Prediction of obstructive coronary artery disease in people living with HIV: value of machine learning incorporating HIV-specific factors

Hui Liu¹Haibo Ding^2,3,4,5Yue Zheng¹Yue Li¹Yang Yu⁶Zhaodi Geng¹Jie Zhou¹Huaibi Huo¹Han Li⁷Xin Peng^1,8Zhaoxin Tian¹Xiaolin Li^3,4,5,9Hong Shang^2,3,4,5Ting Liu^1*

• ¹The First Hospital of China Medical University, Shenyang, China
• ²National Clinical Research Center for Laboratory Medicine, The First Hospital of China Medical University, shenyang, China
• ³Chinese Academy of Medical Sciences, shenyang, China
• ⁴Key Laboratory of AIDS Immunology of Liaoning Province, shenyang, China
• ⁵Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, hangzhou, China
• ⁶Siemens Healthineers China, Shanghai, China
• ⁷Jing'an District Centre Hospital of Shanghai, Fudan University, shanghai, China
• ⁸The Third People's Hospital of Chengdu, chengdu, China
• ⁹The First Hospital of China Medical University, China Medical University, shenyang, China

Objectives: To explore the value of machine learning (ML) model in conjunction with HIV-specific risk factors to predict obstructive coronary artery disease (CAD) (≥50% stenosis) on coronary CT angiography (CTA) in the asymptomatic people living with HIV (PLWH). Methods: In this cross-sectional study, we prospectively analyzed 304 PLWH without chest pain (age 48±11 years, 91% males). The dataset was randomly divided into training and held-out test sets in an 8:2 ratio. The ML model established by random forest was compared with traditional models, including CAD consortium clinical score, CONFIRM score, and Genders clinical model, as well as logistic regression model. The coronary artery calcium score (CACS) was added to the above five models to establish new models. Predictive performance of the models was evaluated according to Delong test.Results: Obstructive CAD occurred in 64 of 304 PLWH (21%). The ML model (AUC of 0.946) had the highest discrimination for obstructive CAD compared with above models (AUC of 0.734, 0.736, 0.737, and 0.782, respectively; P < 0.05 for all comparisons). ML model showed the best calibration and clinical decision-making capability. Moreover, the ML model showed the best predictive performance compared with models after adding the CACS (AUC of 0.772, 0.740, 0.742, 0.750, and 0.798, respectively; P < 0.05 for all comparisons). 5 Conclusions: The ML model incorporating cardiovascular risk factors and HIV-specific factors can more accurately estimate the pretest likelihood of obstructive CAD in PLWH than traditional models. ML improves risk stratification in HIV populations and may help guide management.