Development of a prediction model for residual disease in newly diagnosed advanced ovarian cancer

Objectives. To construct a tool, using computed tomography (CT) imaging and preoperative clinical variables, to estimate successful primary cytoreduction for advanced epithelial ovarian cancer (EOC). Methods. Women who underwent primary cytoreductive surgery for stage IIIC/IV EOC at Mayo Clinic between 1/2/2003 and 12/30/2011 and had preoperative CT images of the abdomen and pelvis within 90 days prior to their surgery available for review were included. CT images were reviewed for large-volume ascites, diffuse peritoneal thickening (DPT), omental cake, lymphadenopathy (LP), and spleen or liver involvement. Preoperative factors included age, body mass index (BMI), Eastern Cooperative Oncology Group performance status (ECOG PS), American Society of Anesthesiologists (ASA) score, albumin, CA-125, and thrombocytosis. Two prediction models were developed to estimate the probability of (i) complete and (ii) suboptimal cytoreduction (residual disease (RD) >1 cm) using multivariable logistic analysis with backward and stepwise variable selection methods. Internal validation was assessed using bootstrap resampling to derive an optimism-corrected estimate of the c-index. Results. 279 patients met inclusion criteria: 143 had complete cytoreduction, 26 had suboptimal cytoreduction (RD > 1 cm), and 110 had measurable RD cm. On multivariable analysis, age, absence of ascites, omental cake, and DPT on CT imaging independently predicted complete cytoreduction (c-index = 0.748). Conversely, predictors of suboptimal cytoreduction were ECOG PS, OFT, and LP on preoperative CT imaging (c-index = 0.685). Conclusions. The generated models serve as preoperative evaluation tools that may improve counseling and selection for primary surgery, but need to be externally validated. (C) 2015 Elsevier Inc. All rights reserved.