The Centers for Medicare & Medicaid Services estimated that $17 billion were spent on unplanned readmissions after hospital discharge in 2004.1 As a result, in an effort to reduce the burden of readmissions, the Affordable Care Act created the Hospital Readmission Reduction Program.2 In this program, hospitals with a higher-than-expected readmission rate could face financial penalties of up to 3% of the hospital's Medicare reimbursement. These actions were largely designed to encourage hospitals to improve their postacute transition of care and improve care coordination in patients with chronic medical conditions to prevent readmissions.3,4 Over time, surgical quality improvement programs such as the National Surgical Quality Improvement Project started collecting data on 30-day readmissions after surgery. Studies examining the reasons for readmissions using these quality improvement databases elucidated that postoperative complications (such as surgical site infections, bowel obstruction, or ileus), both occurring during the postoperative hospitalization and developing postdischarge, were a major cause of readmissions after surgery.5,6 These studies established the credibility of readmission rates to assess surgical quality. As a result, physician and hospital performance reporting websites are now routinely using the 30-day readmission rate as one of the metrics for the development of their rankings.7
Despite the ever increasing use of 30-day readmission rates as a proxy for the quality of surgical care, there are reasons to believe that extrapolating this association to surgical cases performed for oncologic indications is problematic. Complete removal of the tumor (which may need an appropriately aggressive surgery) and timely delivery of adjuvant chemotherapy have been shown to improve overall survival in several cancers. However, both factors may increase readmission rates.
We hypothesized that high-volume hospitals (which previously have been shown in ovarian cancer to be more likely to deliver guideline-based care and perform highly complex oncologic procedures) would have a higher risk-adjusted readmission rate compared with low-volume hospitals despite an improved overall survival rate.
MATERIALS AND METHODS
We performed a retrospective cohort study of women diagnosed between 2004 and 2013 with high-grade serous carcinoma undergoing cytoreductive surgery during the primary treatment course using the National Cancer Database (from here on referred to as “the database”). This database is a joint program of the American Cancer Society and the Commission on Cancer of the American College of Surgeons. This database uses a hospital-based registry, and the Commission on Cancer requires approved programs to abstract and follow all malignant tumors diagnosed or treated at the hospital.8 Currently, approximately 70% of the ovarian cancer cases in the United States are reported to the database.9 Data reported are deidentified to ensure confidentiality; therefore, this study is exempt from obtaining informed consent by the study participants and exempt per the University of Michigan institutional review board policies.
To better understand the relationship between the readmission rates and the quality of surgical care, we performed this study in patients with ovarian cancer using the National Cancer Database. We chose to study ovarian cancer because the relationship between optimal surgical effort and overall survival has been demonstrated in this disease, in which each 10% increase in cytoreduction of tumor correlates with a nearly 6% increase in median survival.10,11 Cases were selected using the diagnostic code C56.9 (malignant neoplasm of ovary). We restricted our analyses to patients undergoing debulking surgery for high-grade serous carcinoma and advanced-stage (defined as International Federation of Gynecologists and Obstetricians stages III and IV). We chose this combination of stage histology for three reasons: 1) more than 70% of patients are diagnosed with high-grade serous histology with advanced stage, 2) survival between different histologic subtypes and stages varies significantly in ovarian carcinoma, and 3) the extensive abdominal disease encountered in this histology underscores the role of surgical effort necessitated for optimal debulking. We included both primary debulking cases and those undergoing interval debulking after neoadjuvant chemotherapy.
The primary outcome of interest was the performance of the risk-adjusted readmission rate as a metric of quality of care. Specifically, we compared the performance of the hospital risk-adjusted readmission rate with other quality of care metrics within hospitals categorized by case volume. The other quality of care metrics considered in our analyses were risk-adjusted 30- and 90-day mortality, rates of adherence to guideline-based care, and 5-year overall survival.
The primary independent variable of interest to examine the quality of care was hospital case volume category. The mean hospital volume was calculated by dividing total number of cases reported by the hospital by the time the hospital reported the cases to the database. The mean hospital volume was then divided into four categories of annualized hospital volume (10 cases per year or less, 11–20 cases per year, 21–30 cases per year, and 31 cases per year or more). These cutoffs were chosen based on previous reports noting that hospitals performing more than 20 cases were high volume12,13 and those with ultrahigh volume (31 cases per year or more) have the lowest 30-day and 90-day outcomes.14 Covariates used in the analysis for risk adjustment included patient characteristics: age (categorized as 40 or younger, 41–50, 51–60, 61–70, 71–80, and older than 80 years), Charlson comorbidity index (the database provides a score of 0, 1, or 2 or more),15 race and ethnicity (categorized as a mutually exclusive race–ethnicity variable—non-Hispanic white, non-Hispanic black, Hispanic, other, or unknown), and median household income from zip code of residence, disease characteristics: stage (III or IV), and receipt of neoadjuvant chemotherapy. Other nonclinical factors included insurance status, distance traveled (categorized into four categories by dividing the entire range of values into quartiles), facility type (as categorized by the National Cancer Database—community cancer program, comprehensive community cancer, academic or research program, integrated network cancer program). Length of stay during index hospitalization was calculated from the day of surgery to the day of discharge and used as a continuous variable.
We used the criteria previously established by Bristow et al13 to define the adherence to the National Comprehensive Cancer Network guidelines. Patients were considered to have received treatment in accordance to the National Comprehensive Cancer Network guidelines if they underwent cytoreductive surgery and received multiagent chemotherapy. Major cytoreductive surgery was identified using the Facility Oncology Registry Data Standards as per the database guidelines.8 Instead of using the Current Procedural Terminology codes, the database guidelines require the data abstractors to review the operative procedures and assign the procedure codes listed in the registry manual. These cases included 1) low-complexity procedures, which included patients undergoing partial or total unilateral or bilateral salpingo-oophorectomy with omentectomy, with or without hysterectomy; 2) intermediate-complexity procedures, which included debulking procedure (defined by the database as partial or total removal of the tumor mass and can involve removal of tumor for multiple sites, may include removal of ovaries with or without the uterus); and 3) high-complexity procedures: debulking procedures with colon (including appendix) or small intestine resection or partial resection of urinary tract (not incidental) or anterior, posterior, total, or extended pelvic exenteration.
We compared the covariate distributions across outcome groups using χ2 tests. Factors associated with postoperative mortality based on prior knowledge or clinical plausibility were included in multivariable analyses. To account for clustering of the cases at the hospital level, a random-effects multivariable logistic regression model, with hospital as a random intercept, was used to assess factors associated with mortality at 30 and 90 days. Factors considered in the model included patient medical and sociodemographic factors (age, race, income, Charlson comorbidity index, and insurance status) and treatment factors (hospital volume, distance from place of residence to the hospital, receipt of neoadjuvant chemotherapy, and year of diagnosis). Hospital facility type was noted to have a significant correlation with hospital volume and therefore not included in the final model. Using the odds ratios derived from logistic regression analysis, we calculated the risk-adjusted probability of 30- and 90-day mortality and 30-day readmission rates. This was obtained by postestimation computation of predicted or expected values from the fitted logistic regression model using methodology previously described.16,17 These risk-adjusted rates were then compared across the four categories of hospital volumes.
After verifying the assumption of proportionality, a Cox proportional hazards model was fitted to evaluate the effects of patient, tumor, and treatment factors on 5-year survival. Adjusted hazard ratios (HRs) and 95% CIs were generated. Risk-adjusted survivals were generated based on the fitted models to estimate the independent effect of the covariates studied. Lastly, relative HRs were calculated using the Stata postestimation “margins” commands to assess the independent effect of the hospital case volume on survival. Data management and analyses were performed using STATA 14. All tests were two-tailed, and α was set at 0.05.
We performed several sensitivity analyses to confirm our findings. First, we repeated our entire analysis after excluding patients who died within 90 days after surgery. We did this for two reasons: 1) to confirm that the benefit in overall survival is not just driven by the short-term mortality benefit and 2) to reduce the immortal time bias, that is, those dying within 90 days could represent patients with higher comorbidities and therefore represent a subset less likely to get care that is concordant to National Comprehensive Cancer Network guidelines. Second, we repeated the analysis by excluding patient receiving neoadjuvant chemotherapy. Although we adjusted for receipt of neoadjuvant chemotherapy in our primary analysis, often the decision to start neoadjuvant chemotherapy is multifactorial and therefore might introduce bias with unmeasured confounders in our study. Third, as a result of the fact that the National Cancer Database only records readmissions for the same hospital, we performed another sensitivity analysis. Previous studies have shown that the likelihood of patients getting readmitted to a different hospital (other than the one that performed the index procedure) increases as the distance from the hospital increases. This rate of readmission to a different hospital is lowest for the cohort traveling the least.18 Therefore, we repeated our analysis with patients only traveling less than 5 miles (first quartile) for their initial treatment. Fourth, we repeated our analysis by using hospital volume of ovarian-debulking surgeries performed per year as a continuous variable instead of categories described previously.
A total of 42,931 patients with high-grade serous carcinoma undergoing debulking surgery between 2004 and 2013 were eligible for the final analysis. A detailed flowchart of cohort development is shown in Figure 1. Most of the patients were in the 50- to 59- (25%) and 60- to 69- (31%) year age group, non-Hispanic white (84%), privately insured (48%) or with Medicare coverage (41%), and received care at a comprehensive community cancer program (39%) or at an academic research program (45%). Baseline demographics of the entire cohort and by each hospital case volume category are summarized in Table 1. Despite significant P values (as a result of large sample size), the cohorts were clinically similar.
The overall unplanned 30-day readmission rate was 6.36% (95% CI 6.13–6.59). Readmission rates based on patient, tumor, and treatment factors are presented in Table 2. In patients undergoing primary debulking surgery, 43.1% (95% CI 41.1–45.1) of patients underwent high-complexity surgery in hospitals performing 31 cases per year or more compared with 34.3% (95% CI 33.6–35.1) in hospitals performing less than 10 cases per year (P<.001). A similar trend was seen in patients undergoing interval debulking after neoadjuvant chemotherapy (Table 3). Hospitals performing 31 cases per year or more had a readmission rate of 9.5% (95% CI 7.8–11.4) in patients undergoing high-complexity surgery compared with 7.9% (95% CI 7.3–8.6) hospitals performing less than 10 cases per year (P=.34). In a multivariate logistic regression model, hospitals performing 31 cases per year or more had a 24% higher likelihood of readmission when compared with those performing 10 cases per year or less (adjusted odds ratio [OR] 1.25, 95% CI 1.06–1.46). Other factors significantly associated with higher odds of readmission included increasing comorbidity score and stage IV disease. Factors associated with lower readmission odds included increasing distance from the hospital (fourth quartile compared with the first quartile, adjusted OR 0.66, 95% CI 0.59–0.75) and patients receiving neoadjuvant chemotherapy (adjusted OR 0.49, 95% CI 0.43–0.56).
Despite the higher odds of readmission, hospitals performing 31 cases per year or more had a significantly lower risk adjusted 30-day mortality (adjusted OR 0.69, 95% CI 0.50–0.96), 90-day mortality (adjusted OR 0.74, 95% CI 0.60–0.91), and a higher adherence to National Comprehensive Cancer Network guidelines (86% vs 77%, P<.001). Comparative performance of the risk-adjusted 30-day readmission rate to other quality of care metrics (risk-adjusted 30- and 90-day mortality, adherence to National Comprehensive Cancer Network guidelines) is shown in Figure 2. Length of stay was not significantly different between hospital volumes overall and when analyzed between the primary debulking and neoadjuvant chemotherapy cohorts separately (Table 3). Overall, patients who were readmitted had a longer length of stay (8.8 days [95% CI 2–22]) compared with those who did not (7.38 days [95% CI 1–18]; P<.001).
Based on hospital volume, survival analysis revealed a median overall survival of 41.2 months (95% CI 40.5–41.9 months) for 10 cases per year or less, 43.3 months (95% CI 42.6–44.1 months) for 11–20 cases per year, 42.2 months (95% CI 40.1–43.9 months) for 21–30 cases, and 49.1 months (46.8–50.6 months) for 31 cases per year or more. Based on the Cox proportional hazards model, care at a high-volume hospital was independently predictive of lower hazard of death (adjusted HR 0.86, 95% CI 0.82–0.90). Increasing age, non-Hispanic black race, Medicaid insurance, receipt of neoadjuvant chemotherapy, and stage IV disease were associated with higher hazards of death (Table 3). Other factors included in the model and the adjusted HR for those factors are presented in Table 4. The relative hazard of death based on the hospital volume of ovarian-debulking surgeries performed per year (as a continuous variable) by a facility, after adjusting of all the other covariates, is shown in Figure 3.
Our results remained unchanged on the sensitivity analysis. After excluding patients who died within 90 days after surgery, hospitals with 31 cases per year or more continued to have a higher risk-adjusted readmission rate, higher adherence to National Comprehensive Cancer Network guideline-based care, and improved 5-year overall survival (results not shown). In the additional sensitivity analyses, we found the following: 1) after excluding those receiving neoadjuvant chemotherapy; 2) and the subgroup of patients traveling less than 5 miles to the hospital. The findings were similar to the overall results reported previously (Appendix 1, available online at http://links.lww.com/AOG/B108).
Emerging evidence supports the notion that readmission rates do in fact reflect the quality of surgical care. Patients experiencing postoperative complications are much more likely to be readmitted.5,19,20 In a recent study of Medicare patients, high-performing hospitals (in terms of lower 30-day mortality and higher procedure volume) had a lower surgical readmission rate.21 However, the majority of research evaluating the performance of readmission rates as a meaningful metric of quality has not included patients undergoing cancer surgery. Although the Hospital Readmission Reduction Program currently does not include oncologic cases, it is still important to address the concerns extrapolating these data to include patients undergoing surgery for a malignancy.
First, a singular focus on readmission rates without consideration on the delivery of guideline-based care largely ignores the fact that in surgical oncology, quality is not only defined by 30-day outcomes, but also by the effect of surgery on the overall patient survival. Guideline-based resection of malignancies often involves extensive surgery in an effort to achieve negative resection margins, dissecting minimum number of lymph nodes, or maximal surgical effort to reduce the cancer burden.22,23 Adherence to these guidelines has been shown to improve survival.24–26 Therefore, financial penalties and hospital rankings based on short-term outcome measures such as 30-day readmission rates may unfairly target surgeons and hospitals undertaking complex oncologic procedures, in which higher initial morbidity may result in higher readmission rates but ultimately lead to an improvement in overall survival.27 Specifically in ovarian cancer, Barber et al28 analyzed the ovarian cancer cases in the National Cancer Database and reported that the receipt of neoadjuvant chemotherapy was associated with a reduction in readmission rates, but these patients had a decreased overall survival.
Second, in patients with a complex medical condition such as a malignancy, the reasons for readmission are often multifactorial with a complex interplay of medical and social factors. There is a limited number of studies examining the reasons for readmission after cancer surgery. In a nationwide study that included patients with cancer, after discharge from a surgical service, the most common reasons included infectious causes (46.3%), nausea or vomiting or dehydration (26.8%), and pain (6.1%).29 For many of these conditions, patients of lower socioeconomic status often prefer care in a hospital setting as a result of its relative ease of access and perceived superior technical quality.30 These readmissions are often preventable and result in hospital stays that are short and less expensive.30 However, studies have not examined the reasons for readmission or the severity of the readmission episode based on hospital characteristics. The lack of this information is a critical barrier in designing tailored strategies to prevent readmissions in different hospital settings.
Lastly, the notion of readmission reflecting a “failed discharge” has been challenged by studies arguing that certain readmissions are appropriate and necessary. In an analysis of the colectomy cases from the regional registry, Brown et al31 highlighted the fact that surgeons with the lowest patient mortality rates after colectomy also had the highest readmission rate. Similarly, analysis of the Medicare data of patients undergoing pancreaticoduodenectomy has shown an inverse relationship between the hospital readmission rate and 30-day mortality.31 These data along with our findings of lower 30- and 90-day readmission rates in high-volume hospitals with high readmission rates further support the hypothesis that higher readmission rates may represent earlier detection and rescue of the patient from severe complications and thereby reflect good clinical judgment, which potentially lower 30- and 90-day mortality rates.
Strengths of this study include an analysis of multiple academic and community hospitals with varying patient volumes across the nation. In addition, the National Cancer Database minimizes the bias by having data entered by certified tumor registrars trained in the acquisition of clinical information. Last, the National Cancer Database is the only data set in which short-term outcomes (30- and 90-day mortality, readmission) as well as long-term survival are available at a nationwide level.
There are some important weaknesses in this current study. First, the National Cancer Database only records readmission at the same facility and does not record the readmissions at different hospitals. Previous studies have shown that readmission to a different hospital is more likely when patients travel long distances.18 We therefore performed sensitivity analysis by restricting the data to patients residing within 5 miles of the treating hospital. This sensitivity analysis does not completely eliminate the possibility of readmission to a different hospital. High-volume hospitals are often referral centers and the actual readmission rate might be 12–20% higher than reported in the database.32 However, this is likely to strengthen the conclusions of our study because the high-volume hospitals might have even higher readmission rates than our results and, as a result, more likelihood of penalties under the hospital readmission reduction program. Second, we focused only on stage III and IV high-grade serous carcinomas in this analysis because the role of aggressive surgical debulking and chemotherapy is most well established in this histology. This group represents roughly 70% of the entire ovarian carcinoma spectrum, and the mortality rates are the highest in this subgroup. Analysis of the entire ovarian cancer group is unlikely to affect the findings of this study substantially.
In summary, in ovarian cancer surgery, hospitals with the highest volume of cases per year achieved higher overall survival, higher adherence to guideline-based care, and had lower 30- and 90-day mortality yet may still be unfairly penalized by the hospital ranking systems if 30-day readmission rates are the singular metric by which surgical quality is measured. Although patients with ovarian cancer represent a minority of surgical cases in the United States, our analysis highlights the need for further studies to quantify the effect of the recent emphasis on reduction of postsurgical complications on the overall survival in patients with cancer.
1. Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med 2009;360:1418–28.
2. Centers for Medicare and Medicaid Services. Hospital readmissions reduction program. Available at: https://www.cms.gov/medicare/medicare-fee-for-service-payment/acuteinpatientpps/readmissions-reduction-program.html
. Retrieved December 25, 2015.
3. Zuckerman RB, Sheingold SH, Orav EJ, Ruhter J, Epstein AM. Readmissions, observation, and the hospital readmissions reduction program. N Engl J Med 2016;374:1543–51.
4. Joynt KE, Jha AK. Characteristics of hospitals receiving penalties under the Hospital Readmissions Reduction Program. JAMA 2013;309:342–3.
5. Morris MS, Deierhoi RJ, Richman JS, Altom LK, Hawn MT. The relationship between timing of surgical complications and hospital readmission. JAMA Surg 2014;149:348–54.
6. Merkow RP, Ju MH, Chung JW, Hall BL, Cohen ME, Williams MV, et al. Underlying reasons associated with hospital readmission following surgery in the United States. JAMA 2015;313:483–95.
7. Wei S, Pierce O, Allen M; ProPublica. Surgeon scorecard. Available at: https://projects.propublica.org/surgeons/
. Retrieved January 21, 2017.
8. American College of Surgeons. Facility Oncology Registry Data Standards (FORDS): revised for 2015. Available at: https://www.facs.org/quality-programs/cancer/ncdb/registrymanuals/cocmanuals/fordsmanual
. Retrieved November 30, 2015.
9. Bilimoria KY, Stewart AK, Winchester DP, Ko CY. The National Cancer Data Base: a powerful initiative to improve cancer care in the United States. Ann Surg Oncol 2008;15:683–90.
10. Bristow RE, Tomacruz RS, Armstrong DK, Trimble EL, Montz FJ. Survival effect of maximal cytoreductive surgery for advanced ovarian carcinoma during the platinum era: a meta-analysis. J Clin Oncol 2002;20:1248–59.
11. Chang SJ, Hodeib M, Chang J, Bristow RE. Survival impact of complete cytoreduction to no gross residual disease for advanced-stage ovarian cancer: a meta-analysis. Gynecol Oncol 2013;130:493–8.
12. Wright JD, Chen L, Hou JY, Burke WM, Tergas AI, Ananth CV, et al. Association of hospital volume and quality of care with survival for ovarian cancer. Obstet Gynecol 2017;130:545–53.
13. Bristow RE, Powell MA, Al-Hammadi N, Chen L, Miller JP, Roland PY, et al. Disparities in ovarian cancer care quality and survival according to race and socioeconomic status. J Natl Cancer Inst 2013;105:823–32.
14. Spencer RJ, Hacker KE, Griggs JJ, Rice LW, Reynolds RK, Uppal S. Ninety-day mortality as a reporting parameter for high-grade serous ovarian cancer cytoreduction surgery. Obstet Gynecol 2017;130:305–14.
15. Rauh-Hain JA, Pepin KJ, Meyer LA, Clemmer JT, Lu KH, Rice LW, et al. Management for elderly women with advanced-stage, high-grade endometrial cancer. Obstet Gynecol 2015;126:1198–206.
16. Onukwugha E, Bergtold J, Jain R. A primer on marginal effects—part II: health services research applications. Pharmacoeconomics 2015;33:97–103.
17. Williams R. Using Stata's margins command to estimate and interpret adjusted predictions and marginal effects. Available at: http://www3.nd.edu/∼rwilliam/stats3/Margins01.pdf
. Retrieved November 30, 2016.
18. Stitzenberg KB, Chang Y, Smith AB, Nielsen ME. Exploring the burden of inpatient readmissions after major cancer surgery. J Clin Oncol 2015;33:455–64.
19. Uppal S, Penn C, Del Carmen MG, Rauh-Hain JA, Reynolds RK, Rice LW. Readmissions after major gynecologic oncology surgery. Gynecol Oncol 2016;141:287–92.
20. Dimick JB, Ghaferi AA. Hospital readmission as a quality measure in surgery. JAMA 2015;313:512–3.
21. Tsai TC, Joynt KE, Orav EJ, Gawande AA, Jha AK. Variation in surgical-readmission rates and quality of hospital care. N Engl J Med 2013;369:1134–42.
22. Morgan RJ Jr, Alvarez RD, Armstrong DK, Burger RA, Chen LM, Copeland L, et al. Ovarian cancer, version 2.2013. J Natl Compr Canc Netw 2013;11:1199–209.
23. Mathoulin-Pélissier S, Bécouarn Y, Belleannée G, Pinon E, Jaffré A, Coureau G, et al. Quality indicators for colorectal cancer surgery and care according to patient-, tumor-, and hospital-related factors. BMC Cancer 2012;12:297.
24. Boland GM, Chang GJ, Haynes AB, Chiang YJ, Chagpar R, Xing Y, et al. Association between adherence to National Comprehensive Cancer Network treatment guidelines and improved survival in patients with colon cancer. Cancer 2013;119:1593–601.
25. Galvan-Turner VB, Chang J, Ziogas A, Bristow RE. Observed-to-expected ratio for adherence to treatment guidelines as a quality of care indicator for ovarian cancer. Gynecol Oncol 2015;139:495–9.
26. Visser BC, Ma Y, Zak Y, Poultsides GA, Norton JA, Rhoads KF. Failure to comply with NCCN guidelines for the management of pancreatic cancer compromises outcomes. HPB (Oxford) 2012;14:539–47.
27. Aletti GD, Dowdy SC, Podratz KC, Cliby WA. Relationship among surgical complexity, short-term morbidity, and overall survival in primary surgery for advanced ovarian cancer. Am J Obstet Gynecol 2007;197:676.e1–7.
28. Barber EL, Rossi EC, Gehrig PA. Surgical readmission and survival in women with ovarian cancer: are short-term quality metrics incentivizing decreased long-term survival? Gynecol Oncol 2017;147:607–11.
29. Brown EG, Burgess D, Li CS, Canter RJ, Bold RJ. Hospital readmissions: necessary evil or preventable target for quality improvement. Ann Surg 2014;260:583–9.
30. Kangovi S, Barg FK, Carter T, Long JA, Shannon R, Grande D. Understanding why patients of low socioeconomic status prefer hospitals over ambulatory care. Health Aff (Millwood) 2013;32:1196–203.
31. Brown RE, Qadan M, Martin RC 2nd, Polk HC Jr. The evolving importance of readmission data to the practicing surgeon. J Am Coll Surg 2010;211:558–60.
32. Boffa DJ, Rosen JE, Mallin K, Loomis A, Gay G, Palis B, et al. Using the national cancer database for outcomes research: a review. JAMA Oncol 2017;3:1722–8.