Lim Ji Young, Kim Eun Joo, Kim Seong Kwang
Inha University, Incheon, Korea (the Republic of).
Gangneung-Wonju National University, Wonju-si, Korea (the Republic of).
BMJ Open. 2025 Aug 31;15(8):e093591. doi: 10.1136/bmjopen-2024-093591.
To identify the key factors associated with admission to elderly medical-welfare facilities in South Korea and to evaluate their relative importance using machine learning techniques, providing an evidence base for policy in a rapidly ageing society.
A cross-sectional secondary data analysis.
The analysis was conducted using the National Health Insurance Service -Senior database, a large-scale public dataset constructed to be statistically representative of the elderly population in South Korea.
A total of 48 614 elderly individuals aged 60-80 years, selected through a scientifically rigorous stratified random sampling method.
The primary outcome was the binary classification of an individual's long-term care arrangement: admission to a facility-based service versus utilisation of home-based care. The secondary outcome was the relative importance and contribution of a wide range of demographic, health-status and care-related variables in predicting facility admission.
After addressing severe class imbalance with the synthetic minority over-sampling technique, our final random forest model demonstrated excellent predictive power, achieving a sensitivity of 0.856 and a balanced accuracy of 0.921. An analysis of feature importance using Shapley Additive Explanations (SHAP) revealed that the most dominant predictor was an individual's cohabitation with institutional staff (mean SHAP value ≈ 0.180), indicating pre-existing contact with the formal care system. Other critical factors included the absence of a primary caregiver (≈ 0.056), being fully dependent due to dementia (≈ 0.017) and the degree of functional impairment as measured by the activities of daily living (ADL) score (≈ 0.017).
Admission to medical-welfare facilities in South Korea is a multifactorial issue, most strongly driven by the erosion of informal care support systems combined with severe health decline, particularly in cognitive and physical function. These evidence-based findings highlight the need for policy interventions that strengthen community-based integrated care and enhance support for family caregivers. Such strategies are essential for promoting ageing-in-place, ensuring the long-term sustainability of the public long-term care system and ultimately improving the quality of life for the nation's growing elderly population.
确定与韩国老年医疗福利设施入住相关的关键因素，并使用机器学习技术评估其相对重要性，为快速老龄化社会的政策提供证据基础。
横断面二次数据分析。
分析使用国民健康保险服务-老年人数据库进行，该大型公共数据集在统计学上代表了韩国老年人口。
通过科学严谨的分层随机抽样方法选取了48614名60至80岁的老年人。
主要结局是个人长期护理安排的二元分类：入住机构服务与使用居家护理。次要结局是一系列人口统计学、健康状况和护理相关变量在预测机构入住方面的相对重要性和贡献。
使用合成少数过采样技术解决严重的类别不平衡问题后，我们最终的随机森林模型显示出出色的预测能力，灵敏度达到0.856，平衡准确率达到0.921。使用夏普利值法（SHAP）进行的特征重要性分析表明，最主要的预测因素是个人与机构工作人员的同居情况（平均SHAP值≈0.180），这表明与正式护理系统有预先存在的接触。其他关键因素包括没有主要照顾者（≈0.056）、因痴呆症而完全依赖他人（≈0.017）以及通过日常生活活动（ADL）评分衡量的功能受损程度（≈0.017）。
韩国老年医疗福利设施的入住是一个多因素问题，最主要的驱动因素是非正式护理支持系统的侵蚀以及严重的健康衰退，尤其是认知和身体功能方面。这些基于证据的发现凸显了政策干预的必要性，即加强基于社区的综合护理并增强对家庭照顾者的支持。此类策略对于促进就地养老、确保公共长期护理系统的长期可持续性以及最终改善该国不断增长的老年人口的生活质量至关重要。
