Deep learning prediction of peak oxygen uptake in patients with coronary heart disease: a retrospective study.

Abstract

Objective: To develop and validate prediction models for peak oxygen uptake (VO₂peak) in patients with coronary heart disease (CHD) using submaximal cardiopulmonary exercise testing (CPET) indicators and deep learning methods.

Design: Retrospective model development and validation study.

Setting: Cardiac Rehabilitation Centre, Peking University Third Hospital, China.

Participants: A total of 10 538 patients with CHD who underwent CPET between January 2014 and December 2019.

Methods: Clinical data and CPET indicators were collected. Multiple machine learning and deep learning models were developed and compared. Model performance was assessed using R², mean absolute error (MAE), bias, Bland-Altman analysis and SHapley Additive exPlanations (SHAP) feature importance ranking.

Results: The neural network model achieved the best performance (R² = 0.82, MAE=1.55 mL/kg/min, bias=0.08). XGBoost was the best-performing traditional machine learning model (R² = 0.74). SHAP analysis identified eight top-ranked features, including VO₂@AT, OUES, weight, VE/VCO₂ slope, VE/VCO₂@AT, age, gender and HR@AT.

Conclusion: The CPET deep learning model shows potential for predicting VO₂peak in CHD patients, but further external validation and prospective studies are required before clinical application.