Exploring the role of artificial intelligence in predicting anesthesia-induced hypotension: A comprehensive review

Can artificial intelligence reliably forecast intraoperative hypotension before it happens? This research examined the predictive performance of machine learning algorithms for anticipating anesthesia-induced hypotension across multiple clinical settings in Germany. We conducted a two-center prospective observational investigation at Ludwig Maximilian University of Munich and Heidelberg University Hospital between March 2021 and November 2022. A total of 847 adult patients undergoing elective noncardiac surgery were enrolled and monitored continuously. Hemodynamic parameters including mean arterial pressure, heart rate, stroke volume variation, and cardiac index were recorded at five-second intervals using the Edwards Lifesciences EV1000 monitoring platform connected to radial arterial catheters. Six machine learning models were trained and validated using stratified cross-validation: logistic regression, random forest, gradient boosting, feed-forward neural network, long short-term memory recurrent network, and an ensemble approach combining all five base classifiers into a unified prediction framework. Hypotension was defined as mean arterial pressure below 65 mmHg persisting for at least one minute during the intraoperative period. The ensemble model achieved the highest area under the receiver operating characteristic curve at 0.91 when predicting events five minutes in advance. Sensitivity reached 84.1% and specificity 81.3% at this prediction horizon. Neural network and LSTM models performed comparably, with AUROC values of 0.86 and 0.88 respectively. Simpler approaches like logistic regression lagged behind at 0.72 AUROC, confirming the need for nonlinear modeling of hemodynamic data. The AI-guided clinical decision support group experienced significantly fewer hypotensive episodes (12.3% versus 23.8%, p