The heterogeneity of the homebound: A latent class analysis of a national sample of homebound older adults

Abstract

Background:

Homebound status is a final common pathway for people with a variety of diseases and conditions. There are 7 million homebound older adults in the United States. Despite concerns regarding their high healthcare costs and utilization and limited access to care, the unique subsets within the homebound population are understudied. Better understanding of distinct homebound groups may enable more targeted and tailored approaches to care delivery. Therefore, in a nationally representative sample of homebound older adults we used latent class analysis (LCA) to examine distinct homebound subgroups based on clinical and sociodemographic characteristics.

Materials and Methods:

Using data from the National Health and Aging Trends Study (NHATS) 2011–2019, we identified 901 newly homebound persons (defined as never/rarely leaving home or leaving home only with assistance and/or difficulty). Sociodemographic, caregiving context, health and function, and geographic covariates were derived from NHATS via self-report. LCA was used to identify the existence of distinct subgroups within the homebound population. Indices of model fit were compared for models testing 1–5 latent classes. Association between latent class membership and 1 year mortality was examined using a logistic regression.

Results:

We identified four classes of homebound individuals differentiated by their health, function, sociodemographic characteristics, and caregiving context: (i) Resource constrained (n = 264); (ii) Multimorbid/high symptom burden (n = 216); (iii) Dementia/functionally impaired (n = 307); (iv) Older/assisted living (n = 114). One year mortality was highest among the older/assisted living subgroup (32.4%) and lowest among the resource constrained (8.2%).

Conclusions:

This study identifies subgroups of homebound older adults characterized by distinct sociodemographic and clinical characteristics. These findings will support policymakers, payers, and providers in targeting and tailoring care to the needs of this growing population.

INTRODUCTION

The homebound are a large and growing population and can benefit from home-based medical care

There are over 7 million homebound U.S. older adults who never or rarely leave home or do so only with assistance or difficulty, thereby meeting the Medicare definition of homebound.¹ Although homebound individuals account for 5.7% of adults over 70, they incur around 11.0% of all Medicare spending.² In 2020, the proportion of older adults who are homebound doubled, likely due to the COVID-19 public health emergency.^2,3 Despite having significant multimorbidity, poor function, high risk of hospitalization and mortality, and being less likely to access office-based primary and specialty care, just 12% of this population receive longitudinal primary care in the home.^2,4 In response to their complex medical and social needs coupled with high acute care expenditures, multiple initiatives have begun to develop and scale models of home-based medical care for this growing population.⁵ These include home-based primary care, which focuses on longitudinal primary/preventive care; home-based palliative care, which encompasses episodic complex symptom management and advance care planning; and other episodic care models to respond to clinical crises, including mobile integrated health-community paramedicine, emergency department care at home, hospital at home, rehabilitation/skilled nursing at home, and transitional care.⁶ Although these home-based care models have been tested individually with promising findings, they are yet to be evaluated as an integrated system and implemented at scale.⁷

The homebound population is likely heterogeneous in terms of sociodemographic and clinical factors

Recent studies have identified a range of sociodemographic factors (e.g., income, race, and ethnicity) and clinical factors (e.g., comorbidity burden, pain) that independently contribute to an individual’s risk of becoming homebound.^8–10 Moreover, the risk of outcomes after becoming homebound, including mortality, differs by age, race and ethnicity, function, and comorbidities.^11,12 Collectively, these studies signal the possible existence of distinct subgroups among the population of homebound older adults; characterization of latent heterogeneity among the homebound will help inform efforts to develop and tailor home-based care interventions that meet the needs of homebound older adults, better target existing home-based care interventions, and plan home-based medical care around the needs of this population. Our aim is to determine the existence of distinct subgroups of homebound older adults using latent class analysis (LCA) and determine differential association with mortality by subgroup.

MATERIALS AND METHODS

Study population

Our data were drawn from the nationally representative National Health and Aging Trends Study (NHATS), a longitudinal study of late-life disability and function among Medicare beneficiaries 65 years of age and older.¹³ NHATS conducts annual in-person interviews that began in 2011. Study participants are asked detailed questions about how they performed activities in the month before interview, as well as their socioeconomic status, medical comorbidities, and home environment. We included individuals who became newly homebound between 2012 and 2019; we included individuals in the sample once, the first time they became homebound. (We excluded individuals newly homebound in 2020, given the confounding impact of the COVID-19 pandemic which resulted in an unprecedented increase in the number of homebound¹¹). We identified 7042 participants from the NHATS cohort who were community dwelling and not homebound in 2011. Individuals were identified as newly homebound the first year they met the following criteria using previously published constructs: (i) never or rarely (once a week or fewer) leaving their home in the past month or (ii) leaving their home at least twice a week but needing help and/or experiencing difficulty. These criteria are consistent with the Medicare definition of homebound.¹ We identified 901 NHATS older adults who became homebound between 2012 and 2019.

Measures

All participant measures were taken from self or proxy report at the interview at which homebound status was identified. We included factors that have previously been shown to be associated with risk of becoming homebound and/or outcomes after becoming homebound, based on the existing literature.^8–11 For the purpose of the LCA, we transformed continuous measures into categorical measures.

Sociodemographic characteristics

We included sex, age (85 or above vs. less than 85), race and ethnicity (white non-Hispanic, Black non-Hispanic, Hispanic, other), marital status (married vs. all else), education (high school or above vs. less than high school), income (below 100% federal poverty level vs. all else), Medicaid status, and financial strain. Financial strain was defined as self-report of any one of the following: lacking the money to pay rent or the mortgage, or utility bills, or medication and prescription bills, or skipping meals because there was not enough money to buy food.¹⁴

Caregiving context characteristics

Caregiving context variables included living arrangement (assisted living facility, living alone), presence of a paid caregiver, and caregiver network size (count). Assisted living included: assisted living facility/continuing care retirement community, group homes, board and care homes, and supervised housing. If an NHATS participant indicated that help was needed with a particular self-care, mobility, or household task for health or functioning reasons, information about each caregiver is obtained, from which we were able to establish the presence of a paid caregiver and the caregiver network size.

Health status and function characteristics

Self-rated health was dichotomized as fair or poor versus excellent, very good, or good. Frailty was defined based on assessment of weight loss, exhaustion, low physical activity, slow walking speed, and low grip strength¹⁵; participants who met three out of five criteria were classified as frail. Dependence in two or more activities of daily living (ADL) (eating, getting out of bed, getting around inside the house, toileting, dressing, showering/bathing) was based on the report that the participant received help with the activity in the prior month. Sensory impairment was defined as self-report of current vision and/or hearing difficulty. Probable dementia was determined according to NHATS comprehensive and validated criteria, which includes self-report of dementia, proxy responses to the Alzheimer’s Disease (AD)-8 screening tool, and cognitive testing that assesses memory, orientation, and executive function.¹⁶ Participants with a score of three or more out of six on a depression screen based on the PHQ-2 screening questions (i.e., “Over the last month, how often have you had little interest or pleasure in doing things, and/or felt, down, depressed, or hopeless?”) were classified as having probable depression.¹⁷ We included count of self-reported chronic conditions from the following list: heart attack, stroke, cancer, hip fracture, heart disease, high blood pressure, arthritis, osteoporosis, diabetes, lung disease, anxiety. We included report of activity-limiting pain, breathing problems, fall in the past month, and hospitalization in the past year. We constructed “activity limiting pain” from two items in NHATS. NHATS asks participants “In the last month, {have you/has {he/she}} been bothered by pain?”. If they respond affirmatively, participants are asked “In the last month, has pain ever limited your activities?”. We included all participants who answered affirmatively to both these items as having activity-limiting pain. All other respondents were recorded as not having activity-limiting pain; this included participants not bothered by pain in the last month and participants bothered by pain that never limited activities.

Geographic characteristics

NHATS participants were classified into census region based on their county of residence. We also included measures of interviewer observed household and neighborhood physical disorder¹⁸ as follows: report of any of peeling paint, pests, broken furniture or floorboards, tripping hazards was classified as household disorder (inside); report of any of broken windows, foundation, bricks/siding, roof, steps was classified as household disorder (outside); report of any of litter/glass/graffiti/vacant houses/vacant stores/foreclosures was classified as neighborhood physical disorder. We included a measure of neighborhood social cohesion (self/proxy reported) derived from responses to three items: people in this community can be trusted; people in this community are willing to help one another; people in this community know each other very well.¹⁸ Social cohesion was defined as responding “agree a lot” to all three items. Finally, we recorded neighborhood ties as the number of years in residence.

Statistical analysis

We used LCA to identify distinct subgroups within the population of homebound older adults. LCA is a form of person-centered mixture modeling which identifies latent subpopulations (classes) based on the pattern of their responses to observed categorical variables.¹⁹ To inform our final selection of indicator variables to define latent classes, we included factors associated with risk of homeboundness and/or outcomes in homebound older adults based on the literature and our clinical experience; these included sociodemographic characteristics, caregiving context characteristics, and health status and function characteristics, as described previously.^8–11 To determine the best class solution, we compared one-class to five-class models, and then examined model fit based on two indices of model fit (Aikake Information Criterion and Bayesian Information Criterion), model interpretability, and clinical judgment. After selecting the latent class model, we assigned each participant to a specific class based on their posterior class membership probabilities. We assigned names to the latent classes based on the members’ characteristics. We conducted a logistic regression to examine the association between latent class subgroup and mortality 1 year after becoming homebound. We used the group with the lowest average mortality as the reference group. All analyses were conducted using Stata, version 16 (StataCorp). All analyses used survey weights provided by NHATS to account for differential probabilities of selection and non-response.

The Johns Hopkins University Institutional Review Board approved the NHATS protocol. The Icahn School of Medicine at Mount Sinai’s Institutional Review Board (IRB) approved the present study.

RESULTS

Baseline sample characteristics

Consistent with other studies of homebound persons,¹ the sample was predominantly female (69%). Seventy three percent were white, non-Hispanic; 39% were aged 85 and older (Table S1). The mean number of comorbidities was 4.48; 40% had probable dementia, and 34% had depression; 50% reported having had activity-limiting pain in the past month. Although 73% of the sample met the definition of frailty, 44% required assistance in 2 or more ADL. Thirty seven percent lived alone, 11% in an assisted living facility, and 21% received paid care. Half had been hospitalized in the past year, and almost one fifth reported a fall in the past month.

LCA reveals a four-class solution

The LCA demonstrated the existence of distinct subgroups within the population of homebound older adults and indices of model fit supported a four-class model (Table 1). The heat map (Figure 1) illustrates the characteristics of members of the four classes based on the variables included in the LCA (i.e., probability of a given characteristic conditional on class membership). Table S1 includes sample characteristics by latent class for all variables. We assigned names to the four classes based on these characteristics as described below:

TABLE 1.

Model fit criteria for different class solutions from latent class analysis.

Model	n	Log-likelihood	df	AIC	BIC
1 class	901	−8798.9	17	17631.8	17713.5
2 class	901	−8550.5	35	17171.0	17339.1
3 class	901	−8396.9	53	16899.7	17154.3
4 class	901	−8311.4	71	16764.9	17105.9

Note: AIC and BIC represent model fit criteria used to inform selection of the final latent class model (the 4 class model having the lowest AIC and BIC).

Abbreviations: AIC, Aikake Information Criterion; BIC, Bayesian Information Criterion; df, degrees of freedom.

FIGURE 1.

Heatmap of latent class analysis (LCA) of newly homebound older adults, n = 901. Data derived from LCA of a sample of 901 adults identified as newly homebound between 2012 and 2019 in the National Health and Aging Trends Study. Participants assigned to a specific class based on their posterior class membership probabilities; color gradient indicates the probability of a given characteristic conditional on class membership.

Resource constrained (n = 264, 29.3%)

Characterized by relatively low overall need for assistance with ADLs, comorbidity (especially cardiopulmonary), and symptom burden, this subgroup also had the lowest prevalence of recent falls and hospitalizations. Nonetheless, 71% of this subgroup were unable to walk 6 blocks, and the prevalence of frailty (49%) and probable dementia (40%) was non-trivial. Individuals in this subgroup were most likely to be from a racially minoritized population (20% Hispanic, 15% Black, non-Hispanic), report low income (45% reported income less than 100% of the federal poverty level), and to live in a home with evidence of household disorder outside.

Multimorbid/high symptom burden (n = 216, 24.0%)

This subgroup was characterized by high multimorbidity (97% with 3 or more comorbidities) and symptom burden (68% with activity-limiting pain, 70% with dyspnea). While individuals in this subgroup were most likely to have cardiopulmonary disease (63% lung disease, 30% heart disease), they were least likely to have probable dementia (11%). Over a third of individuals in this group lived alone; whereas only 12% required assistance with 2 or more ADLs, 89% were unable to walk 6 blocks.

Dementia/functionally impaired (n = 307, 34.0%)

Among this subgroup, 65% had probable dementia and 97% needed assistance in 2 or more ADL; 91% were classified as frail. Physical and psychological symptoms were prevalent in this subgroup: 64% reported activity-limiting pain, 53% had probable depression. Individuals in this subgroup were most likely to have had a fall in the past month (33%) and most likely to have been hospitalized in the prior year (68%). Caregiver support was highest for people in this subgroup: 33% received care from a paid caregiver and average caregiver network size was 3.17.

Older/assisted living (n = 114, 12.7%)

Individuals in this group, 75% of whom were over the age of 85 and 74% of whom were living in an assisted living facility, were the most likely to have probable dementia (70%). While 67% required assistance with 2 or more ADLs and 74% were frail, overall comorbidity and symptom burden were lower than the multimorbid/high symptom burden and the dementia/functionally impaired group.

Association between homebound class and 1 year mortality after becoming homebound

Within a year of becoming homebound, 195/901 (18%) participants died. The older/assisted living subgroup had the highest one-year mortality (32.4%) and the resource-constrained subgroup the lowest (8.2%) (Table S2). Compared with participants in the resource-constrained category, participants in the dementia/functionally impaired (odds ratio 4.30, 95% confidence interval 2.71, 6.84) and the older/assisted living subgroup (4.74, 2.73, 8.25) had significantly higher odds of death within a year of becoming homebound (Figure 2, Table S2).

FIGURE 2.

Plot of association between latent class subgroup and 1 year mortality, n = 901. Data derived from a sample of 901 adults identified as newly homebound between 2012 and 2019 in the National Health and Aging Trends Study; reference category = resource-constrained subgroup. OR, odds ratio, error bars indicate 95% confidence interval, dashed line indicated OR = 1.

DISCUSSION

In a nationally representative sample, we observed four subgroups of homebound older adults differentiated by their health and function characteristics in addition to their sociodemographic and caregiving context. Although others have described distinct subsets among populations of community-dwelling older adults,²⁰ to our knowledge, this is the first study to identify distinct subgroups of homebound older adults.

The findings of this study are highly relevant to medical practices, health systems, payers, and policymakers aiming to improve the value of care for this population. The identification of subgroups with distinct health/function profiles (and thus care needs) aligns with recent calls for the development of a home-based care ecosystem comprised of a diverse suite of integrated clinical services tailored to the heterogeneous needs of this population.⁷ Clinical services within this home-based care delivery system may include home-based primary care, home-based palliative care, rehabilitation-at-home, emergency department care at home, hospital at home, mobile integrated health/community paramedicine, and new models yet to be developed. Clinicians and other stakeholders may consider how the subgroups identified may require a different mix of these clinical services, as proposed below.

The resourceconstrained group may benefit from better access to home-based primary care or a home-based medical co-management model while at the same time accessing in-home rehabilitation services to maintain and improve physical function and prevent falls.^21,22 Social workers within home-based primary care programs would serve to facilitate connections to community-based organizations to address and/or mitigate the impact of resource constraints.²³ Establishing robust and mutually beneficial, bidirectional partnerships between home-based primary care practices and community-based organizations will be central to addressing the needs of this group and advancing health equity.

The multimorbid/high symptom burden group may benefit from home-based primary care for active management of comorbidities with home-based palliative care co-management as needed for symptom management (pain, dyspnea) alongside behavioral health to support symptom management. Mobile integrated health-community paramedicine or hospital at home may be particularly beneficial for this group in managing exacerbations of cardiopulmonary disease and symptom crises.²⁴ As with the resource-constrained group, this group should be screened for unmet need for assistance with ADLs and instrumental ADL and considered for in-home rehabilitation services.

The dementia/functionally impaired subgroup should be considered for home-based palliative care either alongside or as a transition from home-based primary care, to support effective management of prevalent physical and psychological symptoms, facilitate complex medical decision-making, and provide caregiver support.^25,26 Services for this subgroup should also address the high need for assistance with ADLs (e.g., by integrating long-term services and supports). Given the high likelihood of intercurrent clinical crises in this subgroup (e.g., fall, hospitalization), mobile integrated health-community paramedicine may be especially beneficial.

Finally, the older/assisted living subgroup may benefit from initiatives to improve longitudinal medical care within assisted living settings. A recent study has reported significant growth of fee-for-service Medicare home-based medical care within assisted living facilities²⁷; further research is needed to examine the needs of this population and how to effectively implement services within assisted living facilities.

Although some programs have already established comprehensive home-based medical care programs incorporating multiple different service lines, for those planning to develop such a model, the results of this study will be especially timely. In addition to guiding development of services, latent heterogeneity may also contribute to challenges around demonstration of cost savings in evaluation of home-based primary care models, including the Independence at Home demonstration project.²⁸ Future analyses linking NHATS data to Medicare claims would help inform whether the heterogeneity described is accompanied by differences in utilization and cost across subgroups and the resultant implications for economic evaluations.

Activity-limiting pain affected 50% of all homebound older adults in our sample, and 68% of those in the multimorbid/high symptom burden subgroup. Pain significantly impacts emotional wellbeing^29–31 and physical function⁹ among homebound individuals. Homebound older adults are at particular risk of untreated pain given the limited provision of in-home medical and behavioral health care and the challenges in assessing and managing pain in the context of dementia. Given the high prevalence of pain, care models for homebound older adults should both address and be evaluated on their impact on measures of pain³² and other bothersome symptoms; development of reliable and valid patient-proxy reported outcome measures to support assessment of care quality in the context of dementia is an urgent need.

The resource-constrained subgroup had the lowest overall comorbidity and symptom burden and need for assistance with ADLs; nonetheless, frailty prevalence was high (49%). This subgroup was also most likely to report being from a racial or ethnically minoritized group, to live in a low-income household, and a house with household disorder outside (including broken steps), thus raising the possibility that socioeconomic disadvantage imposes homebound status in the context of exacerbating various social determinants of health. Prior studies have demonstrated an elevated risk of becoming homebound among low-income individuals⁸; future research should further explore the mechanism of this possible effect and identify mutable targets for intervention.

Overall, 1 year mortality in the cohort was 18% and among the subgroup with the highest mortality risk, older/assisted living, 32%. This finding is consistent with prior studies,³³ confirms that the majority of homebound older adults may be ineligible for hospice, and furthermore emphasizes the need to develop a home-based care infrastructure with the capacity to support the palliative care needs of this growing population over multiple years.⁷

There are some limitations to report. As the LCA is cross-sectional, the extent to which people move between classes over time and the distribution of class membership among the prevalent (rather than newly/incident) homebound was not established. We were unable to include all caregiver-related and geographic-related variables in modeling which may be important in characterization of subgroups. Finally, LCA is probabilistic, and as such, the number of members in each subgroup are estimates; additionally, these estimates may be sensitive to the thresholds used for derivation of categorical variables. Despite these limitations, this study was conducted in a nationally representative sample, enhancing the generalizability of the findings.

Prior studies have illuminated the multiple clinical and social factors for becoming homebound both among community-dwelling older adults with and without dementia, together with the heterogeneity in outcomes among homebound older adults.^8–11 This study adds to this literature in identifying, for the first time, distinct subgroups in a population of newly homebound older adults and demonstrating their relationship to mortality. In the context of limited resources, these results from a nationally representative sample will support policymakers, payers, medical practices, and health systems to target and tailored services according to population need.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section at the end of this article.

Table S1. Characteristics of newly homebound older adults between 2012 and 2019 overall and by latent class, n = 901

Table S2. Association between latent class subgroup and mortality 1 year after incident homebound status, n = 901

Key points

• There are subgroups of homebound older adults characterized by distinct sociodemographic and clinical characteristics.

Why does this paper matter?

Policymakers, payers, and providers need to consider this heterogeneity when planning services to meet the needs of this growing population.