Heterogeneity in the association between internet use and dementia among older adults: A machine-learning analysis

Abstract

Background & aims

Internet use among older adults may reduce the risk of dementia, but it remains unknown how the effects vary across individuals. The aim of this study was to rigorously examine heterogeneity in the association between internet use and dementia among older adults with a machine learning approach.

Methods

This cohort study used data from the Japan Gerontological Evaluation Study involving functionally independent adults aged 65 or older (n = 5,451). The exposure, internet use a few times a month or more often, was assessed with the 2016 survey (baseline) and covariates (potential confounders and effect modifiers) were assessed with the 2013 survey (pre-baseline). Follow-up continued until 2022, identifying 5.5-year dementia onset (n = 549) using the public long-term care insurance system. Using the generalized random forest algorithm, we estimated how the association between internet use and dementia onset during a 5.5-year follow-up period varies by pre-baseline sociodemographic characteristics and health conditions.

Results

Internet use was on average associated with a lower risk of dementia (estimated population average effect = -0.033; 95 % CI: -0.051, -0.016). However, we found evidence of between-individual heterogeneity in this association, where internet use appeared more beneficial among individuals who reported middle income, higher education levels, and were socially and physically inactive at the pre-baseline wave.

Conclusions

Internet use may disproportionately benefit people based on socioeconomic status, suggesting equity concerns of universal implementation. Understanding such effect heterogeneity can inform more targeted public health interventions.