Machine Learning Approach Using Routine Immediate Postoperative Laboratory Values for Predicting Postoperative...

Area of Science:

• Medical Informatics
• Surgical Outcomes Research
• Machine Learning in Healthcare

Background:

• Preoperative risk stratification models for mortality are established.
• Postoperative risk factors significantly impact patient survival but are less studied.
• Immediate postoperative laboratory values offer a potential data source for risk prediction.

Purpose of the Study:

• To develop and validate machine learning models for predicting postoperative mortality.
• To utilize routine immediate postoperative laboratory values for risk stratification.
• To compare the performance of developed models against the SASA scoring system.

Main Methods:

• Development of prediction models using LASSO logistic regression, random forest, deep neural network, and XGBoost algorithms.
• Training and validation on two tertiary hospital databases.
• Comparison with the SASA scoring system for efficacy assessment.

Main Results:

• All developed models outperformed the SASA model in predicting postoperative mortality.
• The random forest model achieved the highest Area Under the Receiver Operating Characteristic curve (AUROC) of 0.82 and Area Under the Precision-Recall Curve (AUPRC) of 0.13.
• Postoperative phosphorus levels were identified as a key predictor in the random forest model.

Conclusions:

• Machine learning models utilizing immediate postoperative laboratory values demonstrate superior performance in predicting 30-day postoperative mortality.
• These models can effectively identify patients at higher risk of mortality after surgery.
• Routine postoperative lab values are valuable predictors of surgical outcomes.