Overview of models for predicting risk of developing ovarian cancer: a systematic review

Abstract Background Ovarian cancer is the deadliest gynaecological malignancy, and most cases are diagnosed at an advanced stage. There is no population-wide screening; models for predicting risk of ovarian cancer could be useful in decreasing mortality. We aimed to provide the first comprehensive overview of models for predicting the risk of developing ovarian cancer and their performance, risk of bias, and applicability. Methods We performed a systematic search through MEDLINE and Embase to identify articles in which models for predicting risk of developing ovarian cancer were developed and/or validated. Model characteristics and measures of predictive ability were extracted and summarised for each model. The Prediction model Risk of Bias Assessment Tool (PROBAST) was used to assess risk of bias and applicability. Results We identified twenty-eight studies representing twenty-one ovarian cancer risk prediction models. Fifteen risk prediction models were developed or validated in the general population, while six were developed for high-risk populations, defined by family history, and one for patients with endometriosis. Twenty-three studies that reported discrimination and/or calibration had moderate-to-good performance. Risk prediction models with the best predictive ability reported an area under the receiver-operating characteristic curve (AUC) of 0.86 (95% confidence interval (CI) 0.84–0.87) for 2-year and an AUC of 0.77 (95% CI 0.76-0.78) for 10-year ovarian cancer risk in the general population. Conclusions Several models for predicting short- and long-term risk of ovarian cancer have been published, of which a few have been validated in other populations. Some have limited predictive performance. Some models may be suitable for use in primary care or specialty clinics; however, assessment of their utility and validation in diverse cohorts is warranted.