Risk prediction of diabetic nephropathy using machine learning techniques: A pilot study with secondary data

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Nephrology Research

Background:

• Diabetic nephropathy (DN) is a significant complication of diabetes, necessitating improved diagnostic and predictive tools.
• Machine learning (ML) offers potential for analyzing complex health data to identify risk factors and predict disease.
• Principal Component Analysis (PCA) is a dimensionality reduction technique useful for feature selection in ML models.

Purpose of the Study:

• To determine risk factors for diabetic nephropathy (DN) using Principal Component Analysis (PCA) with varying cutoffs.
• To predict DN patients utilizing various machine learning (ML) techniques.
• To identify the optimal combination of PCA and ML for accurate DN prediction.

Main Methods:

• Implemented a combined PCA and ML approach to select optimal features at different PCA cutoffs.
• Evaluated six ML techniques: linear discriminant analysis, SVM, logistic regression, K-nearest neighborhood, naïve Bayes, and artificial neural network.
• Utilized leave-one-out cross-validation to compare ML model performance based on accuracy and Area Under the Curve (AUC).

Main Results:

• The study included 133 respondents, with 73 diagnosed with DN (54.2% female, average age 69.6±10.2 years).
• PCA combined with the SVM-RBF classifier achieved the highest accuracy (88.7%) and AUC (0.91) at a PCA cutoff of 0.96.
• This specific model outperformed other tested ML techniques in predicting DN.

Conclusions:

• The PCA-SVM-RBF model demonstrates high accuracy in classifying DN patients, potentially surpassing existing research models.
• This approach effectively identifies key risk factors and predicts DN, offering a valuable tool for clinical settings.
• Further prospective studies are recommended to validate the clinical applicability of this predictive model.