VITA-D: radiomic web tool to predict vitamin D levels and their relationship to metabolic risk factors

Background: Vitamin D deficiency is a significant risk factor for several chronic conditions. The study aims to predict metabolic risk factors associated with low vitamin D levels in Universidad Técnica Particular de Loja (UTPL) patients in the south America part of Loja-Ecuador using a graphical web interface tool based on ML algorithms. Methods: Two databases (Private and public) were processed using ML training models (SVM, Random Forest and Linear Logistic Regression). (i) Private data collection was obtained from 465 UTPL patients, where vitamin D levels was measured through a blood sample collection to calculate the concentration of 25-hydroxy vitamin D in plasma and determine by enzyme-linked immunosorbent assay (ELISA), and (ii)Public data collection was obtained from FigShare database. Then, a survey was conducted from April 2022 to June 2023 to collect sociodemographic data, identifying 157 variables but only 18 of these were used for ML training models. After statistical analyses, Random Forest was selected as the best performance model in binary levels classification: Class 0 (Deficiency): Vitamin D levels below 15 ng/ml and Class 1 (Sufficiency): Vitamin D levels above 15 ng/ml, during web interface design. Results: Vitamin D deficiency is % in the private database and 18 % in the public database. The Random Forest algorithm achieved higher accuracy (87.73%) in classifying both classes, concerning SVM (80.0%) and LR (70.70%). Thus, the online web application was designed based on Random Forest model as a class predictor. Conclusion: Database's information provides a basis for training ML algorithms and developing a web tool based on the best one ML model for predicting Vitamin D deficiency levels, especially in UTPL staff.