Machine learning prediction of future peripheral neuropathy in type 2 diabetics with percussion entropy and body mass indices

Abstract This study was designed to evaluate the clinical applications of body mass index (BMI) and a percussion-entropy-based index (PEINEW) for predicting the development of diabetic peripheral neuropathy (DPN) in a group of type 2 diabetes mellitus (DM) patients. The study population comprised a sample of 90 subjects with diabetics (aged 37–86 years), who went through a blood test and photoplethysmography (PPG) measurement and were then followed for 5.5 years. Conventional parameters, including the small-scale multiscale entropy index (MEISS), pulse wave velocity with electrocardiogram located (PWVmean), and PEIoriginal, were computed and compared. A logistic regression model with PEINEW and a single categorical variable (BMI) showed a graded association between the diabetics, with a high BMI (i.e., “high” category) associated with a 12.53-fold greater risk of developing DPN relative to the diabetics with a low BMI (i.e., “low” category) (p = 0.001). The odds ratio for PEINEW was 0.893. The Kaplan-Meier survival analysis showed that the diabetic patients with BMI ＞ 30 had a significantly higher cumulative incidence of PN on follow-up than those with BMI ≦ 30 (log-rank test, p