Dynamic Prediction of Resectability for Patients with Advanced Ovarian Cancer Undergoing Neo-Adjuvant Chemotherapy: Application of Joint Model for Longitudinal CA-125 Levels

In patients with advanced ovarian cancer (AOC) receiving neoadjuvant chemotherapy (NAC), predicting the feasibility of complete interval cytoreductive surgery (ICRS) is helpful and may avoid unnecessary laparotomy. A joint model (JM) is a dynamic individual predictive model. The aim of this study was to develop a predictive JM combining CA-125 kinetics during NAC with patients' and clinical factors to predict resectability after NAC in patients with AOC. A retrospective study included 77 patients with AOC treated with NAC. A linear mixed effect (LME) sub-model was used to describe the evolution of CA-125 during NAC considering factors influencing the biomarker levels. A Cox sub-model screened the covariates associated with resectability. The JM combined the LME sub-model with the Cox sub-model. Using the LME sub-model, we observed that CA-125 levels were influenced by the number of NAC cycles and the performance of paracentesis. In the Cox sub-model, complete resectability was associated with Performance Status (HR = 0.57, [0.34-0.95], p = 0.03) and the presence of peritoneal carcinomatosis in the epigastric region (HR = 0.39, [0.19-0.80], p = 0.01). The JM accuracy to predict complete ICRS was 88% [82-100] with a predictive error of 2.24% [0-2.32]. Using a JM of a longitudinal CA-125 level during NAC could be a reliable predictor of complete ICRS.