&get_box_var;
ORIGINAL ARTICLE
The Relationship between Hospital Volume and Mortality in Severe Sepsis David F. Gaieski1, J. Matthew Edwards2, Michael J. Kallan3, Mark E. Mikkelsen3,4, Munish Goyal5, and Brendan G. Carr1,3,6 1
Department of Emergency Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; 2Department of Emergency Medicine, Kaiser Permanente San Diego Medical Center, San Diego, California; 3Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine, and 4Pulmonary, Allergy, and Critical Care Division, Department of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; 5Department of Emergency Medicine, Medstar Washington Hospital Center, Washington, DC; and 6Leonard Davis Institute of Health Economics, University of Pennsylvania, Philadelphia, Pennsylvania
Abstract Rationale: Severe sepsis is increasing in incidence and has a high rate
of inpatient mortality. Hospitals that treat a larger number of patients with severe sepsis may offer a survival advantage. Objectives: We sought to assess the effect of severe sepsis case volume on mortality, hypothesizing that higher volume centers would have lower rates of inpatient death. Methods: We performed a retrospective cohort study over a 7-year
period (2004–2010), using a nationally representative sample of hospital admissions, examining the relation between volume, urban location, organ dysfunction, and survival. Measurements and Main Results: To identify potential differences in outcomes, hospitals were divided into five categories (,50, 50–99, 100–249, 250–499, and 5001 annual cases) and adjusted mortality was compared by volume. A total of 914,200
Severe sepsis, defined as infection accompanied by inflammation and acute organ dysfunction or tissue hypoperfusion (1), is increasing in incidence (2, 3), has an
patients with severe sepsis were identified over a 7-year period (2004–2010). Overall in-hospital mortality was 28.1%. In a fully adjusted model, there was an inverse relationship between severe sepsis case volume and inpatient mortality. Hospitals in the highest volume category had substantially improved survival compared with hospitals with the lowest case volume (adjusted odds ratio, 0.64; 95% confidence interval, 0.60–0.69). In cases of severe sepsis with one reported organ dysfunction, a mortality of 18.9% was found in hospitals with fewer than 50 annual cases compared with 10.4% in hospitals treating 5001 cases (adjusted odds ratio, 0.54; 95% confidence interval, 0.49–0.59). Similar differences were found in patients with up to three total organ dysfunctions. Conclusions: Patients with severe sepsis treated in hospitals with
higher case volumes had improved adjusted outcomes. Keywords: severe sepsis; hospital case volume; organ dysfunction
in-hospital mortality as high as 38% (4), and costs the U.S. health system $24 billion annually (5). The Centers for Disease Control and Prevention (CDC, Atlanta, GA) currently
lists septicemia as the eleventh leading cause of death in the United States (6), and the burden of this disease is expected to increase as the population ages (7).
( Received in original form February 16, 2014; accepted in final form August 10, 2014 ) Supported by an unrestricted educational grant from the Beatrice Wind Gift Fund. Role of funding agency: The Beatrice Wind Gift Fund had no role in the conception of this study, the collection or analysis of the data, or the writing, revision, and submission of the manuscript. No one from the Beatrice Wind Gift Fund has reviewed the manuscript before submission. Author Contributions: D.F.G. had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: D.F.G., J.M.E., M.J.K., B.G.C., M.G., and M.E.M.; acquisition of data: M.J.K. and B.G.C.; analysis and interpretation of data: D.F.G., J.M.E., M.J.K., B.G.C., and M.E.M.; drafting of manuscript: J.M.E., M.J.K., and D.F.G.; critical revision of the manuscript for important intellectual content: D.F.G., B.G.C., M.G., and M.E.M.; statistical analysis: M.J.K.; obtaining funding: D.F.G.; administrative, technical, or material support: D.F.G., M.J.K., and B.G.C.; study supervision: D.F.G. Correspondence and requests for reprints should be addressed to David F. Gaieski, M.D., Perelman School of Medicine at the University of Pennsylvania, Department of Emergency Medicine, Center for Resuscitation Science, Ground Ravdin, 3400 Spruce Street, Philadelphia, PA 19104. E-mail: gaieskid@uphs. upenn.edu This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org Am J Respir Crit Care Med Vol 190, Iss 6, pp 665–674, Sep 15, 2014 Copyright © 2014 by the American Thoracic Society Originally Published in Press as DOI: 10.1164/rccm.201402-0289OC on August 12, 2014 Internet address: www.atsjournals.org
Gaieski, Edwards, Kallan, et al.: Hospital Factors and Mortality in Severe Sepsis
665
ORIGINAL ARTICLE
At a Glance Commentary Scientific Knowledge on the Subject: Little is known about the
relationship between severe sepsis case volume at hospitals and mortality outcomes. We hypothesized that hospitals treating larger numbers of patients with severe sepsis would have lower in-hospital severe sepsis mortality. What This Study Adds to the Field: Our research confirmed our
hypothesis and has important implications about regionalization of severe sepsis care and optimizing costs and outcomes for severe sepsis. Survival from severe sepsis and septic shock requires early recognition and treatment (8–14). Patients with other timesensitive emergency conditions including trauma, stroke, cardiac arrest, and myocardial infarction are known to have better outcomes when treated in highvolume centers (15–19), which typically provide both a high level of resources and clinical experience. Given the variability in both capacity and capabilities at U.S. hospitals, disparities may exist in the care and outcomes for patients with severe sepsis and septic shock. Previous work in the United States suggests that patients with severe sepsis seen in emergency departments (EDs) treating higher numbers of sepsis patients have lower adjusted mortality rates (20). In contrast, a British study found no mortality difference in patients with severe sepsis treated in highversus low-volume intensive care units (ICUs) (21). Walkey and Wiener demonstrated that U.S. academic hospitals treating the highest volumes of patients with severe sepsis had 7% lower adjusted absolute in-hospital mortality when compared with the lowest volume academic hospitals (22). It is not known whether U.S. hospitals in general with higher severe sepsis case volumes provide a survival advantage and, if they do, which subgroups derive the greatest benefit from their expertise. In this study, we used a large nationally representative sample and hypothesized that higher volume centers seeing more severe sepsis cases would offer a survival advantage compared with lower case volume centers. We examined the association overall as well as 666
among patients with varying severity of illness, using the number of organ dysfunctions as a proxy. Some of the results of this study have been previously reported in the form of an abstract (23).
Methods Data
We performed a nationally representative retrospective cohort study from 2004 to 2010 using the Nationwide Inpatient Sample (NIS), the largest all-payer publicly available inpatient database in the United States. The NIS, developed as part of the Healthcare Cost and Utilization Project (HCUP) by the Agency for Healthcare Research and Quality (AHRQ), is a 20% stratified sample of U.S. acute care hospitals, and contains approximately 8 million hospital stays annually, broadly representing admissions to regular hospital wards as well as ICUs. The stratified sampling methodology allows for the generation of nationally representative estimates of incidence and mortality. More than 100 clinical and nonclinical elements are available from each hospital stay, including primary and secondary diagnoses, procedures, discharge status, patient demographics, and total hospital length of stay. Identification of patients as admitted to a hospital ward or an ICU is not available in the NIS. Hospital characteristics including geographic region, urban versus rural location, ownership, teaching status, and bed size are used to ensure data are nationally representative. In the NIS, urban and rural hospital designation is determined via the U.S. Census Bureau’s Core Based Statistical Area (CBSA) codes. Rural and micropolitan regions, defined as at least one population center with a population less than 50,000, are considered rural in the NIS (for further details, see http:// www.hcup-us.ahrq.gov/nisoverview.jsp). Given the deidentified nature of the administrative data, this study was declared exempt by the Institutional Review Board at the University of Pennsylvania (Philadelphia, PA). Patients
Severe sepsis was defined in accordance with consensus guidelines (1) as suspected or documented infection, systemic inflammation, and acute organ dysfunction. There are several published methods using differing approaches to capture cases of severe sepsis using administrative data (2, 4, 24, 25). However, research has demonstrated that there is significant variation in incidence, severity of
illness, number of organ dysfunctions, and mortality depending on the capture method used (3, 26). A method employing a limited number of infection and septicemia codes was found to better approximate epidemiologic chart-based studies (27). Considering these findings, we chose to employ a previously validated method, applying a limited number of infection codes (4, 5) and identifying acute organ dysfunctions that develop during the hospitalization. Patients with severe sepsis (assigned a primary or secondary International Classification of Disease, Ninth Revision [ICD-9] diagnostic code for infection and an additional ICD-9 code for acute organ dysfunction [including dysfunction of the respiratory, cardiovascular, renal, hepatic, coagulation, and central nervous systems] or patients assigned the ICD-9 codes for severe sepsis and septic shock) were collected from the NIS for analysis (see the online supplement). Patients transferred to or from another facility and those not more than 18 years of age were excluded from this analysis. We used ICD-9 codes to identify comorbidities as described by Elixhauser and colleagues, and All Patient Refined Diagnosis Related Groups (APRDRG) to adjust for severity of illness (28–30). Our primary end point was inpatient mortality. As severe sepsis can include various types and numbers of organ dysfunctions, we examined the association between the type of organ dysfunction (e.g., respiratory, renal, etc.), total number of organ systems involved, case volume, and mortality. Statistical Analyses
Descriptive statistics are presented as frequencies and percentages, with comparisons between groups made using the chi-squared (x2) test. We performed unadjusted and adjusted logistic regression modeling, with results expressed as odds ratios (ORs) with 95% confidence intervals (CIs). To account for the complex survey design and clustering within hospitals in the NIS, we used the specific statistical methodology recommended in the HCUP data documentation guide (31). This included survey statistics and Taylor series linearization methodology that estimates the covariance matrix for the unadjusted and adjusted logistic regression model coefficients. Our main set of multivariate logistic regression models was created to study the relationship between annual hospital severe sepsis case volume and inpatient mortality.
American Journal of Respiratory and Critical Care Medicine Volume 190 Number 6 | September 15 2014
ORIGINAL ARTICLE Considerations in creating annual case volume divisions included case-level versus hospital-level quartiles and ensuring a good mix of rural versus urban hospitals in each volume division. After sensitivity analyses were performed, the following annual case volume divisions were created for outcomes comparisons: fewer than 50, 50–99, 100–249, 250–499, and 5001 cases/year. We applied a stepwise modeling methodology to investigate the effect of additional groups of covariates on the (adjusted) relationship between annual hospital severe sepsis volume and mortality. This methodology increases transparency and allows the reader to appreciate how the entry of new variables into the model affects the point estimates of the preceding model. In stepwise fashion, additional adjustments were made (building on the variables introduced in the previous models) for hospital location and teaching status (model 1); age, sex, hospital region, payer status, discharge year, and race (model 2); each of the 29 AHRQ comorbidities (model 3); specific organ dysfunction information (model 4); and APR-DRG risk of mortality (model 5). A P value less than 0.05 was considered significant. To quantify the added benefit by model, likelihood statistics, presented as 2 3 normalized log-likelihood and corresponding chi-squared statistics, were employed. Each of the adjusted models is compared with the one immediately preceding it, demonstrating the incremental value of the additional variables. To address potential limitations of using categorical variables in regression models, we performed secondary analyses to examine the relationship between volume of cases as a continuous variable and outcome (32). This analysis assumes that the change is the same for any 100-case increase in severe sepsis volume (e.g., from 1 to 101 cases or from 1,401 to 1,501 cases). Because the data are skewed with few hospitals reporting more than 1,000 cases of severe sepsis per year, we also performed the secondary analysis by capping the analysis at a maximal value of 1,000 cases/year, thus preventing undue outlier effects. In addition, we performed an analysis on the fully adjusted logistic regression model that examined all combinations of annual severe sepsis volume in both rural and urban hospitals. Finally, to determine the relationship between volume and mortality for specific combinations of organ
dysfunctions, we performed a series of analyses stratified by specific organ dysfunction or groups of dysfunction; these analyses were limited to the respective subpopulations and examined the adjusted relationship of annual hospital severe sepsis volume and mortality in the fully adjusted model (model 5). In the analyses examining a specific organ dysfunction (e.g., respiratory only) or pair of organ dysfunctions (e.g., respiratory and cardiovascular), model 5 was modified by removing the adjustment for the specific organ dysfunction or dysfunctions of interest. This approach was taken to isolate the combination of organ dysfunction(s) in the context of hospital case volume, avoiding the confusion of comparing the patients with the specific organ dysfunction to the rest of the cohort, which would include patients with different combinations and/or number of organ dysfunctions. In all models, we investigated whether the interaction of rural versus urban and hospital case volume improved the fit of the fully adjusted model. All analyses were performed with SAScallable SUDAAN statistical software, version 11.0.0 (Research Triangle Institute, Research Triangle Park, NC).
Results Patient Characteristics
A total of 914,200 patients with severe sepsis was identified over a 7-year period (2004–2010), representing a weighted estimate of 4,506,106 cases nationally (Table 1). Mean in-hospital mortality was 28.1% across the study period and decreased over time (from 34.4% in 2004 to 22.9% in 2010; P , 0.001). The number of severe sepsis cases treated per year at different hospitals ranged from 1 to 2,415. Hospitals that reported fewer than 50 cases/year had unadjusted mortality rates identical to those of hospitals that treated 5001 cases/year (26.8%). The vast majority of comorbid conditions demonstrated small (in absolute terms) but statistically significant differences across volume groups (P , 0.001), with the exception of congestive heart failure (P = 0.14). Of note, hospitals in the highest volume group were much more likely to treat patients with chronic renal disease compared with the lowest volume group (27.0 vs. 17.5%, respectively; P , 0.001). For those patients determined to have severe sepsis, almost half
Gaieski, Edwards, Kallan, et al.: Hospital Factors and Mortality in Severe Sepsis
had acute dysfunction of the cardiovascular, respiratory, or renal systems (44.4, 49.5, and 49.8%, respectively). Hospitals in the highest volume group treated 19.3% more patients with respiratory dysfunction and 15.7% more with renal dysfunction than hospitals in the lowest volume group. Adjusted Mortality
In model 1, adjusting solely for hospital factors (location and teaching status), the highest volume hospitals (5001 cases/year) had lower mortality when compared with the lowest volume (,50 cases/year) hospitals, but hospitals with moderate volume (50–99, 100–249, and 250–499) were not significantly different from lowest volume hospitals. This relationship was similar in model 2 after adjustments for demographic factors (age, sex, hospital region, payer status, discharge year, and race) and in model 3 with further adjustments for individual comorbidities. After adjustment for organ dysfunction in model 4 we demonstrated significantly reduced mortality at higher versus lowest case volume hospitals (Table 2). In the fully adjusted analysis, model 5, there was a substantial inverse relationship between severe sepsis case volume and inpatient mortality. Compared with patients at hospitals that treated fewer than 50 cases of severe sepsis annually (46% of hospitals in our sample), patients at hospitals that treated 5001 cases annually have a 36% improved odds of inpatient survival (adjusted OR, 0.64; 95% CI, 0.60–0.69). Likelihood statistics demonstrated the significant improvement in model fit contributed by the respective sets of specific variables added in stepwise fashion for each of the subsequent models (model 1 vs. unadjusted, model 2 vs. model 1, model 3 vs. model 2, model 4 vs. model 3, and model 5 vs. model 4; all P , 0.001). As described previously, we performed a fully adjusted analysis to determine whether the volume–outcome relationship identified existed in both rural and urban facilities (Table 3). The substantial inverse relationship observed between severe sepsis case volume and inpatient mortality in the overall population was observed in both rural and urban hospitals. The addition of an interaction term between rural–urban and annual volume of severe sepsis to the fully adjusted model (model 5) did not significantly improve its fit (P = 0.10), demonstrating similar relationships between volume and mortality in rural and urban hospitals. We therefore used and 667
ORIGINAL ARTICLE Table 1. Baseline Characteristics of Cases of Severe Sepsis Identified from 2004 to 2010
ORIGINAL ARTICLE
The Relationship between Hospital Volume and Mortality in Severe Sepsis David F. Gaieski1, J. Matthew Edwards2, Michael J. Kallan3, Mark E. Mikkelsen3,4, Munish Goyal5, and Brendan G. Carr1,3,6 1
Department of Emergency Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; 2Department of Emergency Medicine, Kaiser Permanente San Diego Medical Center, San Diego, California; 3Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine, and 4Pulmonary, Allergy, and Critical Care Division, Department of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania; 5Department of Emergency Medicine, Medstar Washington Hospital Center, Washington, DC; and 6Leonard Davis Institute of Health Economics, University of Pennsylvania, Philadelphia, Pennsylvania
Abstract Rationale: Severe sepsis is increasing in incidence and has a high rate
of inpatient mortality. Hospitals that treat a larger number of patients with severe sepsis may offer a survival advantage. Objectives: We sought to assess the effect of severe sepsis case volume on mortality, hypothesizing that higher volume centers would have lower rates of inpatient death. Methods: We performed a retrospective cohort study over a 7-year
period (2004–2010), using a nationally representative sample of hospital admissions, examining the relation between volume, urban location, organ dysfunction, and survival. Measurements and Main Results: To identify potential differences in outcomes, hospitals were divided into five categories (,50, 50–99, 100–249, 250–499, and 5001 annual cases) and adjusted mortality was compared by volume. A total of 914,200
Severe sepsis, defined as infection accompanied by inflammation and acute organ dysfunction or tissue hypoperfusion (1), is increasing in incidence (2, 3), has an
patients with severe sepsis were identified over a 7-year period (2004–2010). Overall in-hospital mortality was 28.1%. In a fully adjusted model, there was an inverse relationship between severe sepsis case volume and inpatient mortality. Hospitals in the highest volume category had substantially improved survival compared with hospitals with the lowest case volume (adjusted odds ratio, 0.64; 95% confidence interval, 0.60–0.69). In cases of severe sepsis with one reported organ dysfunction, a mortality of 18.9% was found in hospitals with fewer than 50 annual cases compared with 10.4% in hospitals treating 5001 cases (adjusted odds ratio, 0.54; 95% confidence interval, 0.49–0.59). Similar differences were found in patients with up to three total organ dysfunctions. Conclusions: Patients with severe sepsis treated in hospitals with
higher case volumes had improved adjusted outcomes. Keywords: severe sepsis; hospital case volume; organ dysfunction
in-hospital mortality as high as 38% (4), and costs the U.S. health system $24 billion annually (5). The Centers for Disease Control and Prevention (CDC, Atlanta, GA) currently
lists septicemia as the eleventh leading cause of death in the United States (6), and the burden of this disease is expected to increase as the population ages (7).
( Received in original form February 16, 2014; accepted in final form August 10, 2014 ) Supported by an unrestricted educational grant from the Beatrice Wind Gift Fund. Role of funding agency: The Beatrice Wind Gift Fund had no role in the conception of this study, the collection or analysis of the data, or the writing, revision, and submission of the manuscript. No one from the Beatrice Wind Gift Fund has reviewed the manuscript before submission. Author Contributions: D.F.G. had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: D.F.G., J.M.E., M.J.K., B.G.C., M.G., and M.E.M.; acquisition of data: M.J.K. and B.G.C.; analysis and interpretation of data: D.F.G., J.M.E., M.J.K., B.G.C., and M.E.M.; drafting of manuscript: J.M.E., M.J.K., and D.F.G.; critical revision of the manuscript for important intellectual content: D.F.G., B.G.C., M.G., and M.E.M.; statistical analysis: M.J.K.; obtaining funding: D.F.G.; administrative, technical, or material support: D.F.G., M.J.K., and B.G.C.; study supervision: D.F.G. Correspondence and requests for reprints should be addressed to David F. Gaieski, M.D., Perelman School of Medicine at the University of Pennsylvania, Department of Emergency Medicine, Center for Resuscitation Science, Ground Ravdin, 3400 Spruce Street, Philadelphia, PA 19104. E-mail: gaieskid@uphs. upenn.edu This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org Am J Respir Crit Care Med Vol 190, Iss 6, pp 665–674, Sep 15, 2014 Copyright © 2014 by the American Thoracic Society Originally Published in Press as DOI: 10.1164/rccm.201402-0289OC on August 12, 2014 Internet address: www.atsjournals.org
Gaieski, Edwards, Kallan, et al.: Hospital Factors and Mortality in Severe Sepsis
665
ORIGINAL ARTICLE
At a Glance Commentary Scientific Knowledge on the Subject: Little is known about the
relationship between severe sepsis case volume at hospitals and mortality outcomes. We hypothesized that hospitals treating larger numbers of patients with severe sepsis would have lower in-hospital severe sepsis mortality. What This Study Adds to the Field: Our research confirmed our
hypothesis and has important implications about regionalization of severe sepsis care and optimizing costs and outcomes for severe sepsis. Survival from severe sepsis and septic shock requires early recognition and treatment (8–14). Patients with other timesensitive emergency conditions including trauma, stroke, cardiac arrest, and myocardial infarction are known to have better outcomes when treated in highvolume centers (15–19), which typically provide both a high level of resources and clinical experience. Given the variability in both capacity and capabilities at U.S. hospitals, disparities may exist in the care and outcomes for patients with severe sepsis and septic shock. Previous work in the United States suggests that patients with severe sepsis seen in emergency departments (EDs) treating higher numbers of sepsis patients have lower adjusted mortality rates (20). In contrast, a British study found no mortality difference in patients with severe sepsis treated in highversus low-volume intensive care units (ICUs) (21). Walkey and Wiener demonstrated that U.S. academic hospitals treating the highest volumes of patients with severe sepsis had 7% lower adjusted absolute in-hospital mortality when compared with the lowest volume academic hospitals (22). It is not known whether U.S. hospitals in general with higher severe sepsis case volumes provide a survival advantage and, if they do, which subgroups derive the greatest benefit from their expertise. In this study, we used a large nationally representative sample and hypothesized that higher volume centers seeing more severe sepsis cases would offer a survival advantage compared with lower case volume centers. We examined the association overall as well as 666
among patients with varying severity of illness, using the number of organ dysfunctions as a proxy. Some of the results of this study have been previously reported in the form of an abstract (23).
Methods Data
We performed a nationally representative retrospective cohort study from 2004 to 2010 using the Nationwide Inpatient Sample (NIS), the largest all-payer publicly available inpatient database in the United States. The NIS, developed as part of the Healthcare Cost and Utilization Project (HCUP) by the Agency for Healthcare Research and Quality (AHRQ), is a 20% stratified sample of U.S. acute care hospitals, and contains approximately 8 million hospital stays annually, broadly representing admissions to regular hospital wards as well as ICUs. The stratified sampling methodology allows for the generation of nationally representative estimates of incidence and mortality. More than 100 clinical and nonclinical elements are available from each hospital stay, including primary and secondary diagnoses, procedures, discharge status, patient demographics, and total hospital length of stay. Identification of patients as admitted to a hospital ward or an ICU is not available in the NIS. Hospital characteristics including geographic region, urban versus rural location, ownership, teaching status, and bed size are used to ensure data are nationally representative. In the NIS, urban and rural hospital designation is determined via the U.S. Census Bureau’s Core Based Statistical Area (CBSA) codes. Rural and micropolitan regions, defined as at least one population center with a population less than 50,000, are considered rural in the NIS (for further details, see http:// www.hcup-us.ahrq.gov/nisoverview.jsp). Given the deidentified nature of the administrative data, this study was declared exempt by the Institutional Review Board at the University of Pennsylvania (Philadelphia, PA). Patients
Severe sepsis was defined in accordance with consensus guidelines (1) as suspected or documented infection, systemic inflammation, and acute organ dysfunction. There are several published methods using differing approaches to capture cases of severe sepsis using administrative data (2, 4, 24, 25). However, research has demonstrated that there is significant variation in incidence, severity of
illness, number of organ dysfunctions, and mortality depending on the capture method used (3, 26). A method employing a limited number of infection and septicemia codes was found to better approximate epidemiologic chart-based studies (27). Considering these findings, we chose to employ a previously validated method, applying a limited number of infection codes (4, 5) and identifying acute organ dysfunctions that develop during the hospitalization. Patients with severe sepsis (assigned a primary or secondary International Classification of Disease, Ninth Revision [ICD-9] diagnostic code for infection and an additional ICD-9 code for acute organ dysfunction [including dysfunction of the respiratory, cardiovascular, renal, hepatic, coagulation, and central nervous systems] or patients assigned the ICD-9 codes for severe sepsis and septic shock) were collected from the NIS for analysis (see the online supplement). Patients transferred to or from another facility and those not more than 18 years of age were excluded from this analysis. We used ICD-9 codes to identify comorbidities as described by Elixhauser and colleagues, and All Patient Refined Diagnosis Related Groups (APRDRG) to adjust for severity of illness (28–30). Our primary end point was inpatient mortality. As severe sepsis can include various types and numbers of organ dysfunctions, we examined the association between the type of organ dysfunction (e.g., respiratory, renal, etc.), total number of organ systems involved, case volume, and mortality. Statistical Analyses
Descriptive statistics are presented as frequencies and percentages, with comparisons between groups made using the chi-squared (x2) test. We performed unadjusted and adjusted logistic regression modeling, with results expressed as odds ratios (ORs) with 95% confidence intervals (CIs). To account for the complex survey design and clustering within hospitals in the NIS, we used the specific statistical methodology recommended in the HCUP data documentation guide (31). This included survey statistics and Taylor series linearization methodology that estimates the covariance matrix for the unadjusted and adjusted logistic regression model coefficients. Our main set of multivariate logistic regression models was created to study the relationship between annual hospital severe sepsis case volume and inpatient mortality.
American Journal of Respiratory and Critical Care Medicine Volume 190 Number 6 | September 15 2014
ORIGINAL ARTICLE Considerations in creating annual case volume divisions included case-level versus hospital-level quartiles and ensuring a good mix of rural versus urban hospitals in each volume division. After sensitivity analyses were performed, the following annual case volume divisions were created for outcomes comparisons: fewer than 50, 50–99, 100–249, 250–499, and 5001 cases/year. We applied a stepwise modeling methodology to investigate the effect of additional groups of covariates on the (adjusted) relationship between annual hospital severe sepsis volume and mortality. This methodology increases transparency and allows the reader to appreciate how the entry of new variables into the model affects the point estimates of the preceding model. In stepwise fashion, additional adjustments were made (building on the variables introduced in the previous models) for hospital location and teaching status (model 1); age, sex, hospital region, payer status, discharge year, and race (model 2); each of the 29 AHRQ comorbidities (model 3); specific organ dysfunction information (model 4); and APR-DRG risk of mortality (model 5). A P value less than 0.05 was considered significant. To quantify the added benefit by model, likelihood statistics, presented as 2 3 normalized log-likelihood and corresponding chi-squared statistics, were employed. Each of the adjusted models is compared with the one immediately preceding it, demonstrating the incremental value of the additional variables. To address potential limitations of using categorical variables in regression models, we performed secondary analyses to examine the relationship between volume of cases as a continuous variable and outcome (32). This analysis assumes that the change is the same for any 100-case increase in severe sepsis volume (e.g., from 1 to 101 cases or from 1,401 to 1,501 cases). Because the data are skewed with few hospitals reporting more than 1,000 cases of severe sepsis per year, we also performed the secondary analysis by capping the analysis at a maximal value of 1,000 cases/year, thus preventing undue outlier effects. In addition, we performed an analysis on the fully adjusted logistic regression model that examined all combinations of annual severe sepsis volume in both rural and urban hospitals. Finally, to determine the relationship between volume and mortality for specific combinations of organ
dysfunctions, we performed a series of analyses stratified by specific organ dysfunction or groups of dysfunction; these analyses were limited to the respective subpopulations and examined the adjusted relationship of annual hospital severe sepsis volume and mortality in the fully adjusted model (model 5). In the analyses examining a specific organ dysfunction (e.g., respiratory only) or pair of organ dysfunctions (e.g., respiratory and cardiovascular), model 5 was modified by removing the adjustment for the specific organ dysfunction or dysfunctions of interest. This approach was taken to isolate the combination of organ dysfunction(s) in the context of hospital case volume, avoiding the confusion of comparing the patients with the specific organ dysfunction to the rest of the cohort, which would include patients with different combinations and/or number of organ dysfunctions. In all models, we investigated whether the interaction of rural versus urban and hospital case volume improved the fit of the fully adjusted model. All analyses were performed with SAScallable SUDAAN statistical software, version 11.0.0 (Research Triangle Institute, Research Triangle Park, NC).
Results Patient Characteristics
A total of 914,200 patients with severe sepsis was identified over a 7-year period (2004–2010), representing a weighted estimate of 4,506,106 cases nationally (Table 1). Mean in-hospital mortality was 28.1% across the study period and decreased over time (from 34.4% in 2004 to 22.9% in 2010; P , 0.001). The number of severe sepsis cases treated per year at different hospitals ranged from 1 to 2,415. Hospitals that reported fewer than 50 cases/year had unadjusted mortality rates identical to those of hospitals that treated 5001 cases/year (26.8%). The vast majority of comorbid conditions demonstrated small (in absolute terms) but statistically significant differences across volume groups (P , 0.001), with the exception of congestive heart failure (P = 0.14). Of note, hospitals in the highest volume group were much more likely to treat patients with chronic renal disease compared with the lowest volume group (27.0 vs. 17.5%, respectively; P , 0.001). For those patients determined to have severe sepsis, almost half
Gaieski, Edwards, Kallan, et al.: Hospital Factors and Mortality in Severe Sepsis
had acute dysfunction of the cardiovascular, respiratory, or renal systems (44.4, 49.5, and 49.8%, respectively). Hospitals in the highest volume group treated 19.3% more patients with respiratory dysfunction and 15.7% more with renal dysfunction than hospitals in the lowest volume group. Adjusted Mortality
In model 1, adjusting solely for hospital factors (location and teaching status), the highest volume hospitals (5001 cases/year) had lower mortality when compared with the lowest volume (,50 cases/year) hospitals, but hospitals with moderate volume (50–99, 100–249, and 250–499) were not significantly different from lowest volume hospitals. This relationship was similar in model 2 after adjustments for demographic factors (age, sex, hospital region, payer status, discharge year, and race) and in model 3 with further adjustments for individual comorbidities. After adjustment for organ dysfunction in model 4 we demonstrated significantly reduced mortality at higher versus lowest case volume hospitals (Table 2). In the fully adjusted analysis, model 5, there was a substantial inverse relationship between severe sepsis case volume and inpatient mortality. Compared with patients at hospitals that treated fewer than 50 cases of severe sepsis annually (46% of hospitals in our sample), patients at hospitals that treated 5001 cases annually have a 36% improved odds of inpatient survival (adjusted OR, 0.64; 95% CI, 0.60–0.69). Likelihood statistics demonstrated the significant improvement in model fit contributed by the respective sets of specific variables added in stepwise fashion for each of the subsequent models (model 1 vs. unadjusted, model 2 vs. model 1, model 3 vs. model 2, model 4 vs. model 3, and model 5 vs. model 4; all P , 0.001). As described previously, we performed a fully adjusted analysis to determine whether the volume–outcome relationship identified existed in both rural and urban facilities (Table 3). The substantial inverse relationship observed between severe sepsis case volume and inpatient mortality in the overall population was observed in both rural and urban hospitals. The addition of an interaction term between rural–urban and annual volume of severe sepsis to the fully adjusted model (model 5) did not significantly improve its fit (P = 0.10), demonstrating similar relationships between volume and mortality in rural and urban hospitals. We therefore used and 667
ORIGINAL ARTICLE Table 1. Baseline Characteristics of Cases of Severe Sepsis Identified from 2004 to 2010
