The Association Between Social Isolation and Incident Dementia Among Older Adults: Evidence From National Health and Aging Trend Study

Abstract

Background and Objectives

Older adults are vulnerable to social isolation, making it crucial to understand its impact on dementia risk. Yet, existing evidence lacks consistency, with studies using varied measures of social isolation and overlooking potential confounders. We aim to investigate the associations between social isolation and dementia risk among older adults, hypothesizing that this association may diminish after adjusting for confounding factors.

Research Design and Methods

We used 2 977 community-dwelling older adults who had no dementia in 2015 from National Health and Aging Trends. Group-based trajectory modeling was used to analyze the trajectories of social isolation, depression, and anxiety from 2011 to 2015. Cox proportional hazards regression models were then employed to estimate the association between social isolation trajectories and incident dementia from 2015 to 2022, adjusting for demographic variables, depression, anxiety, self-rated health, smoking status, and cardiovascular disease-related variables.

Results

Three social isolation trajectories were identified: minimal, moderate, and high levels of social isolation. During a mean follow-up of 3.6 years, 19.0% of participants were diagnosed with dementia. When only demographics were adjusted, individuals in the moderate social isolation group were 22% less likely to develop dementia compared to those with high social isolation. This association between social isolation and incident dementia became nonsignificant after further adjustment for depression, anxiety, and health indicators.

Discussion and Implications

The association between social isolation and dementia risk may be mediated by factors such as depression and other health indicators.

Cognitive impairment, Depression, Longitudinal, Mental health

Translational Significance: Further research with improved methods and measures is needed to deepen our knowledge of social isolation and its impact on dementia. Minorities and people with low social–economic status, chronic diseases, and mental health disorders are disproportionately exposed to social isolation and with higher risk of dementia. Tailored intervention programs are crucial to mitigate social isolation and ameliorate the disparities of social isolation and dementia.

Older adults are particularly vulnerable to social isolation and its associated health consequences (1,2), and this issue has been exacerbated during the coronavirus disease 2019 (COVID-19) pandemic (3,4). Recently, the impact of social relationships on cognitive functions in older adults has been increasingly studied, and a number of studies indicated that social isolation was associated with increasing risks of cognitive decline and dementia (5–9). However, mixed findings were also reported (10,11). The association between social isolation and cognitive decline is likely to be overestimated because of publication bias and methodological issues (6). Particularly, these studies used diverse measures of social isolation, and many of these studies did not sufficiently control possible confounders.

Social isolation is objective and a quantifiable lack of social connectedness. A recent review of the measures of social isolation identified a dozen of measures social isolation, and some of these measures fail to capture crucial perspectives (12,13). Most current measures of social isolation are composed of indicators that capture various domains of an individual’s social connectedness, including social network, social participation, social support, and sense of belonging (12–14). The lack of a standard for social isolation measures makes it difficult to compare across studies.

The association between social isolation and risk of dementia may be mediated by a number of factors, such as behavioral factors (eg, smoking, alcohol use), psychological factors (eg, depression, anxiety), and physical factors (eg, cardiovascular disease) (8,15). Evidence shows socially isolated individuals are less likely to have social support and engage healthy behaviors but more likely to engage unhealthy behaviors such as smoking and substance use (16), are at increased risk for cardiovascular diseases (17–19) and mental health disorders such as loneliness and depression (20,21). These social relations, lifestyle behaviors, mental health disorders, and cardiovascular diseases are well-established determinants of cognitive health and dementia (5,22,23). However, many studies did not adjust these potential cofounders. According to 2 systematic reviews (5,6) that covered 53 studies until 2020, more than 70% of studies did not adjust depression in their examination of the association between social isolation and the risk of dementia.

We aimed to investigate the relationship between social isolation and the risk of dementia in older adults. Our hypothesis was that older adults experiencing social isolation were more likely to develop dementia over time. However, once potential confounding factors such as mental health disorders, smoking, and cardiovascular disease were taken into account, the association between social isolation and dementia risk may diminish or no longer be significant.

Method

Strategy and Sample

The National Health and Aging Trends Study (NHATS) is a nationally representative, longitudinal cohort of the U.S. Medicare beneficiaries aged 65 and above (24). NHATS has conducted interviews with participants each year since 2011, collecting information on participants’ physical and cognitive capacity, daily activities, and conditions of the living environment. NHATS oversampled older adults and Black individuals to maintain a representation of these groups. Besides the first cohort in 2011, the second and third cohorts were added in 2015 and 2022. Our strategy was to examine how the trajectories of social isolation, depression, and anxiety from 2011 (Year 1) to 2015 (Year 5) were associated with the subsequent development of dementia until 2022 (Year 12). Among 8 245 older adults from the first cohort, 4 152 attended the 2015 survey, and 3 618 of them provided valid values for dementia in 2015. Among them, 548 were excluded because of having dementia, and 93 were excluded because they were not community-dwelling, and a total of 2 977 older adults were included in our analyses.

Among 2 977 participants, 58% were females, 73.2% were non-Hispanic Whites, 19.5% were non-Hispanic Blacks, 7.2% were Hispanic, non-Hispanic Asian, and all other races. The mean age was 79.3 [SD = 6.8] in 2015. As for education, 46.3% were less than high school, 25.9% had a high school or some college degree, and 27.8% had a Bachelor’s degree and above. As health indicators, 23.1% self-rated as in either fair or poor health status, 6.4% currently smoke, 1.7% had a heart attack before, 24.7% had heart disease, 74.5% had high blood pressure, 28.6% had diabetes, and 2.3% had a stroke before. Please note all the above demographic statistics were based on 2015.

Measures

Dementia

We used a validated measure of probable dementia for NHATS data (25). During each annual wave, participants who met any of the following criteria were classified as having probable dementia: (a) a diagnosis of dementia made by a physician; (b) a score of 2 or higher on AD8 (26), that is, a validated proxy report assessment of dementia including eight questions for memory loss, judgment, orientation, hobbies/Interests, repetitive questions/statements, daily activities, changes in personality, and personal care; and (c) impairment (ie, ≤1.5 standard deviations below the mean) in at least 2 of the 3 cognitive domains, which were memory, orientation, and executive functioning (25). The measure has been widely applied to estimate the prevalence of dementia among U.S. older adults (27,28).

Social isolation

We used a measure of social isolation that was originated by Berkman and Syme (29) and validated for NHATS data by Pohl et al. (30). This measure includes 4 domains, including marriage or partnership (no = 0; yes = 1), family and friends (0–2 friends and 0–2 relatives = 0; all other scores = 1), church participation (<=every few months = 0; >once or twice a month = 1), and club participation (no = 0; yes = 1). The sum score ranges from 0 to 4. A score of 0 indicates severe isolation, a score of 1 indicates social isolation and a score between 2 and 4 indicates social connection.

Depression and anxiety

NHATS used slightly modified versions of well-established brief screening instruments the Patient Health Questionnaire-2 items (PHQ-2) and the Generalized Anxiety Disorder-2 items (GAD-2) to assess depressive symptoms and anxiety symptoms, respectively (24,31). The 4 questions are: “Over the last month, how often have you: (i) had little interest or pleasure in doing things; (ii) felt down, depressed, or hopeless; (iii) felt nervous, anxious, or on edge; (iv) been unable to stop or control worrying?” Response options are not at all, several days, more than half the days, nearly every day. Questions 1 and 2 form the PHQ-2, and Questions 3 and 4 form the GAD-2. Scores were based on summing scores for the corresponding questions (0 = not at all; 1 = several days; 2 = more than half the days; 3 = nearly every day). Both PHQ-2 and GAD-2 are scored from 0 to 6 with scores of 3 or higher, indicating the presence of clinically significant depressive symptoms or anxiety symptoms.

Health-related variables

Self-rated health was used to measure general health through “Would you say that in general your health is excellent, very good, good, fair, or poor?.” Lower scores represent better self-rated health, with 1 for excellent and 5 for poor. Participants were asked if they currently smoke or not. Participants were also asked if a number of chronic medical conditions had been diagnosed by a doctor. Among these conditions, we chose heart attack, heart disease, high blood pressure, diabetes, and stroke because of the evidence about the association between cardiovascular disease history and risk of dementia (32–34).

Demographics

We included age, gender, race and ethnicity (non-Hispanic White, non-Hispanic Black, all others were combined because of the small size including Hispanic, American Indian, Asian, Native Hawaiian, Pacific Islander, and all other), and the highest education attainment.

Statistical Analyses

First, group-based trajectory modeling was used to assess the trajectories of social isolation, depression, and anxiety from 2011 to 2015. For each outcome, only those participants who had valid values for no fewer than 2 years were used for the trajectory analyses. We explored and evaluated the trajectories by testing linear, quadratic, and cubic terms to determine the shape that best fits the data, by a comparison of AIC and BIC. We also explored both models of CNORM (a normal distribution model) and LOGIT (a generalized logit model). In the former case, the entire range of the total score was utilized, whereas in the latter case, a dichotomous value was employed, determined by the threshold for each outcome. The analyses were conducted using a statistical procedure (SAS proc TRAJ). Chi-square tests were used to determine whether the participant characteristics differed between the trajectory groups for each of the 3 outcomes.

Second, we estimated the association between the trajectory of social isolation from 2011 to 2015 estimated above and incident dementia from 2015 to 2022 using Cox proportional hazards regression models. The date of dementia onset was defined as the date a participant first met the criteria of dementia and the time to event was defined as the time between 2015 and when the participant was either classified as having dementia or censored from observation at the last available contact. The first model adjusted for demographics (age, gender, race/ethnicity, and education), the second model added the trajectories of depression and anxiety, and the third model further adjusted for self-rate health status, if currently smoke, and several variables related to cardiovascular diseases. Survey weights were applied to account for the complex sampling survey design. The analyses were conducted using a statistical procedure (SAS proc PHREG).

Results

As shown in Figure 1, 3 social isolation trajectories were identified, reflecting minimal, moderate, and high levels of social isolation. Among the whole sample (N = 2 977), the predicted group membership for the above 3 trajectories were 16.9%, 28.45%, and 54.71%. Similarly, for both depression and anxiety, 3 trajectories were identified, reflecting minimal, moderate, and high levels. The predicted group memberships for the 3 trajectories (from minimal to high) for depression were 75.9%, 19.7%, and 4.4%. For anxiety, the 3 percentages were 79.7%, 16.8%, and 3.6%.

Figure 1.

Trajectories of social isolation, depression, and anxiety from 2011 to 2015 among 2 977 older adults. For each trajectory, 95% confidence intervals were shown by gray shaded areas.

As shown in Table 1, some demographics and most health indicators significantly differed across trajectory groups. For both social isolation and depression, Non-Hispanic Blacks and participants with low education attainment were more likely to have moderate or high trajectories compared with minimal trajectories. Participants who had high social isolation and depression trajectory groups were more likely to have depression and anxiety, self-rated as fair or poor health, currently smoke, and have cardiovascular diseases or a history of having cardiovascular diseases. Because death is a competing risk for dementia, we further examined if trajectories of social isolation or depression were associated with mortality, either after or in the absence of dementia. Over the study period from 2015 to 2022, a total of 129 participants (4.3%) died. As shown in Table 1, the percentage of deaths did not show significant differences across the trajectory groups for either social isolation or depression.

Table 1.

Characteristics of the Participants by the Groups of Social Isolation Trajectory and Depression Trajectory

Characteristics	Social Isolation Trajectory Group				Depression Trajectory Group
	Minimal (n = 1 623)	Moderate (n = 844)	High (n = 502)	p Value	Minimal (n = 2 259)	Moderate (n = 587)	High (n = 130)	p Value
Mean age in 2015	78.6	79.9	80.4		79.2	79.5	79.9
Female (%)	57.6	59.5	56.8		54.8	68.3	67.7	**
Non-Hispanic Black (%)	15.8	23.2	25.9	**	17.5	26.2	26.9	**
All race/ethnicity except non-Hispanic White and Black	5.6	8.8	8.8	**	6.0	9.2	18.5
Education < high school	36.4	53.6	66.1	**	42	58.9	66.7	**
Education ≥ Bachelor	36.4	20.4	12.5		31.8	15.1	14.7
Has depression	7.1	13.6	17.7	**	3.1	26.6	74.2	**
Has anxiety	5.3	10.3	14.6	**	3.1	19.8	46.5	**
Self-rated fair or poor health	16.0	27.6	38.4	**	15.2	43.3	68.5	**
Currently smoke	3.3	8.4	13.6	**	5.7	8.2	11.5	*
Had heart attack	1.3	2.1	2.6	*	1.6	1.7	3.9
Has heart disease	21.9	28.9	26.6		22.1	31.3	38.8	**
Has high blood pressure	70.9	77.9	79.6	**	71.9	81.6	86.9	**
Has diabetes	26.1	29.0	36.3	**	25.8	36.3	42.3	**
Had stroke	2.2	2.7	2.2		1.9	2.2	9.2	**
Death before 2022, either after or in the absence of dementia	4.7	3.7	3.8		4.5	4.1	3.1

Notes: p Values are for the chi-square test to determine if the characteristics of the participants are different by the trajectory group.

** p < .001.

* p < .05.

During a mean follow-up of 3.6 years, 567 (19.0% of the total sample) were diagnosed with dementia. Table 2 shows the results of 3 Cox proportional hazards regression models. According to Model 1 that adjusted basic demographics, the trajectory of social isolation was significantly associated with the risk of dementia, with individuals in the groups of high level of social isolation were 41% (Hazard ratio = 1.41, 95% CI [1.08, 1.83]) less likely to develop dementia compared with those who had minimal level of social isolation. When the trajectories of depression and anxiety were added in Model 2 and several health indicators were further added in Model 3, the association between social isolation and incident dementia was not significant anymore.

Table 2.

The Associations Between Trajectory of Social Isolation and Incident Dementia: Results of 3 Cox Proportional Hazards Regression Models

Variable	Model 1	Model 2	Model 3
	HR (95% CI)	HR (95% CI)	HR (95% CI)
Age (each year)	1.12 (1.11, 1.13)**	1.12 (1.11, 1.13)**	1.13 (1.11, 1.14)**
Gender
Male (ref)	1	1	1
Female	0.94 (0.75, 1.17)	1.00 (0.79, 1.24)	1.00 (0.78, 1.26)
Race/ethnicity
Non-Hispanic White (ref)	1	1	1
Non-Hispanic Black	1.69 (1.36, 2.11)**	1.61 (1.27,2.03)**	1.36 (1.09, 1.71)**
All race/ethnicity except non-Hispanic White and Black	1.29 (0.92,1.80)	1.20 (0.85, 1.68)	0.98 (0.68, 1.42)
Education
<High school (ref)	1	1	1
High school and some college	0.75 (0.59, 0.96)*	0.78 (0.61, 1.00)	0.79 (0.61, 1.03)
≥Bachelor	0.60 (0.47, 0.77)**	0.62 (0.49, 0.80)**	0.70 (0.55, 0.90)**
Trajectory of social isolation
High	1.41 (1.08, 1.83)*	1.26 (0.98, 1.62)	1.10 (0.88, 1.37)
Moderate	1.28 (1.02, 1.60)*	1.18 (0.95, 1.47)	1.10 (0.85, 1.43)
Minimal (ref)	1	1	1
Trajectory of depression
High		2.38 (1.56, 3.61)**	1.46 (1.12, 1.90)**
Moderate		1.60 (1.24, 2.06)**	1.82 (1.12, 2.97)*
Minimal (ref)		1	1
Trajectory of anxiety
High		1.24 (0.76, 2.02)	0.89 (0.66, 1.19)
Moderate		1.0 (0.74, 1.35)	1.18 (0.71, 1.97)
Minimal (ref)		1	1
Self-rated health†			1.29 (1.15, 1.45)**
If current smoke			1.80 (1.25, 2.58)**
With a history of heart attack			1.59 (0.93, 2.72)
If has heart disease			1.21 (0.97, 1.51)
If has high blood pressure			0.93 (0.73, 1.20)
If has diabetes			1.24 (1.00, 1.53)
With a history of stroke			2.02 (1.33, 3.07)**

Notes: HR = hazards ratio. The first model adjusted for demographics (age, gender, race/ethnicity, and education), the second model added the trajectory groups of depression and anxiety, and the third model further adjusted for self-rate health status, if currently smoke, and several cardiovascular diseases. Values in bold indicate statistical significance (** p < .001; * p < .05).

^†Self-rated health is with 5 levels from 1 for excellent to 5 for poor, with lower scores represent better self-rated health.

In Model 2, the trajectory of depression was significantly associated with the risk of dementia, with the group of high depression was 138% (hazard ratio = 2.38, 95% CI [1.56, 3.61]) more likely to develop dementia compared with the group had a minimal level of depression. In Model 3, the above association to incident dementia was still significant, although the difference decreased. The trajectory of anxiety was not significantly associated with the risk of dementia in both Models 2 and 3.

In Model 3, within the 5 levels assessed, each level worse of self-rated health was associated with a 29% higher risk of dementia. Additionally, individuals who currently smoke, and those who had a history of stroke had 83% and 106% higher risks of dementia, respectively, compared to their counterparts who did not have these conditions. In all 3 models, 1 year older was associated with 12%–13% higher risk of dementia. Non-Hispanic Blacks and those with lower education were associated with higher risk of dementia compared with non-Hispanic Whites and those with higher education. The risk difference between non-Hispanic Blacks and non-Hispanic Whites, and between different educational levels decreased from Model 1 and Model 3, that is, when health indicators were added to the model.

Discussion

Our findings suggest that the association between social isolation and incident dementia may be explained by factors such as depression and other health indicators. Unfortunately, many previous studies didn’t account for potential cofounders that may mediate the above association. Particularly, 1 recent study (35), also using NHATS longitudinal data, found that being socially isolated was associated with a 1.28 higher risk of dementia over 9 years compared with those who were not socially isolated. However, this study did not adjust mental health disorders. On the contrary, several recent studies that adjusted for mental health disorders such as depression, lifestyle behaviors such as smoking and alcohol use, and chronic diseases such as heart disease, still reported a higher risk of dementia among socially isolated individuals compared to those who were not isolated (36–40). To elucidate the impact of social isolation on dementia, rigorous studies are needed from at least the 2 perspectives below.

First, our knowledge of the causal mechanisms from social isolation to dementia remains incomplete. A number of plausible pathways have been proposed and discussed in prior literature (5–9). For example, social isolation may impact dementia risk through factors such as social support, lifestyle behaviors, mental health disorders, and cardiovascular diseases, as outlined in the introduction. Neurobiological mechanisms are of particular interest, as they can directly explain how and why social connections may “stimulate” cognitive function. Two neurobiological theories stand out: the “use it or lose it” theory suggests that insufficient engagement in social or cognitive activities leads to brain atrophy, while the cognitive reserve theory posits that social connections may increase neurogenesis and synaptic density in the brain (41,42). Consequently, isolation and lack of cognitive stimulation may lead to the atrophy of hippocampus (a crucial structure involved in a range of cognitive functions) or diminish neural reserve, ultimately impairing social cognition. We need more fundamental research to understand the neurobiological mechanisms. Furthermore, bidirectional relations exist between social isolation and dementia. It may be challenging for people with dementia to engage in social interactions and maintain relationships, due to loss of independence, communication difficulties, leading to social withdrawal and isolation. Also, the social stigma of dementia may discourage people with dementia from participating in social activities and gatherings. The increased availability of longitudinal cohort data and the application of advanced methods such as structural equations model and survival analysis may help to address the inverse causality. An understanding of the causal mechanism is a prerequisite to identify potential confounders for the association between social isolation and dementia.

Second, despite various measures of social isolation that have been proposed and applied, there is no established standard on how to measure social isolation. The consensus is that the construct of social isolation is multifaceted, and a measure of social isolation should comprise multiple domains to capture an individual’s relationships with other individuals, groups, and community organizations (43). However, some domains may be overlooked or only partially addressed by most existing social isolation measures, for example, social connectedness in the neighborhood and through digital communications. Neighborhood provides opportunities for a resident to socially interact with neighbors and other persons. The geographical closeness of individuals within a neighborhood can facilitate face-to-face interactions, social support, and group activities. Neighborhood social cohesion, that is, feeling of belonging to own neighborhood and the level of connectedness, trust, and mutual support among residents, was missed from most current social isolation measures (43). Neighborhood social cohesion may be more important for older adults because older adults tend to have decreased mobility and may spend more time in their neighborhood (44,45). Another missed domain is digital communication. Digital communication can not only enable regular contact with friends and family, especially when physical distance is a barrier, but also extend and enhance relationships, provide emotional support, offer information and resources. Evidence shows that digital communication can reduce social isolation and loneliness in older people (46–48), and a growing of interventions used digital communication as an approach to promote social interactions for older adults (49,50). However, most current social isolation measures do not explicitly include digital communication or internet use.

Highlighting the disproportionate risk burden of incident dementia among older adults, our findings indicate the importance of multifaceted approaches to dementia prevention, with considerations of racial and ethnic disparities, educational opportunities, health behaviors, and broader socioeconomic factors. Most previous studies showed that non-Hispanic Blacks and Hispanics have higher risks of dementia than non-Hispanic Whites (51). Our results confirmed the higher risk of dementia among non-Hispanic Blacks but not among Hispanic. This is because we combined Hispanics with all other race/ethnicities groups due to the small size. Consistent with the prior literature, our results indicated that low educational attainment (52), smoking (53), and a history of stroke (54) were associated with higher risks of incident dementia.

Leveraging on a nationally representative sample of older adults, we used survival analysis to examine the longitudinal association between social isolation and incident dementia. One strength of this study was that we split the 12 follow-up years into 2 periods and used the trajectories of social isolation during the first 5 years to predict the risk of dementia during the following 7 years. Compared to using a single value at 1 point, the approach may better encapsulate the accumulative impact of social isolation and facilitate the establishment of the temporal causality. This approach was used to study the association between trajectories of depressive symptoms during the first 5 years and the risk of dementia afterward among older adults (55). Additionally, the trajectories of depression and anxiety were assessed using the full score rather than the dichotomous threshold for each outcome, this may better capture their graduate change over time.

However, while our approach is novel in capturing the cumulative impact of social isolation during the first 5 years, it did not account for the impact of social isolation after 2015, and the 5-year window was chosen arbitrarily. The second limitation of our study was the use of simple although established measures of dementia, social isolation, depression, and anxiety. More comprehensive measures may provide higher levels of reliability and validity. Furthermore, the mean age of our participants was 79.3 years old in 2015. Therefore, our findings may not be applicable to relatively younger older adults.

Further research with improved methods and measures is needed to deepen our knowledge of social isolation and its impact on dementia. Minorities and people with low social–economic status, chronic diseases, and mental health disorders are disproportionately exposed to social isolation and with higher risk of dementia. Tailored intervention programs are crucial to mitigate social isolation and ameliorate the disparities of social isolation and dementia.

Funding

None.

Conflict of Interest

None.

Data Availability

This study was based on publicly available data from the National Health and Aging Trends Study data set (data available at: https://www.nhats.org/researcher/nhats). The data analyzed for this study are available from the corresponding author on reasonable request. This study was not preregistered.