The term “drug shortage” lacks a globally standardized definition, with significant variations or no definitions at all across countries (World Health Organization, 2015). From a supply perspective, drug shortages refer to the inadequate availability of medicines, health-related products, and vaccines required to meet public health and patient needs within the healthcare system. On the demand side, shortages arise when demand exceeds supply at any point in the supply chain, potentially leading to stockouts and an inability to meet clinical needs if not addressed promptly (Acosta et al., 2019; Health Canada, 2024).

Drug shortages impose significant societal costs on the healthcare sector, including reduced patient access to medicines, increased healthcare expenses, disruptions in the pharmaceutical markets, heightened anxiety among healthcare providers, and delays in developing innovative drugs (Fox et al., 2014; Pauwels et al., 2015; Shukar et al., 2021). During the COVID-19 pandemic, the global value chain for biopharmaceuticals faced unprecedented threats, intensifying global concerns about drug shortages. One study reported findings from a five-year study that emphasized the growing challenge of drug shortages and pointed the importance of proactive strategies, including investing in technology, strengthening supplier relationships, and advocating for policy reforms (Ayati et al., 2020; Sallam et al., 2024). Addressing this challenge is a critical issue for national healthcare systems (Ivanov and Das, 2020; Ayati et al., 2020; Piatek et al., 2020).

The prediction of drug shortages, time to shortage, and recovery period following a shortage, depending on the characteristics of the drug, are key factors in managing drug shortages (Tucker and Daskin, 2021). In Canada, a study was conducted to develop a machine-learning model to predict drug shortages using data from 22 pharmacy sales and historical drug shortage records (Pall et al., 2023). The model was able to predict shortage classes (none, low, medium, and high) with 69% accuracy 1 month in advance. Despite the lack of inventory data from drug manufacturers and suppliers, this demonstrates meaningful performance given limited data.

In South Korea, the pharmaceutical industry predominantly produces generic drugs and relies heavily on imported active pharmaceutical ingredients (APIs) as well as new drugs, making the supply chain sensitive and vulnerable to shortages. To address these challenges, in 2019, the government developed a government-led drug supply disruption prediction model in collaboration with the Korea Orphan and Essential Drug Center, using data from reports on drug supply interruptions and shortages to address drug shortages (National Institute of Food and Drug Safety Evaluation, 2019). In addition, to proactively identify supply and demand imbalances, they implemented a supply and demand forecasting project utilizing artificial intelligence. To address these imbalances, they expanded the list of nationally essential drugs and reinforced support for production (Seoul Pharmaceutical Association, 2023). However, the limitations of this model-designed only to provide real-time alerts for potential shortages-prevented it from addressing the underlying problems, leading to a significant increase in the number of drugs with unstable supply despite the expansion of national essential drugs (National Institute of Food and Drug Safety Evaluation, 2019; Seoul Pharmaceutical Association, 2023).

Thus, a retrospective analysis of 1,054 regulatory-reported drug shortages predictive system is urgently required to effectively address drug shortages. While the previous model aimed to provide real-time alerts for potential shortages, we analyze drugs that have already undergone shortages, examining their shortage duration range and cause to identify key features relevant for prediction. Also, unlike the previous model, our dataset includes information not only on the causes of shortage but also on their duration ranges. Therefore, this study aims to develop two types of models: one for predicting the causes of drug shortages and another for estimating their duration range. Identifying key features with high importance in each model could help to determine the factors crucial for predicting the occurrence of shortages based on their causes and estimating their duration in South Korea. By providing not only early indications of whether a shortage may occur but also insights into why it happens and how long it is likely to persist, our models allow decision-makers to implement more strategic and timely interventions. These actionable insights support more targeted policy responses and early risk mitigation, ultimately contributing to a more resilient drug supply system in South Korea.

Table 1 shows a summary of 1,054 cases of drug shortages reported by pharmaceutical companies to the KMFDS between 2018 and 2024. The data were categorized based on key attributes, including those related to drug shortage events, drug supply monitoring system, drug manufacturing, and drug characteristics.

The study found that the primary factor for predicting the shortage duration range in South Korea is the Shortage Incidence frequency. A repeated history of shortages reflects persistent supply-side vulnerabilities—such as limited manufacturing capacity, unstable access to APIs, insufficient redundancy across suppliers, or recurring quality-control issues—which remain unresolved over time and increase the likelihood that the same product will experience future disruptions.

The results indicated that the Shortage Incidence frequency, Business size of the Marketing Authorized Company, and Existence of alternative drugs with same ingredients were highly important in most shortage causes. These findings suggest that products marketed by larger Marketing Authorized Companies tend to have more stable production capability, whereas drugs that do not have therapeutically equivalent alternatives face greater risk during supply interruptions. Because these features are all supply-side characteristics, the results indicate that production and supply-chain constraints exert a stronger influence on drug shortages in South Korea than demand-side factors.

The study period (2018–2024) encompassed both pre-pandemic and post-pandemic environments, enabling evaluation of whether COVID-19–related supply-chain disruptions altered shortage determinants. Although national shortage counts fluctuated during the pandemic, the ranking of key predictors remained stable, indicating that the determinants identified in this study represent underlying structural characteristics rather than temporary pandemic-specific effects.

Drug shortages arise from a complex interplay of factors, encompassing supply and demand issues, regulatory challenges, and quality concerns, with various risks and frequencies depending on the drug characteristics (Mazer-Amirshahi et al., 2014). These factors vary depending on national healthcare systems, pharmaceutical market structures, and government regulations (Gray and Manasse, 2012; Acosta et al., 2019). The risk and occurrence patterns of drug shortages also differ according to manufacturer characteristics, domestic production versus importation status, therapeutic class, route of administration, and whether a drug is branded or a generic product (Fox et al., 2014; Giammona et al., 2020). Generic drugs, in particular, are more susceptible to shortages due to low profitability and limited production capacity, and this susceptibility is further exacerbated by quality-control issues and the fragile supply chain of APIs (Fox et al., 2014; United States Government Accountability Office, 2016).

The U.S. pharmaceutical market faces challenges (factors) that drive drug shortages, including limited incentives for producing low-margin medicines, inadequate evaluation and compensation for quality control in manufacturing, and logistical and regulatory challenges in ensuring a stable drug supply (U.S. Food and Drug Administration, 2020). In contrast, most drug supply disruptions in Canada arise from supply-related issues, including pharmaceutical quality control problems, production delays, recalls, regulatory actions, product suspensions, and the unavailability of raw materials (The Multi-Stakeholder Steering Committee on Drug Shortages, 2017). In China, drug supply disruptions are primarily attributed to unprofitable pricing. Price competition in the drug market and government-imposed price reduction policies, which lower prices to unprofitable levels, are identified as key contributors to drug shortages (Yang et al., 2016). Previous research also reports that middle-income countries experience additional challenges such as licensing delays and limited raw materials, while low-income countries face more severe shortages due to insufficient research capacity and weak policy frameworks (Gray and Manasse, 2012). These findings show that drug shortages in many countries are driven predominantly by supply-side factors, which is consistent with the pattern observed in South Korea.

In high-income nations, manufacturing issues, business decisions, raw material shortages, and regulatory challenges are the primary factors contributing to drug shortages. In middle-income countries, similar issues prevail, but additional factors, such as licensing delays and insufficient raw materials for local manufacturers, also play a significant role. In contrast, low-income countries face exacerbated shortages due to limited research, insufficient data, and weak policy frameworks (Gray and Manasse, 2012). These international patterns demonstrate that drug shortages are largely driven by supply-side vulnerabilities, which aligns with the situation in South Korea, where supply-side factors were also found to be highly influential in this study.

Research on drug shortages has been more extensive in developed nations than in South Korea, and many of the insights identified in these countries are relevant to the Korean context. Therefore, addressing drug shortages in South Korea may benefit from applying and adapting strategies that have been effective in other high-income settings.

Complex machine-learning models often show strong internal fit but are vulnerable to overfitting, particularly when class distributions are imbalanced or when the number of predictors is large relative to the sample size. The risk of overfitting is especially important, as specific patterns learned from past data could hinder early recognition of new or atypical shortage signals. To mitigate this risk, the study incorporated multiple safeguards. First, a stratified 70/30 train–test split was used to preserve class proportions, and the test set was kept entirely independent during model development to prevent information leakage. Second, Random Forest–specific techniques such as bootstrap aggregation and out-of-bag (OOB) validation helped assess model performance from quasi-independent samples. Third, Bayesian optimization was applied to tune key hyperparameters including tree depth, sample size, and feature sampling rate, thereby preventing overly complex trees and improving generalizability. Collectively, these techniques reduced the likelihood of overfitting and contributed to the robustness of the final prediction models.

Since drug shortages are closely related to both supply and demand factors, the feature analysis of the study would be enhanced by integrating data from the supply side (i.e., pharmaceutical companies) for modeling. In addition, when the distribution of categorical features is skewed toward “'0,” indicating “unknown,” using out-of-bag bootstrap sampling may lead to the construction of a random forest model where many trees exclude data other than “0” from their samples. Among the 1,054 products analyzed in this study, only seven (0.66%) had an “unknown” route-of-administration label, a level of missingness too small to meaningfully influence model performance. Therefore, the low feature importance of Routes of drug administration in the Shortage Occurrence Prediction Models is likely attributable to the inherently limited relevance of the route-of-administration variable to shortage mechanisms rather than distortions caused by missing data.

This study showed features with high importance for predicting shortage duration causes based on drug shortage data in South Korea. Identifying the importance of each factor for predicting shortage duration helps to identify the duration when a shortage occurs and facilitates the rapid normalization of the drug supply. Analyzing the key features with high importance for each shortage cause would help establish strategies to identify the root cause of future shortages. With additional integration of supply-chain and production data from manufacturers, the proposed models could be interfaced with South Korea’s national drug shortage monitoring system to support real-time risk identification and regulatory response. Such applications would contribute to healthcare system resilience by shifting shortage management from reactive to proactive approaches.

Conducting shortage pattern research using data from various organizations within the drug market, such as pharmaceutical companies, community pharmacies, and hospital pharmacies, along with publicly available data, may contribute to the stability of drug supply in South Korea. Future refinement and practical implementation of the models will require collaboration among government regulators, marketing authorization holders and manufacturers, wholesalers, hospital and community pharmacies, and academic researchers. Integrating data across these stakeholders will help establish a more coordinated national framework for drug shortage prevention and management.

This study was conducted to analyze drug shortages in South Korea in depth but has some limitations. First, this study selected 18 features based on data accessibility and relevance, but potentially important factors such as changes in patient behavior (e.g., changes in purchasing patterns or drug compliance), global supply-chain disruptions, and policy changes were not included due to data collection constraints. These factors were difficult to measure consistently and were not available in standardized, detailed datasets. Second, this study utilized data from 2018 to 2024, and data collection within a limited period may not fully reflect long-term trends in drug shortages. Third, the prediction model developed in this study mainly focused on supply-side factors, so interactions with demand-side factors could not be fully considered. Fourth, the distributions of shortage duration and shortage-cause categories were uneven, which may have biased the prediction models toward majority classes. We acknowledge that the current model does not incorporate specific imbalance-handling techniques (e.g., class-weight adjustments or resampling), which may limit predictive performance for rare but meaningful shortage types. Future work should explore tailored imbalance mitigation strategies to improve minority-class sensitivity. Fifth, external validation using independent datasets—such as those from hospitals, wholesalers, or community pharmacies—was not performed, limiting the ability to assess model performance in real-world or cross-institutional settings. Sixth, because the model was trained solely on South Korean regulatory data, its generalizability to other health systems with different pharmaceutical structures may be limited. Lastly, the predictive model proposed in this study requires additional empirical studies to verify real-world applicability and effectiveness. These limitations are expected to be supplemented in follow-up studies, thereby enhancing the accuracy and practicality of drug shortage prediction models and contributing to the development of more effective policies.

This study developed two machine-learning models—one predicting shortage duration ranges and another predicting shortage occurrence by cause—and identified shortage incidence frequency as the most important predictor of future shortage duration. South Korea must implement an integrated data-sharing system among stakeholders to enable a swift response to future shortages and ensure their rapid resolution. Data should be shared transparently and consistently to complement the current incomplete reporting system. Drugs that have experienced several shortages should have their supply chain continuously monitored. Furthermore, the shortage prediction program in South Korea, which focuses on supply-side features such as economic viability, may help prevent shortages or reduce their duration and contribute to strengthening healthcare system resilience. Effective nationwide implementation will require collaboration among government regulators, marketing authorization holders, manufacturers, wholesalers, hospital and community pharmacies, and academic researchers.