444-P: Machine Learning Driven Prediction of Hyperglycemia, Acute Kidney Injury, and Atrial Fibrillation after Cardiac Surgery Using High-Resolution Metabolomics

Stress hyperglycemia (SH), acute kidney injury (AKI), and atrial fibrillation (AF) are common complications after cardiac surgery. We examined the predictive value of metabolite changes on inpatient outcomes in participants undergoing coronary artery bypass graft surgery. Methods: We analyzed 8,631 and 10,919 metabolites from C18 chromatography and hydrophilic interaction liquid chromatography (HILIC) to predict SH, AKI, and AF. The study included 61 participants (mean age: 62.5; 16.1% female) with one missing SH. We imputed zero values in the metabolites with half the minimum value. For SH and AF, we utilized the log-ratios of metabolite measured during and before surgery as features, while for AKI, we employed those measured exclusively during surgery. We used the Mann-Whitney U-test to select significant features and the Partial Least Squares (PLS) regression for dimension reduction. PLS components, along with covariates such as age, sex, and BMI, were incorporated into machine learning (ML) classifiers, and the Random Forest classifier emerged as the most effective based on leave-one-out cross-validation (LOOCV) performance. Finally, we aggregated significant feature contributions across all LOOCV iterations to identify the most influential features.

Results: To predict SH, our ML pipeline achieved an AUC of 0.82, accuracy of 0.77, sensitivity of 0.84, and specificity of 0.71 identifying five significant features across all LOOCV iterations. For AF prediction, we achieved an AUC of 0.85, accuracy of 0.82, sensitivity of 0.91, and specificity of 0.77, with 97 significant features identified. For AKI, we attained an AUC of 0.74, accuracy of 0.72, sensitivity of 0.79, and specificity of 0.70 identifying 20 significant features.

Conclusion: Machine learning algorithms may aid in the identification of multiple potentially relevant biomarkers linked to adverse cardiometabolic outcomes after cardiac surgery.