Prediction of microvascular complications in diabetic patients without obstructive coronary stenosis based on peri-coronary adipose tissue attenuation model.

Abstract

To investigate the predictive value of peri-coronary adipose tissue (PCAT) attenuation for microvascular complications in diabetic patients without significant stenosis and to develop a prediction model for early risk stratification. This study retrospectively included patients clinically identified for coronary computed tomography angiography (CCTA) and type 2 diabetes between January 2017 and December 2020. All patients were followed up for at least 1 year. The clinical data and CCTA-based imaging characteristics (including PCAT of major epicardial vessels, high-risk plaque features) were recorded. In the training cohort comprising of 579 patients, two models were developed: model 1 with the inclusion of clinical factors and model 2 incorporating clinical factors + RCAPCAT using multivariable logistic regression analysis. An internal validation cohort comprising 249 patients and an independent external validation cohort of 269 patients were used to validate the proposed models. Microvascular complications occurred in 69.1% (758/1097) of the current cohort during follow-up. In the training cohort, model 2 exhibited improved predictive power over model 1 based on clinical factors (AUC = 0.820 versus 0.781, p = 0.003) with lower prediction error (Brier score = 0.146 versus 0.164) compared to model 1. Model 2 accurately categorized 78.58% of patients with diabetic microvascular complications. Similar performance of model 2 in the internal validation cohort and the external validation cohort was further confirmed. The model incorporating clinical factors and RCAPCAT predicts the development of microvascular complications in diabetic patients without significant coronary stenosis. • Hypertension, HbA1c, duration of diabetes, and RCAPCAT were independent risk factors for microvascular complications. • The prediction model integrating RCAPCAT exhibited improved predictive power over the model only based on clinical factors (AUC = 0.820 versus 0.781, p = 0.003) and showed lower prediction error (Brier score=0.146 versus 0.164).