Comparison of three machine learning methods in identification of major bleeding events in postoperative patients with malignant tumors

Abstract Background To develop a machine learning model tool for identifying postoperative patients with major bleeding based on electronic medical record system. Methods This study used the available information in the National Health and Medical Big Data (Eastern) Center in Jiangsu Province of China. We randomly selected the medical records of 2,000 patients who underwent in-hospital tumor resection surgery between January 2018 and December 2021 from the database. Physicians classified each note as present or absent for a major bleeding event during the postoperative hospital stay. Feature engineering was created by bleeding expressions, high frequency related expressions and quantitative logical judgment. Logistic regression (LR), K-nearest neighbor (KNN), and convolutional neural network(CNN) were developed and trained using the 1600-note training set. The main outcomes were accuracy, sensitivity, specificity, positive predictive value, and negative predictive value for each model. Results Major bleeding was present in 4.31% of training set and 4.75% of test set. For the training set, LR method has the sensitivity of 1.0000 and specificity of 0.8152 while CNN method has the sensitivity of 0.9710 and specificity of 0.9027. LR and CNN methods both perform well in the sensitivity and specificity in the test set. Although the KNN method has high specificity in the training set and test set, its sensitivity is very low in both sets. Conclusions Both LR method and CNN method perform well in identifying major bleeding occurring in postoperative patients with malignant tumors, with high sensitivity and specificity.