The effect of social relationships on cognitive decline in older adults: an updated systematic review and meta-analysis of longitudinal cohort studies

Abstract

Abstract Background A previous meta-analysis (Kuiper et al., 2016) has shown that multiple aspects of social relationships are associated with cognitive decline in older adults. Yet, results indicated possible bias in estimations of statistical effects due to the heterogeneity of study design and measurements. We have updated this meta-analysis adding all relevant publications from 2012 to 2020 and performed a cumulative meta-analysis to map the evolution of this growing field of research (+80% of studies from 2012-2020 compared to the period considered in the previous meta-analysis). Methods Scopus and Web of Science were searched for longitudinal cohort studies examining structural, functional and combined effects of social relationships. We combined Odds Ratios (OR) with 95% confidence intervals (CI) using random effects meta-analysis and assessed sources of heterogeneity and the likelihood of publication bias. The risk of bias was evaluated with the Quality of Prognosis Studies in Systematic Reviews (QUIPS) tool. Results The review was prospectively registered on PROSPERO (ID: CRD42019130667). We identified 34 new articles published in 2012-2020. Poor social relationships were associated with cognitive decline with increasing precision of estimates compared to previously reviewed studies [(for structural, 17 articles, OR: 1.11; 95% CI: 1.08; 1.14) (for functional, 16 articles, OR: 1.12; 95% CI: 1.05; 1.20) (for combined, 5 articles, OR: 1.15; 95% CI: 1.06; 1.24)]. Meta-regression, risk and subgroup analyses showed that the precision of estimations improved in recent studies mostly due to increased sample sizes. Conclusions Our cumulative meta-analysis would confirm that multiple aspects of social relationships are associated with cognitive decline. Yet, there is still evidence of publication bias and relevant information on study design is often missing, which could lead to an over-estimation of their statistical effects.