Machine learning algorithms for population-specific risk score in coronary artery bypass grafting

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Health Outcomes Research

Background:

• Existing risk prediction scores for coronary artery bypass grafting (CABG) may not accurately reflect the outcomes in the Indian population.
• Development of a population-specific risk score is crucial for improving surgical decision-making and patient management.

Purpose of the Study:

• To develop and validate a novel risk prediction score (NH Score) tailored for Indian patients undergoing CABG.
• To compare the performance of the NH Score against established scores, namely the Society of Thoracic Surgeon (STS) Score and EuroSCORE II.

Main Methods:

• Utilized baseline data from 6703 adult patients who underwent CABG between 2015 and 2021, splitting it into training (n=5561) and validation (n=1142) sets.
• Employed the CatBoost algorithm for training the NH Score, assessing performance using Precision-Recall Curve, F1 Score, Brier Score, and calibration errors on the validation set.

Main Results:

• The NH Score demonstrated superior performance over both STS Score and EuroSCORE II across all evaluated outcomes, including mortality and morbidity.
• For mortality prediction, the NH Score achieved a significantly higher PR AUC (0.463) compared to STS (0.113) and EuroSCORE II (0.146) (p < 0.0001).
• The NH Score also showed improved prediction accuracy for morbidity and prolonged ventilation compared to the STS Score.

Conclusions:

• The NH Score represents a significant advancement in risk prediction for CABG patients within the Indian population, outperforming current international scores.
• Further validation of the NH Score on larger datasets is recommended to solidify its clinical utility.