JMIR Publications

ABSTRACT

Background:

Postoperative delirium (POD) following cardiac surgery is a severe complication, and early identification of the risk of delirium remains a challenge in clinical practice. Recent investigations have explored the effectiveness of machine learning (ML)-based methods for identifying the risk of POD in cardiac surgery patients. However, more evidence is required to validate the feasibility of these methods.

Objective:

Consequently, this study aims to ascertain the performance of ML in identifying the risk of POD following cardiac surgery, providing evidence for the development or update of future ML-based assessment tools.

Methods:

A comprehensive literature search covered four electronic databases: PubMed, the Cochrane Library, Embase, and Web of Science through August 30, 2024. The risk of bias of the models in the included studies was assessed leveraging the prediction model bias risk assessment tool. Subgroup analyses were implemented to examine the datasets and model types.

Results:

The analysis incorporated 32 original studies, comprising 86,185 cardiac surgery patients, 6,131 of whom developed POD. The meta-analysis unraveled that for the training set, the c-index, sensitivity (Sen), and specificity (Spe) for delirium prediction using ML were 0.818 (95% CI: 0.796–0.840), 0.75 (95% CI: 0.71–0.79), and 0.82 (95% CI: 0.77–0.85), respectively. In the validation set, the c-index, Sen, and Spe for delirium prediction reached 0.777 (95% CI: 0.742–0.811), 0.71 (95% CI: 0.65–0.76), and 0.76 (95% CI: 0.72–0.80), respectively. Logistic regression was the primary modeling method. In the training set, the c-index, Sen, and Spe for delirium prediction reached 0.799 (95% CI: 0.761–0.837), 0.75 (95% CI: 0.68–0.80), and 0.77 (95% CI: 0.73–0.81), respectively. In the validation set, the c-index, Sen, and Spe reached 0.760 (95% CI: 0.720–0.800), 0.69 (95% CI: 0.62–0.76), and 0.73 (95% CI: 0.67–0.79), respectively.

Conclusions:

ML-based methods are viable for predicting the risk of POD following cardiac surgery. Future studies should focus on readily available, minimally invasive, and efficient predictive factors to develop high-performance ML methods for constructing simple, widely applicable clinical assessment tools. This approach enables tailored prevention strategies for patients identified as high-risk for POD.