FEATURE

Variation in Hospital Mortality Rates With Inpatient Cancer Surgery Sandra L. Wong, MD, MS,∗ Sha’Shonda L. Revels, MD, MS,∗ Huiying Yin, MS,∗ Andrew K. Stewart, MA,† Andrea McVeigh, MS,∗ Mousumi Banerjee, PhD,‡ and John D. Birkmeyer, MD∗ Objective: To elucidate clinical mechanisms underlying variation in hospital mortality after cancer surgery Background: Thousands of Americans die every year undergoing elective cancer surgery. Wide variation in hospital mortality rates suggest opportunities for improvement, but these efforts are limited by uncertainty about why some hospitals have poorer outcomes than others. Methods: Using data from the 2006–2007 National Cancer Data Base, we ranked 1279 hospitals according to a composite measure of perioperative mortality after operations for bladder, esophagus, colon, lung, pancreas, and stomach cancers. We then conducted detailed medical record review of 5632 patients at 1 of 19 hospitals with low mortality rates (2.1%) or 30 hospitals with high mortality rates (9.1%). Hierarchical logistic regression analyses were used to compare risk-adjusted complication incidence and case-fatality rates among patients experiencing serious complications. Results: The 7.0% absolute mortality difference between the 2 hospital groups could be attributed to higher mortality from surgical site, pulmonary, thromboembolic, and other complications. The overall incidence of complications was not different between hospital groups [21.2% vs 17.8%; adjusted odds ratio (OR) = 1.34, 95% confidence interval (CI): 0.93–1.94]. In contrast, casefatality after complications was more than threefold higher at high mortality hospitals than at low mortality hospitals (25.9% vs 13.6%; adjusted OR = 3.23, 95% CI: 1.56–6.69). Conclusions: Low mortality and high mortality hospitals are distinguished less by their complication rates than by how frequently patients die after a complication. Strategies for ensuring the timely recognition and effective management of postoperative complications will be essential in reducing mortality after cancer surgery.

mortality rates across hospitals continues to suggest opportunities for improvement. Unfortunately, efforts to reduce disparities in performance have been limited by a lack of understanding about exactly why some hospitals have lower mortality rates than others. One obvious hypothesis is that low mortality hospitals are more adept at avoiding postoperative complications. If true, the most effective strategies for reducing variation in hospital mortality would focus on surgeon proficiency and potentially related technical complications, such as bleeding and anastomotic leak. Like the Surgical Care Improvement Project, strategies could also focus on hospital compliance with evidence-based processes of care associated with specific medical complications, including venous thromboembolism (VTE) and myocardial ischemia. A competing hypothesis, suggested by a recent study of patients undergoing general and vascular surgery, is that low-mortality and high-mortality hospitals are distinguished by their case-fatality rates among patients with complications, or by how well they “rescue” patients after a serious complication.3 If true for cancer surgery, the optimal strategy for reducing disparities in hospital mortality would focus on ensuring the timely recognition and effective management of patients after complications. The purpose of this national study was to elucidate clinical mechanisms underlying variation in hospital mortality with major cancer surgery. Collecting detailed clinical data on patients undergoing surgery at very low- and very high-mortality hospitals, we sought to determine whether differences in hospital mortality could be attributed to specific types of medical and surgical complications. We also examined the extent to which so-called “failure to rescue” might be responsible for variation in cancer surgery mortality.

Keywords: cancer outcomes, complications, mortality, surgical oncology (Ann Surg 2015;261:632–636)

O

n the basis of data from the most recent Nationwide Inpatient Sample, more than 250,000 patients undergo major cancer operations each year in the United States.1,2 Many of these procedures are relatively complex and associated with high rates of mortality and even higher rates of morbidity. Although outcomes with cancer surgery have improved over the past decade, wide variation in

From the ∗ Center for Healthcare Outcomes and Policy and Department of Surgery, University of Michigan Medical School, Ann Arbor; †Remedy Informatics, Chicago, IL; and ‡Department of Biostatistics, University of Michigan School of Public Health, Ann Arbor. Disclosures: This work was supported by the National Cancer Institute at the National Institutes of Health (2 R01 CA098481-05A1 to J.D.B.) and the Agency for Healthcare Research and Quality (K08 HS020937-01 to S.L.W.). J.D.B. has a financial interest in ArborMetrix, Inc, a clinical analytics company focused on hospital- and specialty-based care. The company was not involved with the study herein in any way. Other authors have no conflicts of interest. Reprints: Sandra L. Wong, MD, MS, Center for Healthcare Outcomes and Policy, Building 16, 2800 Plymouth Road, Ann Arbor, MI 48109. E-mail: wongsl@ umich.edu. C 2014 Wolters Kluwer Health, Inc. All rights reserved. Copyright  ISSN: 0003-4932/14/26104-0632 DOI: 10.1097/SLA.0000000000000690

632 | www.annalsofsurgery.com

METHODS Database and Study Population We performed a retrospective cohort study of 49 hospitals accredited by American College of Surgeons’ Commission on Cancer (CoC) that submit data to the National Cancer Database (NCDB). The NCDB is a nationwide oncology registry and outcomes database of more than 1500 cancer programs in the United States and Puerto Rico, representing 70% of all newly diagnosed cases of cancer in the United States.4 The database is maintained by the CoC and the American Cancer Society. Data on all types of cancer are tracked and analyzed prospectively by cancer registrars and submitted annually to the NCDB. The database includes information on patient, tumor, treatment, survival, and hospital characteristics. This study, designated a CoC-approved special study, and examined 2708 patients at 19 low-mortality hospitals and 2924 patients at 30 high-mortality hospitals (Fig. 1). To identify the hospitals, we first ranked 1279 CoC hospitals by a composite measure of mortality, as previously described by Dimick et al.5 In brief, this measure calculates a hospital’s predicted “true” mortality, based on its observed mortality, volume, case mix, and procedure mix. We excluded hospitals with fewer than 100 cancer cases over the 2-year period to ensure sufficient caseloads and reduce the likelihood that chance accounted for performance rankings. Annals of Surgery r Volume 261, Number 4, April 2015

Copyright © 2014 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

Annals of Surgery r Volume 261, Number 4, April 2015

Variation in Hospital Mortality Rates With Cancer Surgery

All National Cancer Database Hospitals (n=1279)

Low-mortality hospitals

Ranking on Mortality Composite Measure

Declined invitation to participate (n= 22)

High-mortality hospitals Declined invitation to participate (n= 47)

Agreed to participate (n=30)

Agreed to participate (n=19) Onsite Data Abstraction Patients at low-mortality hospitals (n=2708)

Patients at high-mortality hospitals (n= 2924)

FIGURE 1. Study design and enrollment schema. This study included 2708 patients at 19 low-mortality hospitals and 2924 patients at 30 high-mortality hospitals. Investigators and hospitals were blinded as to rankings, which were only known to CoC staff members. Starting at the top and bottom of our ranking list, we then began inviting the lowest and highest mortality hospitals to participate in this voluntary study until we had sufficient numbers, based on pre hoc sample size calculations. In the end, 19 hospitals were enrolled in the study from among the 41 lowest mortality hospitals; 30 hospitals were similarly enrolled from among the 77 highest mortality hospitals. Mortality rates were not significantly different among the hospitals that participated and those that did not, in either group. Because low-mortality hospitals tended to be larger, higher volume hospitals, high-mortality hospitals were oversampled to achieve a similar number of patients in both groups of hospitals. Data collection was capped at not more than 150 patients per hospital to minimize data collection burden at individual centers and to avoid oversampling at higher volume hospitals. In those hospitals, the 150 patients were a randomly selected sample.

Data Abstraction To supplement NCDB data on demographic variables and cancer characteristics (including stage), we used trained data abstractors to perform detailed retrospective chart review at participating hospitals. Onsite chart reviews on patients who underwent major resections for bladder, colon, esophageal, gastric, lung, and pancreatic cancers in 2006–2007 were used to capture variables for risk adjustment, including American Society of Anesthesiology (ASA) Physical Status classification, functional status, emergency case, dyspnea, cardiac disease, diabetes, body mass index (BMI), albumin, creatinine, and platelet count. The cancer types were selected to represent procedures associated with the greatest number of perioperative deaths. We also captured data on postoperative complications using a previously developed and validated instrument.6 The occurrence and date of 27 unique complication events were captured and categorized as surgical complications (anastomotic leak, wound infection, fascial dehiscence, bowel obstruction, gastrointestinal bleeding and reoperation), cardiac complications (acute myocardial infarction, cardiac arrhythmia, unexplained cardiac arrest), pulmonary complications (pneumonia, intubation), VTE/stroke, and other medical issues (acute renal failure, admissions to intensive care units). Cause of death was assigned by our clinical endpoints committee to 1 of 5 categories including surgical site, cardiac, respiratory, VTE, and other.  C 2014 Wolters Kluwer Health, Inc. All rights reserved.

Outcomes and Statistical Analysis For this study, we compared low- and high-mortality hospitals with regard to 3 main outcomes: cause-specific mortality (before discharge or within 30 days of the operation), incidence of complications, and case-fatality rates in patients with complications (or “failure to rescue” rates). For the first 2 outcomes, hierarchical logistic regression models were used to compare risk- and reliability-adjusted rates between hospital groups. The model adjusted for race, sex, ASA class, functional status, dyspnea, ischemic heart disease, congestive heart failure, diabetes mellitus, albumin, creatinine, BMI, hematocrit, platelets, cancer type, cancer stage, emergency surgery, and comorbidity. The model also accounted for the clustering of patients within hospitals. Specifically, we used logistic regression to model the binary cause-specific 30-day mortality, with hospital included as a random effect to capture the heterogeneity across hospitals. On the basis of this model, hospital groups were compared using the observed-to-expected mortality ratio. We used analogous modeling strategies to assess case fatality outcomes across the 2 hospital groups. In this case, however, the denominator was patients experiencing complications. All statistical analyses were performed using SAS 9.2 software (SAS institute, Cary, NC) to fit the hierarchical logistic regression models. P < 0.05 was considered statistically significant, and all tests were 2-sided. The institutional review board of the University of Michigan approved the study protocol.

RESULTS Patients at high-mortality hospitals tended to higher risk than those at lower mortality hospitals. As seen in Table 1, high-mortality hospitals had older patients, more minorities, and higher prevalence rates of comorbidities. Procedures at high-mortality hospitals were more likely to be emergent (6.3% vs 3.3%, P < 0.0001). In terms of procedure mix, colectomy accounted for a higher proportion of cases at high-mortality hospitals than at low-mortality hospitals (69% vs 37%). High-mortality hospitals performed commensurately fewer lung resections, esophagectomies, gastrectomies, pancreatectomies, and cystectomies. Unadjusted overall mortality rates were 9.1% at the highmortality hospitals and 2.1% at low-mortality hospitals. After risk adjustment, however, disparities in mortality between the 2 groups shrunk considerably. Risk-adjusted mortality rates were 7.4% and 2.8% at high- and low-mortality hospitals, respectively. www.annalsofsurgery.com | 633

Copyright © 2014 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

Annals of Surgery r Volume 261, Number 4, April 2015

Wong et al

Overall, medical causes of death (cardiac, pulmonary, and VTE/stroke) were responsible for about half of all deaths. Surgical causes accounted for 19%; for the remainder, the cause of death could not be determined definitively. As seen in Table 2, cause-specific mortality rates were higher in all categories for the high-mortality hospitals. After risk adjustment, high-mortality hospitals tended to have higher overall complication rates than low-mortality hospitals

[21.2% vs 17.8%, adjusted odds ratio (OR) = 1.34, 95% confidence interval (CI): 0.93–1.94], but this difference was not statistically significant (Table 3). Similarly, differences in rates of specific types of medical and surgical complications were small and not statistically significant. In contrast, high-mortality hospitals had considerably higher “failure to rescue” rates. Risk-adjusted case-fatality rates among

TABLE 1. Patient Characteristics and Association With Mortality Patient Characteristics Age, yr ≤64 65–74 ≥75 Race White Black Other/Unknown Sex Female Comorbidities 0 1 >2 ASA class 1/2 3 4/5 Functional Status Independent Partially/totally dependent Heart disease No Yes Operation Lung resection Colectomy Esophagectomy Gastrectomy Pancreatectomy Radical cystectomy Emergency surgery No Yes

Low-mortality Hospitals

High-mortality Hospitals

983 (36.3%) 827 (30.5%) 898 (33.2%)

983 (33.6%) 850 (29.1%) 1091 (37.3%)

0.0046

2349 (86.7%) 136 (5.0%) 223 (8.2%)

2490 (85.2%) 324 (11.1%) 110 (3.8%)

Variation in hospital mortality rates with inpatient cancer surgery.

To elucidate clinical mechanisms underlying variation in hospital mortality after cancer surgery...
178KB Sizes 1 Downloads 3 Views