimately- improved patient safety (R. Ulrich, Zimring,

Quan, Joseph, & Choudhary, 2004). However, much remains to be understood. For instance, how does patient movement through this space affect patient propensity to get a hospital-onset infection (HOI).

Despite the advances made in hospital design and numerous guidelines implemented to improve patient outcomes, 25% of hospitalized patients still suffer from HOIs; making it an immediate issue that needs to be addressed (Yokoe et al., 2014).

Based upon these facts, this research explores the following research question:

Due to the key role that hospital environments play in the acquisition of HOIs, do patients exposed to a greater number of differences in their environments through intra-hospital transfers face a higher risk of acquiring the infection?

To answer this question, an analysis of intra-hospital transfers among patients with a HOI known as

Clostridium difficile infection (CDI) will be completed through a case-control study design. This analysis will be based on the postulation that those who are diagnosed with hospital-onset CDI (HO-CDI) show a statistically significantly higher number of intra-hospital transfers when compared to a similar control population. A secondary analysis will utilize social network analysis (SNA) and geographical information systems (GIS) to both gain an understanding of the differences in intra-hospital transfer network structures that patients diagnosed with HO-CDI have, compared to the control population, and how

1 visualization of these networks, both as social networks and through geographic representations, can allow for better mitigation of HO-CDI outbreaks and introduction of new policy.

Background

Hospital Infection Control

Hospitals often rely on several lines of defense when it comes to protecting their patients from HOIs.

These range from training staff on correct hand hygiene protocols to environmental cleaning regimes

(Boyce & Pittet, 2002; Sydnor & Perl, 2011). Surveillance of patients also has contributed to a hospital’s ability to increase patient safety in the face of harmful infections, through the development of sophisticated electronic health records. However, complete infection control within hospitals is elusive.

Understanding the multi-faceted nature of HOIs is key to controlling them, but is further complicated by the unique makeup of each individual disease. To better mitigate infection transfer within hospitals, in this case- HO-CDI, it is important to use available resources developed over time in the field of hospital infection control in a synergistic manner.

A Brief Historical Perspective Infection control began with the idea of disease theory and more specifically, the idea that germs were natural living agents and ultimately the cause of illness. This idea was proposed first by Fracastorius, an

Italian physician, in his book De Contagione published in the mid sixteenth century. It was not until the nineteenth century that this was further conceptualized into Henle’s Postulates and several decades later, into Koch’s germ theory of disease (Kimble & White, 1981).

An epidemiological perspective of disease control emerged later in the nineteenth century from Ignaz

Semmelweiss, a Hungarian physician. He observed that there were significant differences in puerperal fever, also known as childbed fever, rates between two different wards in the hospital that he worked within (Shorter, 1984). He eventually discovered that the ward with practicing medical students had mortality rates almost four times higher than the ward with midwives practicing. Further, he linked this to 2 his observation that the medical students were not washing their hands after doing autopsies, instead simply heading straight to aid in delivery. Despite his landmark discovery, Semmelweis’ theory that the lack of handwashing lead to increased mortality was ultimately ignored until much later. In today’s medical world Semmelweis is noted as the father of hospital epidemiology (Pittet & Boyce, 2001), a field now growing with research focused around understanding and mitigating dangerous HOIs.

Hospital Epidemiology Modern hospital epidemiology has a multitude of definitions primarily focused around the central idea of investigating infectious disease outbreaks within the institution (Felsen & Wolarsky, 1940). However, there has been criticism that this is too focused, and instead hospital epidemiology should be viewed as the medical discipline concerned with the evolution, distribution and characteristics of disease in population groups within the institution (Fuerst, Lichtman, & James, 1965). It is the second definition for hospital epidemiology that is utilized in this paper, due to its more flexible nature.

Increased attention was given to epidemiology, and more broadly infection control, within hospitals beginning in the 1960s when a large outbreak of staphylococcal occurred (Nahmias & Eickhoff, 1957).

Soon following this epidemic, the CDC became involved, sending out a call for hospitals to have physicians conduct surveillance on infectious diseases happening within their walls to create realistic control measures (“Surveillance of Communicable Diseases of National Importance,” 1963). Eventually the burden of surveillance was taken off physicians because this duty could be accomplished better by trained infection control professionals (Wenzel, Osterman, Hunting, & Gwaltney, 1976).

Quickly following this period, infection surveillance and control programs became widespread across the

United States, eventually evolving into the National Health Safety Network (NHSN), a method of passive surveillance that tracks nosocomial diseases (Sydnor & Perl, 2011). The NHSN still operates today; however, many hospitals have their own surveillance programs, often nested under the Patient Quality and Safety branches. 3

The hospital epidemiology concepts used for this research are based upon the idea that HO-CDI is often found within the hospital environment, particularly in areas where infected patients may have visited. It can be distributed throughout the hospital by the movement of both people and equipment, causing quick and hard to track spreading of the bacteria. However, better understanding of how patient exposure relates to environments that may contain HO-CDI bacteria will enable mitigation efforts to be pursued more effectively. Therefore, this research is most concerned with the ‘distribution and characteristic’ categories of hospital epidemiology.

Nosocomial Infections

According to the CDC, nosocomial infections are a localized or systemic condition resulting from an adverse reaction to the presence of an infectious agent or its toxin (Garner, Jarvis, Emori, Horan, &

Hughes, 1988). Further, there must be no evidence that the infection was present before the time of admission to the hospital or healthcare environment (Garner et al., 1988). Today, the term nosocomial infection is largely interchangeable with hospital (or healthcare) onset infection or HOI. Many organisms which can cause HOIs also present themselves in a community acquired form; however, this research is strictly involved in those that appear in acute care settings. HO-CDI is one such HOI and is the infection of interest in this research.

Clostridium Difficile CDI is the HOI examined in this research. HO-CDI presents a complex problem for hospitals due to its long-living bacteria and potentially fatal outcomes. It is important to understand the epidemiology, associated risks, and potential prevention for HO-CDI to help lessen outbreaks within hospital environments.

Epidemiology CDI is a spore-forming, gram-positive anaerobic bacillus associated with diarrhea, which can vary from mild to life-threatening (Cohen et al., 2010). Diagnoses of HO-CDI is made typically through a combination of clinical observations and laboratory testing conclusions, including the occurrence of 4 diarrhea, a positive stool sample of toxigenic Clostridium difficile , or colonscopic or histopathologic evidence of pseudomembranous colitis (Barbut & Petit, 2001). It spreads through direct or indirect contact. HO-CDI has been increasing over the past decades which can be partly attributed to the increase in hospital surveillance; however, it has also been noted that an increase in antibiotic usage and inconsistent antimicrobial stewardship programs could be contributing factors (Lessa et al., 2015).

Risks Antibiotic use is one of the most notable patient risk factors for HO-CDI due to their impact normal intestinal flora; with almost all antibiotics being linked to increased risk of HO-CDI (Bignardi, 1998).

Further risks include patient age, underlying ailments or comorbidities, and any other health procedures that could lead to disruption of intestinal flora and ultimately a compromised immune system (Barbut &

Petit, 2001).

Any discussion of HO-CDI risk factors must also consider environmental influences. Duration of stay within a hospital has been noted as increasing risk, implying that increased exposure to HO-CDI bacteria harbored within the hospital environment may play a key role (Barbut & Petit, 2001). The risk factors associated with increased exposure to HO-CDI are of interest for this research, since it proposes a related factor yet to be investigated---the number of intra-hospital transfers.

Intra-hospital transfers refer to the movement of patients from one room or space within a hospital to another. These are logged within the electronic health record (EHR) databases utilized by hospitals and often occur when patients receive tests, procedures, or are moved into a different ward. These movements expose patients to new environments which could potentially harbor HO-CDI bacteria. Therefore, we postulated in this research that an increased number of exposures to new environments increases the risk that patients have of encountering and contracting HO-CDI.

Prevention CDI provides a complicated case for elimination from hospital environments. The spores can survive up to five months on surfaces and cannot be removed by alcohol-based cleaners which are often the norm

(Pittet et al., 2006). Hand hygiene is the first line of defense when it comes to reducing HO-CDI cases, including wearing gloves, vigorous hand washing, and not relying solely upon alcohol based hand sanitizers (Pittet, Allegranzi, & Boyce, 2009; Vonberg et al., 2008). Another important aspect of prevention is cleaning the hospital environment itself. Typical cleaners often do not destroy the spore form of HO-CDI. Therefore, chlorine-based germicides have been recommended (Fawley et al., 2007).

Ultraviolet light cleaning is another method that has shown decreases in HO-CDI when implemented

(Levin, Riley, Parrish, English, & Ahn, 2013). Prevention is a multifront issue in the case of HOIs and so should be considered from multiple angles.

Previous Research Previous research in this field can be divided into two main subcategories. The first is centered around intra-hospital transfers. Intra-hospital transfer research varies, from infectious disease spread to general physical and psychiatric wellbeing of the patient. Intra-hospital transfers of patients later diagnosed with methicillin-resistant staphylococcus aureus (MRSA), a common HOI, have been indicated as being an independent risk factor for the infection (Sax et al., 2005). Further, patients with critical illnesses or chronic illnesses often experience negative outcomes when subjected to intra-hospital transfers (Waydhas,

Schneck, & Duswald, 1995). Higher patient acuity and longer intra-hospital transport durations also have been linked to more frequent and serious negative effects on patient health, leading for a call in the reduction of time spent in transfer and the overall number of unnecessary transfers (R. S. Ulrich & Zhu,

2007). Despite the rich variety of research, intra-hospital transfers have yet to be explored as a risk factor for HO-CDI. Due to HO-CDI’s longevity and high contagious nature, this is an important facet to consider, particularly for those looking for interventions in hospital infection control.

The second area of related research focuses on inter-hospital networks and their role in the transmission of HO-CDI. Inter-hospital transfer network structures have been identified as risk factors for HO-CDI, with particular importance given to the number of other hospitals patients transfer to and from (DiDiodato

& McArthur, 2017; Simmering, Polgreen, Campbell, Cavanaugh, & Polgreen, 2015). The networks that govern inter-hospital patient sharing create highways for HO-CDI to travel between hospital populations that are often difficult to predict. However, recent research has indicated that using network analysis to find hospitals with high centrality can help to predict a significant percent of HO-CDI incidence at the hospital (Fernández-Gracia, Onnela, Barnett, Eguíluz, & Christakis, 2017). This research proposes focusing on a on smaller scale of disease spread which is not well understood, intra-hospital transfers.

This scale is of equal importance, as it is key for hospitals to control disease within their own facilities as well as between themselves and affiliate hospitals.

Previous research leaves an identifiable gap that this research seeks to fill. Intra-hospital transfers have been linked to MRSA and overall patient health, but not to HO-CDI. Further, hospital networks have been shown to be highly related to outbreaks of HO-CDI, through social network analysis methods; however, no previous studies have harnessed these techniques at an individual hospital scale to understand how intra-hospital transfers can propagate HO-CDI spread within hospitals. By utilizing a combination of methods, this research provides this knowledge, while also suggesting evidence-based solutions.

Research Approach

Hypotheses This research seeks to answer the following question: Do patients exposed to a greater number of differences in their environments through intra-hospital transfers face a higher risk of acquiring HO-

CDI? To answer this question, the primary hypotheses are: patients diagnosed with HO-CDI will display higher numbers of intra-hospital transfers when compared to similar patients not diagnosed with HO-CDI and patients diagnosed with HO-CDI will show different intra-hospital patient network patterns when

7 compared to similar patients not diagnosed with HO-CDI. Together, these hypotheses will illustrate the fundamental differences that HO-CDI patients experience regarding intra-hospital transfers.

Theoretical Framework The guiding theoretical framework for this research is disease ecology (Meade, 1977). Disease ecology utilizes the idea of interactions between population, environment, and culture to understand how disease occurs. Figure 1 adapts the original disease ecology diagram to help further the understanding of the multitude of connections that lead to HO-CDI.

Figure 1. Disease Ecology Framework adapted from Meade 1977

Environment in the disease ecology framework can be broken into stimuli, such as pathogens and drugs, and health services, which for this study is hospitals. Pathogens such as HO-CDI are harbored within the hospital environment and therefore, an important consideration when conceptualizing how the infection may spread. The population side of the interaction includes host variables. In this study these are patient variables such as genetics, immunology, nutrition and demographics. These are all key factors for understanding the baseline health status of individuals susceptible to HO-CDI, as their risk changes accordingly. The final, and most complex, interaction in disease ecology is culture. Encompassed in culture includes behavior, practices, norms, and the movement and creation of environmental elements.

Within this facet, intra-hospital transfers occur, since it is a medical practice to move patients between spaces in hospitals to provide them with the best care. This research challenges current ideas for best care, by asking if this side of the interaction can’t be targeted to reduce excess patient movements. The additional movement of already high-risk patients could in fact be hindering their health, necessitating a reexamination of these engrained practices. Instead, this study suggests that by reducing patient movement we cut off a significant environment, population, and cultural interaction that could lead to increased rates of HO-CDI.

Methodology Logistic Regression A logistic regression model is a commonly used statistical method in the exploration of an outcome variable and was developed in the mid-20th century by David Cox (Cox, 1958). It measures association between the independent variables and the log odds of the model outcome, and can be interpreted as how much a unit increase in a predictor variable affects the odds of the specific outcome being modeled. The glm() function within R Software was used, specifying the binary nature of the dependent variable (R

Core Team, 2017).

Social Network Analysis Social network analysis (SNA) often is used to describe the connections between people, through methods of communication or even relationships that may bind them together. SNA was the creation of a social psychologist named Moreno (Moreno, 1934); however, it did not become quantifiable until the

1960s and 1970s, when mathematicians became interested in the topic (Scott, 1988). Today, it is widespread, from analysis of animal behavior patterns to examination of how information and knowledge transfer occurs through innovation networks (Fritsch & Kauffeld-Monz, 2010; Wey, Blumstein, Shen, &

Jordán, 2008). Public health also has utilized SNA to examine patient sharing and inter-hospital transfers.

These studies indicated the importance of hospital resources and geographic location in hospital centrality measures (Iwashyna, Christie, Moody, Kahn, & Asch, 2009; Lee et al., 2011).

The networks used in SNAs are composed of two main parts, the nodes and edges. In a visualization of social networks, nodes are points. They often represent individuals or in the case of this research, the rooms that individuals occupied in the hospital. Edges can be described as the lines, or pathways, that connect two nodes. For this research, edges represent the physical movement of a patient from one node

(hospital location) to another thereby linking the starting and ending nodes.

The software used to complete SNA for this research was NodeXL- an add-on for Microsoft Excel

(Hansen, Shneiderman, & Smith, 2011). NodeXL provides an analyst with ‘Overall Metrics’ which include vertices (number of nodes), total edges (number of intra-hospital transfers), average geodesic distance (average length of paths taken), in-degree (number of times an edge ends at a particular node), out-degree (number of times edges originate from a particular node), betweenness centrality (number of times a node lies on the shortest path between other nodes), closeness centrality (score based on the sum of all of a node’s shortest paths), eigenvector centrality (number of links a node has to other nodes within the network).

In the current research, SNA methods describe the movement networks of patients diagnosed with HO-

CDI and compares those with the movement network of similar patients without the HOI. This

10 comparison will help draw conclusions on whether the network structure of HO-CDI patients is different than non-HO-CDI patients and provide insight into the differences.

Geographic Information Systems Geographic Information systems or GIS are powerful tools that can aid in both analysis and visualization of spatial data. GIS can be used in a multitude of ways such as the storage, analysis or management of data that is geographically referenced. Originally, GIS was utilized for land use potential in Canada.

However, others saw potential applications for the technology and it quickly spread to other fields. In the

1990’s, the public health community saw GIS as an opportunity to link health information to its geographical locations to help recognize patterns in disease (Richards, Croner, Rushton, Brown, &

Fowler, 1999).

The creation of user-friendly software that facilitated the usage of GIS furthered the interest of researchers immensely. Today, ArcGIS, a GIS software developed by Environmental Systems Research

Institute, INC (ESRI), is one of the most commonly used. First released in 2001, it provides storage, management, analysis, and visualization technique to its users (ESRI, 2014). In its most basic form, a

GIS utilizes computer software, such as ArcGIS, to link where things are in space to what they are. This information is stored in layers within the GIS and allows the analyst to perform complex visualization and statistical techniques upon it by linking them through relational databases. This research utilizes ArcGIS to both visualize and analyze the SNA data outputs.

For analysis, we are concerned with spatial autocorrelation, or the measurement of correlation between observations located in space, of the SNA data outputs. Spatial autocorrelation originates from Tobler’s

First Law of Geography, which states that ‘Everything is related to everything else, but near things are more related than distant things’ (Tobler, 1970). Based upon this, our study hypothesizes that we will see similar values of SNA metrics near one another when represented in geographic space, and consider the implications related to disease spread.

Study Structure This research is structured as a two-part exploration and analysis of patient intra-hospital transfer data; which is divided between two separate papers. The first paper analyzes the patient data taken from the

EHR system, including general demographics, medication, and locational data. The research objective of the first paper is to find if intra-hospital transfers are positively associated with the acquisition of HO-

CDI. This will be accomplished through a case-control study design. The intended journal of publication for this paper is The Journal of Hospital Infection, and is formatted to their specifications as a short report.

The second paper will utilize both social network analysis (SNA) and geographical information systems

(GIS) to initially investigate the differences between the structure of the intra-hospital transfer networks.

That data will then be used to map the metrics in the hospital environment to demonstrate how it could be used to make intervention decisions. The intended journal of publication for this paper is The

International Journal of Medical Informatics and is formatted to their specifications.

Institutional Review Board Approval of Research This study was approved by the IRB under Study ID 2015H0424.

Chapter Two: Analysis of Intra-Hospital Transfers and Clostridium Difficile For Submission to: The Journal of Hospital Infection

Summary

The spread of infectious disease within the hospital setting is a pervasive and growing problem. To examine potential mechanisms which may influence nosocomial Clostridium difficile infection, we used electronic health records from a large regional hospital to examine whether individuals who experienced more intra-hospital transfers, and therefore more environments, would exhibit a higher rate of hospital- onset infection. Findings suggest that one possible strategy for reducing nosocomial Clostridium difficile infection would be to reduce excess intra-hospital transfers, particularly for people with the highest risk for infection, such as those of advanced age or with compromised immune systems.

Introduction

Clostridium Difficile infection (CDI) is a common nosocomial disease often associated with both antibiotics and prolonged hospital stays (Kelly, Pothoulakis, & LaMont, 1994). Researchers recommend that nosocomial CDI be diagnosed through a combination of clinical and laboratory conclusions such as occurrence of diarrhea, a positive stool sample test result for toxigenic Clostridium Difficile , or colonoscopic or histopathologic evidence of pseudomembranous colitis (Cohen et al., 2010). Hospital environments play an important role in the spread of nosocomial CDI and studies show that the infectious agent can survive on surfaces for up to five months (Caroff, Yokoe, & Klompas, 2017; Kaatz et al.,

1988). While there is a significant body of research examining the role of inter-hospital transfers in the spread of infectious disease (Huang et al., 2010; Simmering et al., 2015); there are fewer hospital-level studies that examine the role of intra-hospital transfers on transmission dynamics. There is also very little research which specifically examines the impact of patient movement on nosocomial CDI. This study examines the relationship between patient intra-hospital transfers and risk of nosocomial CDI infection. 13

We hypothesized that patients who experienced a greater number of intra-hospital transfers would exhibit a greater risk of nosocomial CDI diagnoses.

Methods

Data were obtained from the electronic health records (EHR) of adult patients admitted to The Ohio State

University Wexner Medical Center from December 1st, 2013 through January 1st, 2016. During this period, the department of Clinical Epidemiology identified 502 cases of nosocomial CDI using NHSN surveillance definitions. Our study employed a matched case-control design. Three controls per nosocomial CDI case were selected using exact matching on two characteristics with known correlations with nosocomial CDI: whether a patient was on antibiotics while hospitalized and birth year (Jump,

2013). We also used nearest neighbor matching on the admittance department to ensure a similar distribution of patient health conditions. Matching was done using the R Software package ‘MatchIt’ propensity score matching algorithm, and provided equal sized control and test cases (Ho, Imai, King, &

Stuart, 2011).

Once case and control groups were identified, admission, discharge and transfer data were processed to provide the variables used in modeling (Table 1). Intra-hospital transfers were identified in the EHR by coding for each time a patient was removed from a room. The final variables used in modeling were: days to onset of HO-CDI infection and number of intra-hospital transfer during the hospital stay.

Variable Description

Days to Control: Number of days in the hospital Onset Case: Number of days in the hospital before diagnosis with nosocomial CDI

Transfer Control: Number of intra-hospital transfers during hospital stay Number Case: Number of intra-hospital transfers during hospital stay before nosocomial CDI diagnosis

Different Control: Number of different rooms visited during hospital stay Rooms Case: Number of different rooms visited during hospital stay before nosocomial CDI diagnosis Table 1. Definition of Variables

We used multiple logistic regression to model the risk of nosocomial CDI infection as a function of the days to onset and number of transfers. We calculated variance inflation factors (VIFs) for the variables to ensure no multicollinearity and examined goodness of fit using the Hosmer-Lemeshow test and specificity using ROC curves. Modeling was done in the R Software package (R Core Team, 2017).

Results

Table 2 shows descriptive statistics for the study sample. There was no significant difference between cases and controls for antibiotic usage, the most important risk factor for nosocomial CDI. Figure 2 provides the distribution of propensity scores (where units indicated patients), the means of birth year were both 1956 and the mean of control use of antibiotics was .86, while the mean of case use of antibiotics was .84. The matched treatment and control groups have close distribution, indicating a balanced sample and good overall results.

Population Variables Control Case (N=502) Univariate T-Test (N= 1506) (p-value < 0.05) Antibiotic Use 0.097

YES 1304 419

NO 202 83

Years of Age 0.57

98 - 79 175 50

78 - 59 768 263

58 - 39 400 132

38 - 19 163 57 Days to Onset < 0.001* Mean 7.19 10.04 Transfer Number < 0.001* Mean 2.59 3.37 Different Rooms < 0.001* Mean 2.55 3.78 Table 2. Case and Control Descriptive Statistics

Figure 2. Matching Propensity Jitter Graph

Results from the logistic regression model, found in Table 3, suggest a statistically significant difference between both onset days and number of transfers and nosocomial CDI risk. For each additional day in the hospital, the odds of nosocomial CDI infection increase by about 3%. In addition, each additional transfer increases the odds of nosocomial CDI infection by approximately 11%.

Coefficients Odds Confidence Interval STD. Z- Value Pr(>|z|) VIF Ratio (95%) Error

(Intercept) .194 0.194-0.162 .0912 -17.964 <0.001* -

Days to Onset 1.03 1.017-1.043 .006 4.731 <0.001* 1.070

Transfer Days 1.105 1.058-1.154 .022 4.522 <0.001* 1.070

Table 3. Logistic Regression Results

Results from the Hosmer-Lemeshow test indicate that the overall model has poor fit, as the p-value was below 0.05, and the AROC’s value was .637, providing an acceptable value for model discrimination.

Discussion and Conclusion

This study examined if patients diagnosed with nosocomial CDI experienced more intra-hospital transfers prior to onset than their similar counterparts. Exact matching of control and test cases was done on two important correlates to nosocomial CDI, age and antibiotic usage, to minimize bias. Through the application of a multiple logistic regression model we found that cases do have more intra-hospital transfers when compared to controls. This is the first investigation of intra-hospital transfers as a potential risk factor for nosocomial CDI.

Our methodology focused on the use of existing EHR data to help track patient movements within the hospital environment. This new application introduced no technology beyond what most hospitals already have and therefore presents a unique opportunity to mitigate nosocomial CDI infections through the reduction of intra-hospital transfers and movements, particularly for patients who may already be at an elevated risk due to age, compromised immune systems, or pharmacological risks. We assert that the intra-hospital transfers expose patients to more environments which may harbor the C. difficile spores; putting those that experience more intra-hospital transfers at an elevated risk. This is the first study of how intra-hospital transfer effect HOI and therefore, due to the statistically significant results, warrants further investigation.

The limitations of this study include the potential for factors that were not accounted for that may be associated with both intra-hospital transfers and nosocomial CDI, such as severity of illness. A further limitation is that patients may leave a room without being logged as leaving in the EHR system. These undocumented movements would be difficult to capture without GPS patient tracking. The goodness of fit values for the model were weak; however, the objective of this research was not to create a predictive model for nosocomial CDI. Rather, this study’s purpose was to discover if intra-hospital transfers were

17 associated to nosocomial CDI. Future research could use our findings to justify the inclusion of intra- hospital patient transfers for predictive modeling.

Our findings are unique and important to healthcare systems and to decision makers seeking to reduce infection spread. Nosocomial CDI is a threat in all healthcare settings and therefore it reduction is a high priority, particularly to those involved in patient safety. This research supports ideas such as bringing testing equipment to patients, to reduce unnecessary patient movement within hospitals. Beyond potentially reducing nosocomial CDI, this could help reduce patient stress and discomfort. By providing more insight into the mechanisms that propagate nosocomial CDI, it will allow for better decision making and the overall reduction of outbreaks.

This project was supported by the Institute for the Design of Environments Aligned for Patient Safety

(IDEA4PS) at The Ohio State University which is sponsored by the Agency for Healthcare Research &

Quality (P30HS024379). The authors’ views do not necessarily represent the views of AHRQ.

This study was approved by the IRB under Study ID 2015H0424.

Bridge

With the results indicating that there is a statistically significant relationship between the number of intra- hospital transfers and HO-CDI, the next objective of this research is to further explore the actual structure of the intra-hospital transfer networks. Understanding the movements of patients within hospitals through social network analysis provides a unique perspective, highlighting how they connect spaces and identifying rooms most crucial to the HO-CDI intra-hospital transfer network. This could provide important information to hospital administration regarding the types and locations of interventions that should be implemented. Therefore, the objective of the next piece of research is to utilize SNA to analyze intra-hospital patient transfer networks prior to diagnosis with HO-CDI, further investigate the results produced by the SNA in a GIS to identify if spatial variation and correlation between rooms most central to the networks are associated and ultimately produce a methodology that could be used by hospital epidemiologists to implement targeted interventions.

Chapter Three: Network Analysis of Intra-Hospital transfers and Clostridium Difficile For Submission to: The International Journal of Medical Informatics

Structured Abstract

Purpose

To explore how social network analysis (SNA) can be used to analyze intra-hospital patient networks of individuals with a hospital onset infection (HOI) to produce results that can be further visualized and analyzed in a geographical information system (GIS) environment. The goal of this research was to produce results that identified areas of the hospital central to the spread of the hospital acquired infection so that it can be further targeted by intervention programs.

Methods

A case and control study design was used to select 2,008 individuals that were matched on age, antibiotic usage, and admittance department. We retrieved locational data for the patients that showed their path through the hospital, which was then translated into a network with the SNA software. Overall metrics were calculated for the SNA based on three datasets, one comparing the case and control intra-hospital transfer networks, a second looking at room associated with the onset of disease, and a third focusing on the room prior to onset room. Once these metrics were calculated, the data was moved into the GIS for visualization, and global and local spatial autocorrelation analysis.

Results

The SNA analysis comparing cases to controls indicated significant differences in the overall structure of the networks. This implied that patient movement structure was inherently different and could be a contributing factor to the acquisition of the HOI. The analysis into the two additional datasets identified

‘high risk rooms’, which were key to the intra-hospital transfer network. These rooms would be ideal for 20 targeted interventions. A GIS visual representation of these metrics was developed, showing spatial variation across the example hospital floor. This indicated the intra-hospital transfer networks vary spatially and needed to be further examined with statistical measures of spatial autocorrelation. The spatial autocorrelation analysis revealed no significant results at the global level, but statistically significant results for local correlation measures.

Conclusions

Utilizing SNA and GIS analysis in conjunction with one another provided a data rich environment in which the risk inherent in intra-hospital transfer networks was quantified, visualized, and interpreted for potential interventions. The key information provided by this study included the significant differences in case and control intra-hospital patient transfer networks, the identification of high risk rooms and pathways, and statistically significant measures of local spatial autocorrelation. Each of these confirmed the importance that intra-hospital patient networks play in the transmission of HOIs, highlighting opportunities for interventions utilizing these data. Due to spatial variation differences, further research is necessary to confirm this is not a localized phenomenon, but instead a common situation occurring within many hospitals.

1. Introduction

Hospitals are unique, complex, and constantly changing, constructed environments. These structures contain flows of people and objects which potentially act as vectors for the diseases hospitals seek to prevent or treat. New technologies, such as Real Time Locating Systems can successfully track material objects within hospital environments (Kamel Boulos & Berry, 2012). Likewise, data systems such as electronic health records (EHRs) or Radio Frequency Identification (RFID) can be used to virtually track individuals as they move throughout the hospital (Wang, Chen, Ong, Liu, & Chuang, 2006). Despite these advances, tracking hospital acquired infections remains a difficult task. In fact, a quarter of hospital patients will contract a hospital onset infection (HOI) during their stay, making control of these pathogens a top concern for all hospitals (Yokoe et al., 2014).

Clostridium Difficile infection (CDI) is a nosocomial disease associated with prolonged hospital stays and antibiotic usage. The most common symptom is mild to severe diarrhea, which can be fatal. Physicians diagnose this infection through a combination of clinical observations and laboratory results (Cohen et al.,

2010; Kelly et al., 1994). An important aspect of Clostridium difficile is its ability to live on surfaces for up to five months, meaning that hospital environments are a key factor when considering risk (Cohen et al., 2010; Kaatz et al., 1988). As patients move between rooms and spaces during intra-hospital transfers they are exposed to more environments, and in turn could encounter more bacteria harboring surfaces.

Research shows the number of intra-hospital transfers are positively, statistically significantly associated with rates of hospital-onset CDI (HO-CDI), indicating the importance of understanding patient movements within the hospital environment in HO-CDI acquisition (Author Citation Withheld).

From a geographic perspective, patient intra-hospital transfers connect hospital spaces, through both associated individuals and equipment. A series of analysis techniques and methodologies make it possible to visualize potential vehicles of HO-CDI in new ways. Among these are social network analysis (SNA) and geographical information systems (GIS). The current study leverages these techniques/methodologies to understand which rooms are associated with infection acquisition. We believe this information is crucial to the institution of impactful interventions attempting to reduce HO-CDI, because it provides empirical evidence on which rooms should be targeted for intensive cleaning.

1.1 CDI

Clostridium Difficile is a spore-forming gram-positive anaerobic bacillus associated with diarrhea, which can vary from mild to life-threatening and spreads through indirect or direct contact with the bacteria

(Cohen et al., 2010). HO-CDI has been increasing over the past decades which can be partly attributed to the increase in hospital surveillance; however, it has also been noted that an increase in antibiotic usage and inconsistent antimicrobial stewardship programs could be contributing factors (Lessa et al., 2015).

1.2 SNA

Social network analysis or SNA is traditionally a methodology used to describe, visualize, and quantify connections between people. SNA techniques are widespread and have been used to examine topics ranging from animal behavior patterns to how information and knowledge transfer occur through innovation networks (Fritsch & Kauffeld-Monz, 2010; Wey et al., 2008). Public health also has utilized

SNA. Most relevant to the current research are studies conducted on patient sharing and inter-hospital transfers. These studies have indicated the importance of hospital resources and geographic location in hospital centrality to hospital acquired infections(Iwashyna et al., 2009; Lee et al., 2011). However, no previous studies, to our knowledge, have examined the social network structure of intra-hospital transfers.

Networks used in SNA are composed of two main parts, the nodes and edges. In the visualization of social networks, the nodes are represented as points. They often represent individuals, or in the case of this research, the rooms that individuals occupy during a hospital stay. The edges can be described as the lines, or relationships, that connect two nodes. Here, edges represent the physical movement of a patient from one node (hospital location) to another thereby linking the starting and ending nodes.

A number of SNA software packages exist. We used NodeXL, a widely used add-on for Microsoft Excel

(Hansen et al., 2011). NodeXL provides an analyst with ‘Overall Metrics’ which include vertices (number of nodes), total edges (number of intra-hospital transfers), average geodesic distance (average length of paths taken), in-degree (number of times an edge ends at a particular node), out-degree (number of times an edge originates from a particular node) and eigenvector centrality (number of links it has to other nodes within the network).

We used SNA to compare the movement networks of patients diagnosed with HO-CDI to similar patients without HO-CDI to examine whether the network structure of HO-CDI patients differed from non-HO-

CDI patients.

1.3 GIS

Geographic Information systems (GIS) refers to analytical methods and tools used to describe and integrate data spatially. In most instances, object attributes are assigned to features located within space to provide the basis for further statistical analysis. This delineates two important parts of GIS: the spatial data which is geographically referenced through coordinates; and, the attribute data linked to this spatial data, often through common fields stored within the data (ex. place names). In this way, GIS makes it possible to store, manage and spatially analyze data to reveal key spatial interactions. According to Parker

(1988, p. 1548), “GIS analytical operations can be broken into two classes, primary and compound.”.

Primary operations relate to basic functionality like area computation, distance measurement, buffer

24 generation, clustering, and reclassification while compound operations use two or more primary operations to enable complex analysis. In public health, GIS often relies on compound operations to analyze the need for health care based upon neighborhood characteristics or access to healthcare considering geographic variables such as distance and public transportation availability (Mavoa, Witten,

McCreanor, & O’Sullivan, 2012; Root, 2012).

In this research, compound operations were used to combine the spatial data of the hospital floorplan and the attribute data for the hospital rooms (nodes) output by the SNA and to analyze if spatial autocorrelation was present in the data through spatial statistical methods. ArcGIS, the leading GIS software, was utilized to store, analyze and visualize the results (ESRI, 2014).

1.4 Desired Outcome

Clostridium Difficile represents a difficult HOI for hospitals to control due to its longevity, virulence and reoccurrence rate. By using both social network analysis methods and geographic information systems, this research seeks to develop a set of steps that hospitals can use to combat HO-CDI. These two methods allow for identification of areas at a high risk of propagating transmission within hospital settings and visualize the data to help target those locations in an efficient manner.

2. Materials and Methods

2.1 Research Question and Hypotheses

We developed an overarching research question to guide our study:

Are the network structures of HO-CDI patients different than the network patterns of similar patients without HO-CDI; and can this information be transformed into a visualization that can be utilized by hospital epidemiologists for intervention?

To address the first portion of our research question, we sought to understand if network graphs varied between patients that acquired HO-CDIs and those that did not. We expected patients that acquired HO-

CDIs to have more complex network graphs indicating more movement, exposing them to more environments and making them more vulnerable to infection. This hypothesis was based on prior studies indicated Clostridium Difficile spores can live up to five months on hospital surfaces and that cleaning regimes produce notable reductions in HO-CDI cases (Fawley et al., 2007; Levin et al., 2013; Pittet et al.,

2009; Vonberg et al., 2008). These studies indicate that the role that environmental exposure plays in the infection of patients with HO-CDI is significant and would likely be magnified with increased exposures to different environments. To test those ideas, we formulated the following hypotheses:

H1a: SNAs of patients infected with HO-CDI will have more vertices than those of patients not infected with HO-CDI.

H1b: SNAs of patients infected with HO-CDI will have more total edges than those of patients not infected with HO-CDI.

H1c: SNAs of patients infected with HO-CDI can be used to identify high risk edges leading to infection with HO-CDI.

To address the second part of our research question, we wanted to translate the SNA analysis into something that hospital epidemiologists and decision-makers could use. To accomplish this, we explored whether GIS could provide further insight via spatial analysis of the SNA data to identify high-risk rooms that were spatially correlated. This is particularly important due to the longevity of Clostridium Difficile bacteria and because it spreads through both direct and indirect contact. It is possible that during room cleaning, or by simply touching communal equipment outside of adjacent rooms, that bacteria would spread through space. While spatial autocorrelation was the main process that we wanted to explore statistically; we further postulated that HO-CDI spread was not a spatially static process. Instead, we

26 expected to see local differences in HO-CDI spread along floors. To address these ideas, the following hypotheses were formulated:

H2a: In a specific hospital environment, SNA metrics show systematic variation over space.

H2b: Spatial autocorrelation of HO-CDI is present within hospital floors.

The results of these two hypotheses would enable us to recommend ways hospital administration can utilize the results from the GIS analysis to potentially implement new interventions to mitigate HO-CDI spread within the hospital environment.

2.2 Data Acquisition and Preparation

We used a multi-step process to acquire and prepare the data to test our hypotheses. We obtained patient data, including locational information from The Ohio State University Wexner Medical center electronic medical record (EHR) system. The data was kept on a secure server to protect patient health information.

From this data, 502 cases of HO-CDI were identified spanning from late 2013 through 2015. We selected

502 matching controls from the hospital population. Exact matching was completed on birth year and antibiotic usage, which are both predictors of HO-CDI (Fekety et al., 1981). We applied nearest neighbor matching for the admittance department of the patients to ensure a similar distribution and type of patient

(Rubin & Society, 1973). We used R software to complete the matching and specifically used the R software package ‘Matchit’ for this task (Ho et al., 2011).

2.3 Primary Analysis

The data was transformed to permit SNA using NodeXL. The rooms visited by patients were identified as

‘nodes’, while their movement between the rooms created ‘edges’. This was done for both the cases and controls; ultimately resulting in two SNA datasets. Additionally, two other datasets were created for the cases only: one looking at only the rooms that were attributed to HO-CDI within the EHRs (rooms prior 27 to ‘onset’ status) and a second looking at the rooms prior to onset leading to the rooms of onset (rooms prior to rooms of onset). In these subsequent data analyses, we conducted SNA analysis with NodeXL software examining overall network metrics, most commonly used edges, and nodes most associated with

HO-CDI.

To further visualize the SNA data, we created a geographic representation of the hospital by gathering shapefiles, a form of vector data that allows storage of an object’s locational, shape, and attribute characteristics. We then joined SNA output metrics to these shapefiles using hospital room number, common to both the GIS shapefiles and the SNA metrics. One floor was chosen as a case study for how to apply geographic statistical methods to the output derived from a SNA. First, a simple map was created that identified nodes (rooms) that had the highest numbers of HO-CDI associated with them. From there, statistical tests of spatial autocorrelation were applied- including both a global Moran’s I and a Local

Indicator of Spatial Autocorrelation (LISA). The global Moran’s I takes into account the entire dataset and the arrangement of values, to identify if any areas of statistically significant values are located near one another. The LISA statistic instead considers the dataset from a more localized perspective, identifying areas of statistically significant values based upon only the values surrounding them.

Neighbors for these spatial statistics were defined using the queen’s contiguity network (neighboring rooms as those that either share a side or corner with the room of interest) A hospital routing network

(defined connections based on realistic pathways i.e. doors, hallways) was additionally calculated to identify if results of the global Moran’s I or LISA map would change with a different weighting matrix.

3. Results

3.1 SNA

The SNA provided a unique visualization of the entire case and control networks, represented by nodes

(the rooms the patients visited in the hospital) and edges (the patient movements from room to room)

(Figure 3). Figure 3 shows edges taken more than one time in each network, with darker lines denoting more used pathways. The notable difference between the two is the centralized pattern of the control group and the seemingly more dispersed pattern of the case group.

Figure 3. Case and Control SNA Visualization

In addition, the SNA produced overall metrics for both the controls and the cases (Table 4). In nearly every measure, the SNA for cases had higher values than the controls. We used these values to evaluate our hypotheses.

Graph Metric CASE CONTROL Graph Type Directed Directed Difference T-Test (p <0.05) Vertices 770 692 Case + N/A Total Edges 2466 1477 Case + N/A Average Geodesic Distance 3.149376 3.162013 Case + N/A Average In-Degree 2.5688 1.8699 Case + N/A Average Out-Degree 2.5688 1.8699 Case + p < 0.002 Average Eigenvector Centrality 0.0013 0.0014 Control + p < 0.001 Table 4. SNA Overall Metrics

Following the overall metrics analysis, the differences between the two datasets were examined more closely. To address hypothesis H1a, the counts for vertices were inspected. The vertices count for the case dataset was higher than the control, indicating that infected patients visited a substantially higher number of rooms. Therefore, we accepted H1a. For H1b we examined edges, the number of intra-hospital transfers. Once again, the count was higher for the case group, showing that more intra-hospital transfers occurred in the case network, leading to the acceptance of hypothesis H1b. Finally, for hypothesis H1c, metrics which measured between two specific vertices were investigated. The case group’s highest edge weight (number of times a pathway between two rooms was taken) was 14 while the control group’s highest edge weight was 8. Further, nearly all of the control groups’ 16 highest edgeweight pathways led to ‘exit’ from the hospital, while none of 16 highest edgeweight pathways in the case led to ‘exit’. This identified high risk pathways for the case dataset and indicated that hypothesis H1c could also be confirmed.

Additional testing was done to further explore differences between the two networks, including T-tests to see if any of measures were significantly different between case and control groups. T-tests were performed upon edge weight, in-degree, out-degree and eigenvector centrality. Cases had statistically significantly higher values of edge weight (p-value < 0.001) and out-degree (p-value = 0.002); while, controls had statistically significantly higher values for eigenvector centrality (p-value < 0.001).

Two final analyses were completed on the datasets of rooms prior to ‘onset’ status (RPOS) and the room prior to room of ‘onset’ status (RPROS) (Figure 4).

Figure 4. RPOS and RPROS

SNA visualization and calculation of overall metrics were again completed using the NodeXL software.

The initial visualization provided results that were further interpreted as concentric rings of risk with the rooms with the most attributed with HO-CDI in the innermost ring (Figure 5). The central node that all of the edges lead to in Figure 5 represent the ‘onset’ status of HO-CDI, therefore the nodes surrounding it represent the rooms attributed with the onset of disease.

Figure 5. RPOS 'Onset' Risk Rings

The analytical focus of exploring these two datasets were the edge weight and eigenvector values. Edge weight for the RPOS dataset provided rooms with the highest number of HO-CDI diagnoses. In this dataset, the highest room value had an edge weight value of 49. Edge weight for the RPROS dataset provided identifications of pathways most used prior to onset. The highest use pathway for this dataset had an edge weight value of 3. Eigenvector values could also be calculated for RPROS dataset. It was found that the three rooms with the highest eigenvector values were also among the highest 16 edge weight pathways.

3.2 GIS

Figure 6 shows a visualization of the SNA metrics utilizing the spatially referenced hospital floor. The eigenvector of the nodes (rooms) in RPROS is shown on the left, while a LISA map of the same area identifies clusters of similar values. The visualization of the SNA metrics (Figure 5) projected onto the

32 study hospital floor confirmed hypothesis H2a, by showing that the eigenvector values created by the

SNA vary over space. This is important, as it shows that the locational distribution of individuals diagnosed with HO-CDI is not uniform. This heterogeneity in the data indicates that there may be rooms that have a higher risk associated with HO-CDI. Further, hypothesis H2b was confirmed for local spatial autocorrelation measures (Figure 5). The LISA map can highlight rooms that have a high value of eigenvector and are surrounded by other high values (high-high in red), rooms with high values surrounded by rooms with low values (high-low in pink), rooms with low values surrounded by rooms with high values (low-high in light blue) and rooms with low values surrounded by rooms with low values (low-low in dark blue). In our results, the LISA map only identified high-high and low-high clusters. This indicates that there are hotspots of rooms that are key to the intra-hospital transfer network, rather than the RPROS being spatially random. The spatial correlation of these rooms is noteworthy due to the idea that HO-CDI can survive for long periods of time on surfaces and is easily spread through both direct and indirect contact. Areas highlighted by the LISA map could reveal rooms that have not received adequate sanitization and require further intervention by hospital administration. These are important points of intervention that could lead to a reduction in HO-CDI diagnosis.

Figure 6. RPROS Eigenvector and LISA Map

However, hypothesis H2b was rejected for the global measure of spatial autocorrelation as the Moran’s I p-value was insignificant for all metrics tested. This may have occurred because hospitals are split into wards, meaning that they are often partitioned from one another, which could lead to local phenomena but creating insignificant results when the entire study area was considered. Another potential reason for insignificant global spatial autocorrelation could be that the HO-CDI database suffered from the issue of small numbers as it was a rare outcome. The dataset of RPOS also was brought into the GIS environment to examine how its edgeweight values can be harnessed to inform policy. The edgeweight values were

34 mapped for the same floor as the RPROS dataset, and a LISA map was calculated for the measure. The global Moran’s I was once again insignificant.

4. Discussion

4.1 SNA

The guiding question of this study was, are the network structures (as defined by intra-hospital transfers between rooms) of patients with HO-CDI different than similar patients without HO-CDI and can this be translated into information for hospital epidemiologist’s prevention efforts? To answer the first part of our research question, regarding if there were significant differences in the network structure, we used SNA.

The SNA visualization of the entire patient intra-hospital network showed that the case network had several key nodes; while, in comparison, the control network had only one main node. Visually, the differences in the two networks were obvious and the statistical analysis supported this difference. This indicates that if the key nodes of the case network could be identified, they would be ideal for supplementary cleaning---an important step for control of hospital acquired infections. Further in the analysis of the entire network, the case network had higher values for nearly every overall metric, and more importantly had statistically significantly higher numbers for edgeweight. Higher rates for edgeweight indicate that more similar pathways are being taken for the cases than for the controls, providing important information on areas of the hospital that should receive focus for more cleaning to remove the Clostridium Difficile spores. The control network showed a statistically significantly higher value for eigenvector, however value emerged because control network had a single, centralized node

(exit), and patients moved through fewer parts of the hospital and encountered fewer opportunities for infection. The comparison of these two networks confirmed that the intra-hospital transfer network structure of patients with HO-CDI and those without is different. The most important aspect of these differences was the multi-central-nodal pattern of intra-hospital transfers of the cases. This is applicable beyond the hospital network at The Ohio State University and for HO-CDI research in general, as many 35 hospital acquired infections present difficulties when hospitals work to eliminate them from their networks.

The most important result of examining the RPOS dataset through SNA was identification of high risk rooms. Figure 2 was arranged with rooms having the highest edgeweight being located in the center, therefore identifying ‘risk rings’. The centermost ring would be the most opportune location for intervention. A recommended intervention would be targeted cleaning via chlorine releasing disinfectants and with UV light cleaning equipment (Fawley et al., 2007; Levin et al., 2013). A secondary possible recommendation would be to identify these rooms to hospital staff, to reduce the assignment of high risk patients to them.

The RPROS dataset allowed for edgeweight and eigenvector values to be calculated. By cross examining these outputs, arranged from highest to lowest values, it becomes possible to identify the highest risk rooms. Edge weight provided the pathways most taken. while eigenvector centrality identified the nodes most central to the network. Similarly to the ‘risk rings’, those rooms identified as having high edgeweight and eigenvector centrality should be targeted for additional cleaning, as they could have large impacts on reducing the spread of infection.

4.2 GIS

The secondary part of the research question was answered through GIS. Maps are a universal language, and the GIS visualization allowed the data to be transferred from the realm of connections simply based upon patient movements and into a physical, geographic framework. This migration of the data from a

SNA framework into a GIS revealed key information in the form of local spatial autocorrelation. The

LISA statistic identified high eigenvector rooms on a hospital floor that were surrounded by other high eigenvector rooms for the RPROS dataset. This provides two important pieces of information. The first is that by highlighting particular rooms, or groups of rooms, of concern it provides a beginning point for

36 further investigation. These rooms could have characteristics within them, or in the areas around them, that could be propagating HO-CDI since they are crucial to the intra-hospital transfer network. It could also be indicating a diffusion of the disease over small spaces through potential vectors such as patients, employees or equipment. The second is that it provides hospital administration with visualized data to make decisions regarding interventions. Since these rooms are most associated with the intra-hospital transfer network it could identify rooms that are not receiving adequate sanitization and make help to delegate resources more efficiently, ultimately eliminating key rooms in the network.

The GIS visualization for the RPROS dataset considered primarily eigenvector centrality. Eigenvector centrality identified the rooms most crucial to the intra-hospital HO-CDI patient transfer network. The

GIS identified the rooms with the highest values alone and in regard to their neighbors. The measurement of local spatial autocorrelation, LISA, again proved the most applicable. This is likely due to the data itself, as it suffers from a small numbers issue because of the rare outcome of HO-CDI. Further investigation of the room characteristics and targeted cleaning programs for the rooms identified as being significant by the LISA statistic is recommended, particularly focusing on high valued rooms surrounded by high values.

4.3 Policy

The information provided by both the SNA and GIS analysis could inform actions that could be taken to potentially reduce HO-CDI. Improved cleaning practices, or more severe regimens, could be instituted for rooms identified as being highly associated to HO-CDI patients. These rooms were first found by the

SNA, then further delineated by the GIS analysis. Of highest concern should be the rooms that the LISA map showed as high valued surrounded by high values. Cleaning that has been identified as HO-CDI bacteria include chlorine releasing products, bleach based products, and UV light sanitization (Fawley et al., 2007; Levin et al., 2013). More than just identifying the most high-risk rooms, the GIS allowed

37 identification of these rooms in their geographic reality. This is key for the allocation of cleaning staff or assignment of patients to certain wards. Further, this GIS could be integrated into applications for staff that can identify the rooms and patient pathways of high risk in real time to help with decision making as outbreaks unfold.

4.4 Limitations

The data was collected over several years, while HO-CDI spores have been known to live up to five months on surfaces. Therefore, an appropriate time lag would need to be created if the ultimate goal was the creation of a real-time application in order to account for how long bacteria could live in a room post

HO-CDI patient occupancy. However, due to the fact that HO-CDI is a rare outcome this could prove difficult. A limitation in the GIS analysis is having data grouped by floor. Preliminary results indicated that between floor correlation of eigenvector and edgeweight values were insignificant for the hospital studied, however this could be highly dependent on the space and hospital being analyzed, and configuration of air ducts and other non-obvious connections. The final limitation of this study was the reliance on hospital staff entries for time and location of patients. Personal or administrative difference in entering this data could cause potential inaccuracies.

5. Conclusions

The results from both the SNA and the GIS analysis can help to inform hospital staff on which rooms are highly associated to HO-CDI patients and which rooms are becoming key nodes in the overall intra- hospital transfer network. Rooms experiencing high traffic (high eigenvector or high edgeweight) should be the focus of more intense cleaning via chlorine releasing products, bleach cleaning, or UV light cleaning. Rooms presenting high eigenvector centrality values also need to be targeted, as they can affect many more individuals within the intra-hospital transfer network than those rooms with low eigenvector centrality values. By reducing the potential risk of these rooms it is possible to reduce HO-CDI

38 transmission. Analyzing the SNA metrics in a GIS provides further information, such as areas of high-risk rooms which should receive targeted cleaning interventions along with the reduced assignment of patients more vulnerable to HO-CDI. The next step for combination SNA and GIS analysis is the implementation of real-time applications to analyze and visualize patient data to halt outbreaks before they begin.

This project was supported by the Institute for the Design of Environments Aligned for Patient Safety

(IDEA4PS) at The Ohio State University which is sponsored by the Agency for Healthcare Research &

Quality (P30HS024379). The authors’ views do not necessarily represent the views of AHRQ.

Chapter Four: Conclusion

The findings of this research provide important insights into HO-CDI. The initial objective was to identify if patients exposed to a greater number of differences in their environments, through intra- hospital transfers, faced a higher risk of acquiring HO-CDI. The hypotheses for this question were: patients diagnosed with HO-CDI will display higher numbers of intra-hospital transfers when compared to similar patients not diagnosed with HO-CDI and patients diagnosed with HO-CDI will show different intra-hospital patient network patterns when compared to similar patients not diagnosed with HO-CDI.

Initially, it was confirmed that patients diagnosed with HO-CDI had a significantly higher number of intra-hospital transfers. After confirming the primary hypothesis, locational data collected from the patients’ EHRs was utilized to create a social network based upon their intra-hospital transfers. SNA revealed that intra-hospital transfer networks significantly differed between case and controls, based on calculated metrics. Further, the analysis helped identify rooms key to the structure of the infected patients network. These rooms present opportunities for cleaning interventions to aid in the reduction of HO-CDI.

In a final step, local spatial correlation was found for the example floor’s overall SNA metrics utilizing

GIS analysis.

The findings provide empirical data that could be applied in the development of interventions to reduce

HO-CDI. The statistically significant correlation between intra-hospital patient transfers and infection with HO-CDI indicates a possible patient level intervention. Reduction of unnecessary transfers through methods such as bringing portable equipment to the patient instead of taking them to additional procedure or testing rooms; or paying attention to the number of transfers of patients at high-risk for HO-CDI

(advanced age or antibiotic use) could reduce the number of HO-CDI cases. The SNA analysis indicated a fundamental difference in the network structure between patients with and without HO-CDI. This difference highlighted rooms of high traffic within the HO-CDI patient intra-hospital transfer network, along with pathways of high traffic, that could be potential sites of harbored bacteria leading to infection. 40

This further observation provides an opportunity for a hospital level intervention through more stringent room cleaning regimes. The analysis of the SNA metrics in the GIS environment highlighted other important points that could guide interventions. The spatial dependency found in rooms important to the

HO-CDI intra-hospital patient transfer network further validates the idea of introducing rigorous cleaning, due to the indication that bacteria linked to HO-CDI resides within the rooms. By geographically displaying the results, these processes can be visualized by hospital policy makers. This means resources

(cleaning staff and supplies) can be dispersed more efficiently to areas of greatest need.

Future research in this area could be approached from several fronts. The first of these are expanding the entire study to another system of hospitals to test if the results were outliers. We postulate that other hospital networks would see similar associations between HO-CDI and intra-hospital transfers. The second recommendation to further this research would be to incorporate more factors into the SNA, such as finer temporality, which could lead to the implementation of real-time warning systems for rooms or pathways experiencing high intra-hospital patient transfer loads.

With multiple levels of the hospital working together to combine their efforts in lowering the rates of HO-

CDI with an evidence-based intervention, there is the potential for large reductions in infection rates. This idea of collaborative action to reduce HO-CDI rates within hospitals reflects the fundamental ideas of the disease ecology model. The interweaving of factors that affect HO-CDI is complex; however, by targeting several parts simultaneously it will be possible to see wide-spread positive results.

This project was supported by the Institute for the Design of Environments Aligned for Patient Safety

(IDEA4PS) at The Ohio State University which is sponsored by the Agency for Healthcare Research &

Quality (P30HS024379). The authors’ views do not necessarily represent the views of AHRQ.

This study was approved by the IRB under Study ID 2015H0424.

References

Barbut, F., & Petit, J. C. (2001). Epidemiology of Clostridium difficile-associated infections. Clinical Microbiology and Infection, 7(8), 405–410. https://doi.org/10.1046/j.1198-743X.2001.00289.x

Bignardi, G. E. (1998). Risk factors for Clostridium difficile infection. Journal of Hospital Infection, 40(1), 1–15. https://doi.org/10.1016/S0195-6701(98)90019-6

Boyce, J. M., & Pittet, D. (2002). Guideline for Hand Hygiene in Health-Care Settings. Recommendations of the Healthcare Infection Control Practices Advisory Committee and the HIPAC/SHEA/APIC/IDSA Hand Hygiene Task Force. American Journal of Infection Control, 30(8), S1-46. https://doi.org/12461507

Caroff, D. A., Yokoe, D. S., & Klompas, M. (2017). Evolving Insights into the Epidemiology and Control of Clostridium difficile in Hospitals. Clinical Infectious Diseases, 65(7), 1232–1238. https://doi.org/10.1093/cid/cix456

Cohen, S. H., Gerding, D. N., Johnson, S., Kelly, C. P., Loo, V. G., McDonald, L. C., … Wilcox, M. H. (2010). Clinical practice guidelines for Clostridium difficile infection in adults: 2010 update by the society for healthcare epidemiology of America (SHEA) and the infectious diseases society of America (IDSA). Infection Control and Hospital Epidemiology : The Official Journal of the Society of Hospital Epidemiologists of America, 31(5), 431–455. https://doi.org/10.1086/651706

Cox, D. R. (1958). The Regression Analysis of Binary Sequences. Journal of the Royal Statistical Society, 20(2), 215–242. https://doi.org/10.2307/2983890

DiDiodato, G., & McArthur, L. (2017). Interhospital patient transfers between Ontario’s academic and large community hospitals increase the risk of Clostridium difficile infection. American Journal of Infection Control. https://doi.org/10.1016/j.ajic.2017.08.018

ESRI. (2014). ArcGIS Desktop: Release 10.3. Redlands CA.

Fawley, W. N., Underwood, S., Freeman, J., Baines, S. D., Saxton, K., Stephenson, K., … Wilcox, M. H. (2007). Efficacy of Hospital Cleaning Agents and Germicides Against Epidemic Clostridium difficile Strains. Infection Control & Hospital Epidemiology, 28(8), 920–925. https://doi.org/10.1086/519201

Fekety, R., Kim, K. H., Brown, D., Batts, D. H., Cudmore, M., & Silva, J. (1981). Epidemiology of antibiotic-associated colitis. Isolation of clostridium difficile from the hospital environment. The American Journal of Medicine, 70(4), 906–908. https://doi.org/10.1016/0002-9343(81)90553-2

Felsen, J., & Wolarsky, W. (1940). The Hospital Epidemiologist. Hospitals, 14(41).

Fernández-Gracia, J., Onnela, J. P., Barnett, M. L., Eguíluz, V. M., & Christakis, N. A. (2017). Influence of a patient transfer network of US inpatient facilities on the incidence of nosocomial infections. Scientific Reports, 7(1). https://doi.org/10.1038/s41598-017-02245-7

Fritsch, M., & Kauffeld-Monz, M. (2010). The impact of network structure on knowledge transfer: an 42

application of social network analysis in the context of regional innovation networks. The Annals of Regional Science, 44(1), 21–38. https://doi.org/10.1007/s00168-008-0245-8

Fuerst, H. T., Lichtman, H. S., & James, G. (1965). No Title. JAMA, 194(4), 329–332.

Garner, J. S., Jarvis, W. R., Emori, T. G., Horan, T. C., & Hughes, J. M. (1988). CDC definitions for nosocomial infections, 1988. AJIC: American Journal of Infection Control, 16(3), 128–140. https://doi.org/10.1016/0196-6553(88)90053-3

Hansen, D. L., Shneiderman, B., & Smith, M. A. (2011). Analyzing Social Media Networks With NodeXL. Analyzing Social Media Networks with NodeXL. https://doi.org/10.1016/C2009-0-64028-9

Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2011). MatchIt : Nonparametric Preprocessing for Parametric Causal Inference. Journal of Statistical Software, 42(8). https://doi.org/10.18637/jss.v042.i08

Huang, S. S., Avery, T. R., Song, Y., Elkins, K. R., Nguyen, C. C., Nutter, S. K., … Lee, B. Y. (2010). Quantifying Interhospital Patient Sharing as a Mechanism for Infectious Disease Spread. Infection Control & Hospital Epidemiology, 31(11), 1160–1169. https://doi.org/10.1086/656747

Huisman, E. R. C. M., Morales, E., van Hoof, J., & Kort, H. S. M. (2012). Healing environment: A review of the impact of physical environmental factors on users. Building and Environment, 58, 70– 80. https://doi.org/10.1016/j.buildenv.2012.06.016

Iwashyna, T. J., Christie, J. D., Moody, J., Kahn, J. M., & Asch, D. A. (2009). The structure of critical care transfer networks. Medical Care, 47(7), 787–793. https://doi.org/10.1097/MLR.0b013e318197b1f5

Jump, R. L. P. (2013). Clostridium difficile infection in older adults. Aging Health, 9(4), 403–414. https://doi.org/10.2217/ahe.13.37

Kaatz, G. W., Gitlin, S. D., Schaberg, D. R., Wilson, K. H., Kauffman, C. a, Seo, S. M., & Fekety, R. (1988). Acquisition of Clostridium difficile from the hospital environment. American Journal of Epidemiology, 127(6), 1289–1294. https://doi.org/10.1093/oxfordjournals.aje.a114921

Kamel Boulos, M. N., & Berry, G. (2012). Real-time locating systems (RTLS) in healthcare: A condensed primer. International Journal of Health Geographics. https://doi.org/10.1186/1476- 072X-11-25

Kelly, C. P., Pothoulakis, C., & LaMont, J. T. (1994). Clostridium difficile colitis. The New England Journal of Medicine, 330(4), 257–262. https://doi.org/10.1056/NEJM199401273300406

Kimble, J. E., & White, J. G. (1981). On the control of germ cell development in Caenorhabditis elegans. Developmental Biology, 81(2), 208–219. https://doi.org/10.1016/0012-1606(81)90284-0

Lee, B. Y., McGlone, S. M., Song, Y., Avery, T. R., Eubank, S., Chang, C. C., … Huang, S. S. (2011). Social network analysis of patient sharing among hospitals in Orange County, California. American Journal of Public Health, 101(4), 707–713. https://doi.org/10.2105/AJPH.2010.202754

Lessa, F. C., Mu, Y., Bamberg, W. M., Beldavs, Z. G., Dumyati, G. K., Dunn, J. R., … McDonald, L. C. (2015). Burden of Clostridium difficile Infection in the United States. New England Journal of Medicine, 372(9), 825–834. https://doi.org/10.1056/NEJMoa1408913

Levin, J., Riley, L. S., Parrish, C., English, D., & Ahn, S. (2013). The effect of portable pulsed xenon ultraviolet light after terminal cleaning on hospital-associated Clostridium difficile infection in a community hospital. American Journal of Infection Control, 41(8), 746–748. https://doi.org/10.1016/j.ajic.2013.02.010

Mavoa, S., Witten, K., McCreanor, T., & O’Sullivan, D. (2012). GIS based destination accessibility via public transit and walking in Auckland, New Zealand. Journal of Transport Geography, 20(1), 15– 22. https://doi.org/10.1016/j.jtrangeo.2011.10.001

Meade, M. S. (1977). Medical geography as human ecology: the dimension of population movement. Geographical Review, 67(4), 379–393. https://doi.org/10.2307/213623

Moreno, J. L. (1934). Who Shall Survive? A New Approach to the Problem of Human Interrelations. Nervous and mental disease monograph series no 58 (Vol. 58). https://doi.org/10.2307/2084777

Nahmias, A., & Eickhoff, T. (1957). Staphylococcal infections in hospitals: recent developments in epidemiologic and laboratory investigations. New England Journal of Medicine, (265), 74–81.

Parker, H. D. (1988). The unique qualities of a geographic information system: A commentary. American Society for Photogrammetry and Remote Sensing, 54(11), 1547–1549. Retrieved from https://eserv.asprs.org/PERS/1988journal/nov/1988_nov_1547-1549.pdf

Pittet, D., Allegranzi, B., & Boyce, J. (2009). The World Health Organization Guidelines on Hand Hygiene in Health Care and their Consensus Recommendations. Infection Control and Hospital Epidemiology, 30(7), 611–622. https://doi.org/10.1086/600379

Pittet, D., Allegranzi, B., Sax, H., Dharan, S., Pessoa-Silva, C. L., Donaldson, L., & Boyce, J. M. (2006). Evidence-based model for hand transmission during patient care and the role of improved practices. Lancet Infectious Diseases. https://doi.org/10.1016/S1473-3099(06)70600-4

Pittet, D., & Boyce, J. M. (2001). Hand hygiene and patient care: pursuing the Semmelweis legacy. The Lancet Infectious Diseases, 1, 9–20. https://doi.org/10.1016/S1473-3099(09)70295-6

R Core Team. (2017). R Development Core Team. R: A Language and Environment for Statistical Computing. https://doi.org/http://www.R-project.org

Richards, T. B., Croner, C. M., Rushton, G., Brown, C. K., & Fowler, L. (1999). Geographic information systems and public health: Mapping the future. Public Health Reports. https://doi.org/10.1093/phr/114.4.359

Root, E. D. (2012). Moving Neighborhoods and Health Research Forward: Using Geographic Methods to Examine the Role of Spatial Scale in Neighborhood Effects on Health. Annals of the Association of American Geographers, 102(5), 986–995. https://doi.org/10.1080/00045608.2012.659621

Rubin, D. B. ., & Society, I. B. (1973). Matching to Remove Bias in Observational Studies Author ( s ):

MATCHING TO REMOVE BIAS IN OBSERVATIONAL. Biometric S, 29(1), 159–183.

Sax, H., Harbarth, S., Gavazzi, G., Henry, N., Schrenzel, J., Rohner, P., … Pittet, D. (2005). Prevalence and prediction of previously unknown MRSA carriage on admission to a geriatric hospital. Age and Ageing, 34(5), 456–462. https://doi.org/10.1093/ageing/afi135

Scott, J. (1988). Social Network Analysis. Sociology, 22(1), 109–127. https://doi.org/10.1177/0038038588022001007

Shorter, E. (1984). Ignaz Semmelweis: The etiology, concept, and prophylaxis of childbed fever. Medical History, 28(3), 334. https://doi.org/10.2307/3342284

Simmering, J. E., Polgreen, L. A., Campbell, D. R., Cavanaugh, J. E., & Polgreen, P. M. (2015). Hospital Transfer Network Structure as a Risk Factor for Clostridium difficile Infection. Infection Control & Hospital Epidemiology, 36(9), 1031–1037. https://doi.org/10.1017/ice.2015.130

Surveillance of Communicable Diseases of National Importance. (1963). JAMA: The Journal of the American Medical Association, 183(8), 227–229. https://doi.org/10.1001/jama.1963.03700080161155

Sydnor, E. R. M., & Perl, T. M. (2011). Hospital epidemiology and infection control in acute-care settings. Clinical Microbiology Reviews, 24(1), 141–173. https://doi.org/10.1128/CMR.00027-10

Tobler, W. R. (1970). A Computer Movie Simulating Urban Growth in the Detroit Region. Economic Geography, 46, 234. https://doi.org/10.2307/143141

Ulrich, R. S., & Zhu, X. (2007). Medical complications of intra-hospital patient transports: implications for architectural design and research. HERD, 1(1), 31–43. https://doi.org/10.1177/193758670700100113

Ulrich, R., Zimring, C., Quan, X., Joseph, A., & Choudhary, R. (2004). The Role of the Physical Environment in the Hospital of the 21st Century. Report Sponsored by The Robert Wood Johnson Foundation and The Center for Health Design. Retrieved from http://www.healthdesign.org/chd/research/role-physical-environment-hospital-21st-century?page=1

Vonberg, R. P., Kuijper, E. J., Wilcox, M. H., Barbut, F., Tüll, P., Gastmeier, P., … Wiuff, C. (2008). Infection control measures to limit the spread of Clostridium difficile. Clinical Microbiology and Infection. https://doi.org/10.1111/j.1469-0691.2008.01992.x

Wang, S. W., Chen, W. H., Ong, C. S., Liu, L., & Chuang, Y. W. (2006). RFID applications in hospitals: A case study on a demonstration RFID project in a Taiwan hospital. In Proceedings of the Annual Hawaii International Conference on System Sciences (Vol. 8). https://doi.org/10.1109/HICSS.2006.422

Waydhas, C., Schneck, G., & Duswald, K. H. (1995). Deterioration of respiratory function after intra- hospital transport of critically ill surgical patients. Intensive Care Medicine, 21(10), 784–789. https://doi.org/10.1007/BF01700959

Wenzel, R. P., Osterman, C. A., Hunting, K. J., & Gwaltney, J. M. (1976). Hospital-acquired infections I.

Surveillance in a university hospital. American Journal of Epidemiology, 103(3), 251–260. https://doi.org/10.1093/oxfordjournals.aje.a112223

Wey, T., Blumstein, D. T., Shen, W., & Jordán, F. (2008). Social network analysis of animal behaviour: a promising tool for the study of sociality. Animal Behaviour. https://doi.org/10.1016/j.anbehav.2007.06.020

Yokoe, D. S., Anderson, D. J., Berenholtz, S. M., Calfee, D. P., Dubberke, E. R., Ellingson, K. D., … Maragakis, L. L. (2014). A compendium of strategies to prevent healthcare-associated infections in acute care hospitals: 2014 Updates. American Journal of Infection Control. https://doi.org/10.1016/j.ajic.2014.07.002