Social Isolation Changes and Long-Term Outcomes Among Older Adults

Key Points Question Is social isolation change associated with long-term outcomes in older adults? Findings In this cohort study using a national longitudinal health survey of 13 649 adults aged 50 years or older in the US, data revealed that increased isolation was associated with an increased risk of mortality, disability, and dementia. Decreased isolation was associated with a lower risk of mortality only among individuals who were nonisolated at baseline. Meaning These results underscore the importance of interventions targeting the prevention of increased isolation among older adults to mitigate its adverse effects on mortality, as well as physical and cognitive function decline.


Introduction
7][8] An estimated 20% to 25% of community-dwelling older adults are categorized as socially isolated, including 4% experiencing severe social isolation. 4,7,9[18][19][20][21] The dynamic nature of mental health undergoes constant change.However, much of the existing research on social isolation relies on cross-sectional measurements, overlooking the association between changes in social isolation and subsequent health outcomes.While a few studies have reported associations, such as increased social isolation was associated with functional limitations and memory decline, 22,23 this area remains underevaluated.This gap in knowledge impedes our ability to assess interventions aimed at preventing increased isolation or promoting social connections to improve health outcomes.
In this study, we estimated the association between social isolation change and risks of mortality and other health outcomes, including disability, dementia, cardiovascular disease (CVD), and stroke, using the Health and Retirement Study (HRS), a longitudinal, population-based survey study of older adults. 24We analyzed the associations in individuals with and without baseline social isolation.To our knowledge, this study represents the largest of its kind to estimate the association between social isolation changes and health outcomes stratified by baseline isolation status.

Data and Study Design
We analyzed data from 6 birth cohorts of HRS participants from the 2006 to 2020 waves. 24The HRS is a nationally representative, biennial, longitudinal health interview survey of adults aged 50 years and older in the US. 24,25All respondents provide informed consent and receive token payment on their entry into the HRS.The HRS is sponsored by the National Institute on Aging and is performed by the Institute for Social Research at the University of Michigan, Ann Arbor; it has been approved by the University of Michigan Health Sciences Institutional Review Board.The data used in this analysis are retrieved from RAND HRS 26 and contain no unique identifiers.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.Data were analyzed from October 11, 2023, to April 26, 2024.

Social Isolation Exposure
Leave-Behind Questionnaires (LBQ) were included in the HRS biennial core survey starting from 2006-2008, which included the Steptoe 5-item Social Isolation Index (SII) to measure social isolation status. 27,28With SII, each respondent was assigned a positive response to each item, including if they (1) were unmarried/living alone, (2) had less than monthly contact with children (all contacts, including face-to-face meet up, speak on the phone, write, or email), (3) had less than monthly contact with other family members, (4) had less than monthly contact with friends, and (5) did not participate monthly in any groups, clubs, or other social organizations.The final score of the SII was the sum of the 5 items, ranging from 0 to 5, with higher scores indicating more isolated status.
Respondents scoring 2 or above were categorized as socially isolated. 9,16Numerous studies have confirmed the reliability and validity of SII use for older adults. 16,29lternatively, half of HRS respondents were chosen for LBQ, resulting in their assessments of SII every 4 years.We defined each respondent's baseline assessment as the year of their initial SII measurement, with the second SII measurement occurring 4 years after baseline.The primary exposure was the change in the SII score from the initial baseline assessment to the second measurement.Respondents were then categorized as (1) decreased isolation if the score of their second measurement decreased by 1 unit or more from baseline, (2) stable if the score of their second measurement remained unchanged from baseline, and (3) increased isolation if the score of their second measurement increased by 1 unit or more from baseline.

JAMA Network Open | Geriatrics
For our sensitivity analysis, we used an alternative approach, defining changes in social isolation based on binary isolation status transitions.Specifically, for individuals initially isolated, we categorized change groups as transitioning from isolation to nonisolation vs remaining isolated.
Conversely, for initially nonisolated individuals, change groups were defined as transitioning from nonisolation to isolation vs remaining nonisolated.

Outcomes
We investigated 5 incident outcomes: mortality, disability, dementia, CVD, and stroke.The starting time for all outcomes was defined as the second SII measurement, 4 years after baseline.The time of the event was defined as the earliest occurrence of an event after the starting time (second SII measurement).Death and the year of death in the HRS were confirmed using the National Death Index 30 and the Social Security Death Index. 31Disability was assessed in the HRS core survey through self-reported activities of daily living (ADLs) 32 dependencies (walking across a room, dressing, bathing, eating, getting in or out of bed, and using the toilet) exceeding 0. The year of incident disability onset was estimated as the first year when a respondent reported at least 1 ADL dependency.Participants with no events for death or disability were censored at their last HRS interview up to 2020.

Statistical Analysis
We first performed descriptive analyses to characterize the groups of social isolation change.
Descriptive comparisons were performed using analysis of variance or a Wilcoxon rank sum test for continuous measures and χ 2 test for categorical measures.Cumulative incidence curves and incidence rates (IRs) of each outcome were estimated.Time-to-event was defined from the year of the second SII measurement to the time of event if the participants had the event or to the time of the last HRS interview if the participants did not have the outcome.Death was treated as a competing outcome for disability, dementia, CVD, and stroke, analyzed by the Fine-Gray model.
To mitigate potential confounding effects, we used inverse probability for treatment weights (IPTW). 35To calculate the denominator of the IPTW, we used multinomial logistic regression models to model the 3 groups as a function of both baseline covariates and changes in covariates between baseline and the second SII measurement. 35We included the same covariates as potential confounders regardless of their significance in the multinomial model.These baseline covariates included age, sex, HRS cohort, race and ethnicity, educational level, total assets, marital status, body mass index, smoking status, CES-D scores, ADL dependence, cognition, and comorbidities; the change in covariates included changes in CES-D, ADL, cognition, and comorbidities from baseline to the second SII measurement.To assess the IPTW, we conducted a comparison between the absolute standardized differences in covariates for the unweighted and weighted samples.Following that, we used IPTW-weighted Cox proportional hazards regression models to analyze each time-to-event outcome, adjusting for all covariates.Separate analysis was performed for baseline nonisolated and isolated groups.
We performed 2 additional sensitivity analyses.First, we excluded individuals who died within 2 years after the second SII measurement.Second, we incorporated HRS sampling weights at baseline.All analyses were performed in R, version 4.3.1, with package IPTW (R Project for Statistical Computing).Significance was defined as P < .05,using 2-sided tests.

Results
The analysis cohort comprised be older, female, married, and non-Hispanic White and had lower levels of education.At baseline, they also tended to have higher total assets, poorer physical and cognition function, and a higher prevalence of comorbid conditions (eg, hypertension, CVD, and arthritis).Between baseline and the second SII measurement, they experienced higher incidences of stroke and psychiatric problems, increased levels of depression, and worsened physical and cognitive function declines.The demographic characteristics stratified by baseline social isolation status are summarized in eTable 3 in Supplement 1.There were no significant differences in follow-up periods among the 3 comparison groups for any outcome.
Cumulative incidence curves of mortality, disability, and dementia outcomes are shown in Figure 2.For mortality and disability, the increased isolation group exhibited the highest cumulative incidence, followed by the stable group, and then the decreased isolation group.These patterns were consistent for both baseline nonisolated and isolated respondents (Figure 2A-D).A higher incidence of dementia was observed for the increased isolation group, while the stable and decreased isolation groups had similar rates (Figure 2E-F).Minimal differences were observed for CVD and stroke among the 3 groups (eFigure 3 in Supplement 1).
Figure 3 presents the number of events per the number at risk and IRs of the studied outcomes for each isolation change group, stratified by isolation status at baseline.The increased isolation group exhibited higher incidences of mortality, disability, and dementia compared with the stable status and decreased isolation groups, regardless of baseline isolation status.For example, among the respondents who were not isolated at baseline, the estimated mortality IR was 20.With the application of IPTW, adjusted covariates were well balanced among the 3 isolation change groups, with no significant standardized mean difference (eFigure 2 in Supplement 1).

Discussion
In our study of a national cohort of US individuals aged 50 years or older, we found that changes in social isolation during 4 years had a long-term association with distal outcomes.In contrast to existing studies that often use cross-sectional measurements of social isolation, our study captures Disability was measured as an activities of daily living score greater than 0, and dementia was defined as Alzheimer disease or Alzheimer disease-related dementia.The cumulative incidences for cardiovascular disease and stroke are listed in eFigure 3 in Supplement 1. SII indicates Social Isolation Index.
changes in isolation levels and categorizes these changes into groups: decreased isolation, stable isolation, and increased isolation groups.Increased isolation was consistently associated with increased risks of mortality, disability, and dementia, irrespective of the individual's isolation status at baseline.These results suggest a need for interventions aimed at averting increases in isolation among older adults as a means to mitigate its adverse outcomes regarding mortality, as well as physical and cognitive function decline.9) 2918 ( 8) 2335 ( 13) 460 (21)   1924 (23)   1415 ( 26) 633 (7)   2776 (7)   2217 (9)   Incidence, 1000 person-years (95% CI) To our knowledge, our study represents the largest analysis to estimate the association between changes in social isolation and health outcomes.Our results align with the limited existing literature on this topic.For instance, a study analyzing 11 234 participants from the English Longitudinal Study of Aging concluded that increased social isolation predicted memory decline over 6 waves in women. 23Another US cohort-based analysis found that severe isolation over 8 years had the worst heath outcome. 22485 (29)   4553 (21)   1055 ( 15) 3596 (20)   2526 ( 26) 856 ( 18) 2918 (8)   2335 (13)   648 (9)   1924 (23)   1415 ( 26) 460 (21)   2776 (7)   2217 (9)   633 (7)   AHR (95% CI)  26) 1200 (10)   454 ( 15) 1325 (10)   748 (26) 257 (24)   849 (22)   1135 (7)   446 (7)   1283 ( 8 Increased social isolation is associated with worse health outcomes through several potential biological, behavioral, and psychological mechanisms.Cole et al 36 reported that individuals with more social isolation had increased expression of genes related to proinflammatory cytokine signaling and prostaglandin synthesis.2][43] Additionally, social isolation can activate the hypothalamic-pituitary-adrenal axis and the sympathetic nervous system, leading to behavioral alteration, such as physical inactivity, smoking, and disrupted sleep. 37Moreover, a bidirectional association between social isolation and dementia through neurogenesis related to α-amino-3hydroxy-5-methyl-4-isoxazolepropionic acid receptor and brain-derived neurotrophic factor proteins 44 has been reported. 45,46Behaviorally, increased social isolation may result in decreased engagement in social activities, reduced physical exercise, changes in dietary habits, and increased stress, depression, and anxiety, all of which can exacerbate its negative impact on health.
Conversely, our analysis revealed that a decrease in isolation was not associated with a lower risk of any studied outcomes except for mortality among respondents who initially were not isolated.
The decrease in isolation did not show significance with any outcomes among individuals who were initially isolated.This finding suggests potential complexities in the association between changes in isolation and health outcomes.Factors such as baseline isolation status, duration of isolation, and magnitude of isolation decrease may influence the observed lack of associations.Methodological limitations in measuring isolation and health outcomes further underscore the need for additional research to better understand these dynamics.

Strengths and Limitations
Our study has several unique strengths.First, the use of the HRS, with a large sample size and multiple variables collected longitudinally on respondents over 2 decades, enables powerful analysis of the long-term association between social isolation changes and distal outcomes among older adults.Second, the definitions of the social isolation change (from initial baseline SII measurement to the second SII measurement 4 years after baseline) and survival outcomes (from the second SII measurement to events) naturally establish the temporal sequence between exposure and outcome, ensuring the direction of the association.Third, through the application of IPTW, we ensure that the 3 change groups are balanced with respect to the analyzed covariates at the onset of the time-toevent outcomes.Fourth, the analysis was stratified by the social isolation status at baseline, allowing for exploration of heterogeneity associations based on the initial status of social isolation.
Our study also has limitations.Despite rigorous analysis, the HRS is a cohort study, which cannot establish causality.Additionally, the change in social isolation is limited to a 4-year span.Furthermore, the SII is defined based on only 5 questions, which may result in a ceiling effect due to its narrow range.The outcome events may be underestimated as Medicare may not fully capture CVD, stroke, and dementia diagnoses.Future studies would benefit from using finer measurements to capture a more comprehensive understanding of the impact of interventions targeting changes in social isolation.Interventions targeting social isolation in older adults are inherently complex and may have limitations in their efficacy or scope.For example, life events, such as the death of a spouse, can lead to increased social isolation and cannot be prevented.Our analysis indicates that interventions, such as increased community outreach and psychological therapies, aimed at avoiding increased social isolation in the presence of such life events may mitigate the risk of adverse outcomes.Addressing the inevitability of certain life events and incorporating this reality into intervention strategies is crucial for reducing social isolation and improving health outcomes.
SUPPLEMENT 1. eFigure 1. Flow chart of the study design eFigure 2. The standardized mean difference for covariates eFigure 3. The cumulative incidence curves of CVD and stroke for social isolation change from baseline to second isolation measure and the social isolation status eFigure 4. Unadjusted hazard ratios of changes in isolation on distal outcomes stratified by the social isolation status eFigure 5. Sensitivity analysis 1: adjusted hazard ratios of changes in binary isolation group on distal outcomes stratified by the social isolation status eFigure 6. Sensitivity analysis 2: adjusted hazard ratios of changes in isolation group on distal outcomes among individuals who did not die within 2 years post second SII measurement eFigure 7. Sensitivity analysis 3: adjusted hazard ratios of changes in isolation group on distal outcomes stratified by the social isolation status incorporating HRS sampling weights eTable 1. ICD-

Figure 2 .
Figure 2. Cumulative Incidence Curves of Mortality, Disability, and Dementia for Social Isolation Change Groups, Stratified by Baseline Social Isolation Status

Figure 3 .
Figure 3. Incidence of Distal Outcomes for Social Isolation Change Groups, Stratified by Baseline Social Isolation Status Not isolated at baseline A
Social Isolation Changes and Long-Term Outcomes Among Older Adults JAMA Network Open.2024;7(7):e2424519.doi:10.1001/jamanetworkopen.2024.24519(Reprinted) July 24, 2024 2/15 Downloaded from jamanetwork.comby guest on 07/27/2024 Among the 42 406 Health and Retirement Study (HRS) respondents in 2006 to 2020, we excluded 28 757 individuals.For disability, dementia, cardiovascular disease, and stroke outcomes, we further excluded individuals with no linked Medicare claims or with the outcome missing or the outcome before first measurement of social isolation.The flowcharts for the other 4 outcomes are listed in eFigure 1 in Supplement 1. LBQ indicates Leave-Behind Questionnaires.

Table .
Demographic and Clinical Characteristics of the HRS Analysis Cohort by Social Isolation Group From Baseline to Second Social Isolation Index Measurement (continued) d Model-adjusted 27-point cognition score.
10diagnosis codes from Medicare chronic conditions warehouse eTable 2. Social isolation index change from baseline to second social isolation measurement stratified by isolation statusJAMA Network Open | GeriatricsSocial Isolation Changes and Long-Term Outcomes Among Older Adults Demographic and clinical characteristics of the HRS analysis cohort by social isolation group from baseline to second social isolation measure eTable 4. The hazard ratios of full covariates on distal outcomes stratified by the social isolation status JAMA Network Open.2024;7(7):e2424519.doi:10.1001/jamanetworkopen.2024.24519(Reprinted) July 24, 2024 14/15 Downloaded from jamanetwork.comby guest on 07/27/2024 eTable 3.