Comprehensive Summary
This study identified key factors associated with elderly medical-welfare facilities in South Korea to enhance machine learning techniques for improving elderly care. This cross-sectional study used secondary data analysis with the National Health Insurance Service-Senior database, which consists of 60-80 year old individuals in South Korea, and a synthetic minority over-sampling technique (SMOTE) was applied to balance the dataset by socioeconomic status for training random forest models. Using the Shapley Additive Explanations (SHAP), cohabitation with institutional staff, absence of primary caregiver, full dependency due to dementia based on ease of activities of daily living were associated with increased likelihood of admission to medical-welfare facilities. With these, policy interventions would strengthen community-based integrated care to provide better support for families to promote aging-in-place. One limitation is the cross-sectional design, which prevents conclusions about causality, especially the factor of cohabitation with staff. Currently, South Korea has a shifting population pyramid of increased older populations, and as such, may have to change social structures to align more closely with countries like Japan and Germany to accommodate for population shifts.
Outcomes and Implications
This study incorporated a machine learning approach, using Random Forest and SHAP analysis for large representative national datasets to identify key determinants of South Korean elderly admission. Overall, institutionalization was found to be a multifactorial issue driven by the erosion of support systems along with cognitive and physical decline in older populations. One of the most influential predictors was an individual's cohabitation with institutional staff along with an absence of primary caregiver, which could inform policy care and improve healthcare policies in South Korea. By looking beyond current systems in the country, healthcare professionals can expand on their standard of care using a holistic approach and specialize community-based integrated care systems to target underserved groups more effectively. While the results may be correlational, this research prompts further studies to look into social factors to improve patient care for at-risk groups, especially in diverse populations that are relatively understudied.