A machine learning approach to developing an accurate stratification of type 2 diabetes mellitus based on heart rate variability parameters using the k-means clustering technique in elderly women

BACKGROUND: Type 2 DM (T2DM) is increasingly frequent in the world in association with the increase of sedentary behaviors, unhealthy diet, obesity and metabolic syndrome and a public health concern. The aim of the present study was to stratify T2DM using the k-means clustering technique based on parameters derived from heart rate variability (HRV).METHODS: This study was designed as a cross-sectional study consisting of 74 elderly women aged 60 to 70 years and diagnosed with T2DM. The volunteers were instructed to remain in a supine position for 5 minutes at rest while breathing normally with a Polar V800 heart rate monitor, working at a sampling rate of 1000 Hz, to record RR intervals (RRi). HRV analysis in the time and frequency domain was performed to obtain the classical parameters pNN50, SDNN, RMSSD, MeanRRi, LF, HF and HF/LF.RESULTS: In machine learning through the technique called K-means, there was a random initialization with 10 repetitions, which made it possible to find three clusters, the silhouette index was used with two clusters (0.55) and with three clusters (0.58). The silhouette index showed significant values, indicating minimizing variability intra-group and maximizing inter-group for the MeanRRi. The MeanRRi parameter was stratified into three clusters, showing a cardiac clinical with cardiovascular autonomic predisposition in T2DM with a tendency in reduced HRV.CONCLUSIONS: The MeanRRi is a low cost and accurate biomarker of HRV for cardiovascular analysis, when associated with machine learning by K-means technique it becomes efficient to stratify T2DM adding even more value to the field of biotechnology.