AAA indicates abdominal aortic aneurysm; AHA, American Hospital Association; and NIS, Nationwide Inpatient Sample. aMajor complications defined as deep vein/pulmonary embolism, surgical site infection, acute renal failure, pneumonia, respiratory failure, acute myocardial infarction/cardiac event, shock/hypotension/sepsis, and other infection.
AAA indicates abdominal aortic aneurysm; HBHs, high-burden hospitals; LBHs, low-burden hospitals; and MBHs, moderate-burden hospitals.
eTable 1. Primary Procedure and Complication Codes
eTable 2. FTR Model 2
eTable 3. Full FTR Model 3
Elliot Wakeam, Nathanael D. Hevelone, Rebecca Maine, Jabaris Swain, Stuart A. Lipsitz, Samuel R. G. Finlayson, Stanley W. Ashley, Joel S. Weissman. Failure to Rescue in Safety-Net HospitalsAvailability of Hospital Resources and Differences in Performance. JAMA Surg. 2014;149(3):229–235. doi:10.1001/jamasurg.2013.3566
Failure to rescue (FTR), the mortality rate among surgical patients with complications, is an emerging quality indicator. Hospitals with a high safety-net burden, defined as the proportion of patients covered by Medicaid or uninsured, provide a disproportionate share of medical care to vulnerable populations. Given the financial strains on hospitals with a high safety-net burden, availability of clinical resources may have a role in outcome disparities.
To assess the association between safety-net burden and FTR and to evaluate the effect of clinical resources on this relationship.
Design, Setting, and Participants
A retrospective cohort of 46 519 patients who underwent high-risk inpatient surgery between January 1, 2007, and December 31, 2010, was assembled using the Nationwide Inpatient Sample. Hospitals were divided into the following 3 safety-net categories: high-burden hospitals (HBHs), moderate-burden hospitals (MBHs), and low-burden hospitals (LBHs). Bivariate and multivariate analyses controlling for patient, procedural, and hospital characteristics, as well as clinical resources, were used to evaluate the relationship between safety-net burden and FTR.
Main Outcomes and Measures
Patients in HBHs were younger (mean age, 65.2 vs 68.2 years; P = .001), more likely to be of black race (11.3% vs 4.2%, P < .001), and less likely to undergo an elective procedure (39.3% vs 48.6%, P = .002) compared with patients in LBHs. The HBHs were more likely to be large, major teaching facilities and to have high levels of technology (8.6% vs 4.0%, P = .02), sophisticated internal medicine (7.7% vs 4.3%, P = .10), and high ratios of respiratory therapists to beds (39.7% vs 21.1%, P < .001). However, HBHs had lower proportions of registered nurses (27.9% vs 38.8%, P = .02) and were less likely to have a positron emission tomographic scanner (15.4% vs 22.0%, P = .03) and a fully implemented electronic medical record (12.6% vs 17.8%, P = .03). Multivariate analyses showed that HBHs (adjusted odds ratio, 1.35; 95% CI, 1.19-1.53; P < .001) and MBHs (adjusted odds ratio, 1.15; 95% CI, 1.05-1.27; P = .005) were associated with higher odds of FTR compared with LBHs, even after adjustment for clinical resources.
Conclusions and Relevance
Despite access to resources that can improve patient rescue rates, HBHs had higher odds of FTR, suggesting that availability of hospital clinical resources alone does not explain increased FTR rates.
The Institute of Medicine1 defines a safety-net hospital as one that has an explicit commitment to serving vulnerable patient populations or that cares for a disproportionate share of individuals covered by Medicaid or who are underinsured or uninsured. These hospitals are found in all regions of the country and typically serve minority and low-income populations. The safety-net burden of a hospital has been defined as the proportion of discharged patients insured by Medicaid or uninsured, and studies2- 6 have documented poorer quality of care at hospitals with a high safety-net burden. Furthermore, Rhoads and colleagues7 examined surgical mortality among patients with colorectal cancer in California undergoing resection in hospitals with high proportions of Medicaid patients and found significantly worse 30-day and 1-year mortality.
Quiz Ref IDFailure to rescue (FTR), defined as death after postoperative complications, has been shown to account for a significant proportion of the variability in surgical mortality across US hospitals.8,9 It has been suggested that effective rescue may require increased nursing surveillance and training as well as hospital clinical resources, such as critical care and interventional or high-technology services.10- 14 Specifically, these services may modulate the pathway of care involved in the rescue of patients with complications, and those hospitals and teams with appropriate access to these services may be able to intervene in a more timely and effective manner.
Hospital clinical resource availability depends on a hospital’s financial health, which is affected by the patient population it serves, market conditions, and payment rates. Safety-net facilities may be less able than other hospitals to invest in quality improvement,2 and their lack of financial resources may limit access to clinical resources that can affect rates of FTR. Therefore, we hypothesized that safety-net hospitals would have decreased access to clinical resources and as a result might have higher rates of FTR.
The study was approved by the Partner’s Health System institutional review board. Patient-level discharge data were obtained using the Nationwide Inpatient Sample from January 1, 2007, to December 31, 2010. The Nationwide Inpatient Sample is a stratified, survey-weighted 20% sample of all US hospitals provided by the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality.15 It contains data on procedures, comorbid conditions, insurance status, and demographics. Hospital characteristics and clinical resources were obtained from the 2009 American Hospital Association (AHA) annual survey (http://www.aha.org/research/rc/stat-studies/index.shtml).
Quiz Ref IDA hospital’s safety-net burden was calculated as the proportion of patients who were insured by Medicaid or were uninsured.2,16 All hospitals were classified into 1 of the following 3 categories: low-burden hospitals (LBHs) were in the lowest quartile of safety-net burden (<16.1%), moderate-burden hospitals (MBHs) were in the middle 2 quartiles (16.1%-32.5%), and high-burden hospitals (HBHs) were in the highest quartile (>32.5%).
Quiz Ref IDPatients (n = 260 020) included in the study underwent any of 8 high-risk general, thoracic, and vascular operations (esophagectomy, gastrectomy, open abdominal aortic aneurysm repair, pneumonectomy, lung lobectomy, colectomy, hepatectomy, and pancreatectomy) during the study period. Primary procedure codes are listed in eTable 1 in the Supplement.17 Patients were excluded if they were older than 90 years or younger than 18 years, had missing sex information, had missing urgency of the operation information, were coded as transfers or admissions for trauma, or were treated at hospitals with missing AHA identification numbers (Figure 1). This formed our general surgery cohort of 163 963 patients and was used to calculate overall complication rates. Major complications were identified as in previous studies9,12,17- 19 using International Classification of Diseases, Ninth Revision coding and included acute myocardial infarction or cardiac event, deep vein thrombosis or pulmonary embolism, gastrointestinal hemorrhage, pneumonia, respiratory failure, postoperative hemorrhage, surgical site infection, other postoperative infection, or shock, hypotension, or hypovolemia (eTable 1 in the Supplement lists the codes). The overall major complication rate among the general surgery cohort was 28.4% and varied by safety-net burden (29.7%, 28.8%, and 26.6% for HBHs, MBHs, and LBHs, respectively; P = .002). The overall mortality rate was 4.4%.
To perform our analyses of FTR, the sample was further restricted to patients experiencing a major complication (n = 46 519). Of all deaths in the general surgery cohort, 91.0% were captured in the final study population (ie, patients with major complications).
The AHA annual survey assesses facilities and services owned or provided by the responding hospital. Fourteen clinical resources were selected a priori from the previous literature7,10,12,13,20- 22 on the basis of their potential to moderate the rescue pathway for the included complications. Because of concerns regarding collinearity, these services were collapsed into the following 10 hospital clinical resource groups: (1) advanced cardiology (electrophysiology, interventional cardiology, and cardiac intensive care), (2) sophisticated internal medicine services (presence of bone marrow transplantation), (3) fully implemented electronic medical record, (4) intensivists provide care in the intensive care unit, (5) high hospital technology (solid-organ transplantation and cardiac surgery), (6) high ratio of respiratory therapists to beds, (7) high ratio of nurses to patients, (8) percentage of nurses who are RNs, (9) advanced endoscopy (endoscopic retrograde cholangiopancreatography and ablation of Barrett esophagus), and (10) advanced diagnostic radiology (presence of a positron emission tomographic scanner). Hospitals were considered to have advanced cardiology, high hospital technology, and advanced endoscopy if all components were present. Continuous variables were inspected for normality and were subsequently dichotomized at the median to create binary variables. Major teaching status was defined by membership in the Council of Teaching Hospitals. Hospitals were also characterized by US region and by control (government nonfederal, private nonprofit, or private for profit).
Our primary analysis at the patient level sought to determine the association between safety-net burden and FTR. We began by calculating unadjusted odds ratios (ORs) and used the Rao-Scott χ2 test to compare demographic and comorbidity status, accounting for survey weights and clustering. Next, 2 logistic regression models were fit. Model 1 estimated the likelihood of death (among patients with a major complication) for categories of safety-net burden, adjusting for patient demographics only (age, sex, race/ethnicity, urgency of the operation, payer, comorbidity, and procedure). Patient comorbidities were scored using the method by Elixhauser et al.23 Model 2 adjusted for patient and hospital characteristics (hospital size, teaching status, hospital region, and hospital control). All models used survey weights and controlled for clustering of effects by hospital.
Our secondary analysis investigated the role of hospital clinical resources as a confounder of the relationship between safety-net burden and FTR. We first examined the distribution of each clinical resource group across hospitals classified by safety-net burden. The Pearson χ2 test was used to compare hospital-level proportions. Next, we evaluated the association between each of 10 hospital clinical resource groups and FTR, adjusted for patient demographics but not hospital characteristics, using multivariate logistic regression. Seven hospital clinical resources with P < .10 after adjustment for patient demographics were selected for further analysis. We then fit model 3 with all patient demographics and hospital characteristics, as well as these 7 hospital clinical resources.
We repeated these analyses after stratifying by procedure. For ease of presentation, we calculated adjusted FTR rates using generalized estimating equation regression models with the potential confounders set to their mean values in the study population. All P values were 2-sided and considered significant at .05. Analyses were performed using available software (SAS, version 9.3; SAS Institute Inc).
In total, 7632 patients in the FTR analyses (16.4%) were treated at hospitals with missing information from the AHA annual survey. Our main analyses excluded these patients. To assess the sensitivity of our results to excluding these patients with missing data, we used the missing data technique of reweighted estimating equations.24 Using reweighted estimating equations, the probability of having all covariates observed was calculated using a logistic regression model with all nonmissing variables, including the outcome (FTR), as predictors. In the reweighted estimating equations approach, each patient in the study is weighted by the inverse probability of having no missing data. Using reweighted estimating equations, the results were almost identical to those obtained by excluding any patients with missing data from the analysis, so for simplicity we present the results obtained after exclusion of patients with missing data.
Table 1 summarizes patient demographics stratified by safety-net status of the hospital in which they underwent an operation. Quiz Ref IDPatients in HBHs were younger (65.2 vs 68.2 years, P = .001), more likely to be of black race/ethnicity (11.3% vs 4.2%, P < .001), and less likely to undergo an elective procedure (39.3% vs 48.6%, P = .002) compared with patients in LBHs. Surgical case mix also differed by safety-net burden, with different distributions for hepatectomy, pancreatectomy, gastrectomy, esophagectomy, and lung lobectomy. A total of 6491 deaths occurred in the final study cohort, giving a mean FTR rate of 14.0%.
A total of 1299 hospitals contributed at least 1 patient who experienced a complication to the final study population. The HBHs were more likely to have more than 200 beds (22.9% vs 16.5%, P = .10) and to be major teaching facilities (16.3% vs 5.9%, P < .001) compared with LBHs (Table 1).
Compared with LBHs, HBHs had greater access to high hospital technology (8.6% vs 4.0%, P = .02), sophisticated internal medicine (7.7% vs 4.3%, P = .10), and high ratios of respiratory therapists to beds (39.7% vs 21.1%, P < .001). However, HBHs also had lower proportions of RNs among nurses (27.9% vs 38.8%, P = .02), less frequent access to a positron emission tomographic scanner (15.4% vs 22.0%, P = .03), and lower rates of a fully implemented electronic medical record (12.6% vs 17.8%, P = .03).
As summarized in Table 2, patients treated at HBHs had higher unadjusted rates of FTR (15.4%) compared with patients treated at LBHs (12.4%) or MBHs (14.1%). Teaching hospitals (OR, 0.92; 95% CI, 0.81-1.03; P = .16) and large hospitals (OR, 0.92; 95% CI, 0.83-1.01; P = .09) had lower odds of FTR (model 2 in eTable 2 in the Supplement). After adjustment for patient demographics and hospital characteristics, HBHs had significantly increased odds of FTR compared with LBHs (OR, 1.39; 95% CI, 1.23-1.57, P < .001); similarly, MBHs had higher odds of FTR (OR, 1.16; 95% CI, 1.05-1.28; P = .004) (Table 2). The C statistic for both models 1 and 2 was 0.77, indicating good discriminatory power of the model.
Seven of ten hospital clinical resources were associated with lower FTR rates after controlling for patient demographics (Table 3). Hence, a third model was fitted that included patient demographics and hospital characteristics (covariates from model 2), as well as indicator variables for these 7 hospital clinical resources. After controlling for patient, hospital, and clinical resource variables, safety-net status was still associated with increased odds of FTR for HBHs (OR, 1.35; 95% CI, 1.19-1.53; P < .001) and MBHs (OR, 1.15; 95% CI, 1.05-1.27; P = .005) compared with LBHs (Table 2). The C statistic for the final FTR model was 0.77 (full model in eTable 3 in the Supplement).
We repeated the analyses above for each procedure to confirm the observed relationships. This subanalysis controlled for all patient demographics and hospital characteristics and showed similar trends for the association between safety-net burden and FTR (Figure 2).
The implications of health care reform for safety-net hospitals are as yet unknown. While many uninsured persons will obtain coverage under the Patient Protection and Affordable Care Act, historical lack of resources and the introduction of value-based payments may adversely affect some hospitals.2Quiz Ref IDWe found that hospital safety-net burden was an independent predictor of FTR after controlling for hospital and patient factors. Hospitals with a high safety-net burden (HBHs) were more likely to be large teaching facilities with sophisticated internal medicine services and high technology but had lower proportions of RNs among nurses, electronic medical record implementation, and a positron emission tomographic scanner. The differences in resource distributions that we observe may relate to compromises made under financial pressures at HBHs and are consistent with previously reported trends in safety-net hospitals.3,25 Despite the presence of several protective clinical resources, the odds of FTR remained significantly higher in HBHs. Although we cannot precisely identify the cause of this disparity, it may relate to culture, teamwork, or the way in which resources are mobilized and used to provide care for patients with complications in a timely fashion. It is also likely that the population of HBHs is heterogeneous, and divergent outcomes may exist even within this group. Furthermore, many safety-net hospitals are publicly owned and governed. They may be subject to political and managerial constraints that could affect their ability to provide high-quality care in ways that we were unable to measure. Further study using national data will be needed to answer these questions.
Safety-net hospitals have been examined previously with respect to FTR in a limited study26 that included all medical and surgical patients at 54 hospitals in the University HealthSystem Consortium (Chicago, Illinois), showing that safety-net status was associated with higher congestive heart failure mortality and FTR and had a significant interaction with nurse staffing. Our results build on this by rigorously defining FTR and focusing on a surgical cohort in a national sample of hospitals, while also investigating the role of hospital clinical resources. In addition, our study extends prior work relating to the protective effect of such clinical resources.12,14,20,27 In addition to resources included in other studies, we obtained data on intensivists, fully implemented electronic medical records, and advanced cardiology and found all of them to be significantly associated with lower FTR rates. Intensivists have a critical role in the rescue pathway for surgical patients who require intensive care unit admission. The use of electronic medical records, including computerized order entry, may improve communication, safety, and timeliness of care. Access to advanced cardiology services, including electrophysiology, interventional cardiology, and cardiac critical care, is increasingly important to provide appropriate care for complex surgical patients. These specialists are needed to ensure that patients are treated appropriately in an environment of increasingly nuanced and complex evidence-based guidelines for acute myocardial infarction and arrhythmias.
Our study has several limitations. It relies on administrative data, which can lack completeness and accuracy of coding, leading to difficulty with adequate risk adjustment. We attempted to overcome this by using comorbidity codes by Elixhauser et al,23 which have been well validated, and by adjusting for characteristics known to affect FTR, such as age and urgency of the operation, among others. Coding deficiencies could explain similar comorbidity profiles between hospitals of differing safety-net burden; however, this is thought to be more of an issue in hospitals with high safety-net burden and should, if anything, bias our results against the observed finding.7,28 Accurate coding of complications is another difficulty in using administrative data; however, we tried to overcome this by using validated codes as described above. The use of the AHA annual survey could potentially introduce a bias given that it is self-reported; however, it remains the most comprehensive and accurate survey of hospital information in the United States. Furthermore, the missing data within the AHA could introduce bias; however, a missing data analysis produced no changes in the results. The AHA data are also limited in that they do not allow us to analyze the efficacy with which resources are used and how they are distributed between departments within a hospital. Last, the AHA data do not allow us to examine the influence of provider experience or supervision on our findings.
Our work identifies a significant disparity in surgical safety among a national sample of patients undergoing high-risk general, thoracic, and vascular operations. Furthermore, although the distribution of hospital clinical resources varied somewhat by safety-net burden, this does not explain the higher rates of FTR in high safety-net burden hospitals. With impending expansions in Medicaid as provided under the Patient Protection and Affordable Care Act, the societal importance of safety-net hospitals may increase. The Supreme Court decision to allow states to opt out of Patient Protection and Affordable Care Act–associated Medicaid expansions, while leaving in place the plan to decrease disproportionate share payments (designed to offset hospital costs associated with uncompensated care), may exacerbate resource shortages in safety-net hospitals among these states.29,30 The quality of care in these institutions is critical, and FTR may represent an important area of potential quality improvement.
Accepted for Publication: May 10, 2013.
Corresponding Author: Elliot Wakeam, MD, Center for Surgery and Public Health, Department of Surgery, Brigham and Women’s Hospital, One Brigham Circle, Boston, MA 02130 (email@example.com).
Published Online: January 15, 2014. doi:10.1001/jamasurg.2013.3566.
Author Contributions: Drs Wakeam and Weissman had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Wakeam, Weissman.
Acquisition of data: Wakeam, Hevelone, Maine, Swain.
Analysis and interpretation of data: Wakeam, Hevelone, Maine, Swain, Lipsitz, Finlayson, Weissman.
Drafting of the manuscript: Wakeam, Maine, Swain.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Wakeam, Hevelone, Lipsitz.
Obtained funding: Wakeam, Ashley, Weissman.
Administrative, technical, or material support: Finlayson, Ashley, Weissman.
Study supervision: Finlayson, Weissman.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was funded by grant 20120369 from The Commonwealth Fund.
Role of the Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the sponsor.
Correction: This article was corrected for an error in an author name on January 28, 2014.