Rev Cardiovasc Med. 2026 Jun 15;27(6):46777. doi: 10.31083/RCM46777. eCollection 2026 Jun.
ABSTRACT
BACKGROUND: The dependence of the acquisition of the coronary artery calcification score (CACS) on computed tomography (CT) has drawbacks, including the ethical concerns of radiation exposure in the care of patients with non-cardiovascular diseases, where CACS has been shown to correlate with its prognosis. Significant heterogeneities exist between patients with and without coronary artery calcification (CAC). Mathematical formulae using medical history and common, non-invasive test results enable cheap, ready assessment of CAC and subsequent research into how it can be used for clinical decision making.
METHODS: 694 patient records of visits to Fuwai Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College from 2009 to 2023 were partitioned into a training (visited before 2023) and an independent validation set (visited in 2023). With age, gender, current smoking, diabetes, low-density lipoprotein cholesterol (LDL-C), reduced renal function, usage of statins and aspirin as candidate predictors, five logistic regression models were built under two paradigms. Bootstrap resampling was employed for internal validation, followed by external validation and calibration on the validation set. Models built under each paradigm were compared, followed by head-to-head comparison of the "best" models built under each paradigm with a comprehensive criteria involving both model performance and predictor parsimony.
RESULTS: 694 records were used for modeling, with 536 and 158 records in the training and validation set respectively. Model 1 (c statistic upon external validation: 0.77) outperformed other models built under Paradigm 1 while Models 4 (c statistic upon external validation: 0.79) and 5 (c statistic upon external validation: 0.79) built under Paradigm 2 outperformed Model 1. Model 5 was more parsimonious in predictors. All models were well calibrated.
CONCLUSION: With gender, current smoking, LDL-C, age, diabetes and reduced renal function as predictors, Model 5 outperformed other models and was hence recommended for further use. By assessing the presence of CAC with medical history and blood test results instead of CT, our model offers an approach to immediate, radiation-free assessment of CAC, which may further unleash the clinical utility of CAC in clinical practice that may have remained unraveled.
PMID:42416588 | PMC:PMC13339184 | DOI:10.31083/RCM46777