Development and validation of a histology-specific natural history model of ovarian cancer.

backgroundOvarian cancer is the second leading cause of death from gynecologic cancers, yet no effective screening program exists for the general population. Past screening trials evaluated the effectiveness of annual ovarian cancer screening and concluded that it does not yield substantial mortality reduction. Future investments on ovarian cancer screening trials would require convincing preliminary evidence on the effectiveness of interventions of interest. Simulation modeling is an effective, fast, cost-efficient, and safe approach to gain insights on the effectiveness of interventions, that is increasingly being used to inform guidelines for cancer screening programs. Models that simulate the natural progression of diseases in the absence of any intervention, commonly referred to as natural history models, are the cornerstone of such analyses, because they provide a reference point for evaluating interventions. Currently, no histology-specific natural history model exists for ovarian cancer despite major differences among subtypes.

objectiveDevelop and validate a histology-specific ovarian cancer natural history model. STUDY

designWe developed natural history models for the most common histological subtypes of epithelial ovarian cancer: high-grade serous carcinoma, low-grade serous carcinoma, mucinous carcinoma, clear cell carcinoma, endometrioid carcinoma, carcinosarcoma, and not otherwise specified. Each natural history model simulates the natural progression of ovarian cancer from disease's onset until death from any cause. We modeled ovarian cancer progression as a state-transition model comprising of 13 mutually exclusive and collectively exhaustive health states. We informed the model input parameters using observed, nationally representative estimates, whenever possible. Unobserved parameters (eg, preclinical transitions) were estimated through calibration to histology-specific data from the Surveillance, Epidemiology, and End Results registry. We validated the natural history models on the control arms of the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial and the United Kingdom Collaborative Trial of Ovarian Cancer Screening trials, in terms of ovarian cancer incidence and mortality rates, and stage distribution at diagnosis. Differences between observed and estimated outcomes were assessed using traditional statistical tests.

resultsThe calibrated natural history models reproduced the observed Surveillance, Epidemiology, and End Results data (range of weighted root mean square error across histological subtypes: 0.0081-0.0185) as well as individual calibration targets; survival after diagnosis, stage distribution at diagnosis, and age distribution at diagnosis (ranges of mean square error across histological subtypes: 0.0029-0.0204, 0.0005-0.0203, and 0.0637-0.0816, respectively). The natural history models reproduced the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial's observed incidence and mortality rates, and stage at diagnosis (P value=.411 for incidence, P value=.195 for mortality, and P value=.200 for stage distribution at diagnosis) and the United Kingdom Collaborative Trial of Ovarian Cancer Screening's observed ovarian cancer incidence (P value=.607) and mortality (P value=.624) rates. The average duration of the preclinical phase ranges between 1 and 3 years, which partly explains screening's failure to yield mortality reduction. Moreover, stage II ovarian cancer, independent of histological subtype, is a transient state characterized by noticeably shorter average duration when compared to stages I, III, and IV.

conclusionThe developed natural history models accurately describe the histology-specific natural progression of ovarian cancer and provide important insights into the natural history of the disease. The models may be used to evaluate the impact of future and emerging ovarian cancer interventions, thus providing valuable information to decision-makers and policymakers.