Digital participation and socioeconomic inequalities in later-life health: a cross-national analysis of six harmonized aging cohorts

Digital participation is associated with later-life health and varies across socioeconomic groups. We examined how the fitted SES–health association changed after including reported participation and whether cross-cohort differences in this participation-related portion reflected socioeconomic patterning of participation, adjusted health differences between participants and non-participants, or both. We analyzed 449,861 person-waves from six harmonized aging cohorts of adults aged 50 years or older. Unweighted complete-case regressions related standardized socioeconomic status (SES) to a health index combining cognition, functioning, depressive symptoms, and self-rated health, adjusting for demographics, available behavioral covariates, and wave, with respondent-clustered standard errors. Descriptive mediation and cross-cohort decompositions examined internet use in HRS, ELSA, SHARE, and CHARLS, with a broader communication indicator analyzed separately in MHAS; KLoSA contributed health-gradient comparisons. Mediation intervals used 1,000 respondent-level bootstrap resamples. Adjacent-wave models adjusted for baseline health and SES. Secondary analyses examined participation breadth, education, geography, weights, and age definitions. Higher SES was associated with better composite health in every cohort. On cohort-standardized scales, the adjusted health difference per one-SD higher SES ranged from 0.185 SD in KLoSA to 0.411 SD in HRS. After adding reported internet use, the fitted SES–health coefficient was 11.3%–24.2% smaller in the four internet-specific cohorts and 6.3% smaller for MHAS’s broader measure. In comparisons with HRS, socioeconomic patterning of participation made the larger contribution. After accounting for access and internet use, the ELSA activity measure was associated with a further 5.6-percentage-point reduction in the fitted SES–health coefficient (95% confidence interval [CI] 4.4–6.7), whereas the CHARLS device-count measure changed it by 0.2 points (95% CI − 0.3 to 0.6). Baseline digital participants had higher adjusted health scores at the adjacent wave; in MHAS, the survey-weighted participant–non-participant difference was 0.021 SD (95% CI − 0.024 to 0.067). Education-stratified estimates indicated cohort-specific patterns. These cohort-specific analyses show that reported digital participation is socially patterned and associated with part of the SES–health gradient. Given the observational design, the findings point to priorities for direct evaluation: whether affordable, accessible digital support can help disadvantaged older adults sustain participation and how any health-related effects are distributed across socioeconomic groups.