Abstract
Background/Objectives: Vitamin D deficiency (VDD) is a global health concern associated with metabolic disease and immune dysfunction. Despite known risk factors like limited sun exposure, diet, and lifestyle, few studies have explored these factors comprehensively on a large scale. This cross-sectional study aimed to identify VDD-associated factors in South Korea via an integrative approach of machine learning and statistical analyses using Korea National Health and Nutrition Examination Survey (KNHANES) IX-1 data. Methods: Using the KNHANES dataset, six machine learning algorithms were applied to evaluate VDD (serum 25[OH]D3 < 20 ng/mL)-associated factors through feature importance scores. Thereafter, multivariate linear and logistic regression models were applied to the dataset-stratified by sex and age. Results: Among 583 variables, 17 VDD-associated factors were identified using the CatBoost model, which achieved the highest F1 score. When these factors were assessed through statistical analysis, dietary supplement use emerged as a consistent factor associated with VDD across all subgroups (younger men, younger women, older men, and older women). In younger adults, HDL cholesterol, blood and urinary creatinine, water intake, urban residence, and breakfast frequency were significantly associated with VDD. Additionally, blood urea nitrogen and fasting plasma glucose in men and urinary sodium in women showed sex-specific associations with serum 25(OH)D levels. Conclusions: This study identified key VDD-associated factors in the South Korean population, which varied by age or sex. These findings highlight the multifaceted nature of VDD, influenced by dietary, lifestyle, and biochemical factors and underscore the need for strategies integrating machine learning and statistical analysis.