Predictors of Long-Term Survival among High-Grade Serous Ovarian Cancer Patients

Despite intense study, relatively little is known about factors associated with long-term survival (LTS) from ovarian cancer (1). Ovarian cancer is rare and survival is poor (47% 5-year survival; ref. 2). These characteristics make it difficult to study in a single institution or cohort. We conducted a retrospective observational study of high-grade serous ovarian cancer (HGSOC) to explore predictors of LTS, using data from five integrated health systems in the Cancer Research Network (CRN, https://crn.cancer.gov; ref. 3).

All data for this analysis were derived from the Virtual Data Warehouse (VDW), a common data model that includes standardized, individual-level data extracted from the electronic medical record from the participating sites (4). We identified all incident cases of ovarian cancer diagnosed between 2000 and 2008 at five integrated health systems affiliated with the CRN: Health Partners (Minnesota), and four Kaiser Permanente regions: Colorado, Northern California, Northwest, and Washington state. This study received Institutional Review Board approval and a waiver of informed consent.

We limited this analysis to cases of incident high-grade (grade 3 or 4) serous (morphology codes between 8440 and 8490) ovarian cancer, stages I–IV. We defined LTS as survival ≥7 years and used the VDW to follow patients through December 31, 2015 for death, disenrollment, or end of follow-up. Cases were included if they were age 18 years or older at diagnosis and enrolled continuously with a drug coverage in their health care plan for at least 1 year prior to diagnosis. Cases were excluded if they were known to be alive but disenrolled before 7 years of follow-up.

Logistic regression was employed to model associations between covariates and the outcome of LTS. All covariates were assessed in the year prior to diagnosis. We used forward-selection to build the model and retained all variables with a P < 0.3. We decided a priori that age, stage, and CA-125 level at diagnosis would be in the final model as these factors have been previously reported to be associated with survival among ovarian cancer cases (5, 6). We computed the Elixhauser comorbidity score (7) for each case and included individual components of the score in the model-building process if at least 20 patients had the condition. All analyses were conducted using SAS/STAT software, Version 9.4 (SAS Institute Inc.). We conducted a post hoc power calculation for our study. We had 80% power, using a two-sided test with an α = 0.05, to detect an OR >1.6 with one primary predictor. Power calculations were performed in PASS.

We identified 2,388 incident stage I to IV ovarian cancers. Of these, we excluded 298 patients who were not enrolled at least 1 year prior to diagnosis with a drug coverage benefit, and 160 who did not have a record of death and were not continuously enrolled at least 7 years post diagnosis. There were an additional 1,288 cancers that were not high-grade serous tumors, of those, 62% (n = 578) were missing information to classify the tumor grade. This left a total of 642 high-grade serous ovarian cancers for analysis.

Table 1 shows the characteristics that were considered for inclusion in the logistic regression model. The median survival time was 38 months [95% confidence interval (CI): 35–42 months]; 148 (23%) survived at least 7 years and were considered LTSs, while the remainder (N = 494, 77%) survived fewer than 7 years. Our final logistic regression model included study site, age, stage at diagnosis, CA-125 (categorized as <35 or ≥35 U/mL), comorbidity score (categorized as 0, 1, 2, or ≥3), receipt of chemotherapy, body mass index (BMI, in categories), and four separate comorbid conditions: weight loss, depression, hypothyroidism, and liver disease (see Fig. 1 for ORs and 95% CIs). Information on BMI and smoking status were missing in approximately half of the population. Patients diagnosed at an earlier stage had statistically significantly higher odds of LTS than patients diagnosed at a later stage (P < 0.0001). Younger patients were statistically more likely to survive than older patients (P = 0.009), and a diagnosis of depression within the year prior to a cancer diagnosis was inversely associated with LTS (P = 0.011). Our final model had a c-statistic of 0.806 suggesting a strong model fit.

In this exploratory analysis using electronic medical record data, we did not identify any new medical conditions or characteristics that may be associated with LTS of HGSOC. Others (8) have reported on comorbidities associated with ovarian cancer survival with mixed results. The strengths of this study include a rich data resource of an insured population that allows for detailed capture of comorbidities, treatment, and follow-up. However, we lacked information on BMI and smoking status on a large proportion of the study population. We had no data on family history or genetic susceptibility that would have strengthened our analysis.

No potential conflicts of interest were disclosed.

Conception and design: L.H. Kushi, P.A. Pawloski, B. Powell, C.L. Pearce, H.S. Feigelson

Development of methodology: C.L. Clarke, L.H. Kushi, P.A. Pawloski, J.E. Bulkley, A.N. Burnett-Hartman, B. Powell, H.S. Feigelson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.L. Clarke, L.H. Kushi, J. Chubak, P.A. Pawloski, J.E. Bulkley, M.M. Epstein, B. Powell, H.S. Feigelson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.L. Clarke, J. Chubak, P.A. Pawloski, J.E. Bulkley, M.M. Epstein, B. Powell, H.S. Feigelson

Writing, review, and/or revision of the manuscript: C.L. Clarke, L.H. Kushi, J. Chubak, P.A. Pawloski, J.E. Bulkley, M.M. Epstein, A.N. Burnett-Hartman, B. Powell, C.L. Pearce, H.S. Feigelson

Study supervision: H.S. Feigelson

This work was conducted in the Cancer Research Network (CRN) funded by the NCI (U24CA171524). M.M. Epstein received additional support by a grant from The University of Massachusetts Center for Clinical and Translational Science (KL2TR01455). C.L. Pearce received grant funding from the NCI (P30 CA046592).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.