Heart Rate Turbulence modeling using Boosted Regression Trees

Heart Rate Turbulence (HRT) is a relevant cardiac risk stratification criterion. It is accepted the baroreflex hypothesis as a source of the HRT. However, several studies showed different results about the relationship between coupling interval (CI) and HRT turbulence slope (TS) parameter. Our aim was to propose a nonparametric model based on Boosted Regression Trees (BRT) of TS as a function of CI, heart rate (quantified by sinus cardiac length SCL in ms), Age and Sex. We used a set of 11 patients with normal heart undergoing electrophysiological study (EPS) and 61 holters from actue myocardial infarction (AMI) patients (Hospital Arrixaca de Murcia). The AMI set was split into: AMI low-risk, and AMI high-risk according to HRT. We propose to model TS using BRT, which is an ensemble approach to build a regression model using several trees. SCL was the explicative variable with the highest importance both in EPS and AMI low-risk. TS correlated nonlinearly with SCL, and negatively with CI both in EPS and AMI low-risk. The model was completely different for AMI-HR. R2 was higher for EPS (0.63) and AMI-LR (0.38) than for AMI-HR (0.28). The model was in agreement with the baroreflex hypothesis, and the role of age and sex agrees with previous results for EPS and AMI-LR. CI was the most important variable and positively correlated with TS in AMI-HR.