Aims: Existing analysis methods about the relationships between the long-term care (LTC) risks and determinants of the elderly often experience issues of independence, linearity and homogeneity. Such methodological issues are particularly problematic when the relationships are complicated, making the prioritization of LTC interventions to groups with different care needs difficult. To overcome these shortcomings, this study made an initial attempt to identify representative patterns of elderly people's LTC risks associated with different determinants, from a methodological perspective.
Methods: A data mining approach (Exhaustive CHAID) is applied to data collected from 26 357 elderly people in a Japanese city in 2016. It can automatically segment the samples into a tree structure that accommodates nonlinear, heterogeneous and sequential relationships, without caring about issues of independence across determinants.
Results: In total, 16 representative patterns are derived, among which four patterns are considerably riskier (the unhealthy percentage >10.0%), and five are moderately riskier (the unhealthy percentage is 5.0%-10.0%). Age is the most important determinant, followed by locomotor function, cognitive function, homebound, gender and residential location, in that order, which all show heterogeneous associations with the risk patterns.
Conclusions: The Exhaustive CHAID is suitable to derive scientifically-sound insights into an early screening of risk factors for pre-frail groups and tailor-made preventive LTC interventions for frail groups in a flexible and objective way. Geriatr Gerontol Int 2020; 20: 674-679.
Keywords: Exhaustive CHAID; Japanese elderly people; Kihon Checklist; heterogeneity; risk of long-term care.
© 2020 Japan Geriatrics Society.