A systematic screening and heart team approach contributes to unravel novel risk factors in revascularisation candidates of complex coronary artery disease: a machine learning approach.

Abstract

The baseline characteristics affecting mortality following percutaneous or surgical revascularisation in patients with left main (LM) and/or three-vessel (3V) coronary artery disease (CAD) differ between real-world practice and those established in randomized control trials (RCT) due to the constraints of inclusion/exclusion criteria. This study aimed to assess whether systematic screening identifies novel and registry-specific baseline characteristics influencing long-term mortality. LASSO (Least Absolute Shrinkage and Selection Operator) regression was used to screen 42 baseline characteristics shared by the SYNTAX trial and a single-center Polish registry of 1035 consecutive patients with complex CAD, receiving revascularization and followed up for 5 years. After screening, classical Cox regression analysis was performed to examine the suitability of a Linear model for predicting 5-year mortality, which was then compared to the mortality predicted in the same cohort using the SYNTAX score 2020 (SS2020). Five-year mortality in the registry was 12.3%, with the strongest predictors of pulmonary hypertension, chronic obstructive pulmonary disease and insulin-dependent diabetes. In an internal validation, the linear model constructed after LASSO screening and combined with a classical Cox regression analysis improved the prediction of 5-year mortality compared to the SS2020 (c-index 0.92 and 0.75, respectively). Machine learning improved the detection of registry-specific risk factors in all comers patients amenable to surgical or percutaneous revascularization who were discussed in a heart team. The risk factors identified from RCT are not necessarily the same as those detected in real clinical practice when systematic screening is applied.