Proposal of a Multivariable Prediction Model for Graded Morbidity after Liver Resection for Colorectal Metastases

Abstract

Abstract Background Prognostic models to predict individual early postoperative morbidity after liver resection for colorectal liver metastases (CLM) are not available but could enable optimized preoperative patient selection and postoperative surveillance for patients at greater risk of complications. The aim of this study was to establish a prognostic model for the prediction of morbidity after liver resection graded according to Dindo. Methods N = 679 cases of primary liver resection for CLM were retrospectively analyzed using univariable and multivariable ordinal regression analyses. Receiver operating characteristics curve (ROC) analysis was utilised to assess the sensitivity and specificity of predictions and their potential usefulness as prognostic models. Internal validation of the score was performed using data derived from 129 patients. Results The final multivariable regression model revealed lower preoperative levels, a greater number of units of intraoperatively transfused packed red blood cells (pRBCs), longer duration of surgery, and larger metastases to independently influence postoperatively graded morbidity. ROC curve analysis demonstrated that the multivariable regression model is able to predict each individual grade of postoperative morbidity with high sensitivity and specificity. The areas under the receiver operating curves (AUROC) for all of these predictions of individual grades of morbidity were > 0.700, indicating potential usefulness as a predictive model. Moreover, a consistent concordance in Grades I, II, IV, and V according to the classification proposed by Dindo et al. was observed in the internal validation. Conclusion This study proposes a prognostic model for the prediction of each grade of postoperative morbidity after liver resection for CLM with high sensitivity and specificity using pre- and intraoperatively available variables.