Prediction of the postoperative 90-day mortality after acute colorectal cancer surgery: development and temporal validation of the ACORCA model

Abstract

The aim of this study was to develop and validate a model to predict 90-day mortality after acute colorectal cancer surgery. The model was developed in all patients undergoing acute colorectal cancer surgery in 2014-2016 and validated in a patient group operated in 2017 in Denmark. The outcome was 90-day mortality. Tested predictor variables were age, sex, performance status, BMI, smoking, alcohol, education level, cohabitation status, tumour localization and primary surgical procedure. Variables were selected according to the smallest Akaike information criterion. The model was shrunken by bootstrapping. Discrimination was evaluated with a receiver operated characteristic curve, calibration with a calibration slope and the accuracy with a Brier score. A total of 1450 patients were included for development of the model and 451 patients for validation. The 90-day mortality rate was 19% and 20%, respectively. Age, performance status, alcohol, smoking and primary surgical procedure were the final variables included in the model. Discrimination (AUC = 0.79), calibration (slope = 1.04, intercept = 0.04) and accuracy (brier score = 0.13) were good in the developed model. In the temporal validation, discrimination (AUC = 0.80) and accuracy (brier score = 0.13) were good, and calibration was acceptable (slope = 1.19, intercept = 0.52). We developed prediction model for 90-day mortality after acute colorectal cancer surgery that may be a promising tool for surgeons to identify patients at risk of postoperative mortality.