Development of 5-year risk prediction models for incident dementia and mortality in a community-dwelling older Japanese population: The Japan Prospective Studies Collaboration for Aging and Dementia (JPSC-AD).

Improving cognitively healthy survival is important for achieving healthy aging. Therefore, it would be valuable to estimate the future risk of either incident dementia or death in community-dwelling older adults. This study aimed to develop a set of risk prediction models for either incident dementia or death that can be applied according to data availability across diverse clinical settings, using longitudinal data from community-dwelling older Japanese adults. A total of 8,334 participants aged ≥65 years were prospectively followed up from 2016 to 2023. Logistic regression was used to develop multivariable prediction models. The developed models were translated into simplified scoring systems based on the β coefficients. The discrimination abilities of the models were assessed by C-statistics, and the calibration was assessed by calibration plots. During the follow-up period, 1,151 participants developed either dementia or death. In the multivariable model using potential predictors commonly available in the primary care setting, age, male sex, formal education ≤9 years, diabetes mellitus, use of lipid-lowering agents, leanness, history of stroke, history of heart disease, history of respiratory disease, history of cancer, current smoking, no regular exercise habit, and low frequency of social interactions were selected as predictors. Adding cognitive status, depressive symptoms, apolipoprotein E-ε4 carrier status, and brain MRI markers to the model further enhanced its predictive performance. The developed models and simplified scores showed good discrimination and calibration. Risk stratification with our models may be useful for assessing the risks of incident dementia and death and, consequently, for prolonging cognitively healthy survival.