Can machine learning improve risk prediction and individualise decision making between percutaneous and surgical revascularisation?

Background/Introduction In patients with three-vessel coronary artery disease (CAD) and/or left main (LM) CAD, individual risk prediction plays a key role in deciding between percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG). Purpose The aim of this study was to assess whether these individualised revascularization decisions can be improved by applying machine learning (ML) algorithms and integrating clinical, biological, and anatomical factors. Methods The present study consisted of patients enrolled in the SYNTAX study, which is a randomized trial comparing PCI and CABG in patients with complex coronary artery disease. ML algorithms were used to develop a prognostic index for 5-year death, which was combined, in the second stage, with assigned treatment (PCI or CABG), and pre-specified effect-modifiers: disease type (3-vessel or LMCAD) and anatomical SYNTAX score. The model’s discriminative ability to predict the risk of 5-year death and treatment benefit between PCI and CABG was cross-validated in the SYNTAX trial (n=1800 patients) and externally validated in the CREDO-KYOTO registry (n=7362), and then compared to the original SYNTAX score II 2020 (SS2020). Results The gradient boosting ML model identified the top 10 prognostic factors at the time of decision making as: age, creatinine clearance, left ventricular ejection fraction, peripheral vascular disease, C-reactive protein, hemoglobin, HbA1c, glucose, systolic and diastolic pressure. The ML model performed best for predicting 5-year all-cause death with C-indexes of 0.78 (95% CI 0.75-0.81) in cross-validation and 0.77 (95% CI 0.76-0.79) in external validation. The ML model discriminated 5-year mortality better than the SS2020 in the external validation cohort (vs C-index 0.72 for SS2020, p <0.001) and identified heterogeneity in the treatment benefit of CABG versus PCI (Figure). Conclusions An ML-based approach for identifying individuals who benefit from CABG or PCI is feasible and effective. Implementation of this model in health care systems—trained to collect large numbers of parameters—may harmonize decision making.Figure