Establishment of model for predicting mortality risk after abdominal surgery using preoperative indices based on different machine learning algorithms

Abstract

Objective To establish the model for predicting the mortality risk after abdominal surgery using preoperative indices based on different machine learning algorithms. Methods Fifty patients died after abdominal surgery with general anesthesia from June 2015 to December 2018 in our hospital were enrolled in the study.Based on the types of surgery and age of dead patients, 150 patients who were discharged from hospital upon recovery postoperatively were randomly selected from our database as control cases with a ratio of 1∶3.The total dataset of 200 patients was randomly divided into training dataset (n=140) and testing dataset (n=60). Preoperative indices (each index of baseline characteristics, each index of anesthesia interview information and indices of preoperative examination) were used to develop the model for predicting the mortality risk after abdominal surgery based on four machine learning algorithms AdaBoost, GBDT, LR, and SVM, and the model was evaluated in the testing dataset. Results The area under the receiver operating characteristic curves of models developed using preoperative index based on AdaBoost, GBDT, LR, and SVM for predicting the postoperative mortality risk were 0.796, 0.794, 0.846 and 0.781, respectively.There were no significant differences in area under the receiver operating characteristic curves among different models (P>0.05). Conclusion The model for predicting mortality risk after abdominal surgery using preoperative indicators based on different machine learning algorithms is successfully established. Key words: Artificial intelligence; Machine learning; Forecasting; Death; Postoperative complications