Predicting Diabetic Neuropathy Risk Level Using Artificial Neural Network and Clinical Parameters of Subjects With...

Area of Science:

• Neurology
• Diabetology
• Medical Informatics

Background:

• Diabetic neuropathy risk assessment is crucial for patient management.
• A novel tool utilizes clinical characteristics and vibration perception threshold (VPT) for automated neuropathy estimation.
• Key predictors include age, weight, HbA1c, cholesterol, and diabetes duration.

Purpose of the Study:

• To develop and validate a risk assessment tool for diabetic neuropathy.
• To evaluate the predictive accuracy of clinical variables and VPT.
• To explore advanced modeling techniques for improved diagnostic accuracy.

Main Methods:

• Utilized patient data including age, height, weight, urine albumin-to-creatinine ratio, glycated hemoglobin, total cholesterol, and diabetes duration.
• Vibration perception threshold (VPT) measured using a neurothesiometer, categorized into low, medium, and high risk.
• Developed a Neural Network model based on a Proportional Odds Model for prediction.

Main Results:

• Initial analysis using patient data (n=5088) achieved 54% accuracy.
• The Neural Network model demonstrated superior prediction accuracy (>70%) with simulated patient data (n=4158).
• Clinical parameters alone can provide a reasonable diagnosis of diabetic neuropathy.

Conclusions:

• Diabetic neuropathy can be diagnosed accurately using only clinical parameters, without specialized devices.
• The developed Neural Network model significantly enhances prediction accuracy.
• This approach offers a cost-effective and accessible method for neuropathy assessment.