Machine learning as a new frontier in mitral valve surgical strategy.

Abstract

One of the surgical options available for ischemic mitral regurgitation (MR) is mitral valve repair but is limited by recurrent regurgitation as it is experienced by a significant percentage of patients and has a negative impact on patient outcomes. Efforts to model and identify predictors of recurrent MR rely on complicated echocardiographic and clinical measurements that are subjective and not routinely collected. Kachroo et al. approached this problem in a unique way by using the STS database and machine learning (ML) to develop models that predict recurrent MR or death at 1 year. The STS database contains many routinely collected demographic and clinical parameters but requires a methodology, such as ML, that will accommodate collinearity and the unknown significance of many predictors. Kachroo et al. developed three good ML models with the area under curve 0.72-0.75. Data-driven selection of important predictors showed that three revascularization targets, peripheral vascular disease, and use of β-blockers are most predictive of recurrent MR. We applaud the authors for pioneering a novel methodology and paving the way for a bright future in ML which includes integrating medical imaging, waveform, and genomic data to practice personalized medicine for our patients.