Models to predict outcomes after primary debulking surgery: Independent validation of models to predict suboptimal cytoreduction and gross residual disease

Abstract

ObjectiveTreatment planning requires accurate estimation of surgical complexity (SC) and residual disease (RD) at primary debulking surgery (PDS) for advanced ovarian cancer (OC). We sought to independently validate two published computed tomography (CT) prediction models. MethodsWe included stage IIIC/IV OC patients who underwent PDS from 2003 to 2011. Two prediction models which included imaging and clinical variables to predict RD > 1 and any gross RD, respectively, were applied to our cohort. Two radiologists scored CTs. Discrimination was estimated using the c-index and calibration were assessed by comparing the observed and predicted estimates. ResultsThe validation cohort consisted of 276 patients; median age of the cohort was 64 years old and majority had serous histology. The validation and model development cohorts were similar in terms of baseline characteristics, however the RD rates differed between cohorts (9.4% vs 25.4% had RD >1 cm; 50.7% vs. 66.6% had gross RD). Model 1, the model to predict RD >1 cm, did not validate well. The c-index of 0.653 for the validation cohort was lower than reported in the development cohort (0.758) and the model over-predicted the proportion with RD >1 cm. The second model to predict gross RD had excellent discrimination with a c-index of 0.762. ConclusionsWe are able to validate a CT model to predict presence of gross RD in an independent center; the separate model to predict RD >1 cm did not validate. Application of the model to predict gross RD can help with clinical decision making in advanced ovarian cancer.