Prediction of Diabetic Sensorimotor Polyneuropathy Using Machine Learning Techniques.

Area of Science:

• Medical Informatics
• Neurology
• Endocrinology

Background:

• Diabetic sensorimotor polyneuropathy (DSPN) is a significant complication of diabetes mellitus (DM).
• Early detection of DSPN is crucial for preventing neuropathic pain and foot ulcers.
• Current diagnostic methods may benefit from advanced predictive modeling.

Purpose of the Study:

• To compare the efficacy of machine learning (ML) techniques against traditional statistical methods for predicting DSPN in DM patients.
• To identify key predictors for DSPN using ML models.
• To evaluate the role of electrophysiological findings in DSPN prediction.

Main Methods:

• Utilized three ML methods: XGBoost (XGB), Support Vector Machine (SVM), and Random Forest (RF).
• Analyzed data from 470 DM patients classified into four DSPN severity groups.
• Compared ML model performance against linear regression analysis using Area Under the Curve (AUC).

Main Results:

• Random Forest (RF) achieved the highest AUC (0.8250) in differentiating clinical and electrophysiological DSPN criteria.
• RF performance was significantly superior to linear regression analysis (AUC = 0.6620).
• Serum glucose, IFCC, HbA1c, and albumin levels were identified as the top four DSPN predictors.

Conclusions:

• ML techniques, especially RF, offer effective prediction of DSPN in DM patients.
• Electrophysiological analysis is vital for accurate DSPN diagnosis.
• ML models provide a valuable tool for early DSPN detection and management.