Simple patient and lesion characteristics on coronary angiography can accurately predict FFR-proven ischemia

Abstract Aims Fractional flow reserve (FFR) measured during invasive coronary angiography (ICA), the reference standard for diagnosing coronary ischemia, potentially risks pressure wire-induced arterial wall injury. We aimed to develop a model for diagnosing coronary ischemia based on simple patient and lesion characteristics without invasive intracoronary pressure measurement. Methods We performed quantitative coronary angiography (QCA) analyses on subjects who had undergone FFR measurements during clinically indicated ICA. Clinical, ICA and QCA variables were input into a prediction model for coronary ischemia, defined as FFR ≤0.8. Least absolute shrinkage and selection operator (LASSO) regression with internal ten-fold cross-validation was implemented using R software to minimize multicollinearity and select final model variables, i.e., those corresponding to the nadir of the binomial deviation plot. Per-vessel and per-subject results were reported. Results Among 103 subjects (mean age 61 ±10 years; 74 male, 29 female) recruited from hospitals in Singapore (participants in a prospective study NCT03054324) and China, we analyzed 148 stenotic vessels visually assessed diameter (DS) ≥30%. From a dataset comprising 37 variables including baseline clinical characteristics, coronary artery morphological parameters and plaque characteristics, LASSO selected six final model variables: lesion length, minimal lumen diameter, stenosis flow reserve, DS, systolic blood pressure and diabetes, negating serious multicollinearity. The first three variables were measured on QCA, the third using ICA images acquired at rest. Our model outperformed DS ≥70% for discriminating coronary ischemia at both per-vessel (area-under-curve [AUC] 0.907 vs 0.770, p<0.001) and per-subject levels (AUC 0.882 vs 0.759, p<0.001). Per-vessel model accuracy, sensitivity, specificity, positive and negative predictive values were 85.1%, 82.4%, 93.2%, 87.1% and 83.3%, respectively; and per-subject, 83.4%, 82.0%, 85.7%, 89.3% and 76.6%, respectively. Conclusion Compared with visual diameter assessment, our LASSO-based model attained superior prediction of FFR-defined coronary ischemia using an array of variables without the need for invasive pressure measurement.