Development and Internal Validation of a Clinical Prediction Model for Coronary Slow Flow Phenomenon in Type 2...

Patients with type 2 diabetes mellitus (T2DM) may exhibit coronary slow flow phenomenon (CSFP) despite the absence of obstructive epicardial coronary stenosis. Dedicated prediction tools remain limited. This study aimed to develop and internally validate a model for estimating the probability of TIMI frame count (TFC)-defined CSFP in patients with T2DM.

Among 501 patients with T2DM undergoing coronary angiography at a single center in Xinjiang, China (2021-2025), 429 met eligibility criteria. CSFP was defined as maximum coronary stenosis <40% and TFC >27 frames in at least one major coronary artery. Patients were randomly divided into training (n = 300) and validation (n = 129) cohorts. Predictors selected by penalized regression were entered into multivariable logistic regression. Performance was assessed using area under the receiver operating characteristic curve (ROC-AUC), area under the precision-recall curve (AUC-PR), calibration, bootstrap-corrected concordance index, and decision curve analysis; additional analyses were complementary.

CSFP occurred in 44 patients (14.7%) in the training cohort and 16 (12.4%) in the validation cohort. The final model included smoking history, systolic blood pressure, hemoglobin A1c, lymphocyte count, and uric acid. ROC-AUC was 0.832 (95% confidence interval [CI]: 0.774-0.889) in the training cohort and 0.840 (95% CI: 0.733-0.946) in the validation cohort; AUC-PR was 0.467 and 0.542, respectively. Calibration was acceptable (Brier scores 0.101 and 0.084); bootstrap-corrected concordance indices were 0.827 and 0.846. Decision curve analysis suggested net benefit across clinically relevant thresholds. Given the limited number of CSFP events in the validation cohort, validation estimates should be interpreted cautiously.

We developed and internally validated a five-variable model for estimating the probability of TFC-defined CSFP in patients with T2DM. The model was translated into a nomogram and web-based calculator as a complementary tool for individualized risk characterization. External validation is required before broader clinical application.