To predict or not to predict? The dilemma of predicting the risk of suboptimal cytoreduction in ovarian cancer

Abstract

Although maximal cytoreduction is the cornerstone of current treatment for patients with advanced ovarian cancer, optimal cytoreduction is not always achievable in the clinic. Therefore, using clinical characteristics, diagnostic imaging, serum biomarkers or laparoscopic findings, many studies have attemptesd to find models for predicting surgical resectability. However, most of these prediction models showed limited effectiveness and have not been properly validated. To establish a reliable prediction model, several requirements should be met. First, the goal of surgical cytoreduction should be adequately defined. Second, the desired accuracy for making the model clinically useful should be defined. Third, the model should test all relevant predictors, including clinical, radiological and biochemical predictors, and be developed using a large dataset that provides a sufficient number of events. Fourth, any prediction model should be validated with a relevant external dataset. Lastly, the prediction model should be able to aid decision making and, thereby, improve the outcome of patients. Therefore, randomized clinical trials with decision making based on prediction models are urgently required.