Low Digital Engagement, Depressive Symptoms, and Probable Cognitive Impairment Among Community-Dwelling Older Adults in Shanghai: A Cross-Sectional Path Analysis of Self-Perceived Burden, Uselessness, and Loneliness

Low engagement with digital technologies is increasingly recognized as a health-related correlate in later life, yet the psychosocial variables linking it to mental and cognitive health remain unclear. Guided by the stress-process model, this cross-sectional study examined whether the associations of low digital engagement (LDE) with depressive symptoms and probable cognitive impairment were consistent with indirect pathways through self-perceived burden, self-perceived uselessness, and loneliness. We analyzed data from 1703 community-dwelling adults aged 55 to 85 years in Shanghai, China, collected between 2022 and 2023. Two path models specified self-perceived burden and self-perceived uselessness as parallel first-stage mediators and loneliness as a second-stage mediator: a linear model for depressive symptoms (CES-D-10) and a logit model for probable cognitive impairment (AD8 ≥ 2), each adjusting for sociodemographic, lifestyle, and health covariates (5000 bootstrap replications). Greater LDE was associated with more depressive symptoms (b = 0.062, 95% CI: 0.011, 0.113) and higher odds of probable cognitive impairment (OR = 1.040 per point, 95% CI: 1.014, 1.067). For depressive symptoms, the total indirect association accounted for 75.2% of the total association: significant specific indirect components included burden (30.3%), uselessness (15.6%), and the serial chains via burden and loneliness (8.7%) and via uselessness and loneliness (7.0%). The direct association was not statistically significant. For probable cognitive impairment, the direct association remained significant, and the significant indirect components were limited to the burden route. Formal contrasts indicated no significant difference between the burden and uselessness routes. The association between LDE and loneliness was attenuated to non-significance once the self-evaluative variables were included. Reverse models fit the data equally well. These cross-sectional associations generate hypotheses for future longitudinal and intervention research in rapidly digitizing urban settings.