Machine-learning algorithms for prediction of survival by stress echocardiography in chronic coronary syndromes

Abstract

Abstract Background Stress echocardiography (SE) is based on regional wall motion abnormalities and coronary flow velocity reserve (CFVR). They recognise different and complementary pathophysiological targets, and show independent and incremental value in predicting survival [1]. The approach based on machine-learning (ML) has the recognised potential to identify unsuspected patterns, and this can be especially relevant in the field of risk stratification by cardiac functional stress testing due to multiple parameters used in comprehensive stress testing and the variable weight of covariates [2]. Aim To assess SE outcome data analysis with ML approach. Methods We included 6,881 prospectively recruited and retrospectively analyzed patients (median age 67 years, IQR 59–74) with suspected (n=4,279) or known (n=2,602) coronary artery disease submitted to clinically-driven dipyridamole SE in 5 Italian institutions (Figure 1). The primary outcome measure was all-cause death. A Random Forest Survival model has been implemented to model the survival function according to the patient's characteristics. The Random Forest predicted response dependency on covariates has been investigated by reporting the variable dependence and the partial dependency plot (Figure 2). A web application was developed to predict the survival function according to the patients' characteristics. The external validation cohort was made of additional 1,002 patients recruited by a single, independent center in the same time period. Results During a median duration of follow-up of 3.4 years (IQR 1.6–7.5), 814 (12%) patients died. The mortality risk was higher for patients aged more than 60 years, resting ejection fraction <60%, resting WMSI, positive delta WMSI scores, and CFVR <3.0. The C-index performance (perfect prediction=1) was 0.79 in the internal validation cohort and 0.81 in the external, independent validation data set. Survival functions for individual patient were easily obtained with an open-access web-app. Conclusion An ML approach can be fruitfully applied to outcome data obtained with SE. Survival showed a constantly increasing relationship between survival and CFVR <3.0 and stress-rest wall motion score index >0. Since processing is largely automated, this approach can be easily scaled to larger and more comprehensive data sets to further refine stratification, guide therapy and be ultimately adopted as an open-source on-line decision tool. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): The study was partially funded by CNR-MIUR (National Research Council, Italian Ministry of University and Research) Ageing subproject.