Drug shortage in South Korea: machine learning-based prediction models and analysis of duration and causal factors

Background: Drug shortages remain a critical challenge for healthcare systems in South Korea. This study aimed to develop predictive models to forecast drug shortage duration and identify their underlying causes. Methods: Using 1,054 regulatory-reported drug shortage cases from 2018 to 2024 obtained from the Korean Ministry of Food and Drug Safety (KMFDS), we developed two machine learning models: (1) Model 1 to estimate shortage duration ranges, and (2) Model 2 to classify shortage causes into seven categories. Eighteen features related to drug shortages were included based on relevance and data availability. Key predictors were identified using Random Forest feature importance. Results: Model 1 achieved an accuracy of 62%, with Shortage Incidence Frequency being the most influential variable (importance = 0.152). In Model 2, weighted precision, recall, and F1-score all exceeded 70%, indicating robust performance despite imbalanced class distributions. The most important predictors for cause classification included Shortage Incidence Frequency, Existence of alternative drugs with the same ingredient, and Business size of the Marketing Authorization Holder. Conclusion: The KMFDS should continuously monitor drugs with repeated shortage episodes through regular reporting, early-warning systems, and supply-risk assessments. Incorporating supply-side indicators-particularly those related to economic feasibility-into national surveillance programs may help prevent shortages and mitigate their duration. By identifying key predictors associated with shortage causes, this study provides evidence to guide policy prioritization and targeted interventions.