JMIR Publications

ABSTRACT

Background:

Loneliness is a longstanding and widely acknowledged public health problem among older adults, yet evidence remains mixed regarding how virtual communication is associated with loneliness. Less is known about whether this association varies across baseline profiles among older adults.

Objective:

To examine the association between frequent virtual communication with family, friends, or acquaintances and loneliness among older adults in Japan, and to explore whether this association varied across baseline profiles.

Methods:

Using linked 2019 and 2022 data from the Japan Gerontological Evaluation Study, we conducted an observational analysis of community-dwelling Japanese adults aged 65 years or older (N = 4,807). Baseline covariates were measured in 2019. Frequent virtual communication was assessed in 2022 based on retrospective reports of internet or email use for communication with family, friends, or acquaintances during the preceding year; frequent use was defined as use at least two to three times per week. Participants not meeting this definition were categorized as non-users/infrequent users. Loneliness was measured in 2022 using the 3-item UCLA Loneliness Scale (range 3-9). We used targeted maximum likelihood estimation with SuperLearner and 42 baseline covariates to estimate the sample-average adjusted association, and a generalized random forest approach to explore heterogeneity in the estimated association across baseline profiles. Supplementary analyses examined baseline loneliness adjustment in an alternative linked sample, a dichotomized loneliness outcome, propensity score overlap, and model calibration.

Results:

The analytic sample included 4,807 participants, of whom 1,616 (34%) met the frequent virtual communication definition. After adjusting for observed baseline covariates, frequent virtual communication was associated with −0.136 lower loneliness scores (95% confidence interval −0.194, −0.078). Generalized random forest analyses suggested exploratory heterogeneity in the estimated association across baseline profiles; most individual-level forest predictions for the estimated association were below zero (93.5%). In supplementary analyses, the association remained in the same direction after adjustment for baseline loneliness in an alternative linked sample (adjusted mean difference -0.117, 95% CI -0.173 to -0.061) and when loneliness was dichotomized (odds ratio for loneliness 0.734, 95% CI 0.597 to 0.904). Propensity score diagnostics showed overlap between exposure groups, although overlap was limited in the tails.

Conclusions:

In this observational analysis of older adults in Japan, frequent virtual communication was associated with modestly lower loneliness, and generalized random forest analyses suggested exploratory heterogeneity across baseline profiles. The findings should be interpreted cautiously because the exposure was retrospectively reported, loneliness was measured in the same survey wave, baseline loneliness was not available in the main analytic sample, and sample selection may have affected generalizability. The findings should not be interpreted as evidence of causal effects.