Baroreflex sensitivity with different lags and random forests for staging cardiovascular autonomic neuropathy in subjects with diabetes

• Published in: Computers in biology and medicine Vol. 127; p. 104098
• Main Authors: Petry, Daiana, Mirian de Godoy Marques, Claudia, Brum Marques, Jefferson Luiz
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.12.2020; Elsevier Limited
• Subjects: Autonomic nervous system; Baroreceptors; Baroreflex sensitivity; Biomarkers; Blood pressure; Cardiovascular autonomic neuropathy; Classification; Diabetes; Diabetes mellitus; Diabetic neuropathy; Electrocardiography; Heart rate; Heart rate variability; Indexing; Insulin; Internal Medicine; Machine learning; Negative feedback; Nervous system; Other; Patients; Principal components analysis; Quality of life; Random forest; Reflexes; Respiration; Sensitivity; Software; Subclinical CAN; Time lag; Variables; Variance analysis; Cardiovascular autonomic neuropathy; Diabetes mellitus; Machine learning; Baroreflex sensitivity; Subclinical CAN; Random forest; Heart rate variability; baroreflex sensitivity; subclinical CAN; heart rate variability; diabetes mellitus; random forest; machine learning; cardiovascular autonomic neuropathy
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2020.104098

Impaired baroreflex sensitivity (BRS) may indicate cardiovascular autonomic neuropathy (CAN), which often remains undiagnosed during the initial course of diabetes mellitus. The baroreflex mechanism can be considered negative feedback because of baroreflex delay, the time delay between a change in blood pressure and the counteracting heart rate response. This work sought to analyze BRS considering lags from 1 to 10 RR intervals. We hypothesized that diabetic patients with subclinical CAN (SCAN) have a detectable delay in autonomic nervous system activity and that this would differ from patients without CAN (NCAN) and with established CAN (ECAN). In the first stage, 30 patients were included in an exploratory analysis using the Principal Component Analysis. Six indexes related to the BRS delay were proposed and considered significant for staging diabetic patients. Three indexes allowed for the differentiating of patients with and without CAN, and three indexes distinguished subjects with SCAN from subjects with NCAN or ECAN. Then, in the second stage, a random forest model was developed with 72 subjects, using the variables selected in the first stage. It was possible to detect SCAN, and to point out those subjects with the potential to change the CAN stage, allowing for the tracking of CAN progression. The model achieved a sensitivity of 96% and specificity of 100% to detect SCAN. Thus, the BRS analysis considering delayed reaction in the dynamics of heart rate variability may contribute to an accurate screening tool to staging CAN, in addition to indicating patients with most insidious disease progress. •HR response delay to changes in BP can be used as biomarker of SCAN in diabetics.•BRS indexes at differents lags can detect SCAN and separate from others stages.•Random forest provides screening tool for CAN without active patient collaboration.•Screening with new BRS metrics allow monitoring of autonomic dysfunction progression.•BRS's analysis approach turns simple assessing autonomic function in clinical setting.