Longitudinal association between social interaction and screen-positive cognitive impairment risk in older adults: domain-specific effects and the impact of baseline cognitive status

Abstract

Social relationships are key social determinants of healthy aging. However, the population-level impact of different dimensions of social interaction and their dependence on baseline cognitive status remain insufficiently understood. We used a two-wave longitudinal dataset (2020–2023) of community-dwelling older adults ( N  = 418). Social interaction was assessed using the Index of Social Interaction, covering five domains: independence, social curiosity, interaction, participation in society, and feeling of safety. The outcome was a self-reported Kihon Checklist screen-positive cognitive impairment risk classification rather than a clinical diagnosis. Baseline-cognition-adjusted prospective regression models were designated as the principal analyses, and models without baseline cognition were secondary. Missing values in the derived analytic variables were handled using multiple imputation with 40 completed datasets and pooling according to Rubin’s rules. Linear-model inference used HC3 heteroscedasticity-robust standard errors, and inverse-probability-of-attrition weighting (IPAW) was used as an attrition sensitivity analysis. In the principal baseline-cognition-adjusted model, baseline total ISI was not independently associated with 2023 screen-positive cognitive impairment risk (OR = 0.92, 95% CI 0.79–1.08, p  = 0.309), and none of the five ISI domains was significant. Baseline screen-positive cognitive impairment risk was strongly associated with follow-up screening status (total-ISI model OR = 8.05, 95% CI 4.91–13.21, p  < 0.001). Secondary models without baseline cognition showed inverse associations for total ISI (OR = 0.83, 95% CI 0.72–0.96, p  = 0.011) and social curiosity (OR = 0.72, 95% CI 0.56–0.92, p  = 0.008), but these did not persist in the principal adjusted analyses. Stabilized IPAW yielded a non-significant total-ISI estimate (OR = 0.896, 95% CI 0.797–1.008, p  = 0.067); 1st–99th percentile truncation yielded OR = 0.901 (95% CI 0.800–1.015, p  = 0.086). After accounting for baseline cognitive screening status, we found no statistically significant independent longitudinal association between baseline social interaction and subsequent screen-positive cognitive impairment risk. The attenuation of secondary unadjusted associations suggests that pre-existing cognitive differences and reverse directionality are important considerations. Social-interaction domains should therefore be interpreted as exploratory correlates rather than established protective or preventive targets.