Systematic screening by a heart team and a machine learning approach contribute to unraveling novel risk factors in revascularization candidates with complex coronary artery disease.

ABSTRACT

INTRODUCTION: The baseline characteristics affecting mortality following percutaneous or surgical revascularization in patients with left main and / or 3­vessel coronary artery disease (CAD) observed in real­world practice differ from those established in randomized controlled trials (RCTs) due to the constraints of inclusion / exclusion criteria. OBJECTIVES: This study aimed to assess whether systematic screening enables identification of novel and registry­specific baseline patient characteristics influencing long­term mortality. PATIENT AND METHODS: Least absolute shrinkage and selection operator (LASSO) regression was used to screen 42 baseline patient characteristics shared by the SYNTAX (Synergy between Percutaneous Coronary Intervention with Taxus and Cardiac Surgery) trial and a single­center Polish registry of 1035 consecutive patients with complex CAD who received revascularization and were followed-up for 5 years. After screening, a classic Cox regression analysis was performed to examine the suitability of a linear model for predicting 5­year mortality, which was then compared with the mortality predicted in the same cohort using the SYNTAX score II 2020 (SS2020). RESULTS: The 5­year mortality rate in the registry was 12.3%, and the strongest predictors were pulmonary hypertension, chronic obstructive pulmonary disease, and insulin­dependent diabetes. In an internal validation, the linear model constructed after LASSO screening and combined with a classic Cox regression analysis improved the prediction of 5­year mortality, as compared with the SS2020 (concordance index of 0.92 and 0.75, respectively). CONCLUSIONS: A machine learning approach improved the detection of registry­specific risk factors in all­comer patients amenable to surgical or percutaneous revascularization who were evaluated by a heart team. The risk factors identified in RCTs are not necessarily the same as those detected in real clinical practice when systematic screening is applied.