Subtypes of Older Adults Starting Long-Term Care in Japan: Unsupervised Machine Learning Approach

ObjectivesThe older population requiring long-term care (LTC) exhibits heterogeneity in physical and cognitive functions; however, an established classification is lacking. We aimed to identify distinct subgroups of older adults with LTC needs using unsupervised machine learning and to examine differences in their prognoses. DesignRetrospective cohort study. Setting and participantsUsing survey data for care-need certification (linked to LTC and medical insurance claims) in City A, Japan, we identified community-dwelling adults aged [≥] 65 years who started LTC. Data from City B were used for validation of clustering. MethodsWe applied latent class analysis to group the participants in City A, based on all 74 items (38 on physical functions, 9 on cognitive functions, 15 on behavioral problems, and 12 on medical procedures) in the Japanese standardized care-needs certification survey. Then, we examined the association between the identified subtypes and four outcomes, including death, hospitalization, nursing home admission, and care-need level deterioration, using regression models. ResultsAmong 3,841 participants in City A (median age, 83 years; 59.3% female), five subtypes were identified: (i) mild physical, (ii) mild cognitive, (iii) moderate physical, (iv) moderate multicomponent, and (v) severe multicomponent. The results of clustering were replicated in City B. Compared with the mild physical subtype, the severe multicomponent subtype showed the highest risk of death (adjusted hazard ratio [aHR] 2.56; 95% confidence interval [CI] 2.02-3.24), and nursing home admission (aHR 5.91; 95% CI 4.57-7.63). The moderate physical subtype showed a higher risk of hospitalization (aHR 1.32; 95% CI 1.16-1.49), and the moderate multicomponent subtype was more likely to experience care-need deterioration (adjusted odds ratio 1.67; 95% CI 1.26-2.22). Conclusions and ImplicationsThis study identified five subtypes of older adults who started LTC. These findings inform individualized care decisions and tailored planning of medical and LTC services.