Together, but Isolated: A Dyadic Study of Social Isolation in older Adults and Caregivers

Abstract

Objectives:

Social isolation has negative health implications for older adults and caregivers. We examine the interdependency of social isolation among community-dwelling Medicare beneficiaries and their primary caregivers.

Methods:

Using data from the National Study of Caregiving and the National Health and Aging Trends Study (2015–2017, N = 522 dyads), we conducted actor-partner interdependence models to examine the dyadic associations between baseline social isolation and social isolation at 2-year follow-up among older adults and caregivers. We also tested for effect modification by dementia status and relationship type.

Results:

There was a low but significant correlation of baseline social isolation levels within dyads. Older adults’ baseline social isolation was positively associated with caregivers’ social isolation at follow-up but not vice versa. No evidence of effect modification was found.

Conclusion:

Reducing social isolation among older adults may benefit their caregivers. Future investigation of relationship quality and shared social activities of caregiving dyads is warranted.

Introduction

About one in four community-dwelling older adults experience social isolation in the United States (Cudjoe et al., 2020). Social isolation, defined as an objective lack of or limited social contact, carries myriad adverse implications for health and well-being among older adults (Cudjoe et al., 2020; Gimm et al., 2023; National Academies of Sciences, Engineering, and Medicine, 2020; Pomeroy et al., 2023). Moreover, social isolation can be detrimental, increasing the overall risk for dementia (Huang et al., 2023) and resulting in a heightened risk of hospitalization among persons living with dementia (Pomeroy et al., 2024). Various factors may increase an individual’s risk for social isolation, such as living alone, physical function, or sensory impairment (Huang et al., 2024; Mehrabi et al., 2024; National Academies of Sciences, Engineering, and Medicine, 2020). Yet, while social isolation in late life can be chronic, it can also be dynamic. For example, a longitudinal study from 2011 to 2022 among 7,000 community-dwelling Medicare beneficiaries found that 17% experienced persistent social isolation whereas 27% experienced intermittent social isolation (Umoh et al., 2024). Cognitive decline and progression of dementia, as well as related stigma, may further exacerbate levels of social isolation among older adults (Dyer et al., 2021; Herrmann et al., 2018; Huang et al., 2024).

Increasingly, there are also concerns about social isolation among the 24 million family caregivers of American older adults (Hajek et al., 2021; Liang et al., 2023; Qian et al., 2025; Wolff et al., 2025). Caregiving responsibilities can lead to time constraints on caregivers’ ability and opportunities to maintain social connections (Beeson, 2003; Gray et al., 2020), such as reduced participation in social activities that they value (Patterson et al., 2023), reduced quality time with families, and reduced hours at work (or even early retirement) that reduces time spent interacting with coworkers (Van Houtven et al., 2013). Caregivers may also experience emotional barriers to prioritizing their social and emotional needs. Qualitative evidence suggests that caregivers’ feelings of guilt for participating in social activities without their care recipients can limit social interactions (Kotwal et al., 2024). Caregivers may withdraw from their social networks due to others’ lack of understanding of their challenges (Vasileiou et al., 2017), concerns about burdening others (Lee et al., 2022), and stigma related to their care recipients’ conditions in public (Kotwal et al., 2024), such as behavioral issues of dementia.

Social isolation among caregivers and the older adults they support (i.e., care recipients) can be a shared challenge. For example, there may be a disruption in shared social activities due to changes in cognitive and behavioral functioning that occur along the dementia trajectory (Kotwal et al., 2024). According to the Interdependency Theory (Rusbult & Van Lange, 2003), the behaviors and outcomes of one individual in a close relationship, such as a caregiving relationship (Lyons et al., 2002), can impact the behaviors and outcomes of the other individual, known as the partner, in the dyad (Rusbult & Van Lange, 2003). Caregiving relationships are built upon existing close and long-standing relationships, such as marriage. The caregiving relationships itself evolves through new types of interactions, e.g., providing/receiving personal care, and can result in changing roles, i.e., role reversal for adult children becoming the caregivers of their aging parents (Eifert et al., 2015; Lyons et al., 2002). In the caregiving context, the dyadic relationship (Wilson et al., 2024) typically refers to the care recipient and their primary caregiver, who is either the only caregiver or the one who bears the greatest amount of caregiving work. The care recipient’s behaviors (e.g., attending religious services) and outcomes (e.g., depression) may influence the behaviors and outcomes of their primary caregiver, and the primary caregiver’s behaviors and outcomes may also influence the care recipient as well. Considering the close relationships and significant amount of time spent together between community-dwelling older adults and their primary caregivers, it is possible that one partner’s social isolation may impact the other partner’s social isolation. For instance, primary caregivers for older adults with smaller social support networks (i.e., more isolated) bear higher caregiving responsibilities, such as greater caregiving hours, leading to greater restrictions on their social participation (Patterson et al., 2023). Caregivers’ withdrawals from community activities and family gatherings due to perceptions of dementia stigma can reduce social participation opportunities for their care recipients (Kotwal et al., 2024). Caregiving dyads are also exposed to shared risk factors of social isolation, such as cultural stigma of illnesses (Liu et al., 2008) and rural residence (Freedman & Nicolle, 2020). Moreover, as illness trajectories progress and caregiving responsibilities intensify, the relational dynamics within dyads often shift (Lyons et al., 2002) and the flow of dependency within dyads becomes less reciprocal (Kotwal et al., 2024). Prior dyadic evidence suggests that care recipients’ perceptions of mutuality decline faster than their caregivers (Lyons et al., 2007). As a result, caregivers’ experiences of social isolation are likely to be more influenced by the social isolation of the care recipients than the other way around. Yet quantitative evidence about social isolation in caregiving remains limited (Hajek et al., 2021), and empirical research on the interdependency of social isolation within caregiving dyads is even scarcer.

Importantly, the levels of interdependence in social isolation within the caregiving dyad may vary across relationship types. Dyadic evidence on mental health outcomes, such as depression (Litzelman & Al Nassar, 2022), has shown differing partner effects across relationship types of the caregiving dyads. Despite marriage being a consistent protective factor for social isolation among older adults (National Academies of Sciences, Engineering, and Medicine, 2020), spousal caregivers tend to experience higher caregiving strain and psychosocial impacts of caregiving compared to non-spousal caregivers (Kotwal et al., 2024; Ornstein et al., 2017; Schulz & Sherwood, 2008), including social isolation (Li et al., 2021; Qian et al., 2025). Dyadic evidence further suggests that the gender composition of the spousal dyads may moderate the association between caregivers’ social isolation and care recipients’ quality of life (Rippon et al., 2023). On the other hand, non-spousal caregivers (e.g., adult children, other relatives, and friends) represent not only the majority of caregivers of older adults in the United States (Wolff et al., 2025), but also reflect the variety of caregiver relationship types in later life (Freedman et al., 2024). Comparing the differences in social isolation between spousal versus non-spousal caregiving relationships can inform future interventions and policies that support family caregiving in various social contexts.

Moreover, older adults living with dementia and their caregivers face unique changes related to social isolation (Lee et al., 2022). Dementia caregivers bear a higher burden of care, such as greater number of hours spent on caregiving, which can lead to greater restrictions in social participation (Patterson et al., 2023). The progression of dementia further creates challenges for caregivers to identify strategies that adapt to patients’ cognitive and functional decline while meeting the patients’ and their own social needs (Kotwal et al., 2024; Lee et al., 2022). Despite evidence showing greater risk of social isolation among dementia caregivers in comparison with non-dementia caregivers (Qian et al., 2025), most empirical evidence within caregiving dyads is cross-sectional and limited to specific health conditions, such as dementia (Rippon et al., 2023) or cancer (Kelley et al., 2019; Segrin et al., 2019). A critical gap in the literature is understanding how experiences of social isolation are interdependent among older adults with and without dementia and their caregivers over time. Examining diverse caregiving relationships and considering broad samples of older adults receiving care in the community may generate a more comprehensive understanding of shared social challenges and strategies to enhance social connection, particularly for dementia caregiving dyads.

To capture the effects that occur between two closely related individuals, such as an older adult and their primary caregiver, it is essential to consider both partners in the dyadic relationship, in other words, including both caregivers and care recipients in empirical examinations (Wilson et al., 2024). In this investigation, we followed the analytic framework of the Actor–Partner Interdependence Model (APIM) (Kenny et al., 2020) to examine the dyadic associations of social isolation levels over a 2-year period among dyads of community-dwelling older adults and their primary caregivers. The APIM, which has been widely used in gerontology (Wilson et al., 2024), takes into account the interdependence process by estimating the effect of a person’s own variable (actor effect) and the effect of the same variable from the partner’s standpoint (partner effect) on an outcome variable.

In this study, we used data from the National Study of Caregiving linked with the National Health and Aging Trends Study (2015–2017) to (1) examine the dyadic associations in social isolation levels between community-dwelling older adults (care recipients) and their primary caregivers from 2015 (baseline) to 2017 (2-year follow-up) and (2) explore heterogeneity in partner effects across dementia status and relationship type of dyads. Informed by the Interdependency Theory (Rusbult & Van Lange, 2003) and prior evidence in dyadic health and social isolation, our hypotheses are:

Hypothesis 1.

Older adults’ (care recipients’) social isolation levels at follow-up will be positively associated with their own social isolation levels at baseline (care recipient actor effects, Hypothesis 1a); Caregivers’ social isolation levels at follow-up will be positively associated with their own social isolation levels at baseline (caregiver actor effects, Hypothesis 1b).

Hypothesis 2.

Older adults’ (care recipients’) social isolation levels at follow-up will be positively associated with their caregivers’ social isolation at baseline (caregiver-to-care recipient partner effects, Hypothesis 2a); Caregivers’ social isolation levels at follow-up would be positively associated with their care recipients’ social isolation at baseline (care recipient-to-caregiver partner effects, Hypothesis 2b); The care recipient-to-caregiver partner effects would be stronger than the caregiver-to-care recipient partner effects (Hypothesis 2c).

Hypothesis 3.

The partner effects will be stronger for spousal caregiving dyads versus non-spousal caregiving dyads.

Hypothesis 4.

The partner effects will be stronger for dementia caregiving versus non-dementia caregiving.

Methods

Data Source

This study used data from the 2015 and 2017 National Health and Aging Trends Study (NHATS) and the National Study of Caregiving (NSOC). NHATS is a nationally representative, longitudinal survey of health-related trends among Medicare beneficiaries aged 65 and older. The initial cohort was first interviewed in 2011 and sample replenishment was undertaken in 2015 and 2022/2023. The response rates in 2015 (NHATS Round 5) and 2017 (NHATS Round 7) are 73.6% and 91.9%. (Freedman et al., 2023)

NSOC is a nationally representative survey of family and other unpaid caregivers to NHATS participants who received assistance with self-care, mobility, or household activities. During the NHATS interview, up to five caregivers are identified by older adults from whom they received assistance with self-care, mobility, or household activities in the last month. Their family and other unpaid caregivers are then contacted and interviewed separately in NSOC. NSOC interviews were conducted in conjunction with NHATS in 2011, 2015, 2017, 2021, 2022, and 2023. Importantly, most NSOC rounds are cross-sectional samples, except that in 2017, all caregivers who participated in NSOC in 2015 were eligible to participate in a follow-up interview. The response rate in 2015 was 63.0%, and 74.9% in the 2017 follow-up. (Freedman et al., 2019).

Analytic Sample

Our analytic sample included caregiving dyads who completed both NHATS/NSOC in 2015 and 2017. A caregiving dyad consisted of a care recipient in NHATS and their primary caregiver in NSOC. If the older adult had only one caregiver in NSOC (n = 378) they were considered the primary caregiver. If an older adult had multiple caregivers in NSOC (n = 144) the caregiver who reported providing the greatest hours of care was considered the primary caregiver, as defined in other NSOC analyses (Litzelman & Al Nassar, 2022).

Eligible caregiving dyads (n = 930) were excluded if care recipients were (a) living in nursing homes or other types of residential facilities in 2015 or 2017 (n = 280) or (b) living in a private residence but used proxies for the interviews in 2015 or 2017 (n = 128). The final analytic sample included n = 522 caregiving dyads, including n = 129 dyads in which the care recipients were classified as having “probable dementia” or “possible dementia” in 2015 (Kasper et al., 2013).

Measures

Social Isolation (Baseline and Follow-up).

Social isolation of caregivers and care recipients was measured using a multi-domain social isolation typology (Cudjoe et al., 2020) informed by the Berkman-Syme Social Network Index (Berkman & Syme, 1979) and the Lubben Social Network Scale (Lubben, 1988). The typology considers four key domains of social isolation, including living arrangement, core discussion network size, religious service attendance, and social participation (Cudjoe et al., 2020), and has been widely used in NHATS-based studies (Huang et al., 2023; Pomeroy et al., 2023; Umoh et al., 2024) and been adapted in NSOC-based research among caregivers (Qian et al., 2025).

We adapt the 4-domain social isolation typology (Cudjoe et al., 2020) by reverse coding individual items such that greater scores indicated higher levels of social isolation for ease of interpretation. Specifically, care recipients (i.e., NHATS participants) receive one point for responding to each of the following indicators of social isolation: (1) living alone; (2) talking to fewer than two persons about important matters in the past year (i.e., having a core discussion network size smaller than two persons); (3) not attending religious services in the past month; and (4) not participating in other activities (clubs, meetings, or group activities or doing volunteer work) in the past month (Cudjoe et al., 2020). These four items are summed to create a composite score ranging from 0 (least isolated) to 4 (most isolated).

For caregivers (i.e., NSOC participants), we adopted the four-domain social isolation typology into NSOC with the threshold of core discussion network size as smaller than one person (Qian et al., 2025). This adjustment was necessary due to differences in question wording (“Do you have friends or family that you talk to about important things in your life”) and the unavailability of a measure capturing the number of individuals in the core discussion network in the NSOC survey instrument.

Between-Dyad Covariates: Baseline Caregiving Characteristics.

We considered key caregiving characteristics as between-dyad covariates in the analysis, including relationship type, co-residence status, hours and duration of care, care tasks, number of helpers listed by care recipients (range: 1–5), and dementia classification. Relationship type of the caregiving dyad included spouse/partner or non-spouse (adult children or other relatives such as nieces/nephews and grandchildren or non-relatives such as friends). Caregiving intensity was captured by hours of care per week and duration of care (less than 1 year, 1–5 years, and above 5 years). For assistance or care provided, we calculated the number of types of assistance that a caregiver provided (range: 1–9) (Litzelman & Al Nassar, 2022), capturing the complexity of the nature of caregiving (household chores, shopping, banking, personal care, getting around the home, transportation, medical/nursing care, health behavior assistance, and medical logistics). Dementia status (no dementia vs. possible or probable dementia) was measured using a validated algorithm through the following steps: (1) asking the participant or their proxy whether they have been diagnosed with dementia or Alzheimer’s disease; (2) using a score from the AD8 Dementia Screening Interview, an 8-item instrument designed to assess memory, temporal orientation, judgment, and function; and (3) cognitive testing including testing memory, orientation, and executive function (Kasper et al., 2013). Because of our interest in the unique caregiving experience in the context of cognitive impairment, we broadly captured dementia by combining possible dementia and probable dementia in one group as suggested by the NHATS research team (Kasper et al., 2013).

Within-Dyad Covariates: Baseline Socio-Demographics and Health.

For within-dyad covariates, we considered sociodemographic and health covariates of both the caregivers and care recipients, including a continuous measure of age (in years), gender (female/male), race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and non-Hispanic other), marital status (married/coupled vs. not married/coupled), education (below high school vs. high school and above), employment (working for pay vs. not working for pay), imputed household income (Hu & Freedman, 2024a, 2024b), self-rated health, depressive symptoms (PHQ-2) (Kroenke et al., 2003), and number of health conditions. For health conditions, we calculated the total number of conditions measured in NHATS (excluding dementia) and NSOC. Care recipients’ health conditions (range: 0–8) included a self-reported physician diagnosis of diabetes, hypertension, heart disease, arthritis, osteoporosis, lung diseases like emphysema or asthma, cancer, and stroke. Caregivers’ health conditions (range: 0–10) included heart attack, other heart disease (e.g., congestive heart failure), hypertension, arthritis, osteoporosis, diabetes, lung disease (e.g., asthma), cancer, vision difficulty, and hearing difficulty.

Analytic Approach.

Descriptive statistics of care recipients, caregivers, and dyads were calculated for all variables (percentages or mean/standard deviations/median/IQR). Guided by the actor-partner interdependency model (APIM) framework, we conducted linear dyadic multilevel models with random intercepts (Kenny et al., 2020). APIM is commonly used in dyadic analysis (Wilson et al., 2024), including caregiving dyads (Kelley et al., 2019; Litzelman & Al Nassar, 2022), to examine the effect of a predictor reported by both individuals within a dyad on their own outcome (actor effect) and that of the other individual (partner effects). Specifically, individuals (care recipients and primary caregivers) were nested within the caregiving dyad. We examined the association between care recipient and caregiver social isolation at baseline on both self (actor effects) and partner’s social isolation (partner effects) at follow-up. The hypothesized actor and partner effects are illustrated in Figure 1. To explore effect modification, we conducted stratified analyses by dementia and relationship type and then tested interaction terms for dementia and relationship type. All models controlled for between- and within-dyad covariates. Data management and descriptive statistics were conducted in RStudio 2024.09.0 (Packages: “haven,” “tidyverse”). APIM was conducted in SAS 9.4 using PROC MIXED (Kenny et al., 2020). Missing data was handled using listwise deletion (Kenny et al., 2020), and n = 132 dyads were deleted due to missing values for some variables. Numbers of missing values for each variable are included in Supplemental Table 1.

Figure 1.

Hypothesized Actor and Partner Effects on Care Recipient and Caregiver Social Isolation. Notes: Actor Effects are Represented by Dash Lines and Partner Effects are Represented by Solid Single-Headed Arrows. Double-Headed Arrows Indicate Correlated Variables. CR = Care Recipient; CG = Caregiver

Results

Table 1 presents the baseline characteristics of older adults (care recipients) and their primary caregivers. Care recipients’ mean age was 79.6 (SD = 7.6), and caregivers’ mean age was 62.3 (SD = 14.1). The majority of both care recipients (65%) and caregivers (67%) were female. On a scale ranging from 0 (lower social isolation level) to 4 (higher social isolation level), the mean baseline social isolation level for care recipients (mean = 1.66, SD = 1.02) was higher than caregivers (mean = 1.20, SD = 0.96, p < 0.001). Additionally, the Pearson correlation between the baseline social isolation scores of the two members of a caregiving dyad was low but positive (r = 0.27, p < 0.001). About 25% of care recipients were classified as having dementia (Table 2). Most caregivers were adult children (43.7%) or spouses/partners (39.3%) of their care recipients. More than half of the sample provided care for over 5 years at baseline. The detailed caregiving characteristics are listed in Table 2.

Table 1.

Baseline characteristics of community-dwelling older adults (care recipients) and their primary caregivers (N = 522)

	Care recipients (N = 552)	Caregivers (N = 552)
	%/Mean (SD)/Median (IQR)	%/Mean (SD)/Median (IQR)
Age in years, mean (SD)	79.60 (7.65)	62.70 (14.10)
Gender
Female	64.56%	67.43%
Male	35.44%	32.38%
Race and ethnicity
Non-Hispanic White	64.56%	67.43%
Non-Hispanic Black	35.44%	32.38%
Others	2.30%	6.13%
Hispanic	61.49%	59.20%
Marital status
Married/Coupled	49.81%	64.18%
Unmarried/Uncoupled	50.19%	35.25%
Education
High school and above	67.24%	82.95%
Below high school	32.38%	15.71%
Employment
Not employed	95.21%	68.58%
Employed	4.79%	29.89%
Income, median (IQR)	25,000 (13,979–45,000)	36,178 (20,000–71,768)
Self-rated health
Poor/Fair	50.57%	20.69%
Good/Very good/Excellent	49.23%	78.74%
Depressive symptoms	1.39 (1.44)	1.06 (1.33)
Number of health conditions	3.66 (1.62)	2.02 (1.67)
Social isolation level	1.66 (1.02)	1.20 (0.96)

Notes: Data came from the National Health and Aging Trends Study Round 5 linked to National Study of Caregiving Round 5. The “other” race category encompassed American Indian, Alaska Native, Asian, Native Hawaiian, Pacific Islander, and other races. Depressive symptoms was measured as PHQ-2. To calculate the number of health conditions, we consider all conditions measured in NHATS (excluding dementia) and NSOC. Care recipients’ health conditions (range: 0–8) included a self-reported physician diagnosis of diabetes, hypertension, heart disease, arthritis, osteoporosis, lung diseases like emphysema or asthma, cancer, and stroke. Caregivers’ health conditions (range: 0–10) included heart attack, other heart disease (e.g., congestive heart failure), hypertension, arthritis, osteoporosis, diabetes, lung disease (e.g., asthma), cancer, vision difficulty, and hearing difficulty. The social isolation level (range: 0 to 4) considers four domains, including living arrangement, core discussion network size, religious attendance, and social participation, with a lower score denoting a lower social isolation level.

Table 2.

Baseline caregiving characteristics of dyads of community-dwelling older adults (care recipients) and their primary caregivers (N = 522)

	%/Mean (SD)/Median (IQR)
Caregiver’s relationship to care recipient
Partner/Spouse	39.27%
Non spouse/partner	60.73%
Caregiver-care recipient co-residency
Co-resided	62.45%
Not co-resided	37.55%
Hours of care per week, median (IQR)	11 (5–28)
Duration of care
Less than 1 year	8.62%
1–5 years	37.93%
Above 5 years	52.30%
Assistance provided
Total count, mean (SD)	5.91 (2.29)
Household chores	83.52%
Shopping	86.78%
Banking	66.09%
Personal care	37.93%
Getting around	52.68%
Transportation	76.63%
Medical/Nursing care	52.68%
Health behavior assistance	63.03%
Medical logistics	71.65%
Number of caregivers, mean (SD)	2.41 (1.27)
Dementia classification of care recipient
Probable/Possible dementia	24.71%
No dementia	75.10%

Notes: Data came from the National Health and Aging Trends Study Round 5 linked to National Study of Caregiving Round 5. The “non-spouse/partner” relationship type includes adult children such as daughter/son or daughter/son-in-law, other relatives such as nieces/nephews and grandchildren, as well as non-relatives such as friends. The number of caregivers ranges from 1 to 5 given the study design of NSOC.

Changes in Social Isolation from Baseline to Follow-Up Among Care Recipients and Caregivers

The social isolation levels of about half of care recipients (52%) and caregivers (48%) remained the same from baseline to follow-up. About 23% of care recipients and 19% of caregivers had higher social isolation levels at follow-up. About 24% of care recipients and 18% of caregivers had lower social isolation levels at follow-up. Examining the item-level components that comprise the social isolation typology, we observed both positive and negative transitions of isolation levels among care recipients and caregivers over the 2-year period. For example, about 32% of care recipients with social participation at baseline (n = 165) reported no participation at follow-up, while 13% of those without participation at baseline (n = 357) reported social participation at follow-up. Among caregivers, about 26% of caregivers with social participation at baseline (n = 230) reported no participation at follow-up, while 21% of those without participation at baseline (n = 289) reported social participation at follow-up. Detailed changes in each of the four domains of social isolation are presented as alluvial plots in Supplemental Figure 1.

Actor Effects of Baseline Social Isolation on Social Isolation at Follow-Up (Hypotheses 1a-1b)

Table 3 presents the unadjusted and adjusted actor and partner effects of social isolation at baseline on social isolation 2 years later (follow-up). Full results from the adjusted model are available in Supplemental Table 2. As hypothesized, we found actor effects for both care recipients (Hypothesis 1a: care recipient actor effects) and caregivers (Hypothesis 1b: caregiver actor effects). For care recipients, their social isolation at follow-up was positively associated with their baseline social isolation after adjusting for their baseline sociodemographic, health, and caregiving characteristics (b = 0.61, SE = 0.04, p < 0.0001). Similarly, caregivers’ social isolation at follow-up was positively associated with their baseline social isolation (b = 0.59, SE = 0.04, p < 0.0001).

Table 3.

Actor and partner effects of baseline social isolation levels on social isolation levels over 2 years

	Model 1a: Unadjusted model (N = 442)			Model 1b: Adjusted model (N = 391)
	Beta	SE	p-value	Beta	SE	p-value
Actor effects
Care recipient	0.6881	0.0363	<0.0001	0.6060	0.0422	<0.0001
Caregiver	0.6210	0.0378	<0.0001	0.5947	0.0419	<0.0001
Partner effects
Caregiver-to-care recipient	0.0300	0.0388	0.4392	0.0586	0.0426	0.1699
Care recipient to-caregiver	0.0856	0.0354	0.0160	0.1040	0.0406	0.0107

Notes: Model 1b adjusted for baseline covariates, including care recipients’ and caregivers’ age, gender, race/ethnicity, marital status, education, employment, income, self-rated health, depression mood, number of health conditions excluding dementia, dyad’s relationship type and co-residency, hours of care, duration of care, number of care tasks, dementia classification, and number of caregivers.

Partner Effects of Baseline Social Isolation on Social Isolation at Follow-Up (Hypotheses 2a-2c)

We found a partner effect of care recipients’ baseline social isolation on caregivers’ social isolation at follow-up (Hypothesis 2b: care recipient-to-caregiver partner effect), but no significant partner effects were observed on care recipients’ social isolation at follow-up (Hypothesis 2a: caregiver-to-care recipient partner effect). After adjusting for caregivers’ baseline social isolation and other covariates, caregivers’ social isolation at follow-up was positively associated with care recipients’ baseline social isolation (b = 0.10, SE = 0.04, p < 0.05). Caregiver-to-care recipient partner effects were not significantly different from the care recipient-to-caregiver partner effects (Hypothesis 2c).

Analyses Examining Effect Modifications of Caregiver-Care Recipient Relationship Type (Hypothesis 3)

We found positive actor effects within care recipients and within caregivers across relationship types. There was no moderation effect of relationship type on actor effects (p = 0.7030). We found a significant care recipient-to-caregiver partner effect among non-spousal caregiving dyads (b = 0.11, SE = 0.05, p < 0.05). There was no statistically significant care recipient-to-caregiver partner effect among spousal caregiving dyads (b = 0.10, SE = 0.07, p = 0.1338). No significant moderation effect of relationship type on partner effects was found (p = 0.6534) (Table 4).

Table 4.

Actor and partner effects of baseline social isolation levels on social isolation levels 2 years later, stratified by caregiver-care recipient relationship type

	Model 2a: Spouse/Partner (n = 173)			Model 2b: Non-spouse (n = 220)			Model 2c: Interaction effect
	Beta	SE	p-value	Beta	SE	p-value	p-value
Actor effects							0.7030
Care recipient	0.5832	0.0662	<0.0001	0.6215	0.0571	<0.0001
Caregiver	0.6101	0.0606	<0.0001	0.5628	0.0605	<0.0001
Partner effects							0.6534
Caregiver-to-care recipient	0.0784	0.0636	0.2195	0.0271	0.0607	0.6560
Care recipient to-caregiver	0.0985	0.0654	0.1338	0.1109	0.0533	0.0386

Notes: Models 2a-c adjusted for baseline covariates, including care recipients’ and caregivers’ age, gender, race/ethnicity, marital status (except Model 2a), education, employment, income, self-rated health, depression mood, number of health conditions excluding dementia, dyad’s relationship type and co-residency, hours of care, duration of care, number of care tasks, dementia classification, and number of caregivers. Model 2c tested interaction effects of relationship type actor effects and partner effects in the whole sample.

Analyses Examining Effect Modifications of Dementia Classification (Hypothesis 4)

We found that the actor effects within care recipients and within caregivers were slightly higher among individuals in non-dementia dyads than in dementia caregiving dyads, but the moderation effect of dementia on actor effects was not statistically significant (p = 0.1169). We found a significant care recipient-to-caregiver partner effect among non-dementia caregiving dyads (b = 0.13, SE = 0.05, p < 0.01). There was no statistically significant care recipient-to-caregiver partner effect among dementia caregiving dyads (b = 0.0047, SE = 0.10, p = 0.9625). No significant moderation effect of dementia on partner effects was found (p = 0.3453) (Table 5).

Table 5.

Actor and partner effects of baseline social isolation levels on social isolation levels 2 years later, stratified by dementia classification

	Model 3a: Dementia (n = 94)			Model 3b: Non-dementia (n = 297)			Model 3c: Interaction effect
	Beta	SE	p-value	Beta	SE	p-value	p-value
Actor effects							0.1169
Care recipient	0.4890	0.1060	<0.0001	0.6152	0.0470	<0.0001
Caregiver	0.4599	0.1044	<0.0001	0.6316	0.0457	<0.0001
Partner effects							0.3453
Caregiver-to-care recipient	0.0264	0.1051	0.8023	0.0796	0.0475	0.0951
Care recipient to-caregiver	0.0047	0.0989	0.9625	0.1251	0.0454	0.0062

Notes: Models 3a-c adjusted for baseline covariates, including care recipients’ and caregivers’ age, gender, race/ethnicity, marital status, education, employment, income, self-rated health, depression mood, number of health conditions excluding dementia, dyad’s relationship type and co-residency, hours of care, duration of care, number of care tasks, dementia classification, and number of caregivers. Model 3c tested interaction effects of dementia classification on actor effects and partner effects in the whole sample.

Discussion

This study builds on the growing gerontological research literature on social isolation by examining the dyadic association of social isolation among older adults and caregivers across relationship types over time during the care continuum. By leveraging a 2-year longitudinal sample, we found that social isolation levels among caregiving dyads were not static and could increase or decrease over time. Our work suggests that caregiving relationships contribute to the multilevel mechanisms of social connection (Office of the Surgeon General, 2023). Considering social isolation is a modifiable factor of health (National Academies of Sciences, Engineering, and Medicine, 2020; Office of the Surgeon General, 2023), future work may further explore how these dynamics of social isolation in the caregiving context may impact health outcomes of older adults and their caregivers (Rippon et al., 2023). To the best of our knowledge, this is the first population-based dyadic study on social isolation among US caregiving dyads over a 2-year period, adding to the emerging empirical research in caregiving dyad health that leverages large, national databases (Ferraris et al., 2022; Kelley et al., 2019; Litzelman & Al Nassar, 2022). Our work builds upon recent cross-sectional evidence of social isolation in the US and internationally among caregivers (Li et al., 2021; Qian et al., 2025; Victor et al., 2021) and caregiving dyads (Rippon et al., 2023), pointing to the importance of routine assessment and support for psychosocial challenges along the care continuum, across health conditions, and amid a broad range of dyads’ relationships.

Importantly, our work adds social isolation to the expanding literature on the dyadic association of psychosocial well-being among caregiving dyads (Ferraris et al., 2022; Kelley et al., 2019; Litzelman & Al Nassar, 2022; Rippon et al., 2023; Segrin et al., 2019), particularly care recipient-to-caregiver partner effects. We found that older adults’ social isolation was correlated with their caregivers’ social isolation at baseline and predicted caregivers’ social isolation in 2 years after adjusting for caregivers’ baseline social isolation, other known correlates of social isolation (e.g., gender), and key caregiving characteristics (e.g., hours of care). The process of providing and receiving care and the changes in health and functioning can lead to changes in their daily interactions, perceptions of roles, and thus the nature of the relationships (Eifert et al., 2015; Lyons et al., 2002). It is possible that older adults with fewer social connections (i.e., more socially isolated) are more dependent on their caregivers not only for instrumental support and assistance but also for emotional and social needs. Over time, their caregivers may have less availability and greater emotional strain that makes maintaining other social relationships, such as at workplace, difficult. Moreover, we did not find a significant partner effect of caregivers’ social isolation on their care recipients’ social isolation. Our work adds to mixed empirical evidence on caregiver-care recipient partner effects in terms of social isolation, loneliness, and quality of life (Rippon et al., 2023; Segrin et al., 2019). Echoing the Interdependency Theory (Rusbult & Van Lange, 2003), we found that the social isolation experience of one partner may have different impacts on the other partner in the context of caregiving. It is possible that as care needs increase, relationships change such that the impact of two closely related individuals, including spouses, can be less mutually beneficial. More in-depth examination of how these close relationships change through the care continuum (Lyons et al., 2002) as disease progresses are needed, particularly the assessment of positive qualities of dyadic relationships associated with caregiving, such as shared values and reciprocity (Park & Schumacher, 2014). Future work may consider examining how contextual factors may impact social isolation experiences among caregiving dyads, such as home environments for co-residing dyads (Saylor et al., 2023), cultural stigma of certain illnesses (Liu et al., 2008), as well as their shared socio-geographic vulnerability such as living in isolated rural areas (Freedman & Nicolle, 2020). Furthermore, future work should consider the dyadic association of social isolation across a longer period and other important transitions over the life course of care recipients and caregivers (Keating et al., 2019), such as retirement, transitions to nursing homes, and widowhood.

Contrary to our Hypothesis 3, we did not find a partner effect on caregivers among spousal caregiving dyads, despite previous cross-sectional evidence on the higher risk of social isolation among spousal caregivers in comparison to other relationships (Li et al., 2021; Qian et al., 2025). Because living arrangement has little variation over time among married/coupled individuals, our social isolation measure may be less sensitive to capturing change in social isolation levels among spousal dyads. To address this concern, we conducted a sensitivity analysis using a modified, 3-item social isolation measure excluding living arrangement in APIM models stratified by relationship type. Our findings were consistent (Supplemental Tables 3 and 4). One possible explanation of the null finding of partner effects within spousal caregiving dyads is the negative changes in spousal relationships due to caregiving (Braun et al., 2009). Long-term caregiving can attenuate emotional attachment and mutual influences in older couples (Beeson, 2003; Park & Schumacher, 2014). Future work on social isolation in caregiving dyads and older adult couples should consider the influences of past and changing dyadic relationships along with caregiving trajectories (Beeson, 2003; Braun et al., 2009), such as mutuality (Park & Schumacher, 2014) and intimacy (Waite et al., 2022). Spousal caregivers are at higher risks of social isolation (Li et al., 2021; Qian et al., 2025), especially the disruption of shared social activities due to symptoms of illness (Kotwal et al., 2024), potential changes in their social role and relationship to their spouse, and challenges in isolation and loneliness at critical transition points such as a care recipient’s transition to a nursing home or to the end-of-life (Kotwal et al., 2024; Ornstein et al., 2017). Future research among spousal caregiving dyads should consider more in-depth examinations of dyadic relationships grounded in theories of marriage in later life (Wilson et al., 2024) as well as gendered experiences in isolation during caregiving (Beeson, 2003; Shiu et al., 2016).

Interestingly, our stratified analysis demonstrated a partner effect on caregivers among non-dementia caregiving dyads but not dementia dyads. Cognitive and functional decline of persons living with dementia can lead to negative restructuring of the caregiver-care recipient relationship (Pearlin et al., 1990), such as changed roles and disruption of meaningful shared activities (Kotwal et al., 2024). The null finding of care recipient-to-caregiver partner effects within dementia dyads may reflect dementia caregivers’ relational deprivation as the patients’ dementia progresses (i.e., feeling of being deprived of their past relationship with significant companionship or the loss of the affectional exchange with the care recipient) (Beeson, 2003; Pearlin et al., 1990). Since social isolation is considered a modifiable risk factor for cognitive decline and incident dementia (Livingston et al., 2024), this finding indicates the importance of assessing social isolation among older adults and their caregivers, particularly as a potential means to protect against dementia onset. Importantly, the null finding within dementia caregiving dyads needs to be interpreted with caution due to its small sample size in the current analysis (n = 95). Due to skip patterns of social isolation items in NHATS questionnaires, we were unable to assess the social isolation of community-dwelling older adults who relied on proxies to respond to the survey on their behalf (n = 128). Of note, in sensitivity analyses using the modified 3-item version of social isolation (Supplemental Table 5), we found significant caregiver-to-care recipient partner effects among non-dementia dyads but not in the dementia dyads, suggesting the possibility that cognitive impairment among older adults may influence the mutual impacts on social isolation within caregiving dyads. Considering the higher risks of social isolation among older adults with dementia (Pomeroy et al., 2024) and their close relationships with caregivers, future work should explore social isolation experiences in caregiving dyads along the progression and care continuum of dementia (Dyer et al., 2021).

This study has several limitations. Firstly, our current findings may not be generalizable to certain stages of caregiving with distinct challenges in social isolation. Given our study design, the caregiving dyads in the current sample did not include the older adults who were transitioning to facilities such as assisted living, or who died during the 2-year interval. Our sample has as an average caregiving duration of over 5 years at baseline, and thus, the findings may not capture the onset of caregiving. Secondly, it is possible that the most isolated community-dwelling older adults and their caregivers are not participating in these surveys (Pomeroy et al., 2025). Our findings among dementia caregiving dyads can be biased towards those with better functioning. Prior evidence has suggested that 41% of interviews with persons living with dementia in NHATS 2015 were completed by proxies, who might be less isolated due to the availability of proxies (Pomeroy et al., 2024). More efforts are needed for better measurement approaches and imputation strategies to assess objective social isolation among persons living with dementia and others with functional dependency on proxies in population-based surveys (Pomeroy et al., 2024). Thirdly, we acknowledge the caveat that the social isolation measure is slightly different for caregivers and care recipients due to the unavailability of the exact number of individuals in the core discussion network in the NSOC survey instrument. We use a lower threshold of core discussion network size among caregivers (smaller than one person being isolated) than care recipients (smaller than two persons being isolated) (Cudjoe et al., 2020), which may underestimate social isolation levels among caregivers. The measure among caregivers may be less sensitive to detect changes in core discussion network size (e.g., a change from two to one person in core network size does not yield a change in the score), potentially leading to under-estimations of both actor and partner effects on caregivers. A sensitivity analysis using a harmonized version of social isolation in care recipients shows similar findings (Supplemental Table 6). Fourth, due to unavailability of data, in the current investigation we were not able to assess the dyadic association between social isolation and loneliness and potential impacts of relationship quality on these associations (Rippon et al., 2023). Fifth, we acknowledge that the statistical power in subgroup analyses is limited. The null findings of partner effects within the subgroups of dementia caregiving and spousal caregiving dyads should be interpreted with caution. Lastly, although we consider caregiving as a dyadic process between caregiver and care recipient in the current analysis, it is important to note the dyad is nested within a dynamic care network of families, paid caregivers, and clinicians (Freedman et al., 2024). We consider total number of caregivers in the network as a covariate. Future work on relationships and interactions within caregiving networks, such as coordination for in-home care (Fabius et al., 2023), would enrich the knowledge of the interdependency of caregiver and care recipient’ psychosocial well-being.

Our study has implications for intervention development and policy change. Our findings highlight that reducing social isolation among community-dwelling older adults has the potential to also benefit their caregivers, including non-dementia caregivers and non-spousal caregivers who are underrepresented in existing interventions (Velloze et al., 2022). Future interventions need to strengthen strategies for recruitment and retention to increase the inclusion of non-spousal caregivers, such as adult children, who face different caregiving challenges from spouses, such as travel time to provide care, and competing responsibilities, including employment and child care (Wolff et al., 2025). Training and support for caregivers may incorporate behavioral therapies to help caregivers to identify their individual needs, barriers, and strategies for social connection (Abshire Saylor et al., 2024; Van Orden et al., 2023). Efforts to facilitate restoring and adapting meaningful social activities shared by caregivers and care recipients, such as music and art activities (Allison et al., 2022), may be particularly beneficial for dyads coping with dementia (Kotwal et al., 2024). Technology-based interventions, such as virtual peer support groups and psychoeducational programs, can benefit older adults and caregivers by helping them maintain and develop meaningful social connections (Lee et al., 2022; Williams et al., 2022) while reducing participation barriers and improving retention in interventions (Mao et al., 2023), which may be particularly beneficial to those living in rural and other resource-limited settings (Blusi et al., 2015). Caregiver support programs can also consider providing online resources and referrals to increase social connection. For example, social media platforms that were developed for patients and caregivers, such as CaringBridge and CarePages, can help caregivers seek assistance, receive timely social support, and further develop their social networks offline through in-person visits (Bloom et al., 2021). Moreover, national and state-level policy efforts can help to address the social and emotional needs of older adults and caregivers in community. For example, the newly launched Guiding an Improved Dementia Experience (GUIDE) Model by the Centers for Medicare & Medicaid Services, includes 24/7 access to a support line and assistance to access to respite services for caregivers (Centers for Medicare & Medicaid Services, 2024), providing timely support and more respite opportunities for caregivers to engage in valued social activities (Lee et al., 2022). Future policy efforts should consider more routine assessments of changing and unmet social needs of both caregivers and older adults along with their illness trajectories (Kotwal et al., 2024; Lee et al., 2022).

In conclusion, this study suggests the potential benefits of enhancing social connection among community-dwelling older adults in reducing social isolation for their family caregivers. Ongoing, community-based efforts to promote social connection and address the shared challenges in isolation across the care continuum have the potential to lead to broader health impacts for aging populations in the United States.

Supplementary Material

Supplemental material for this article is available online.