Developing Machine Learning Model for Mortality and Postsurgical Complications Using Preoperative Data

Preoperative assessment is the evaluation carried out before delivering anesthesia and performing surgery. In the preoperative assessment, physicians develop a plan to mitigate any surgical complications by developing a risk management strategy. The objective of this study is to develop a machine learning model to predict the risk of 30-day mortality and postsurgical complications for surgical patients based on preoperative data. Data from the MIMIC-III database was used, which includes over 52,000 ICU admissions from more than 40,000 U.S. patients. The present study applied and compared five supervised machine algorithms with a focus on two outcomes: 30-day mortality and postsurgical complications. A web-based application was developed based on the result of the algorithms. The database was split into 70% training and 30% validation using a 5-fold cross-validation method using the MIMIC preoperative data. We used Accuracy, AUC, Sensitivity, and Specificity for the analysis of the results. The highest performance model in the testing is Logistic regression with the Accuracy of 0.714, AUC of 0.706, a Sensitivity of 0.595, and a Specificity of 0.71. The models with the highest performance in the testing phase were the Decision Tree and Gradient Boosting with equal Accuracy of 0.706, AUC of 0.702, Sensitivity of 0.585, and Specificity of 0.714. The findings of this study suggest that clinical data for EHR can be used to develop prediction models for the risk of 30-day mortality and complications for adult patients undergoing all types of surgery.