Older adults’ care networks and the pathways to unmet needs

Abstract

Background and Objectives

Older adults differ widely both in the care they require and in who provides them care, often reporting significant unmet needs for assistance. Few studies have simultaneously considered the type of disability (self-care, mobility, and household activities) and multisource care networks (kin, extended-kin, non-kin, and paid help) to understand factors influencing unmet care needs among community-living older adults.

Research Design and Methods

Using data from the National Health and Aging Trends Study (2011; N = 3,265; M_Age [SD] = 77 [7.74] years, 62% women), we conducted a latent class analysis to identify care network types based on older adults’ functional limitations and caregiver sources. Multinomial logistic regression models predicted network membership based on personal and structural predictors. Zero-inflated Poisson regression examined the relationship between network type and unmet care needs 1 year later.

Results

Seven distinct care network types emerged, characterized by combinations of caregiving sources and disability domains. Kin caregivers were involved across all network types. Older adults coresiding with kin typically received minimal paid help, which significantly increased their likelihood of unmet care needs in the subsequent year. Networks predominantly relying on non-kin caregivers tended to not use paid services and exhibited higher unmet care needs. Mismatches between disability type and the assistance received (e.g., requiring self-care assistance but primarily receiving household help) were associated with unmet care needs in the subsequent year.

Discussion and Implications

The provision of adequate care was contingent upon the direct alignment of caregiving tasks with the functional limitations of aging adults and the effective coordination of informal and formal care resources. Enhancing care alignment through targeted assessments, supplementing family caregiving with formal services, and promoting coordinated caregiving arrangements could substantially reduce unmet care needs.

Translational Significance.

Even with several caregivers, older adults frequently experience unmet care needs. This study found different care network configurations depending on older adults’ disabilities and the source of care. Results showed that older adults depending mostly on kin or non-kin caregivers generally had more unmet care needs. Importantly, aligning the type of caregiving assistance with the older adults’ actual functional limitations and strategically adding paid services to supplement informal care reduced unmet needs. Results offer practical guidance for families, service providers, and policymakers looking to develop tailored and effective caregiving strategies, ultimately enhancing the daily care experiences and general well-being of community-living older adults.

Families have traditionally served as the primary safety net for older adults experiencing functional and cognitive limitations. However, demographic trends and diversification of family structures have reduced the availability of informal caregivers (Wu et al., 2024). These societal changes have affected who provides care and how care is organized, raising important concerns about the adequacy and sustainability of existing caregiving approaches. Concurrently, the older adult population is rapidly expanding, increasing the number of older adults who require assistance with personal activities of daily living, such as dressing and bathing, and instrumental activities of daily living, such as financial management and grocery shopping. In response, older adults increasingly rely on care networks that blend unpaid informal caregivers with formal long-term services and supports or paid help.

Older adults differ considerably in their care needs and available sources of support (Hu et al., 2023). More importantly, care networks frequently fail to fully address these diverse needs, resulting in unmet care needs that negatively impact physical health, mental health, and quality of life and increase risks of hospitalization and institutionalization (Beach et al., 2020). Prior studies have significantly advanced our understanding by exploring aspects of care network complexity, such as caregiver roles, network composition, and their association with functional limitations and physical and psychological well-being (e.g., Hu et al., 2023; Leggett et al., 2025; Lin, 2024). However, these studies have not simultaneously integrated multiple caregiver types (kin, extended kin, non-kin, and paid help) with multidomain disabilities (self-care, mobility, and household activities) within a single analytic framework to prospectively predict unmet care needs. To better inform care planning and resource allocation decisions, research that comprehensively integrates these dimensions is needed.

To address this gap, we used latent class analysis, a person-centered analytic framework, to simultaneously identify (a) common care network configurations among community-living older adults, (b) factors predicting membership in these networks, and (c) how each network configuration relates prospectively to unmet care needs. Understanding these associations will inform targeted interventions, improve future care planning, and optimize resource allocation within long-term care systems.

Understanding care network models

Researchers have proposed several models to describe how formal (paid) and informal support are combined to care for older adults. Cantor’s (1991) hierarchical compensatory model suggests that older adults have a defined preference order for receiving care that typically begins with a spouse, followed by adult children, extended relatives, and friends, with paid caregivers only if they have no alternative. This model emphasizes the sequential substitution of caregivers based primarily on familial closeness and obligation. By contrast, Litwak (1985) proposed the task-specificity model, which posits that care networks are organized not by caregiver preference but by matching specific caregiving tasks with the appropriate caregiver based on their availability and skill levels. According to this model, family members are more likely to provide nontechnical help (e.g., personal care, basic household chores), friends are primary sources of emotional support, and formal services deliver technical help or specialized assistance such as medical care. A third model, the complementarity model (Chappell & Blandford, 1991), integrates the two previous models by suggesting that formal care services precisely fill gaps in informal caregiving or supplement informal assistance as the care recipient’s needs intensify. In this sense, informal and formal care are not substitutes but complementary components of a more extensive caregiving system designed to meet increasingly complex care needs.

Empirical studies evaluating these models have yielded mixed results (Rook & Schuster, 1996). Moreover, recent policy shifts in the United States have emphasized the role of paid services in supplementing rather than replacing the informal care provided by family members (Gaugler, 2022). Given these complexities, selecting a single theoretical model that is applicable to all care arrangements is less productive. Instead, it is important to understand how caregiving patterns emerge from the interaction of task demands, relational closeness, service availability, and specific contexts to address older adults’ diverse caregiving needs better.

Empirical studies on care networks

Empirical work on care networks has progressed from dyadic snapshots to multidimensional, dynamic analyses of caregiving structures (Freedman et al., 2024). Early studies challenged the assumption that each older adult receives care primarily from a single individual. Using ego-network data, Marcum et al. (2020) took a network-oriented approach, highlighting that the primary caregiver role is not always clear-cut and that only 60% of helpers in the same network identified the same person as the primary caregiver, with nominations frequently overlapping among kin and non-kin. Similarly, research in the Netherlands (Swinkels et al., 2024) identified distinct “main-carer” archetypes that centered around a coresident spouse or partner, informal caregivers, formal caregivers, or privately paid caregivers, underscoring significant contextual variations in primary caregiver configurations and their implications for caregiver and care recipient well-being. Although these studies provide important insights about who leads, they primarily capture leadership with caregiving networks but do not fully explore the broader interaction and coordination among multiple caregivers.

Subsequent research expanded the focus to consider the size of the care network and the composition of caregiving groups explicitly. Hu et al. (2023), using data from the National Health and Aging Trends Study (NHATS), shed light on the prevalence and benefits of shared caregiving situations, highlighting that older adults with higher disability levels experienced fewer unmet needs when supported by multiple caregivers rather than a single primary caregiver. Broese van Groenou et al. (2016) further emphasized this dynamic in mixed informal–formal caregiving networks in the Netherlands, showing that explicit coordination across caregiver types enhanced continuity of care and reduced gaps in support. Racial–ethnic comparisons by Lai et al. (2025) underscored these structural differences, showing that Black and Hispanic older adults often rely on larger, extended-family networks compared to their White counterparts, who predominantly rely on spouses. Collectively, these findings underscore the protective benefits of larger, coordinated caregiving networks, yet they do not explicitly differentiate how such structures specifically affect care outcomes across multiple disability domains, a gap addressed by the current study.

A third strand of research examines task division and caregiver collaboration. Jacobs et al. (2014) differentiated complementary (task-specialized) from supplementary (task-shared) patterns of caregiving, linking complementarity task-sharing to denser and potentially more effective caregiver communication. Ali et al. (2022), using latent class analysis (LCA), identified five distinct caregiving “care bundles,” characterized by different patterns of caregiving intensity, regularity, and specific activity domains such as transportation, household tasks, and personal care activities. Extending this line of inquiry, Leggett et al. (2025) classified dementia caregiving networks based on collaborative patterns of task-sharing into siloed, small-but-mighty, and complex network types. Notably, they found that siloed networks, characterized by limited task-sharing among caregivers, predicted poorer sleep outcomes for care recipients and reduced social support for caregivers. Although these studies introduce nuanced typologies aligning specific caregiving patterns to outcomes, they primarily capture networks and outcomes at a single point in time rather than accounting for how they adapt to evolving care needs.

Recent scholarship addresses this limitation by examining how caregiving networks evolve over time. Spillman et al. (2020), through a four-wave NHATS panel, showed that dementia care networks expanded and increasingly featured generalist caregivers performing multiple tasks as care demands intensified, in contrast with more rapidly changing network compositions among older adults without dementia. Lin (2024) further applied latent transition analysis, highlighting evolving caregiving structures and emphasizing the increased reliance on the use of assistive technologies as supplementary caregivers when health deteriorates. Moreover, Nemmers et al. (2024) found that larger, friend-heavy caregiving networks lacking close familial involvement significantly increased the likelihood of transitioning to residential care. Collectively, these studies illustrate the dynamic adaptation of caregiving networks, yet they also leave unaddressed how specific network configurations systematically affect unmet care needs across diverse disability domains.

None of the studies reviewed here systematically integrates multisource caregiver typologies (i.e., kin, extended-kin, non-kin, paid) with multidomain disability (i.e., self-care [SC], mobility [MO], household activities [HA]) to prospectively assess future unmet care needs. The present study explicitly addresses this gap by employing a person-centered latent class framework that incorporates four caregiver sources across three disability domains to predict unmet care 1 year later. By clarifying how variations in care networks impact the adequacy of support provided (i.e., unmet need), this study offers actionable insights to inform targeted interventions and improve resource allocation strategies to support community-living older adults.

Predictors of care network configurations

Existing theoretical and empirical research underscores the importance of care networks; however, a clearer understanding of the specific factors that shape these networks is still needed. According to Andersen and Newman’s (2005) behavioral model, access to care is jointly determined by three key domains: predisposing, enabling, and need factors. Later adaptations of this model apply the exact three domains to predict long-term care contexts, framing unmet activities of daily living (ADL)/instrumental activities of daily living (IADL) assistance as a failure of realized access to care.

Predisposing

Advanced age and being male predispose older adults toward increased utilization of formal and informal care (Babitsch et al., 2012; Janssen et al., 2016). Race and ethnicity also have been found to be important, as Black and Hispanic care recipients typically rely on larger, less kin-centric networks than their White counterparts (Lai et al., 2025). Previous research has also found that the availability and proximity of kin substantially shape caregiving arrangements. Spouses and adult children typically serve as primary informal caregivers (Davey et al., 2004).

Enabling

When immediate kin are unavailable, extended family or even non-kin may become crucial caregiving resources (Keating & Dosman, 2009). Friend-heavy networks, however, have been linked to higher odds of residential relocation, underscoring limits to purely non-kin support (Nemmers et al., 2024). Moreover, older adults living with kin are often less inclined to utilize formal caregiving services. In contrast, those with extended kin or friend-based networks may more readily integrate formal care to address gaps in informal support (Keating & Dosman, 2009).

Need

Factors that increase the need for care, such as the severity of physical impairment and cognitive decline (Aartsen et al., 2004; Janssen et al., 2016), are strong determinants of care utilization (Lyons et al., 2000; Penning et al., 2016). Longitudinal NHATS analysis shows that as care needs intensify, care networks may shift compositionally, either by expanding or task-sharing, incorporating more informal helpers, technology assistance, or paid helpers to meet the rising care needs better (Lin, 2024; Spillman et al., 2020).

These factors collectively shape care network configurations, determining who provides care, how care is organized, and the adequacy of care provided by operationalizing the presence or absence of unmet needs in the following year.

Care network configurations and unmet needs

Ultimately, the effectiveness of care networks is assessed by their ability to adequately address older adults’ daily needs and forestall unmet needs. When these networks fail to provide sufficient support, older adults face adverse consequences, including poor hygiene, medication errors, injuries, inadequate nutrition, falls, hospitalization, and even mortality (Allen et al., 2014). Approximately, one fourth of community-dwelling Medicare beneficiaries with functional impairments have experienced at least one adverse outcome due to unmet care needs (Allen & Mor, 1997; Allen et al., 2014; Pruchno et al., 2016).

Existing research has consistently linked smaller or role-­siloed care networks (such as living alone or having fewer available helpers) to an elevated risk for adverse outcomes such as sleep disturbances, poor hygiene, missed medications, injuries, and hospitalization. On the other hand, shared-care arrangements with multiple helpers are associated with a lower probability of unmet needs when disability is high (Hu et al., 2023). Additionally, well-coordinated mixed informal–formal networks reduce care gaps (Broese van Groenou et al., 2016).

Task diversity is also important. Specifically, some studies have found that difficulties with IADLs are more strongly associated with unmet needs than with ADLs (Allen & Mor, 1997; Allen et al., 2014). Also, older adults with multiple comorbidities tend to report more unmet needs than those with fewer health limitations (Allen & Mor, 1997; Pruchno et al., 2016). Further, research has shown that additional support from secondary caregivers can substantially reduce the burden on primary caregivers (Liang et al., 2022; Spillman et al., 2020). Despite these insights, prior studies rarely integrate multiple caregiving dimensions simultaneously (e.g., caregiver identity, specific caregiving tasks) with older adults’ unmet needs trajectories, highlighting a critical evidence gap that the present study addresses.

Present study

This study adopted a comprehensive, person-centered analytic approach to address these critical gaps. We simultaneously integrated multiple caregiving sources (kin, extended kin, non-kin, and paid helpers) with distinct disability domains (SC, MO, and HA). Our first objective was to classify distinct care network typologies. Guided by Andersen and Newman framework, we then tested how predisposing (age, gender, race/ethnicity, education), enabling (kin proximity, household income, paid-helper availability), and need factors (disability severity, dementia status) predicted membership in each typology. Finally, we examined how these configurations were linked to older adults’ unmet needs 1 year later, treating unmet needs as a direct indicator of realized access to assistance. Our research questions, therefore, were:

1. What were the predominant care network configurations of community-living older adults living with functional limitations?

2. Which predisposing, enabling, and need variables predicted membership in these care network configurations?

3. How were these care network configurations associated with unmet needs for assistance a year later?

Method

Sample

Data were taken from the NHATS Round 1 (2011) and Round 2 (2012) public use data (www.nhats.org; Montaquila et al., 2012). Round 1 offers the largest baseline cohort, allowing us to examine how the initial network structure predicts next-year unmet needs in Round 2. Moreover, longitudinal analyses using the same cohort show that key network characteristics observed in 2011 continue to shape trajectories through at least 2015 (Lin, 2024; Spillman et al., 2020), supporting this data set’s utility for studying care networks and unmet needs. NHATS is based on a nationally representative sample of Medicare enrollees aged 65 and older. An annual interview was conducted to collect information on older adults’ demographic information, living environment, physical and cognitive limitations, quality of life, and assistance needed and received with ADLs. The original sample in Round 1 included 8,245 older adults. This study included a subsample of 3,265 respondents (see Supplementary Figure 1) who were community-living and had at least one limitation in SC, MO, or HA (as defined in the Measures section) in 2011. We excluded individuals with missing data on key variables (i.e., disability or help indicators) to ensure a more reliable estimation of care network membership.

Measures

Difficulties in performing ADLs

Participants answered a series of questions about SC, MO, and HA that included activities characteristic of the traditional ADL and IADL measures. SC activities included eating, bathing, using the toilet, and dressing. MO activities included leaving the home, walking inside the home, and getting out of bed. HA activities included laundry, grocery shopping, making hot meals, and keeping track of medications. For each activity, participants were first asked whether they had performed this activity in the last month. If they responded no, they were asked why it was not done. Participants who indicated that it was too difficult to do the activity alone and no one was there to help were coded as having disabilities in that activity. If participants responded “yes” to a particular activity, follow-up questions determined whether the activity was performed with or without help and whether it was done with difficulty if performed alone. We coded those who performed the activity alone but indicated that they had had at least a little difficulty with SC and MO activities or some difficulties performing HA activities as having a disability in that activity (following the coding scheme in Allen et al. [2014]). If they indicated that they had performed an activity with assistance from others because of health and functioning reasons, we also defined them as having a disability in that activity. As such, each respondent was assigned a binary indicator (0 = no disability; 1 = has a disability) for each activity, and a summary binary indicator for each of the three domains was calculated (0 = no disabilities in the domain; 1 = if they had any disability in the domain). Finally, the sum of binary indicators for each activity was taken to represent the total number of disabilities in the prior month to the interview in 2011. This approach captures the presence of disability and the degree of functional limitation within and across domains.

Care network

When participants reported receiving assistance with an activity last month, they were asked to indicate who assisted. We categorized responses into four groups: close kin (spouse/partner, daughter/son, daughter-in-law/son-in-law, mother/father, mother-in-law/father-in-law, and granddaughter/grandson), extended kin (stepdaughter/stepson, sister/brother, sister-in-law/brother-in-law, stepmother/stepfather, niece/nephew, aunt/uncle, cousin, ex-wife/husband, boyfriend/girlfriend, other relative), non-kin (stepdaughter’s son/daughter, stepson’s son/daughter, daughter-in-law’s son/daughter, son-in-law’s son/daughter, boarder/renter, roommate, neighbor, friend, coworker, minister/priest/other clergy, other nonrelative), and paid helper (paid aide/housekeeper/employee, psychiatrist, psychologist, counselor, therapist). For each activity, we created four binary indicators (0 = No; 1 = Yes), representing whether kin, extended kin, non-kin, and paid helper helped with each activity in the prior month in 2011. Four summary binary indicators (0 = No; 1 = Yes) were generated to represent which helper category assisted (i.e., kin, extended kin, non-kin, and paid helper). This categorization captures the diverse types of assistance that older adults may receive, enabling a nuanced analysis of care network configurations.

Unmet needs

Participants were asked if they had any undesired or adverse incidents in the last month due to lacking assistance (consistent with the coding scheme in Allen et al. [2014]). We defined such incidents as unmet needs. Incidents included missing a meal, being unable to bathe, having wet or soiled clothes, and being unable to change clothes in the SC domain; being housebound, bedridden, or not moving around inside the home in the MO domain; and not having clean laundry, groceries, a hot meal, and medicine-related mistakes in the HA domain. We created a binary indicator (0 = No; 1 = Yes) for each activity, indicating the presence of unmet needs in each activity in both 2011 and 2012, allowing us to explore how unmet needs evolve over time. The sum of these indicators represented the total number of unmet needs in each year.

Participant’s characteristics

The following covariates were also included in our analyses. Participant age was coded in years, and gender was dummy coded (0 = Male; 1 = Female). Race (0 = Not White; 1 = White) and education (0 = High school or below; 1 = Above high school) were also dummy coded. Annual household income was assessed as a composite of multiple sources. For participants with missing income data, we used imputed values provided in the NHATS public-use files (DeMatteis et al., 2016). Living arrangement was a nominal variable recoded using items from the household roster to define who lived with the participant (0 = Lives alone; 1 = Lives with someone else other than spouse or child; 2 = Lives with child; 3 = Lives with spouse; 4 = Lives with child and spouse). An additional binary variable captured whether the participant had any children living outside the household (0 = No; 1 = Yes) without overlap with the living arrangement categories. Although coresidence with close kin likely increases the probability of receiving assistance, it does not guarantee caregiving support. Conversely, the separate binary variable indicating the presence of children living outside the household specifically reflects the availability of potential caregivers who do not coreside. Thus, these variables represent distinct constructs capturing actual living arrangements versus potential caregiver availability. Finally, cognitive health status (see Kasper et al. [2013] for the coding scheme) was used as a control variable and was coded as 0 (No dementia), 1 (Possible dementia), and 2 (Probable dementia). We included dementia status because cognitive impairment has been shown to substantially influence both care arrangements and the likelihood of unmet needs.

Analytic strategies

Using a LCA approach, we classified care network structures along two dimensions—the older adults’ disabilities and the people who assisted them. The LCA analysis included 15 binary indicators: For each disability domain (i.e., SC, MO, HA), we added (a) one indicator for having a limitation in that domain and (b) four indicators indicating whether help for that domain came from kin, extended kin, non-kin, and paid helper (3 domains × [1 + 4] = 15 indicators). We compared several competing latent class models to determine the number of latent classes that best described the data parsimoniously. More specifically, we fit models with three to nine classes and selected the best model based on Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), entropy, and model interpretability. Class membership probability, which represents the proportion of the sample expected to belong in each latent class, and item–response probabilities, which reflect the likelihood of different responses to the items conditional on class membership, were used to describe the prevalence and characteristics of each latent class. The item–response probabilities were shown using a radar plot to depict the patterns of disability domains and care sources.

After determining the ideal number of care networks, we investigated if older adults’ personal characteristics, family composition, and living arrangements varied by class membership. Specifically, we estimated a multinomial logistic regression using the R3STEP option in Mplus (Asparouhov & Muthén, 2014) with the following predictors: participant’s age, gender, living arrangement, whether he/she has a child outside the household, and dementia status. Finally, we used zero-inflated Poisson regression to examine whether class membership predicted unmet needs in 2012, controlling for inflation in total disabilities and unmet needs in 2011. Zero-inflated models are particularly well suited for count data, where a large portion of the sample reports zero unmet needs, thus helping to account for the overdispersion inherent in this outcome. For this, we assigned each participant to a latent class corresponding to his or her most likely membership based on maximum posterior probability (Nagin, 2005). Thus, we treated class membership as a categorical variable and performed subsequent outcome analyses. All analyses were conducted in STATA (Version SE 14) and Mplus (Version 7.4; Muthén & Muthén, 1998–2015). Analytic weights were used to account for differential selection probabilities and to adjust for potential bias related to unit nonresponse (Montaquila et al., 2012).

Results

Table 1 presents the weighted distribution of sociodemographic factors and health indicators for the selected sample. The average age of the participants in 2011 was 76.99 years (SD = 7.74, range = 65–105 years). The majority of participants were women (62.10%), White (83.97%), lived with others (71.80%), had at least one child living outside the household (86.05%), and did not have dementia (66.63%). Approximately, half of the sample had completed high school or attended college (42.03%). Mean annual household income was $40,420 (SD =$120,290, range = 0K–4,000K). Half of the respondents were currently married (50.92%), and the greater majority (86.05%) had at least one child living outside the household. Two-thirds had no dementia (66.63%), and one-third lived alone (28.20%). Most participants had kin helpers (94.35%), 37% of the older adults had extended kin as helpers, 36% had non-kin helpers, and 10% had paid helpers (not shown in the table). These patterns underscore the diverse support sources for older adults and highlight potential gaps in paid assistance.

Table 1.

Weighted population description on selected sample.

Variables	Proportion (%)	95% Confidence interval	M	SD	Range
Gender
Male	37.90	[35.98, 39.85]
Female	62.10	[60.15, 64.02]
Race
White	83.97	[82.73, 85.14]
Not White	16.03	[14.86, 17.27]
Education
High school or below	57.97	[55.96, 59.95]
Above high school	42.03	[40.05, 44.04]
Marital status
Not married	49.08	[47.10, 51.05]
Married	50.92	[48.95, 52.90]
Living arrangement
Lives with others	71.80	[70.02, 73.52]
Lives alone	28.20	[26.48, 29.98]
Child outside HH
No child outside HH	13.95	[12.62, 15.40]
Has child outside HH	86.05	[84.60, 87.38]
Dementia class
Probable dementia	18.91	[17.52, 20.40]
Possible dementia	14.45	[13.19, 15.81]
No dementia	66.63	[64.82, 68.40]
Age			76.99	7.74	[65, 105]
Income			40.42K	120.29K	[0K, 4,000K]
Total disabilities			3.75	3.06	[1, 11]
Total unmet need 2011			0.66	1.25	[0, 10]
Total unmet need 2012			0.53	1.10	[0, 9]

Note. HH = household; M = mean; SD = standard deviation. Sample size is 3,265; population size is 12,974,853. The Mplus and STATA data files used for analysis can be provided upon request to the corresponding author.

On average, older adults had disabilities in three to four domains, though the extent of disabilities varied considerably (M = 3.75, SD = 3.06; range = 1–11), with some older adults experiencing relatively few functional limitations and others living with multiple comorbidities. Additionally, the proportion of older adults who reported at least one adverse consequence due to unmet needs increased from 34% in 2011 to 39% in 2012. This increase in just 1 year underscores how unmet needs can escalate over time, underscores the importance of proactively identifying care network configurations most at risk, and underscores the importance of enhancing support for older adults needing assistance.

We used LCA to classify care network patterns based on the older adults’ care needs and the sources of support they received. Fit indices indicated that a seven-class model provided the best balance of model fit and parsimony (AIC = 28152.91, BIC = 28829.01, entropy = 0.92; see Table 2). As the number of classes increased, both AIC and BIC decreased, but the improvements in model fit slowed down notably after the six-class model, suggesting little incremental explanatory power obtained from models with more than seven classes.

Table 2.

Latent class analysis: model fit indices and classification quality.

Fit index	3 Classes	4 Classes	5 Classes	6 Classes	7 Classes	8 Classes	9 Classes
AIC	30356.48	29641.19	28939.32	28439.03	28152.91	27932.14	27772.58
BIC	30642.76	30024.92	29420.51	29017.68	28829.01	28705.70	28643.59
Sample adjusted BIC	30493.42	29824.74	29169.50	28715.82	28476.32	28302.16	28189.22
Entropy	0.90	0.85	0.89	0.91	0.92	0.90	0.91

Note. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion. The selected model is shown in boldface.

Table 3 shows the prevalence and conditional item–response probabilities for each of the seven latent care network classes. Figure 1 presents a radar plot of the seven care network classes, highlighting how each class differs based on functional disability and primary source(s) of support.

Table 3.

Class membership prevalence and conditional item–response probabilities of the seven-latent class model.

Disability domain and source of support	Class 1	Class 2	Class 3	Class 4	Class 5	Class 6	Class 7
	Tri-domain Disability, Mostly Paid + Kin Help	Tri-domain Disability, Kin Help Only	Tri-domain Disability, Extended Kin Support	Tri-domain Disability, Non-kin Support	SC Disability, Kin Help Primarily for HA	MO Disability, Kin Help Primarily for HA	HA Disability, Kin Help for HA
Class prevalence (%)	4.48	18.60	2.76	3.15	31.59	29.80	9.62
Conditional item–response probabilities
Disabilities in SC	0.93	1.00	0.79	0.84	1.00	0.09	0.00
Disabilities in MO	0.98	1.00	0.95	0.93	0.49	1.00	0.00
Disabilities in HA	0.88	0.90	0.85	0.89	0.32	0.31	1.00
Kin help for SC	0.39	0.80	0.12	0.14	0.24	0.00	0.00
Extended kin help for SC	0.00	0.00	0.42	0.00	0.01	0.00	0.00
Non-kin help for SC	0.02	0.05	0.04	0.46	0.01	0.00	0.00
Paid help for SC	0.77	0.04	0.12	0.06	0.01	0.00	0.00
Kin help for MO	0.55	0.88	0.19	0.20	0.00	0.18	0.00
Extended kin help for MO	0.01	0.01	0.67	0.04	0.00	0.01	0.00
Non-kin help for MO	0.11	0.04	0.07	0.50	0.00	0.02	0.00
Paid help for MO	0.76	0.01	0.14	0.02	0.00	0.01	0.00
Kin help for HA	0.66	0.99	0.26	0.39	0.68	0.64	0.74
Extended kin help for HA	0.04	0.01	0.98	0.03	0.04	0.04	0.07
Non-kin help for HA	0.09	0.04	0.09	0.94	0.03	0.04	0.08
Paid help for HA	0.88	0.01	0.19	0.13	0.02	0.02	0.05

Note. HA = household activities; MO = mobility; SC = self-care activities. Boldface probabilities (>.40) characterize each class

Figure 1.

Radar-plots depicting item–response probabilities of care needs (SC = self-care; MO = mobility; HA = household activities) and care providers (kin, extended kin, non-kin, and paid helpers) by the latent care network class membership.

Class 1 (4.48%)—Tri-domain Disability, Mostly Paid + Kin Help. Older adults in this group showed high disability across all three domains (SC, MO, and HA), and mostly relied on paid helpers supplemented by kin for assistance, and had minimal involvement of extended kin or non-kin.

Class 2 (18.60%)—Tri-domain Disability, Kin Help Only. Older adults in this class also reported disabilities in all three domains, but they relied exclusively on close kin for assistance. The main difference from Class 1 was the absence of paid helpers, emphasizing a heavy reliance on family care.

Class 3 (2.76%)—Tri-domain Disability, Extended Kin Support. Like Classes 1 and 2, older adults in this class experienced disabilities across SC, MO, and HA. However, they relied primarily on extended kin, with lower probabilities of receiving help from close kin, non-kin, or paid sources.

Class 4 (3.15%)—Tri-domain Disability, Non-kin Support. This group had high disability across all domains and received most of their help from non-kin (e.g., friends or neighbors), with low probabilities of kin or paid help, suggesting that non-kin serve as the primary caregivers in this class.

Class 5 (31.59%)—SC Disability, Kin Help Primarily for HA. Older adults in this class report SC difficulties, but they received help from kin on HA. They received minimal support from extended kin, non-kin, or paid help.

Class 6 (29.80%)—MO Disability, Kin Help Primarily for HA. Older adults in this class reported MO limitations, but they received help from kin on HA. Although there is a small probability (.18) of them receiving help from kin on MO issues, it remained substantially lower than the support provided for HA (.64).

Class 7 (9.62%)—HA Disability, Kin Help for HA. In this class, older adults primarily experienced disabilities in HA and received corresponding kin support. The alignment between the disability domain and the type of assistance suggests a more direct match between need and support.

After establishing the care network class types, multinomial logistic regressions were conducted to examine whether personal and structural characteristics of older adults predicted class membership (see Table 4). Because Class 7 (HA Disability, Kin Help for HA) represented a relatively “lower risk” group—characterized by difficulty carrying out household tasks and matching kin assistance—we selected it as the reference class in our multinomial logistic regression. This choice allowed us to interpret each covariate in terms of how it increases or decreases the likelihood of membership in classes characterized by more extensive or differently aligned care needs and support.

Table 4.

Personal and structural characteristics predicting class membership.

Predictors	Class 1a	Class 2a	Class 3a	Class 4a	Class 5a	Class 6a
	Tri-domain disability	Tri-domain disability	Tri-domain disability	Tri-domain disability	Disability in SC	Disability in MO
	Paid + Kin	Kin help	Extended kin help	Non-kin help	Kin help HA	Kin help HA
Intercept	−6.21 (1.49)***	−2.19 (0.99)*	2.45 (1.76)	−1.13 (1.53)	3.58 (0.92)***	2.79 (0.89)**
Age	0.06 (0.02)**	0.01 (0.01)	−0.01 (0.02)	0.01 (0.02)	−0.02 (0.01)*	−0.02 (0.01)
Income (K)	0.00 (0.00)	−0.01 (0.00)*	−0.04 (0.02)*	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)
Gender
Male	Ref	Ref	Ref	Ref	Ref	Ref
Female	0.32 (0.28)	0.06 (0.19)	−0.49 (0.39)	0.09 (0.35)	−0.61(0.18)***	−0.24 (0.18)
Race
Non-White	Ref	Ref	Ref	Ref	Ref	Ref
White	−0.07 (0.27)	0.11 (0.19)	−0.21 (0.33)	0.29 (0.32)	0.27 (0.19)	0.05 (0.18)
Education
High school or below	Ref	Ref	Ref	Ref	Ref	Ref
Above high school	0.63 (0.25)*	0.15 (0.19)	−0.13 (0.39)	−0.09 (0.30)	0.30 (0.17)	0.14 (0.17)
Living arrangement
Lives alone	Ref	Ref	Ref	Ref	Ref	Ref
Others in HHb	−0.19 (0.43)	0.68 (0.37)	1.00 (0.39)**	0.57 (0.40)	−0.66 (0.43)	−0.26 (0.34)
Child in HH	−0.09 (0.32)	1.83 (0.26)***	−0.96 (0.65)	−1.77 (0.49)***	0.05 (0.26)	0.31 (0.24)
Spouse in HH	−0.39 (0.32)	1.48 (0.24)***	−1.79 (0.75)*	−1.72 (0.51)***	0.18 (0.20)	0.15 (0.19)
Spouse + child in HH	−0.97 (0.55)	1.56 (0.34)***	−3.31 (1.94)	−2.36 (1.08)*	−0.03 (0.32)	−0.08 (0.31)
Has child outside HH
No	Ref	Ref	Ref	Ref	Ref	Ref
Yes	0.21 (0.36)	0.59 (0.25)*	−1.51 (0.37)***	−0.81 (0.35)*	−0.18 (0.25)	0.05 (0.25)
Dementia
No	Ref	Ref	Ref	Ref	Ref	Ref
Possible	0.09 (0.31)	0.33 (0.23)	−0.31 (0.46)	0.23 (0.38)	−0.32 (0.22)	−0.48 (0.21)*
Probable	0.63 (0.27)*	0.76 (0.22)***	−0.12 (0.38)	0.17 (0.38)	−1.21 (0.25)***	−1.06 (0.22)***

Note. HA = household activities; HH = household; Ref = reference group; MO = mobility; SC = self-care activities. The numbers shown are parameter estimates (standard error).

^aCoefficients are estimated from comparing each class with Class 7 (HA Disability, Kin Help for HA).

^bOthers in HH refers to someone other than spouse or child.

^* p < .05;

^** p < .01;

^*** p < .001.

Each additional year of age was associated with an increased likelihood of belonging to Class 1 (Tri-domain Disability, Mostly Paid + Kin Help; b = 0.06, p = .003) and a slightly reduced likelihood of belonging to Class 5 (SC Disability, Kin Help Primarily for HA; b = −0.02, p = .05). Gender also played a role, wherein male participants were more likely than female participants to be in Class 5 (SC Disability, Kin Help Primarily for HA; b = −0.61, p = .001).

We found living arrangements to be consistently influential. Relative to living alone, coresidence with a child (b = 1.83, p < .001), spouses (b = 1.48, p < .001), or both spouse and child (b = 1.56, p < .001) substantially increased the likelihood of membership in Class 2 (Tri-domain Disability, Kin Help Only). Conversely, living with a child (b = −1.77, p < .001), spouse (b = −1.72, p < .001), or both spouse and child (b = −2.36, p = .029) significantly decreased the likelihood of membership in Class 4 (Tri-domain Disability, Non-kin Support). Coresidence with nonspouse/nonchild adults increased the likelihood of membership in Class 3 (Tri-domain Disability, Extended Kin Support; b = 1.00, p = .01).

Dementia status also predicted class membership. Individuals with probable dementia were significantly more likely to belong to Class 1 (Tri-domain Disability, Mostly Paid + Kin Help; b = 0.63, p = .02) and Class 2 (Tri-domain Disability, Kin Help Only; b = 0.76, p < .001), but significantly less likely to be in Class 5 (SC Disability, Kin Help Primarily for HA; b = −1.21, p < .001) or Class 6 (MO Disability, Kin Help Primarily for HA; b = −1.06, p < .001).

Results of the zero-inflated Poisson regression analyses predicting whether care network membership in 2011 predicted subsequent unmet needs in 2012 (controlling for covariates and prior unmet needs; see Table 5 for the Incidence rate ratio (IRR)) indicated significantly greater unmet needs for older adults in Class 2 (Tri-domain Disability, Kin Help Only; IRR = 1.69, p = .01), Class 4 (Tri-domain Disability, Non-kin Support; IRR = 1.89, p = .016), and Class 5 (SC Disability, Kin Help Primarily for HA; IRR = 1.55, p = .007), compared to their counterparts in Class 7 (HA Disability, Kin Help for HA). No significant differences were observed between Class 1 (Tri-domain Disability, Mostly Paid + Kin Help), Class 3 (Tri-domain Disability, Extended Kin Support), and Class 6 (MO Disability, Kin Help Primarily for HA) compared to the reference Class 7.

Table 5.

Zero-inflated Poisson regression predicting unmet needs in 2012.

Predictors	Incidence rate ratio (Standard error)	95% Confidence interval
Intercept	1.40 (0.71)	(0.52, 3.79)
Total number of disabilities in 2011	0.97 (0.03)	(0.91, 1.03)
Total number of unmet needs in 2011	1.24 (0.05)***	(1.14, 1.35)
Age	0.99 (0.01)	(0.98, 1.00)
Income	1.00 (0.00)	(1.00, 1.00)
Gender
Male	Ref	Ref
Female	1.11 (0.11)	(0.92, 1.34)
Race
Non-White	Ref	Ref
White	1.00 (0.10)	(0.83, 1.21)
Education
High school or below	Ref	Ref
Above high school	1.04 (0.09)	(0.88, 1.23)
Living arrangement
Lives alone	Ref	Ref
Others in HHa	1.00 (0.21)	(0.66, 1.51)
Child in HH	0.98 (0.12)	(0.78, 1.24)
Spouse in HH	0.99 (0.12)	(0.78, 1.25)
Spouse + child in HH	1.05 (0.17)	(0.76, 1.43)
Has child outside HH
No	Ref	Ref
Yes	1.10 (0.12)	(0.88, 1.36)
Care network class
Class 1: Tri-domain Disability, Mostly Paid + Kin Help	1.43 (0.31)	(0.94, 2.18)
Class 2: Tri-domain Disability, Kin Help Only	1.69 (0.33)**	(1.15, 2.47)
Class 3: Tri-domain Disability, Extended Kin Support	1.23 (0.31)	(0.76, 2.01)
Class 4: Tri-domain Disability, Non-kin Support	1.89 (0.50)*	(1.13, 3.17)
Class 5: SC Disability, Kin Help Primarily for HA	1.55 (0.25)**	(1.13, 2.14)
Class 6: MO Disability, Kin Help Primarily for HA	1.29 (0.26)	(0.87, 1.92)
Class 7: HA Disability, Kin Help for HA	Ref	Ref
Total number of disabilities in 2011	−0.30 (0.11)**	(−0.51, −0.09)
Total number of unmet needs in 2011	−0.11 (0.17)	(−0.44, 0.22)
Intercept	1.52 (0.26)***	(1.01, 2.03)

Note. HA = household activities; HH = household; MO = mobility; Ref = reference group; SC = self-care activities.

^aOthers in HH refers to someone other than spouse or child.

^* p < .05;

^** p < .01;

^*** p < .001.

Discussion

Our study aimed to identify care network typologies among community-living older adults with functional limitations, examine factors predicting membership in these networks, and assess how network patterns influence prospective unmet care needs. Using NHATS data, we found seven unique care network configurations defined by multiple disability domains (SC, MO, HA) and caregiver types (close kin, extended kin, non-kin, paid help). These findings significantly extend previous research by explicitly combining multidomain disability profiles with multisource care networks, thereby allowing us to assess unmet care needs prospectively.

Our study identified three “high-need” caregiving configurations characterized by multidomain disabilities and four “domain-specific” networks focused on a single functional limitation domain. In line with previous studies (Freedman & Spillman, 2014; Hu et al., 2023; Spillman et al., 2020), our study findings indicate that informal kin caregivers were consistently central across all configurations, whereas formal paid assistance combined with kin care was available to only about 4% of older adults. Our study expands on earlier studies by Ali et al. (2022), Jacobs et al. (2014), and Leggett et al. (2025), confirming that exclusive reliance on informal support, whether from kin or non-kin, significantly increases the risk of unmet needs, especially for older adults with severe or multidomain disabilities.

More importantly, our study presents “care mismatch,” an underexplored concept wherein the kind of help given insufficiently addresses the specific functional limitations of the older adult. For example, we found that having help with household tasks, even though the help needed is for SC, was associated with higher unmet care needs 1 year later. Although earlier studies have generally recognized task mismatches (Jacobs et al., 2014), our study specifically shows their prospective consequences, underscoring the critical need for targeted assessments and tailored interventions by care managers or occupational therapists to ensure that provided support matches functional impairments.

Our findings on the predictors of network membership closely correspond with Andersen and Newman’s behavioral model of health service utilization. Older adults living with close relatives, such as spouses or adult children, predominantly belonged to the high-need, kin-exclusive caregiving networks. Interestingly, despite the anticipated advantage of close familial support, membership in these kin-only networks, characterized by limited paid care use, was associated with higher unmet needs. This paradox fits earlier research showing that caregivers constrained by filial obligations may delay seeking formal services until their caregiving resources are exhausted (Beach & Schulz, 2017). Our study highlights the importance of including paid services, especially for older adults living with significant functional or cognitive impairments.

Additionally, we found that greater educational level, rather than merely financial advantage, significantly predicted the use of paid services. These findings underscore the necessity for information accessibility as a vital enabling resource. Similar to earlier research, dementia also emerged as an important predictor of membership in high-need care networks, highlighting the complex, multifaceted needs associated with cognitive impairment (Freedman et al., 2024; Lai et al., 2025).

Our care network typology aligns with caregiving networks found in other studies. For example, our kin-only care networks (Class 2) resemble the “siloed” caregiving network described by Leggett et al. (2025). Both studies found that these networks had limited sharing of caregiving responsibilities with others and were associated with poorer health outcomes for the care recipient. The non-kin network in our study (Class 4) aligns closely with caregiving arrangements involving primarily neighbors or friends described in European research studies (Broese van Groenou et al., 2016; Nemmers et al., 2024). These networks, characterized by lower use of paid services, were also associated with higher unmet care needs. Furthermore, our results support the complementarity model by demonstrating that paid services can effectively augment informal family care rather than replace it, and when done so effectively, it can significantly lower adverse outcomes due to unmet care needs (Broese van Groenou et al., 2016; Swinkels et al., 2024).

Limitations and future research directions

Our analyses were controlled for various demographic and health variables, yet variables such as regional variations in service availability, caregiver burden, and use of assistive technology may have influenced our results. Future research should consider including these variables to examine these potential influences comprehensively. Additionally, because our study used data from 2011 to 2012 period, the results may not be completely reflective of recent developments in broader cultural developments, and policy shifts around home and community-based services, advancements in caregiver programs, training, and interventions. Future studies should replicate the findings of this study with more recent NHATS waves to investigate whether the care network configurations are stable and are adapting to older adults’ care needs. Also, although selecting a particular class (e.g., Class 7) as a reference category is typical practice in multinomial logistic regression, it may have masked some other comparisons between other network classes. Finally, although pooling immediate relatives into a single “kin” category facilitated stable estimations of latent classes, future research should differentiate among kin caregivers (e.g., adult children, grandchildren, siblings) to better capture variations in caregiving outcomes.

Study implications

Our findings suggest several actionable implications for practice and policy. First, kin-only networks could benefit from formal care coordination and expanded access to respite services to reduce caregiver strain and address unmet needs. Second, focused informational outreach aimed at non-kin caregivers would help raise knowledge of and use of paid services. Third, care mismatch situations emphasize the need for regular care needs assessments by health professionals to ensure alignment of caregiving activities with the older adults’ actual care needs. Fourth, policymakers should consider investing in local care coordination initiatives and funding caregiver education programs. Finally, our care network typology provides a useful framework for assessing caregiving arrangements and can inform more targeted resource allocation and the development of family support services that reflect the specific needs of different care networks.

Conclusion

By explicitly linking multidimensional disabilities and multisource caregiving within a unified analytical framework, our study offers valuable insights to enhance caregiving policy, intervention design, and resource allocation. Given the growing population of older adults with care needs, proactive strategies guided by this study’s findings will be essential to support caregiving networks and optimize care outcomes effectively.

Supplementary material

Supplementary material is available online at Innovation in Aging (https://academic.oup.com/innovateage).

Funding

None declared.

Conflicts of interest

None declared.

Data availability

Data are publicly available at www.nhata.org. This study was not preregistered.