Applying data mining techniques to predict vitamin D deficiency in diabetic patients

Area of Science:

• Endocrinology
• Medical Informatics
• Nutritional Science

Background:

• Vitamin D is crucial for overall health, impacting calcium absorption, immune function, and metabolic processes.
• Diabetes mellitus is a significant health concern, and vitamin D status may influence its pathogenesis and severity.
• The 25-hydroxyvitamin D (25OHD) test for vitamin D levels is costly and not universally covered, necessitating alternative assessment methods.

Purpose of the Study:

• To develop and evaluate a predictive model for identifying vitamin D deficiency in diabetic patients.
• To leverage data mining techniques and feature selection on electronic health records for deficiency prediction.
• To compare the effectiveness of different feature selection algorithms, including relief-F.

Main Methods:

• Utilized historical electronic health records from diabetic patients.
• Applied data mining techniques for classification modeling.
• Employed the relief-F algorithm for feature selection to identify significant predictors of vitamin D deficiency.
• Evaluated model performance using metrics such as accuracy, sensitivity, specificity, F1-score, precision, kappa, and ROC curves.

Main Results:

• The relief-F feature selection method effectively removed non-essential features without compromising classification performance.
• The support vector machines (SVM) model, utilizing a radial kernel and 18 selected features, achieved the highest classification accuracy of 97.044%.
• The study successfully identified key features from electronic health records for predicting vitamin D deficiency in the diabetic population.

Conclusions:

• Data mining techniques, combined with effective feature selection, can accurately predict vitamin D deficiency in diabetic patients.
• The developed SVM model demonstrates high efficacy for identifying individuals at risk of vitamin D deficiency, potentially reducing reliance on expensive direct testing.
• This approach offers a cost-effective strategy for screening and managing vitamin D status in diabetic populations.