OPTIONS & TOOLS
DOI: 10.1200/CCI.16.00053 JCO Clinical Cancer Informatics - published online March 28, 2017
Survival Disparities by Hospital Volume Among American Women With Gynecologic Cancers
We describe survival disparities among women with uterine, ovarian, or cervical cancer by cancer-specific mean annual hospital volume.
National Cancer Database 1998-2011 uterine (n = 441,863), ovarian (n = 223,017), and cervical (n = 146,698) cancer data sets were used. Cancer-specific mean annual hospital volumes were calculated. Overall survival (OS) was plotted by hospital volume using restricted mean OS times from Cox regression.
Uterine, ovarian, and cervical cancers were reported from 1,651, 1,633, and 1,600 hospitals, respectively. Median values of mean annual hospital volumes among hospitals were 8.6 (interquartile range [IQR], 2.6 to 20.8), 4.4 (IQR, 1.4 to 10.3), and 2.4 (IQR, 0.6 to 6.6) for uterine, ovarian, and cervical cancers, respectively. Increased hospital volume was associated with increased OS among women with stage III to IV high-grade serous ovarian cancer, stage II to IV squamous or adenocarcinoma cervical cancer, and stage I to IV endometrioid, clear cell, serous, or carcinosarcoma uterine cancers (all P < .03). Differential OS between women treated at higher- versus lower-volume cancer centers exceeded 5, 5, and 13 months among women with advanced endometrial, ovarian, or cervical cancer, respectively (all P < .001). Hospital volume was not associated with OS among patients with stage II to IV cervical cancer treated with brachytherapy (P = .17). Use of adjuvant therapies decreased OS disparities by hospital volume among women with advanced ovarian or endometrial cancer.
Increased delivery of brachytherapy for treatment of cervical cancer may decrease survival disparities by hospital volume. Standardization of adjuvant therapies may diminish survival disparities by hospital volume among women with advanced ovarian or endometrial cancer. In addition, survival of American women with gynecologic cancer may be increased by centralization of care.
Ovarian and uterine cancers are the fifth and sixth leading causes of cancer mortality among American women.1 Despite screening programs, cervical cancer continues to affect 0.6% of American women (one of 157) and is the second leading cause of cancer mortality among women age 20 to 39 years.1 Of these three gynecologic cancers, population-level overall survival (OS) improvement has been observed only in women with ovarian cancer, a 10% increase from 36% to 46% 5-year OS since the 1970s.1 One opportunity for improving OS is to decrease outcome disparities among cancer centers.
Women who receive care for ovarian cancer at hospitals treating > 20 ovarian cancers per year are more likely to receive care consistent with National Comprehensive Cancer Center (NCCN) guidelines and have increased OS.2,3 High-volume centers more often provide optimal cytoreductive surgery.4 Survival disparities among hospitals are complicated by disparities among women grouped by race or socioeconomic status (SES). Women with ovarian cancer who are black, are uninsured, have Medicare, or who have lower SES are more likely to have incomplete surgical staging.5 Even among black women who receive NCCN-adherent care, the hazard of death is increased compared with that of white women.6 The argument for centralization of ovarian cancer care as a quality intervention was recently reviewed.4
Disparities among women with endometrial or cervical cancer by differences in cancer-specific hospital volume are less reported. Both endometrial and cervical cancer mortality rates were increased for women in the state of New York who lived farther from a cancer center.7 Race, insurance status, and treatment at low-volume hospitals were associated with OS among women with endometrial cancer.8 OS among black women remains decreased compared with that for white women for all common gynecologic cancers.9
All of the above data are from analyses of cancer registries. It remains unclear whether OS increases or plateaus at higher hospital volumes. Cervical cancer survival by hospital volume remains uncharacterized. Here we show survival disparities among women with uterine, ovarian, or cervical cancer by cancer-specific hospital volume.
The National Cancer Database (NCDB) is a hospital-based national cancer registry created by the American College of Surgeons and American Cancer Society, and it includes an estimated 69%, 75%, and 96% of all ovarian, uterine, and cervical malignancies, respectively, diagnosed nationally.10 Individual-level data are entered by professional registrars and are audited.10
The 1998-2011 NCDB uterine (n = 441,863), ovarian (n = 223,017), and cervical (n = 146,698) cancer data sets were used for this study. For patients with ovarian cancer, women with serous (8010, 8140, 8441, 8460, 8461), clear cell (8310), endometrioid (8380), or mucinous (8480) International Classification of Diseases for Oncology, 3rd revision (ICD-O-3) histologies who received systemic chemotherapy and stage-appropriate surgery were included in the overall ovarian cancer cohort used for multivariable survival analysis (n = 104,766). Women with stage III to IV grade 3 or 4 serous ovarian cancer were analyzed separately for restricted mean survival analyses of an advanced-stage, high-grade serous cohort (n = 47,991). For ovarian cancer, stage-appropriate surgery was considered cytoreduction/debulking or hysterectomy with oophorectomy with or without omentectomy. For patients with cervical cancer, women with locally advanced or metastatic stage II to IV squamous (8070-2) or adenocarcinoma (8140) ICD-O-3 histologies were analyzed (n = 57,987). For endometrial cancers, women with endometrioid (8380-3), clear cell (8310, 8313), serous (8441, 8450, 8460, 8461), or carcinosarcoma (8950, 8951, 8980, 8981) ICD-O-3 histologies were included (n = 306,221). Subsets of women with type I (endometrioid stage III [n = 23,953] or stage IV [n = 7,144]) or type II (serous, clear cell, or carcinosarcoma, stage III [n = 11,340] or stage IV [n = 8,757]) were also analyzed as advanced endometrial cancer cohorts.
Cancer-specific mean annual hospital volumes were calculated as the total number of cancer-specific cases from 1998 to 2011 divided by 14 years. Centers were classified as community, academic, or other. For ovarian and endometrial cancers, administration of adjuvant radiation (RT) or chemotherapy was classified as yes, no, or unknown by using the summary treatment variables. For cervical cancer, adjuvant RT was classified as external beam only or brachytherapy with or without external beam RT (EBRT). The NCDB analytic stage group was used for all cohorts. Additional covariates included age at diagnosis, Charlson/Deyo composite comorbidity score, grade of disease, race, insurance status, median household income quartile by ZIP code, and community high school dropout rates. Race was grouped as white, black, other, or unknown. The “other” group included many race codes describing mostly East Asians and South Asians, Pacific Islanders, and Native Americans. The counts of these specific race codes were too low for multivariable regressions. Insurance status was coded by the NCDB as uninsured, private insurance, Medicaid, Medicare, government, or unknown.
Baseline patient and disease characteristics were tabulated. As indicated in Results, the Mann-Whitney U test was used for two-group comparisons, the Kruskal-Wallis test was used for multiple-group comparisons of nonparametric numeric data, and Fisher’s exact test was used to compare categorical variable counts. Survival analyses were performed for women with clinical follow-up data. Multivariable Cox proportional hazards regression models were created by stepwise selection using the Efron approximation for ties of event times. Initial models included all presumed main-effect covariates. The proportional hazards assumption was checked at each step, and models were stratified as needed. Interactions between covariates were tested and are described in Results if they were significant and meaningful. The reported statistical significance and hazard ratio (HR) estimates for hospital volume were robust to differences in model selection. The analysis of deviance table verified that terms in each final model significantly improved the model. Goodness of fit was confirmed by quantitatively evaluating deviance residuals. Restricted mean survival curves and relative hazard estimates with 95% CIs were plotted from univariable and multivariable Cox regressions, respectively, of OS as a function of hospital volume.11 Analyses were performed by using the “survival” and “rms” R packages (http://www.R-project.org).
Baseline patient and disease characteristics are reported in Table 1. Uterine, ovarian, and cervical cancers were reported from 1,651, 1,633, and 1,600 hospitals, respectively. Histograms of hospital counts by cancer-specific hospital volume are shown in Appendix Figure A1 and indicate that most cancer centers treat a low volume of women for each cancer type. Few hospitals are high-volume facilities for each cancer. Median values of mean annual hospital volumes among hospitals were 8.6 (interquartile range [IQR], 2.6 to 20.8), 4.4 (IQR, 1.4 to 10.3), and 2.4 (IQR, 0.6 to 6.6) patients per year for uterine, ovarian, and cervical cancers, respectively. Median values of mean annual hospital volumes among women treated for cancer were 50.9 (IQR, 21.6 to 83.2), 27.0 (IQR, 12.9 to 43.0), and 16.1 (IQR, 6.9 to 29.9) patients per year for uterine, ovarian, and cervical cancers, respectively. Because more women are treated at higher-volume than at lower-volume centers, the median volume values among women treated are higher than the median volume values among cancer centers. Women treated at academic hospitals were treated at higher-volume centers than women treated at community hospitals (P < .001; Appendix Table A1). Hospital volumes were significantly but not largely different among women with different socioeconomic and demographic characteristics (Appendix Table A2).
Among women with ovarian cancer, the hazard of death decreased 3% (95% CI, 1% to 4%) for each 20-patients-per-year increase in mean annual ovarian cancer hospital volume (Table 2). In this model, increased grade, black race, Medicaid or Medicare insurance, and increased comorbidity scores were independently associated with increased hazard of death, and highest income quartile was associated with increased survival (Table 2). Modeling a significant (P < .001) interaction between race and income demonstrated that, compared with white women from communities in the lowest income quartile, black women from communities in the lowest income quartile had increased hazard of death (HR, 1.44; 95% CI, 1.22 to 1.62; P < .001). Black women living in communities from the second, third, and fourth income quartiles had progressively increased OS (second quartile: HR, 1.23; 95% CI, 1.06 to 1.44; P = .007; third quartile: HR, 1.11; 95% CI, 0.96 to 1.28; P = .16; fourth quartile: HR, 0.96; 95% CI, 0.82 to 1.12; P = .62). Among white women, community income quartile was not significantly associated with OS. Among the subset of women with stage III to IV high-grade serous ovarian cancer, hospital volume was significantly associated with OS (Fig 1). Women treated at cancer centers that had 5, 25, or 50 patients per year had OS times of 49.4 months (95% CI, 48.8 to 50.0 months), 51.7 months (95% CI, 51.1 to 52.4 months), and 54.7 months (95% CI, 54.0 to 55.4 months), respectively (P < .001). Receiving surgery and chemotherapy as part of their initial disease management decreased the magnitude of the survival difference between women treated at lower- versus higher-volume hospitals (Fig 1). Women treated with surgery and chemotherapy at cancer centers with 5, 25, or 50 patients per year had OS times of 54.7 months (95% CI, 53.9 to 55.4 months), 56.3 months (95% CI, 55.5 to 57.1 months), and 58.5 months (95% CI, 57.7 to 59.4 months), respectively (P < .001).
Among women with endometrial cancer, the hazard of death decreased 1% to 2% per each 20-patients-per-year increase in mean annual uterine cancer hospital volume (Table 2). Black race was associated with decreased OS, and other race group and increased income quartiles were associated with increased OS (Table 2). Among women with all stages of endometrioid, clear cell, serous, or carcinosarcoma endometrial cancers, increased hospital volume was associated with increased OS (Appendix Fig A2). Hospital volume was significantly associated with OS among subsets of women with stage III or IV type I or type II endometrial cancers (Fig 2). Women with type II stage III disease had OS of 52.8 months (95% CI, 51.2 to 54.4 months), 54.5 months (95% CI, 52.8 to 56.2 months), or 58.0 months (95% CI, 56.1 to 59.8 months) if treated at hospitals with 25, 50, or 100 endometrial cancers per year, respectively (P < .001). Similarly, OS estimates of women with type I stage III endometrial cancer ranged from 96.9 months (95% CI, 95.5 to 98.4 months) to 103.4 months (95% CI, 101.8 to 105.0 months) between women treated at hospitals with a relatively low volume of 25 patients per year compared with hospitals with 100 patients per year (P < .001). Similar differences in OS times by hospital volume were also seen for women with stage IV endometrial cancers (Fig 2). Among subsets of these women who received adjuvant RT and/or chemotherapy, hospital volume often lost its association with OS (Fig 2).
Among women with stage II to IV squamous or adenocarcinoma cervical cancer, the hazard of death decreased 5% (95% CI, 3% to 8%) for each 20-patients-per-year increase in mean annual cervical cancer hospital volume (Table 2). Medicaid, Medicare, and increased comorbidities were associated with decreased OS, other race group was associated with increased OS, and highest income quartile was marginally associated with increased OS (Table 2). Increased hospital volume was associated with increased OS for each stage II, III, or IV group (Fig 3). When women with stage II to IV cervical cancer were grouped by their RT treatment, lower hospital volume was strongly associated with decreased OS for women who received EBRT only (Fig 3). Women with stage II to IV cervical cancer treated with EBRT at cancer centers with 5, 25, or 50 patients per year had OS times of 64.9 months (95% CI, 63.7 to 66.1 months), 70.6 months (95% CI, 69.3 to 72.0 months), and 78.0 months (95% CI, 76.5 to79.5 months), respectively (P < .001). However, hospital volume was not significantly associated with OS among women who received brachytherapy having similar OS times of 88.5 to 92.4 months, presumably reflecting a higher cure rate after brachytherapy across the range of observed hospital volumes (Fig 3). This suggests that the differential OS observed for cervical cancer by hospital volume is partially a result of the differences in access to brachytherapy.
Among women who received EBRT alone or brachytherapy with or without EBRT, the odds ratio for receiving brachytherapy was 0.65 (95% CI, 0.60 to 0.71) among women treated at a hospital with less than the median number (≤ 2.4 patients per year) of mean annual cervical cancer volume (Fisher’s exact test P < .001). Only 36.6% of women (949 of 2,596) treated at these lower-volume facilities received brachytherapy. Only at cancer centers that treated > 20 cervical cancers per year did 50.0% of women (8,154 of 16,303) receive brachytherapy as part of their RT for stage II to IV cervical cancer. Among women who received only EBRT, total regional dose was more often lower at the lower-volume (≤ 2.4 patients per year) hospitals (total regional dose: median, 45 Gy [IQR, 39.6 to 48.95 Gy] v 45 Gy [IQR, 41.4 to 50.4 Gy]); Mann-Whitney U test P < .001). Volumes for ovarian and endometrial cancers are consistently and positively linearly correlated across American cancer centers (Appendix Fig A3). However, some high-volume cervical cancer centers are not particularly high volume for endometrial cancer (Appendix Fig A3).
Appendix Table A3 shows HR estimates for subsets of women treated at community versus academic cancer centers. Hospital volume associations reached significance only for women with ovarian and endometrial cancer treated at academic centers. Histograms show that proportionally fewer women were treated at high-volume community hospitals, which may explain why hospital volume did not reach significance in the community (Appendix Fig A4). Conversely, women with cervical cancer treated at community hospitals had decreased hazard of death with increasing hospital volume (Appendix Table A3). Relative hazard predicted from the multivariable Cox models as functions of volume illustrate these associations (Fig 4).
Survival is significantly increased with increasing hospital volume for women with ovarian, endometrial, or cervical cancer. The hazard of death with decreasing hospital volume may be greatest among women with cervical cancer. However, survival disparities between lower- versus higher-volume cancer centers were not observed among women with stage II to IV cervical cancer who received brachytherapy as part of their management. For advanced-stage ovarian and endometrial cancers, differences in the administration of adjuvant chemotherapy and/or RT partially explained the observed differences in OS by hospital volume. For instance, use of adjuvant chemotherapy diminished or eliminated OS disparities among subsets of women with advanced endometrial cancer (Fig 2), suggesting that differences in use of adjuvant therapies drive hospital volume survival associations among women with endometrial cancer.
Our results are consistent with those previously reported for women with ovarian cancer, namely that survival was increased among women who receive guideline-consistent management and treatment at higher-volume hospitals.2,3,12 One series from a military hospital demonstrated that lower-volume hospitals may provide guideline-consistent ovarian cancer care if care is managed by a gynecologic oncologist.13 Another series of 367 women from an NCCN-designated cancer center reported institutionally specific reasons for failure to adhere to NCCN guidelines for ovarian cancer.14 Adherence failures were typically chemotherapy nonadherence as a result of postoperative death or early discontinuation related to toxicities.14 In addition to guideline nonadherence, smaller centers are less likely to provide intraperitoneal chemotherapy.4
Our observations that women who live in communities in the highest income quartile group and those with private insurance were more often treated at higher-volume centers for endometrial and ovarian cancers may reflect resource seeking among women who are less resource limited. Alternatively, these observations may simply reflect that higher-volume cancer centers are more often located in more heavily populated cities with higher median incomes by ZIP code. An effect of some high-volume referral centers that often do not accept uninsured or underinsured women is also a potential explanation for these findings.
Cancer centers with higher volumes of ovarian and endometrial cancers treat women of a higher SES more often than lower-volume cancer centers do. Black women living in low-income communities are a population especially at-risk for death after diagnosis of ovarian cancer, even among women who receive management consistent with NCCN guidelines in terms of stage-appropriate surgery and treatment with chemotherapy. A gradient effect of increased survival with increased community income was observed among black women but was not observed among white women. This suggests that black communities with differential median incomes have differential access to care or quality of care that white women do not experience.
Conversely, women with cervical cancer and with lower SES based on insurance status and income quartile were more often treated at higher-volume cervical cancer centers. Our results suggest that survival of women with cervical cancer may be increased by increasing access to standard-of-care brachytherapy and standardizing EBRT.
To the best of our knowledge, this study is the largest analysis of survival outcomes by hospital volume in gynecologic cancers. We add to the existing literature by this analysis of women with cervical cancer. Our subset analyses of women with specific adjuvant treatments show that OS disparities by hospital volume may be eliminated or decreased by standardizing therapy. We also analyzed hospital volume as a continuous variable, which allowed for improved estimation of hazards associated with differences in cancer center volume.
Limitations of this study include selection biases that cannot be accounted for with the NCDB variables. For example, women treated at low-volume centers were probably less often treated by surgeons, radiation oncologists, and medical oncologists with gynecologic cancer–focused practices. In addition, higher-volume centers may see more women with incurable disease who are motivated to join clinical trials or receive additional lines of therapy rather than transition to hospice care. Coding of variables in the NCDB is also limited in that the reporting facility may not be the treatment facility for specific therapies. A large number of women in the NCDB have missing data, such as clinical follow-up, that could be used for multivariable regressions. However, if more women with complete information were included in the regression models, it would narrow the CIs of the reported HRs without changing the HR estimates, given the large number of women in the reported analyses and that these estimates were robust to differences in model selection.
More detailed studies are needed to better understand the complex survival disparities among American women treated at higher- versus lower-volume cancer centers. Currently, large cancer administrative data sets remain our best data sources. Public health and health services opportunities for immediate OS improvement for women with gynecologic cancers include centralization and standardization of care. Increased centralization of gynecologic cancer care to higher-volume regional cancer centers is more realistic than increasing standardization of care among a large number of lower-volume centers, many of which may be resource limited. For example, brachytherapy requires specific facilities and credentialing that many low-volume centers may not be able to establish or maintain. Both centralization and standardization should be a priority of health system administrators, cancer-center accreditors, health care policy makers, and payers to realize the opportunities to increase OS among women with gynecologic cancer.
Conception and design: Brandon-Luke L. Seagle, Anna E. Strohl, Shohreh Shahabi
Administrative support: Wilberto Nieves-Neira, Shohreh Shahabi
Provision of study materials or patients: Brandon-Luke L. Seagle, Anna E. Strohl
Collection and assembly of data: Brandon-Luke L. Seagle
Data analysis and interpretation: Brandon-Luke L. Seagle, Monica Dandapani, Wilberto Nieves-Neira
Manuscript writing: All authors
Final approval of manuscript: All authors
Accountable for all aspects of the work: All authors
The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.
No relationship to disclose
No relationship to disclose
No relationship to disclose
Speakers' Bureau: Caris Life Sciences
Travel, Accommodations, Expenses: Caris Life Sciences
No relationship to disclose
We thank the Surgical Quality and Outcomes Improvement Center of Northwestern University. B.-L.L.S. thanks Kevin H. Eng, PhD, Department of Biostatistics and Bioinformatics, Roswell Park Cancer Institute, for statistical discussions and Jonathan B. Strauss, MD, MBA, for discussions regarding radiotherapy.
|1.||Siegel RL, Miller KD, Jemal A: Cancer statistics, 2016. CA Cancer J Clin 66:7-30, 2016 Crossref, Medline, Google Scholar|
|2.||Bristow RE, Palis BE, Chi DS, et al: The National Cancer Database report on advanced-stage epithelial ovarian cancer: Impact of hospital surgical case volume on overall survival and surgical treatment paradigm. Gynecol Oncol 118:262-267, 2010 Crossref, Medline, Google Scholar|
|3.||Bristow RE, Chang J, Ziogas A, et al: Adherence to treatment guidelines for ovarian cancer as a measure of quality care. Obstet Gynecol 121:1226-1234, 2013 Crossref, Medline, Google Scholar|
|4.||Cowan RA, O’Cearbhaill RE, Gardner GJ, et al: Is it time to centralize ovarian cancer care in the United States? Ann Surg Oncol 23:989-993, 2016 Crossref, Medline, Google Scholar|
|5.||Hodeib M, Chang J, Liu F, et al: Socioeconomic status as a predictor of adherence to treatment guidelines for early-stage ovarian cancer. Gynecol Oncol 138:121-127, 2015 Crossref, Medline, Google Scholar|
|6.||Bristow RE, Powell MA, Al-Hammadi N, et al: Disparities in ovarian cancer care quality and survival according to race and socioeconomic status. J Natl Cancer Inst 105:823-832, 2013 Crossref, Medline, Google Scholar|
|7.||Tan W, Stehman FB, Carter RL: Mortality rates due to gynecologic cancers in New York state by demographic factors and proximity to a Gynecologic Oncology Group member treatment center: 1979-2001. Gynecol Oncol 114:346-352, 2009 Crossref, Medline, Google Scholar|
|8.||Fader AN, Habermann EB, Hanson KT, et al: Disparities in treatment and survival for women with endometrial cancer: A contemporary National Cancer Database registry analysis. Gynecol Oncol 143:98-104, 2016 Crossref, Medline, Google Scholar|
|9.||Chatterjee S, Gupta D, Caputo TA, et al: Disparities in gynecological malignancies. Front Oncol 6:36, 2016 Crossref, Medline, Google Scholar|
|10.||Bilimoria KY, Stewart AK, Winchester DP, et al: The National Cancer Data Base: A powerful initiative to improve cancer care in the United States. Ann Surg Oncol 15:683-690, 2008 Crossref, Medline, Google Scholar|
|11.||Eng KH, Seagle BL: Covariate-adjusted restricted mean survival times and curves. J Clin Oncol 35:465-466, 2017 Link, Google Scholar|
|12.||Cliby WA, Powell MA, Al-Hammadi N, et al: Ovarian cancer in the United States: Contemporary patterns of care associated with improved survival. Gynecol Oncol 136:11-17, 2015 Crossref, Medline, Google Scholar|
|13.||Phippen NT, Barnett JC, Lowery WJ, et al: Surgical outcomes and national comprehensive cancer network compliance in advanced ovarian cancer surgery in a low volume military treatment facility. Gynecol Oncol 131:158-162, 2013 Crossref, Medline, Google Scholar|
|14.||Erickson BK, Martin JY, Shah MM, et al: Reasons for failure to deliver National Comprehensive Cancer Network (NCCN)-adherent care in the treatment of epithelial ovarian cancer at an NCCN cancer center. Gynecol Oncol 133:142-146, 2014 Crossref, Medline, Google Scholar|