The Effect of Multidisciplinary Care Teams on Intensive Care Unit Mortality

ORIGINAL INVESTIGATION

HEALTH CARE REFORM The Effect of Multidisciplinary Care Teams on Intensive Care Unit Mortality

Michelle M. Kim, MSc; Amber E. Barnato, MD, MPH; Derek C. Angus, MD, MPH; Lee F. Fleisher, MD; Jeremy M. Kahn, MD, MSc

Background: Critically ill patients are medically coming, the lowest odds of death were in intensive care units plex and may benefit from a multidisciplinary approach (ICUs) with high-intensity physician staffing and mul- to care. tidisciplinary care teams (OR, 0.78; 95% CI, 0.68-0.89 [PϽ.001]), followed by ICUs with low-intensity physi- Methods: We conducted a population-based retrospec- cian staffing and multidisciplinary care teams (OR, 0.88; tive cohort study of medical patients admitted to Penn- 95% CI, 0.79-0.97 [P=.01]), compared with hospitals with sylvania acute care hospitals (N=169) from July 1, 2004, low-intensity physician staffing but without multidisci- to June 30, 2006, linking a statewide hospital organiza- plinary care teams. The effects of multidisciplinary care tional survey to hospital discharge data. Multivariate lo- were consistent across key subgroups including pa- gistic regression was used to determine the indepen- tients with sepsis, patients requiring invasive mechani- dent relationship between daily multidisciplinary rounds cal ventilation, and patients in the highest quartile of se- and 30-day mortality. verity of illness.

Results: A total of 112 hospitals and 107 324 patients Conclusions: Daily rounds by a multidisciplinary team were included in the final analysis. Overall 30-day mor- are associated with lower mortality among medical ICU tality was 18.3%. After adjusting for patient and hospi- patients. The survival benefit of intensivist physician staff- tal characteristics, multidisciplinary care was associated ing is in part explained by the presence of multidisci- with significant reductions in the odds of death (odds ra- plinary teams in high-intensity physician-staffed ICUs. tio [OR], 0.84; 95% confidence interval [CI], 0.76-0.93 [P=.001]). When stratifying by intensivist physician staff- Arch Intern Med. 2010;170(4):369-376

ORE THAN 4 MILLION IN- A potential complement to intensivist tensive care unit (ICU) staffing is a multidisciplinary care model in admissions occur an- which physicians, nurses, respiratory thera- nually in the United pists, clinical pharmacists, and other staff States.1 These patients members provide critical care as a team. A areM often at high risk of death: mortality for multidisciplinary approach acknowledges critical illness syndromes such as acute lung the complexities of modern critical care and injury and sepsis ranges from 25% to 50%, the important role of communication be- and 20% of Americans die while receiving tween health care providers in delivering 2-5 intensive care services. One approach to comprehensive care. Such a model is en- lowering ICU mortality is to optimize the dorsed by the Society of Critical Care Medi- cine and the American Association of Criti- See also pages cal Care Nurses.11,12 Yet, unlike intensivist 319 and 363 physician staffing, little research has sys- tematically evaluated the relationship be- organization of ICU services.6 For ex- tween multidisciplinary care and out- ample, a large body of literature indicates comes, and there are few data to justify that the presence of trained intensivist phy- widespread adoption of this approach. Ex- sicians is associated with improved sur- isting studies are generally single center in vival,7 leading many policy makers to call nature with limited ability to adjust for for expansion of the intensivist-led model variations in case-mix or temporal trends of critical care.8 Unfortunately, there are not between periods.13-15 enough trained intensivists to meet either The objective of our study was to deter- current or future demand, and only a mi- mine the independent effect of multidis- Author Affiliations are listed at nority of ICUs are currently staffed in this ciplinary care teams on the mortality of cri- the end of this article. manner.9,10 tically ill patients, using a multicenter

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 369

©2010 American Medical Association. All rights reserved. hospital-level organizational survey and patient-level out- detailed information on the exact components of the multidis- comes data. We also sought to determine the interaction ciplinary team or the training of the physician that directed rounds. between multidisciplinary care teams and intensivist phy- The secondary exposure of interest was each ICU’s physician staff- sician staffing to see if part of the benefit of intensivist staffing model. Physician staffing models were reported in the sur- ing could be explained by multidisciplinary care. We hy- vey as no intensivist, optional intensivist consult, mandatory intensivist consult, and primary intensivist management. These pothesized that multidisciplinary care teams would be groups were further categorized into high-intensity (manda- associated with improved critical care survival, particu- tory consult or primary intensivist management) and low- larly in settings without high-intensity physician staffing. intensity (optional intensivist consult or no intensivist) physician staffing, according to prior reports.7 METHODS We also sought to evaluate the relationship between multidisciplinary care and intensivist physician staffing with re- spect to their effect on mortality. We created the following 4 STUDY DESIGN AND PATIENTS groups of hospitals based on their combination of multidisciplinary care and physician staffing models: (1) low-intensity We conducted a retrospective cohort study using state dis- staffing without multidisciplinary care teams; (2) low- charge data from the Pennsylvania Health Care Cost Contain- intensity staffing with multidisciplinary care teams; (3) high- ment Council (PHC4). The PHC4 collects clinical and admin- intensity staffing with multidisciplinary care teams; and (4) high- istrative data on all patients discharged from nonfederal hospitals intensity staffing without multidisciplinary care teams. We within the Commonwealth of Pennsylvania. All discharges be- excluded hospitals in the last group (high-intensity staffing with- tween July 1, 2004, and June 30, 2006, were eligible for the analy- out multidisciplinary care teams) because there were not enough sis. We excluded pediatric hospitals, rehabilitation hospitals, hospitals to accurately estimate outcomes in that category. long-term acute care hospitals, and specialty surgery hospi- The primary outcome variable was mortality within 30 days tals. Patient level data were linked to the Pennsylvania Depart- of hospital admission. We controlled for potential confound- ment of Health’s death records to obtain each patient’s vital sta- ing variables that could be related to the multidisciplinary care tus at 30 days after admission. Hospital characteristics were teams, physician staffing model, and mortality. Risk-adjust- obtained from the hospital-level data file maintained by the PHC4 ment variables were selected a priori and included age, sex, ad- and the 2005 American Hospital Association Annual Survey. mission source (emergency department, interhospital trans- Data on ICU care models were obtained from a cross- fer, or direct), chronic conditions as defined by Elixhauser sectional, mixed-mode organizational survey of Pennsylvania hos- comorbidities modeled as indicator covariates, ventilation sta- 16 pitals. The survey was conducted between June 1, 2005, and tus on admission as defined by ICD-9-CM procedure codes, pri- May 31, 2006, and completed by each hospital’s chief nursing mary diagnosis as defined by ICD-9-CM diagnosis codes, hos- officer. A total of 118 hospitals completed the survey (69.8%). pital teaching status as determined by each hospital’s resident Four hospitals completed the survey but did not respond to ques- to bed ratio (nonteaching status: ratio, 0; small teaching: ratio, tions about ICU care models, resulting in 114 hospitals with com- Ͼ0 to 0.2; large teaching status: ratio, Ͼ0.2), ICU type (medi- plete responses. Responding and nonresponding hospitals were cal, mixed medical-surgical, or mixed medical-coronary), re- similar in bed size, community size, teaching status, and other gion of Pennsylvania as defined by the PHC4, and each hospi- 16 key characteristics. All questions about ICU structure and or- tal’s mean annual admission volume.17-20 We further controlled ganization were specific to the hospital’s single ICU that treats for severity of illness using the MediQual Atlas probability of the majority of adult, noncardiac, nonsurgical patients. Conse- in-hospital death, a validated risk-adjustment tool for hospi- quently, this analysis was restricted to medical patients and the talized patients using key clinical and demographic variables single ICU in each hospital that primarily served those patients, measured on admission.21 Reported areas under the curve for typically either a medical or mixed medical-surgical ICU. MediQual Atlas mortality prediction in medical patients range We identified patients admitted to an ICU using revenue codes from 0.837 to 0.874, which are comparable to other common specific to intensive care. We excluded patients younger than ICU risk-adjustment systems.22,23 The MediQual Atlas score is 18 years at the time of admission and patients in hospitals who automatically calculated by PHC4 on patients admitted to Penn- did not fully respond to the survey. In addition, because we only sylvania hospitals but may be absent due to missing clinical data. had organizational data on the single, primary noncardiac nonsurgical ICU at each hospital, we excluded patients with a primary cardiac, surgical, or neurological diagnosis as defined by STATISTICAL ANALYSIS International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) discharge codes and discharge diag- We compared descriptive statistics for hospitals and patients nosis related groups. Because the discharge data do not specify by care model group using the ␹2 test or the t test, as appro- the actual admission ICU, for hospitals with more than 1 ICU priate. To determine the independent effect of multidisci- (27.7% of total) it is possible that some excluded patients were plinary care and high-intensity physician staffing on 30-day mor- actually admitted to the ICU of interest. To evaluate for pos- tality, we created patient-level multivariate logistic regression sible selection bias, we compared the mortality of excluded pa- models controlling for potential confounders described in the tients in hospitals with the different types of multidisciplinary preceding subsection. We modeled categorical variables using care models described in the following subsection. indicator covariates and continuous variables using quadratic splines. We created 3 separate models: a model with multidis- VARIABLES AND RISK ADJUSTMENT ciplinary care teams alone (model 1), a model with physician staffing alone (model 2), and a model with the grouped mul- The primary exposure of interest was each hospital’s response tidisciplinary care team/physician staffing variable (model 3). to the question: “Does the ICU have daily multidisciplinary ICU The last model was designed to evaluate the interaction be- rounds consisting of the physician, nurse, and other health care tween high-intensity physician staffing and multidisciplinary professionals (eg, social worker, respiratory therapist, pharma- care, given that we could not control for both in a single mul- cist)?” This response was coded as either yes or no. Each study tivariate model. In all models we used generalized estimating ICU operated under a single care model. We did not have more equations with robust Huber-White confidence intervals (CIs)

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 370

135 923 Survey nonresponse

7699 Hospitals with high-intensity staffing but no multidisciplinary care teams

327 520 Patients

113 623 Patients: 96 449 Patients: 117 448 Patients: Low-intensity staffing Low-intensity staffing High-intensity staffing No multidisciplinary care teams Multidisciplinary care teams Multidisciplinary care teams

74 074 Excluded 62 101 Excluded 84 021 Excluded 35 790 Surgical 34 901 Surgical 56 189 Surgical 30 396 Cardiac 19 091 Cardiac 16 054 Cardiac 6171 Neuro 5587 Neuro 9157 Neuro 105 OB 703 OB 143 OB 1612 Trauma 1819 Trauma 2478 Trauma

39 549 Medical patients included 34 348 Medical patients included 33 427 Medical patients included in analysis in analysis in analysis

Figure. Flow diagram of patients into the study. OB indicates obstetric; Neuro, neurological. to account for potential clustering within hospitals.24 We as- was low-intensity staffingϩno multidisciplinary care in sessed model discrimination using the C statistic. We also per- 54 hospitals (48.2%); low-intensity staffing and mul- formed 3 prospectively defined subgroup analyses to examine tidisciplinary care in 36 hospitals (32.1%); and high- the effects of staffing models on high-risk patients. Subgroups intensity staffingϩmultidisciplinary care in 22 hospi- of interest were patients in the top quartile of the MediQual tals (19.6%). Among excluded patients, mortality was Atlas score, patients receiving invasive mechanical ventilation similar between the different staffing groups (low- as defined by ICD-9-CM procedure code, and patients with se- ϩ vere sepsis as defined using previously validated criteria.2,25 intensity staffing no multidisciplinary care, 8.0%; and To account for missing MediQual Atlas scores we per- low-intensity staffingϩmultidisciplinary care, 9.3%; and formed multiple imputation using Markov Chain Monte Carlo high-intensity staffingϩmultidisciplinary care, 8.7%). simulation, creating 10 imputed data sets and combing regres- Hospital characteristics are given in Table 1. High- sion results according to a previously described method.26,27 We intensity physician staffing and multidisciplinary care also performed a sensitivity analysis, dropping patients with miss- teams were more common in teaching hospitals and hos- ing MediQual Atlas scores under the assumption that the data pitals with critical care fellowships. High-intensity phy- are missing completely at random (ie, purely for administrative sician staffing and multidisciplinary care teams were also rather than clinical reasons).28 We analyzed only the complete cases for the subgroup of patients in the highest quartile of more common in large hospitals and hospitals with a high MediQual Atlas score. volume of annual admissions. In hospitals with high- The multiple imputation was performed with SAS 9.2 (SAS intensity staffing, the ICU of interest tended to be a medi- Institute Inc, Cary, North Carolina). Other statistical analyses were cal specialty ICU, compared with hospitals with low- performed with Stata 10.0 (StataCorp, College Station, Texas). intensity staffing, where the ICU of interest tended to be All tests were 2 tailed, and PՅ.05 was considered significant. This a mixed medical-surgical or mixed medical-cardiac ICU. research was approved by the institutional review boards of the High-intensity physician-staffed ICUs also tended to be University of Pennsylvania and the University of Pittsburgh. larger than low-intensity staffed ICUs. Patient demographics were generally similar between RESULTS groups (Table 2). Patients in ICUs with high-intensity physician staffing and multidisciplinary care were more likely A total of 471 112 patients were admitted to Pennsylva- to require mechanical ventilation, were more likely to carry nia hospital ICUs during the study period (Figure). We a diagnosis of sepsis, and had a higher probability of in- excluded 55 hospitals with survey nonresponse or in- hospital death. Accordingly, unadjusted in-hospital mor- complete response, and 2 hospitals with high-intensity tality was higher in ICUs with high-intensity physician staff- physician staffing and no multidisciplinary care teams. ing and multidisciplinary care teams (16.4%) compared with The further exclusion of patients with nonmedical diag- hospitals with low-intensity staffing and multidisci- noses resulted in 112 hospitals and 107 324 patients in plinary care teams (13.9%) and low-intensity staffing with- the final analysis. The care model in the ICU of interest out multidisciplinary care teams (11.2%).

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 371

Hospital Characteristics Teaching status, No. (%) Nonteaching 42 (78) 22 (61) 7 (32) Small teaching 11 (20) 9 (25) 8 (36) .001 Large teaching 1 (2) 5 (14) 7 (32) Critical care fellowship, No. (%) 1 (2) 3 (8) 4 (18) .04 No. of beds, median (IQR) 128 (77-208) 198 (84-311) 286 (144-645) Ͻ.001 ICUs in hospital, median (IQR) [range], No. 1 (1-1) [1-5] 1 (1-2) [1-5] 1 (1-3) [1-5] .002 Ownership, No. (%) Nonprofit 52 (96) 36 (100) 19 (86) .06 For profit 2 (4) 0 3 (14) Annual nonsurgical ICU admissions, median (IQR), No. 272 (147-402) 380 (196-705) 588 (304-1103) .04 ICU characteristics ICU type, No. (%) Medical 2 (4) 4 (11) 9 (41) Combined medical-coronary 19 (35) 9 (25) 2 (9) .001 Combined medical-surgical 33 (61) 23 (64) 11 (50) No. of beds, median (IQR) 11 (6-16) 15 (9-29) 21 (16-48) Ͻ.001

Abbreviations: ICU, intensive care unit; IQR, interquartile range.

In the primary analysis all multivariate models had C a stratified model that included intensivist physician staff- statistic values of 0.85 or greater. Controlling for hospital ing, multidisciplinary care was associated with a signifi- and patient characteristics and accounting for clustering cant mortality reduction in ICUs with low-intensity phy- by center, but not accounting for intensivist staffing, mul- sician staffing, conveying a decreased risk of death that tidisciplinary care teams were associated with a 16% re- approached, but did not equal, ICUs with high-intensity duction in the odds of death (odds ratio [OR], 0.84; 95% physician staffing. These results suggest that in hospitals CI, 0.76-0.93 [P=.001]) (Table 3). High-intensity phy- without high-intensity physician staffing multidisci- sician staffing alone was associated with a similar reduc- plinary rounds are likely to improve patient outcomes. tion in the odds of death (OR, 0.84; 95% CI, 0.75-09.4 Several mechanisms may explain these findings. Mul- [P=.002]). When we simultaneously evaluated multidis- tidisciplinary rounds may facilitate implementation of best ciplinary care teams and high-intensity physician staffing clinical practices such as evidence-based treatments for in a stratified model, the lowest odds of death were in ICUs acute lung injury, sepsis, and prevention of ICU com- with both high-intensity physician staffing and multidis- plications.29-31 Pharmacist participation on rounds is as- ciplinary care teams (OR, 0.78; 95% CI, 0.68-0.89 sociated with fewer adverse drug events32 and alone may Ͻ [P .001]), followed by ICUs with multidisciplinary care be associated with lower mortality among ICU pa- teams and low-intensity physician staffing (OR, 0.88; 95% tients.33 Multidisciplinary rounds may also improve com- CI, 0.79-0.97 [P=.01]), compared with hospitals with- munication between health care providers.34 Communi- out either multidisciplinary care teams or low-intensity cation may facilitate implementation of respiratory therapy staffing. and nurse-driven protocols for weaning and sedation, In the subgroup analyses, C statistic values ranged from which can reduce duration of mechanical ventilation and 0.71 to 0.78. Results were similar in the 3 planned sub- shorten ICU length of stay.35,36 group analyses, with significant mortality reductions ob- Our findings have important implications for the or- served with multidisciplinary care teams and high- ganization of critical care services. First, this study pro- intensity physician staffing in patients with sepsis, patients vides empirical evidence to support a multidisciplinary requiring mechanical ventilation, and patients in the high- model of critical care. Based on these results and expert est quartile of severity of illness (Table 4). Odds ratios opinion voiced in consensus guidelines, it is reasonable from the complete case analyses excluding the 21 038 pa- for hospitals to implement routine multidisciplinary tients with missing MediQual Atlas scores (19.6% of total) rounds when staffing capabilities allow.11 In addition, our scores were all within 0.01 of the ORs from the multiple results provide insight into ways to improve mortality imputation analyses, with similar 95% CIs. in ICUs without intensivist physician staffing. Work- force analyses suggest that there are not enough inten- COMMENT sivists to meet demand, and as a consequence only a mi- nority of ICUs in the United States are staffed by trained In a large population-based sample of hospitals, daily intensivists.9,10 Directors of ICUs report that lack of enough rounds by a multidisciplinary care team were indepen- trained of intensivists is a key barrier to implementing dently associated with lower mortality in ICU patients. In an intensivist model of care.37 Our study shows that hos-

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 372

Low-Intensity Low-Intensity High-Intensity Physician Staffing؉No Physician Staffing؉ Physician Staffing؉ Multidisciplinary Care Multidisciplinary Care Multidisciplinary Care Variable (n=39 549) (n=34 348) (n=33 427) P Value Age, mean (SD), y 65.4 (17.9) 64.3 (18.6) 62.0 (17.8) Ͻ.001 Female, % 50.7 50.7 48.9 Ͻ.001 Race, % White 90.6 82.8 72.8 Black 5.8 12.8 18.2 Ͻ.001 Other 3.6 4.4 9.1 Admission source, % Emergency department 76.1 85.1 66.3 Direct admission 18.8 10.2 20.0 Hospital transfer 1.0 2.8 6.5 Ͻ.001 SNF transfer 2.6 1.2 1.6 Other facility 1.5 0.7 5.6 Charlson comorbidity index, % 0 21.4 22.5 20.1 1-2 45.0 43.8 43.9 Ͻ.001 3-4 21.0 20.6 21.3 Ն5 12.6 13.1 14.7 Primary diagnosis, % COPD/asthma 5.8 4.9 3.6 Gastrointestinal—other 6.1 4.9 4.5 Gastrointestinal bleed 4.9 4.9 4.2 HIV 0.2 0.3 0.5 Hypertension 3.1 3.0 2.5 Complication 4.4 4.1 6.8 General 20.1 17.7 16.0 Infection 4.1 4.0 3.3 Liver disease 2.1 2.3 3.7 Ͻ.001 Neurological infection 0.3 0.4 0.7 Neoplasm 3.7 4.0 6.7 Overdose 6.6 8.7 6.3 Pneumonia 9.6 8.8 6.3 Psychiatric 1.6 2.2 1.4 Pulmonary—other 5.2 5.6 5.2 Respiratory failure 10.9 12.9 14.8 Sepsis 8.2 9.0 10.6 Vascular 2.5 2.5 3.0 Requiring mechanical ventilation, % 17.0 26.2 31.3 Ͻ.001 Sepsis, % 7.4 10.8 14.0 Ͻ.001 ICU length of stay, median (IQR), d 2 (1-4) 2 (1-5) 2 (1-5) Ͻ.001 Hospital length of stay, median (IQR), d 5 (3-9) 5 (3-10) 6 (3-11) Ͻ.001 Unadjusted in-hospital mortality, % 11.2 13.9 16.4 Ͻ.001 MediQual Atlas predicted death probability, %a Score 0.00-0.05 5.0 3.7 2.6 Score 0.05-0.10 29.4 26.7 25.0 Score 0.10-0.15 17.1 16.4 16.3 Ͻ.001 Score 0.15-0.25 19.3 19.5 19.3 Score 0.25-1.00 29.2 33.9 36.8 Discharge location for survivors, % Home 62.1 55.8 61.6 Other hospital 10.0 10.1 5.6 SNF 18.5 22.0 19.0 Ͻ.001 LTAC 1.5 2.5 3.7 Hospice 1.9 3.0 2.7 Other 6.0 6.6 7.4

Abbreviations: COPD, chronic obstructive pulmonary disease; HIV, human immunodeficiency virus; IQR, interquartile range; LTAC, long-term acute care hospital; SNF, skilled nursing facility. a Percentages are reflective of patients with nonmissing data (n=86 286). pitals without the ability to implement high-intensity phy- Our results also confirm prior studies showing that sician staffing can still achieve significant mortality re- high-intensity physician staffing lowers mortality in the ductions by implementing a multidisciplinary, team- ICU. Several cohort studies indicate that intensivist-led based approach. critical care is associated with improved outcomes in the

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 373

©2010 American Medical Association. All rights reserved. ICU.7,38,39 The benefit of intensivists was recently called achieve superior outcomes.44 Our results show that the into question by a study suggesting higher mortality in benefit of intensivists is due, at least in part, to the mul- ICUs staffed by intensivists.40 Although the causes of this tidisciplinary care models typically found in intensivist- discrepant finding are unknown, possibilities include the led ICUs. self-selected nature of hospitals in the cohort, selective We did not have detailed information about the char- referral of high-risk patients to intensivists, and use of acteristics of the multidisciplinary team, such as team size, in-hospital rather than 30-day mortality, which can lead the training and experience of the physician leading the to discharge bias.41-43 We demonstrate a benefit from in- team, or the exact ancillary staff members comprising the tensivist staffing in a population-based cohort of hospi- team. Given the wording of our questionnaire, we ex- tals consistent with prior reports. In addition, our study pected that at a minimum the multidisciplinary teams in- expands the literature on intensivist staffing by demon- cluded the primary physician, the bedside nurse, and at strating a potential mechanism for the effect. Despite the least 1 other health care provider. Still, although our study wealth of literature on intensivist physician staffing, few provides empirical support for a multidisciplinary ap- studies are directed at understanding how intensivists proach to care, we are unable to identify either specific attributes of the multidisciplinary team or an optimal team size that may be associated with improved outcomes. Table 3. Association Between Intensivist Physician Staffing Questions remain about how medical teams function, and a and 30-Day Mortality for All Patients even what defines a medical team in practice. These top- ics are important areas for future research. We also do Variable OR (95% CI) P Value not know the ideal role of the physician in multidisci- Model 1: multidisciplinary care plinary rounds. In some ICUs a single intensivist- staffing alone trained physician may direct rounds on all ICU pa- No multidisciplinary care 1 [Reference] Multidisciplinary care 0.84 (0.76-0.93) .001 tients, while in others multiple physicians of varying Model 2: intensivist physician staffing backgrounds may lead rounds on their patients at dif- alone ferent times. Each of these organizational styles would Low intensity 1 [Reference] meet our definition of multidisciplinary rounds. Future High intensity 0.84 (0.75-0.94) .002 work should empirically examine the benefits and limi- Model 3: interaction between intensivist physician staffing and tations of these different care models. multidisciplinary care teams Our analysis has several limitations. First, we only had Low intensityϩno 1 [Reference] data on care models at a single ICU at each hospital— multidisciplinary care the ICU primarily providing care to nonsurgical, non- ϩ Low intensity multidisciplinary 0.88 (0.79-0.97) .01 cardiac patients. We excluded patients unlikely to have care High intensityϩ 0.78 (0.68-0.89) Ͻ.001 received care in that ICU. These results do not necessar- multidisciplinary care ily generalize to surgical, cardiac, and neurological patients, or to specialty ICUs serving those populations. Be- Abbreviations: CI, confidence interval; OR, odds ratio. cause the PHC4 discharge data do not specify exactly in a Estimates are adjusted for age, sex, admission source, Elixhauser comorbidities, mechanical ventilation status, MediQual Atlas severity score, which ICU the patient received care, in hospitals with primary diagnosis, teaching status, intensive care unit type, region, and more than 1 ICU it is possible that we excluded some annual volume (N=107 301 patients). patients who received care in the ICU of interest and in-

Table 4. Planned Subgroup Analyses for Association Between Intensivist Physician Staffing and 30-Day Mortalitya

Subgroup

Top Quartile of MediQual Ventilated Patients Atlas Score Patients With Sepsis (n=26 177) (n=21 591) (n=11 321)

Variable OR (95% CI) P Value OR (95% CI) P Value OR (95% CI) P Value Model 1: multidisciplinary care staffing alone No multidisciplinary care 1 [Reference] 1 [Reference] 1 [Reference] Multidisciplinary care 0.81 (0.71-0.92) .001 0.83 (0.74-0.93) .001 0.80 (0.70-0.92) .002 Model 2: intensivist physician staffing alone Low intensity 1 [Reference] 1 [Reference] 1 [Reference] High intensity 0.83 (0.71-0.96) .02 0.82 (0.73-0.91) Ͻ.001 0.81 (0.70-0.94) .005 Model 3: interaction between intensivist physician staffing and multidisciplinary care Low intensityϩno multidisciplinary care 1 [Reference] 1 [Reference] 1 [Reference] Low intensityϩmultidisciplinary care 0.85 (0.74-0.96) .01 0.88 (0.78-0.99) .03 0.85 (0.73-0.98) .03 High intensityϩmultidisciplinary care 0.74 (0.63-0.88) .001 0.76 (0.67-0.85) Ͻ.001 0.74 (0.63-0.87) Ͻ.001

Abbreviations: CI, confidence interval; OR, odds ratio. a Estimates are adjusted for age, sex, admission source, Elixhauser comorbidities, mechanical ventilation status, MediQual Atlas severity score, primary diagnosis, teaching status, intensive care unit type, region, and annual volume.

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 374

©2010 American Medical Association. All rights reserved. cluded some patients who did not. The finding that mor- Barnato) and grant K23HL082650 from the National tality was similar among excluded patients in each group Heart, Lung, and Blood Institute (Dr Kahn), National In- suggests that any misclassification bias would be mini- stitutes of Health, and a grant from the Leonard Davis mal. Second, we did not have organizational data on 55 Institute of Health Economics (Dr Kahn). hospitals that not fully complete our survey. Although Role of the Sponsors: The sponsors had no role in the survey respondents were similar to nonrespondents, re- design and conduct of the study; collection, manage- sponse bias cannot be ruled out. Finally, we were un- ment, analysis, and interpretation of the data; and prepa- able to observe the effects of high-intensity physician staff- ration, review, or approval of the manuscript. ing models without multidisciplinary care teams owing Disclaimer: The following statement is provided and to a low number of hospitals in this category. We can- required by the Pennsylvania Health Care Cost Con- not comment on whether multidisciplinary care teams tainment Council (PHC4): The PHC4 has provided improve outcomes within high-intensity physician- these data in an effort to further the PHC4’s mission of staffed ICUs. educating the public and containing health care costs With the aging of the population, demand for critical in Pennsylvania. The PHC4, its agents and staff, have care is certain to rise in the coming years. Evidence- made no representation, guarantee, or warranty, ex- based strategies on how to best organize and manage ICUs pressed or implied, that the data—financial-, patient-, are needed.45 We demonstrate that daily rounds by a mul- payer-, and physician-specific information—are error tidisciplinary care team are associated with lower mor- free or that the use of the data will avoid differences of tality in the ICU. Clinicians, hospital administrators, and opinion or interpretation or disputes with those who policy makers can use these results to help optimally or- use published reports or purchased data. The PHC4, ganize critical care services and potentially improve out- its agents and staff, will bear no responsibility or comes for critically ill patients in hospitals where inten- liability for the results of the analysis or consequences sivist staffing is not available. of its use. Previous Presentation: This work was presented in ab- Accepted for Publication: November 9, 2009. stract form at the Seventh World Congress on Health Eco- Author Affiliations: Department of Health Care Man- nomics; July 14, 2009; Beijing, China. agement and Economics, Wharton School of Business (Ms Kim), Leonard Davis Institute of Health Economics (Drs Fleisher and Kahn), Department of Anesthesia and Criti- REFERENCES cal Care, School of Medicine (Dr Fleisher), Center for Clinical Epidemiology and Biostatistics, School of Medi- 1. Halpern NA, Pastores SM, Greenstein RJ. Critical care medicine in the United cine (Drs Fleisher and Kahn), and Division of Pulmo- States 1985-2000: an analysis of bed numbers, use, and costs. Crit Care Med. 2004;32(6):1254-1259. nary, Allergy, and Critical Care, School of Medicine (Dr 2. Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Kahn), University of Pennsylvania, Philadelphia; and Sec- Epidemiology of severe sepsis in the United States: analysis of incidence, out- tion of Decision Sciences and Clinical Systems Model- come, and associated costs of care. Crit Care Med. 2001;29(7):1303-1310. ing, Division of General Internal Medicine, School of 3. Angus DC, Barnato AE, Linde-Zwirble WT, et al; Robert Wood Johnson Founda- Medicine (Dr Barnato), CRISMA Laboratory, Depart- tion ICU End-Of-Life Peer Group. Use of intensive care at the end of life in the United States: an epidemiologic study. Crit Care Med. 2004;32(3):638-643. ment of Critical Care Medicine, School of Medicine (Drs 4. Rubenfeld GD, Caldwell E, Peabody E, et al. Incidence and outcomes of acute Barnato and Angus), and Department of Health Policy lung injury. N Engl J Med. 2005;353(16):1685-1693. and Management, Graduate School of Public Health (Drs 5. Erickson SE, Martin GS, Davis JL, Matthay MA, Eisner MD; NIH NHLBI ARDS Barnato and Angus), University of Pittsburgh, Pitts- Network. Recent trends in acute lung injury mortality: 1996-2005. Crit Care Med. 2009;37(5):1574-1579. burgh, Pennsylvania. 6. Carmel S, Rowan K. Variation in intensive care unit outcomes: a search for the Correspondence: Jeremy M. Kahn, MD, MSc, Center for evidence on organizational factors. Curr Opin Crit Care. 2001;7(4):284-296. Clinical Epidemiology and Biostatistics, University of 7. Pronovost PJ, Angus DC, Dorman T, Robinson KA, Dremsizov TT, Young TL. Pennsylvania School of Medicine, 723 Blockley Hall, 423 Physician staffing patterns and clinical outcomes in critically ill patients: a sys- Guardian Dr, Philadelphia, PA 19104 (jmkahn@mail tematic review. JAMA. 2002;288(17):2151-2162. 8. Milstein A, Galvin RS, Delbanco SF, Salber P, Buck CR Jr. Improving the safety .med.upenn.edu). of health care: the leapfrog initiative. Eff Clin Pract. 2000;3(6):313-316. Author Contributions: Dr Kahn had full access to all the 9. Angus DC, Kelley MA, Schmitz RJ, White A, Popovich J Jr; Committee on Man- data in the study and takes responsibility for the integ- power for Pulmonary and Critical Care Societies (COMPACCS). Current and pro- rity of the data and the accuracy of the data analysis. Study jected workforce requirements for care of the critically ill and patients with pulmonary disease: can we meet the requirements of an aging population? JAMA. concept and design: Kim, Barnato, Angus, Fleisher, and 2000;284(21):2762-2770. Kahn. Acquisition of data: Barnato and Kahn. Analysis and 10. Angus DC, Shorr AF, White A, Dremsizov TT, Schmitz RJ, Kelley MA; Committee interpretation of data: Kim, Barnato, and Kahn. Drafting on Manpower for Pulmonary and Critical Care Societies (COMPACCS). of the manuscript: Kim and Angus. Critical revision of the Critical care delivery in the United States: distribution of services and compli- manuscript for important intellectual content: Barnato, An- ance with Leapfrog recommendations. Crit Care Med. 2006;34(4):1016-1024. 11. Brilli RJ, Spevetz A, Branson RD, et al; American College of Critical Care Medi- gus, Fleisher, and Kahn. Statistical analysis: Kim and Kahn. cine Task Force on Models of Critical Care Delivery; the American College of Criti- Obtained funding: Barnato, Fleisher, and Kahn. Adminis- cal Care Medicine Guidelines for the Definition of an Intensivist and the Practice trative, technical, and material support: Barnato, Angus, of Critical Care Medicine. Critical care delivery in the intensive care unit: defining and Kahn. Study supervision: Kahn. clinical roles and the best practice model. Crit Care Med. 2001;29(10):2007- 2019. Financial Disclosure: None reported. 12. Durbin CG Jr. Team model: advocating for the optimal method of care delivery Funding/Support: This work was funded by grant in the intensive care unit. Crit Care Med. 2006;34(3)(suppl):S12-S17. K08AG21921 from the National Institute on Aging (Dr 13. Burns SM, Earven S, Fisher C, et al; University of Virginia Long Term Mechani-

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 375

©2010 American Medical Association. All rights reserved. cal Ventilation Team. Implementation of an institutional program to improve clini- 30. Ferrer R, Artigas A, Levy MM, et al; Edusepsis Study Group. Improvement in pro- cal and financial outcomes of mechanically ventilated patients: one-year out- cess of care and outcome after a multicenter severe sepsis educational program comes and lessons learned. Crit Care Med. 2003;31(12):2752-2763. in Spain. JAMA. 2008;299(19):2294-2303. 14. Young MP, Gooder VJ, Oltermann MH, Bohman CB, French TK, James BC. 31. Pronovost P, Needham D, Berenholtz S, et al. An intervention to decrease catheter- The impact of a multidisciplinary approach on caring for ventilator-dependent related bloodstream infections in the ICU. N Engl J Med. 2006;355(26):2725- patients. Int J Qual Health Care. 1998;10(1):15-26. 2732. 15. Sebat F, Johnson D, Musthafa AA, et al. A multidisciplinary community hospital 32. Leape LL, Cullen DJ, Clapp MD, et al. Pharmacist participation on physician rounds program for early and rapid resuscitation of shock in nontrauma patients. Chest. and adverse drug events in the intensive care unit. JAMA. 1999;282(3):267- 2005;127(5):1729-1743. 270. 16. Lin CY, Farrell MH, Lave JR, Angus DC, Barnato AE. Organizational determinants of 33. MacLaren R, Bond CA, Martin SJ, Fike D. Clinical and economic outcomes of hospital end-of-life treatment intensity. Med Care. 2009;47(5):524-530. involving pharmacists in the direct care of critically ill patients with infections. 17. Zimmerman JE, Kramer AA, McNair DS, Malila FM. Acute Physiology and Chronic Crit Care Med. 2008;36(12):3184-3189. Health Evaluation (APACHE) IV: hospital mortality assessment for today’s criti- 34. Shortell SM, Zimmerman JE, Rousseau DM, et al. The performance of intensive cally ill patients. Crit Care Med. 2006;34(5):1297-1310. care units: does good management make a difference? Med Care. 1994;32 18. Kahn JM, Goss CH, Heagerty PJ, Kramer AA, O’Brien CR, Rubenfeld GD. Hos- (5):508-525. pital volume and the outcomes of mechanical ventilation. N Engl J Med. 2006; 35. Ely EW, Baker AM, Dunagan DP, et al. Effect on the duration of mechanical ven- 355(1):41-50. tilation of identifying patients capable of breathing spontaneously. NEnglJMed. 19. Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comor- 1996;335(25):1864-1869. bidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11): 36. Kress JP, Pohlman AS, O’Connor MF, Hall JB. Daily interruption of sedative in- 1130-1139. fusions in critically ill patients undergoing mechanical ventilation. NEnglJMed. 20. Quan H, Parsons GA, Ghali WA. Validity of procedure codes in International Clas- 2000;342(20):1471-1477. sification of Diseases, 9th Revision, Clinical Modification administrative data. Med 37. Kahn JM, Matthews FA, Angus DC, Barnato AE, Rubenfeld GD. Barriers to imple- Care. 2004;42(8):801-809. menting the Leapfrog Group recommendations for intensivist physician staff- 21. Iezzoni LI, Moskowitz MA. A clinical assessment of MedisGroups. JAMA. 1988; ing: a survey of intensive care unit directors. J Crit Care. 2007;22(2):97-103. 260(21):3159-3163. 38. Treggiari MM, Martin DP, Yanez ND, Caldwell E, Hudson LD, Rubenfeld GD. Effect 22. Iezzoni LI. The risks of risk adjustment. JAMA. 1997;278(19):1600-1607. of intensive care unit organizational model and structure on outcomes in patients 23. Kuzniewicz MW, Vasilevskis EE, Lane R, et al. Variation in ICU risk-adjusted mor- with acute lung injury. Am J Respir Crit Care Med. 2007;176(7):685-690. tality: impact of methods of assessment and potential confounders. Chest. 2008; 133(6):1319-1327. 39. Nathens AB, Rivara FP, Mackenzie EJ, et al. The impact of an intensivist-model 24. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous ICU on trauma-related mortality. Ann Surg. 2006;244(4):545-554. outcomes. Biometrics. 1986;42(1):121-130. 40. Levy MM, Rapoport J, Lemeshow S, Chalfin DB, Phillips G, Danis M. Associa- 25. Barnato AE, Alexander SL, Linde-Zwirble WT, Angus DC. Racial variation in the tion between critical care physician management and patient mortality in the in- incidence, care, and outcomes of severe sepsis: analysis of population, patient, tensive care unit. Ann Intern Med. 2008;148(11):801-809. and hospital characteristics. Am J Respir Crit Care Med. 2008;177(3):279- 41. Vasilevskis EE, Kuzniewicz MW, Dean ML, et al. Relationship between discharge 284. practices and intensive care unit in-hospital mortality performance: evidence of 26. Ambler G, Omar RZ, Royston P. A comparison of imputation techniques for han- a discharge bias. Med Care. 2009;47(7):803-812. dling missing predictor values in a risk model with a binary outcome. Stat Meth- 42. Kahn JM, Kramer AA, Rubenfeld GD. Transferring critically ill patients out of hos- ods Med Res. 2007;16(3):277-298. pital improves the standardized mortality ratio: a simulation study. Chest. 2007; 27. Rubin DB, Schenker N. Multiple imputation in health-care databases: an over- 131(1):68-75. view and some applications. Stat Med. 1991;10(4):585-598. 43. Rubenfeld GD, Angus DC. Are intensivists safe? Ann Intern Med. 2008;148(11): 28. Greenland S, Finkle WD. A critical look at methods for handling missing covar- 877-879. iates in epidemiologic regression analyses. Am J Epidemiol. 1995;142(12): 44. Kahn JM, Brake H, Steinberg KP. Intensivist physician staffing and the process of 1255-1264. care in academic medical centres. Qual Saf Health Care. 2007;16(5):329-333. 29. The Acute Respiratory Distress Syndrome Network. Ventilation with lower tidal 45. Barnato AE, Kahn JM, Rubenfeld GD, et al. Prioritizing the organization and man- volumes as compared with traditional tidal volumes for acute lung injury and the agement of intensive care services in the United States: the PrOMIS Conference. acute respiratory distress syndrome. N Engl J Med. 2000;342(18):1301-1308. Crit Care Med. 2007;35(4):1003-1011.

(REPRINTED) ARCH INTERN MED/ VOL 170 (NO. 4), FEB 22, 2010 WWW.ARCHINTERNMED.COM 376