Middle-age healthcare spending and utilization at the intersection of comorbidities, functional limitation, and race/ethnicity

Abstract

Background and Objectives

Middle age is a pivotal developmental life stage marked by health transitions, including the onset of major chronic diseases and functional limitations. Social determinants of health, particularly race/ethnicity, can play a significant role in accentuating the effects of comorbidity and functional limitations. This study examines how comorbidity, functional limitations, and race/ethnicity influence patterns of health services use and spending among middle-aged adults.

Research Design and Methods

We used pooled yearly cross-sectional data (Medical Expenditure Panel Survey, 2008-2022; excluding 2020) from middle-aged adults (ages 50-64). We derived comorbidity phenotypes using latent class analyses (LCA) and generalized linear regression to examine their associations with health services outcomes, considering the vulnerabilities introduced by functional limitations and race/ethnicity.

Results

LCA supported a 3-class solution: low prevalence of chronic conditions (63.4%), hypertensive/arthritis/joint pain (29.3%), and complex cardiovascular (C-CVD, 7.3%). Individuals with C-CVD had elevated levels of healthcare spending and use, including problematically high levels of emergency department and inpatient hospitalizations, despite higher use of office-based visits (OBVs). We found that Non-Hispanic Black adults (NHBs) and Hispanic adults had lower health spending compared to Non-Hispanic White adults (NHWs) and were less likely to use OBVs. NHBs had a higher propensity for ED use, whereas Hispanic adults were less likely to be hospitalized. Physical and instrumental/activities of daily living (I/ADL) limitations increased OBVs for both NHBs and NHWs, while I/ADLs increased hospitalizations among Hispanic adults.

Discussion and Implications

These findings underscore the importance of targeted healthcare for high-risk patient populations for mitigating excess health burdens.

Innovation and Translational Significance:

This study adds to current work on the correlates of health disparities by examining how comorbidity, functional limitations, and race/ethnicity influence patterns of health services use and spending among middle-aged adults. In our analysis of a nationally representative sample, we identified that co-occurring comorbidities and functional limitations in midlife significantly impact the utilization of healthcare resources and healthcare spending among U.S. racial/ethnic minorities. Targeted interventions focusing on these intersectional risks, especially those tailored for individuals with co-occurring complex comorbidities and functional limitations, can yield significant efficiencies in the healthcare system and potentially decrease the burden of disease among the impacted.

Introduction

Middle age may be marked by the onset of major chronic health conditions (eg, hypertension and arthritis) along with functional limitations. Recently, we have seen an increase in both overall and cause-specific mortality rates in young and middle-aged adults.¹ Social determinants of health (SDOH), including race/ethnic classifications, can play an important role in accentuating the effects of comorbidity and functional limitations. Few studies, however, have examined healthcare use and spending at the intersection of comorbidity, functional limitations, and race/ethnicity in middle age.

In the United States and globally, comorbidity is increasingly prevalent at a younger age. Quinones, Newsom² found that more than 50% of individuals over age 50 experience multiple co-occurring chronic conditions. These chronic conditions increase the likelihood of functional limitations. Bowling et al. (2019) estimated that slightly more than a third of middle-aged adults have some functional limitations (34%), with 1 in 10, reporting mobility limitations (11%), and nearly 2 in 10 reporting difficulties with basic activities of daily living (BADL) (15%) and instrumental activities of daily living (IADL) (17%). Critically, three-quarters of participants with limitations had 2 or more chronic conditions, and nearly a third had 4 or more.³

Functional limitations, at younger ages of onset, can complicate the effects of chronic conditions by triggering disabling conditions (or vice versa), resulting in early exit from the workforce, threatening overall well-being,⁴ distorting access to health-enabling healthcare resources, and reducing health-related quality of life.⁵

Despite decades of research, disparities in care access and care quality in the United States persist. Individuals with functional limitations are more likely to have restricted access to high-quality care and to face provider bias and reluctance for care.⁶ SDOH, including race/ethnicity and socioeconomic status, can further disadvantage individuals with functional limitations, thereby widening existing health disparities.⁷ Healthcare disparities at the intersection of race/ethnicity, functional limitations, and severe comorbidity are largely understudied. Recent policy efforts (eg, Affordable Care Act [ACA]) have established mandates to help mitigate and relieve some of the barriers to access to high-quality and affordable care for high-risk individuals. Yet, many obstacles and systemic problems endure.⁶

Study purpose and expectations

This study adds to current work on the correlates of health disparities by examining how comorbidity, functional limitations, and race/ethnicity influence patterns of health services use and spending among middle-aged adults. We apply a modified Andersen’s Health Behavior (ABM) framework to examine time-pooled cross-sectional data collected by the Medical Expenditures Panel Survey (MEPS). First, we test variability in use of and spending on healthcare services among middle- and late-middle-aged adults by comorbidity, functional limitations, and race/ethnicity. Second, we examine whether, and to what extent, worse health outcomes and functional limitations status explain differences in use among race/ethnic minorities. Third, we examine the role of predisposing, enabling, and health needs factors in explaining differences by morbidity, functional ­limitations, and race/ethnicity.

Methods

Data

To coincide with the amendment of the Americans with Disabilities Act (ADA) in 2008 and to ensure the availability of consistent data across all applicable cross-sections, we use individual-level data from the yearly household component files of the MEPS (2008-2022, excluding 2020 to avoid direct confounding with COVID data collection). MEPS collects data, publicly available for analyses, on non-institutionalized individuals in the United States, covering demographic, socioeconomic, health insurance, healthcare access, healthcare quality, health services use, and health spending. The data and more details are available at https://meps.ahrq.gov/mepsweb/.

Study subpopulation

Our analyses are restricted to include individuals in middle age and late-middle age defined as 50-64 years. To ensure an appropriate sample size within this age group, we further restrict the analyses to participants in one of 3 racial/ethnic groupings: Non-Hispanic White adults (NHW), Non-Hispanic Black adults (NHB), and Hispanic adults. We further exclude individuals with missing values on any of the covariables of interest (see below). The analytic sample size is N = 63,191 (unweighted). All analyses are weighted by applying the probability weights generated by MEPS, using statistical procedures from the survey analyses suite in Stata 17.1.

Healthcare expenditures and services

Outcomes include total health expenditures (in $2022) and 3 health services use measures: total yearly office-based visits, ED visits, and inpatient hospitalizations. Healthcare expenditures include all categories of spending (inpatient, outpatient, office visits, and emergency room). Additionally, we include these services as they constitute over half of the annual national health expenditures.

Comorbidities

The Agency for Healthcare Research and Quality (AHRQ) identifies 11 conditions as leading causes of burden, cost, and mortality. These conditions include (1) high blood pressure, (2) diabetes, heart disease including (3) coronary, (4) myocardial infarction, (5) angina, and (6) any “Other” heart problems, (7) emphysema, (8) stroke, (9) cancer, and (10) arthritis. Additionally, we measure (11) depression, following existing standards in population survey data, by using threshold values of 3+ on the PHQ-2. We conduct latent class analyses (LCA) to generate data-driven phenotypes, based on the above-listed variables, representing probabilistic population clusters of comorbidities.⁸ We have included more details on LCA in ­Supplementary Figure 1 (see online supplementary material for a color version of this figure).

Functional limitations

We distinguish between 2 functional limitations constructs: physical limitation and limitation in either IADLs or BADLs (henceforth, I/ADLs). We include details on how these 2 constructs are ascertained in the Supplementary Methods section in Supplementary Material.

Race/ethnicity

Due to sample size considerations, we focus on individuals self-classifying as one of 3 groups: NHW, NHB, and Hispanic adults.

Covariables

We use literature review and ABM to guide covariable selection. ABM is a multilevel model that considers individual and contextual level factors related to health services use and spending. ABM groups the factors into predisposing, enabling, and need factors. In our analysis, predisposing factors (in addition to race/ethnicity) include sex, marital status, and geographic region; enabling factors include educational attainment, household income, employment, and insurance status; finally, severity of mental health needs is measured using the mental component scores (MCS) of the 12-Item Short Form Survey (SF-12).

Data analyses

We conduct our analyses in 3 steps. First, we generate descriptive statistics to characterize the target population, overall and by race/ethnicity (Supplementary Table 1) and by comorbidity classifications (Supplementary Table 2). Second, we fit survey generalized linear regression models to estimate the associations between comorbid classification, functional limitations, and race/ethnicity and each of the health services and expenditures outcomes of interest (Tables 1 and 2; detailed in Supplementary Tables 3-6). To examine total healthcare expenditures, we fit 2-part models using logistic regression to test the propensity of spending and a generalized linear model with a Gaussian distribution and a log link to model the extent of spending among spenders. To model the number of yearly office-based visits, ED visits, and inpatient hospitalizations, we use negative binomial regression. For each outcome, we estimate 10 models (5 for each functional limitation measure, that is, physical limitations and I/ADLs): (1) accounting for comorbidity class, functional limitation (physical and I/ADLs, independently), and race/ethnicity to estimate covaried associations with the outcome of interest, (2) adjusting for predisposing and (3) enabling factors, as specified, in the covariables section above, (4) additional adjustment for SF-12 scores, and (5) testing for potential interaction effects/moderation in the associations between each functional limitation measure, independently, and the outcomes by (a) comorbid classification and (b) race/ethnicity. The fully adjusted models (4 and 5) include the fixed effect of a linear yearly indicator. Estimates for model 4 for the health expenditures and health services use outcomes are included in Tables 1 and 2, respectively. Estimates from each of the fit models are included in Supplementary Tables 3-6. Finally, we calculate and plot post-hoc marginal means (for continuous expenditure and count of health services use measures) derived from each of the estimated models. These plots include the point estimates and their 95% confidence intervals (Figures 1 and 2).

Table 1.

Estimates of the association between functional limitations, comorbidity classifications, and race/ethnicity and total health expenditures.

Variable	Any spending: Logistic	Total among spenders: GLM (Gaussian)
	OR [95% CI]	b [95% CI]
A: Two-part model with physical limitations as the exposure
No physical limitation	Ref	Ref
Physical limitation	1.99*** [1.67, 2.38]	8,709*** [7,751, 9,666]
Low prevalence of chronic conditions	Ref	Ref
Hypertensive arthritis	4.08*** [3.58, 4.65]	4,470*** [3,821, 5,119]
Complex cardiovascular	5.47*** [4.04, 7.41]	11,260*** [9,872, 12,648]
Non-Hispanic White	Ref	Ref
Non-Hispanic Black	0.56*** [0.49, 0.63]	-913* [-1,702, -125]
Hispanic	0.66*** [0.58, 0.75]	-1,208** [-1,927, -490]
Constant	-	5,491* [1,147, 9,836]
B: Two-part model with I/ADLs as the exposure
No I/ADLs	Ref	Ref
I/ADLs	5.59*** [3.41, 9.18]	18,343*** [15,313, 21,374]
Low prevalence of chronic conditions	Ref	Ref
Hypertensive arthritis	4.37*** [3.85, 4.96]	5,582*** [4,945, 6,218]
Complex cardiovascular	5.90*** [4.35, 7.99]	12,264*** [10,907, 13,622]
Non-Hispanic White	Ref	Ref
Non-Hispanic Black	0.55*** [0.49, 0.62]	-1,137** [-1,920, -354]
Hispanic	0.65*** [0.57, 0.73]	-1,581*** [-2,294, -867]
Constant	-	5,298* [1,038, 9,558]

Source: Public use data files from the Medical Expenditure Panel Survey (MEPS) for 2008-2022, excluding 2020 to avoid direct confounding with COVID data collection.

Abbreviation: I/ADLs = instrumental/activities of daily living (this is a combined measure of Instrumental and Basic ADLs).

Estimates (odds ratios [OR], and beta coefficients [b]) are derived from 2-part models (logistic and generalized linear model [GLM]) and presented with their 95% confidence intervals (CI). Models include the functional limitations exposure (physical limitation and I/ADLs, independently), comorbidity classifications, and race/ethnicity. The estimates adjust for age, sex, marital status, region of residence, insurance coverage, employment, education, income (poverty to income ratio), SF-12 mental score, and a linear trend for time. * = p-value<0.05; ** = p-value<0.01; *** = p-value<0.001.

Table 2.

Estimates of the association between functional limitations, comorbidity classifications, and race/ethnicity and healthcare use measures.

Variables	Outcomes
	# Office-based provider visits	# ED visits	# Hospital discharges
	IRR [95% CI]	IRR [95% CI]	IRR [95% CI]
A: Physical limitations as the exposure
No physical limitation	Ref	Ref	Ref
Physical limitation	1.70*** [1.62;1.78]	1.56*** [1.43;1.69]	1.88*** [1.71;2.08]
Low prevalence of chronic conditions	Ref	Ref	Ref
Hypertensive arthritis	1.46*** [1.41;1.52]	1.77*** [1.63;1.91]	2.25*** [2.02;2.51]
Complex cardiovascular	1.57*** [1.46;1.68]	2.63*** [2.38;2.91]	4.35*** [3.86;4.91]
Non-Hispanic White	Ref	Ref	Ref
Non-Hispanic Black	0.78*** [0.73;0.84]	1.22*** [1.13;1.32]	1.05 [0.95;1.16]
Hispanic	0.79*** [0.73;0.85]	0.99 [0.90;1.08]	0.86* [0.77;0.97]
B: I/ADLs as the exposure
No I/ADLs	Ref	Ref	Ref
I/ADLs	1.74*** [1.57;1.92]	1.66*** [1.48;1.86]	2.15*** [1.86;2.48]
Low prevalence of chronic conditions	Ref	Ref	Ref
Hypertensive arthritis	1.58*** [1.52;1.64]	1.89*** [1.75;2.05]	2.49*** [2.24;2.77]
Complex cardiovascular	1.69*** [1.58;1.81]	2.80*** [2.54;3.09]	4.79*** [4.25;5.39]
Non-Hispanic White	Ref	Ref	Ref
Non-Hispanic Black	0.77*** [0.72;0.83]	1.21*** [1.12;1.30]	1.03 [0.93;1.15]
Hispanic	0.77*** [0.71;0.83]	0.96 [0.87;1.05]	0.82*** [0.73;0.92]

Source: Public use data files from the Medical Expenditure Panel Survey (MEPS) for 2008-2022, excluding 2020 to avoid direct confounding with COVID data collection.

Abbreviation: I/ADLs = instrumental/activities of daily living (this is a combined measure of instrumental and basic ADLs).

Estimates (incidence rate ratios [IRR]) are derived from negative binomial regressions and presented with their 95% confidence intervals (CI). Models include the functional limitations exposure (physical limitation and I/ADLs, independently), comorbidity classifications, and race/ethnicity. The estimates adjust for age, sex, marital status, region of residence, insurance coverage, employment, education, income (poverty to income ratio), SF-12 mental score, and a linear trend for time. * = p-value<0.05; *** = p=value<0.001

Figure 1.

Post-hoc marginal estimates of the association of instrumental and basic activities of daily living (I/ADL) and health expenditures/use and their 95% confidence intervals. Estimates are based on 2-part models (for expenditures) and negative binomial regression (for office visits, emergency department (ED), and inpatient hospitalizations). LCC = low prevalence of chronic conditions; NHW = non-Hispanic White Adults; NHB = non-Hispanic Black Adults; H = Hispanic adults. The partially adjusted (“P. Adjusted”) model includes instrumental and basic activities of daily living (I/ADL), comorbidity classifications, and race/ethnicity. The “Adjusted” model includes predisposing (age, sex, marital status, region of residence), and enabling factors (insurance coverage, employment, education, income (poverty to income ratio) (we estimate an additional model with predisposing factors only, not presented in this figure, but shown as M2 results in supplementary tables). The “F. Adjusted” additionally adds the SF-12 mental score and a linear trend for time. Finally, the “Adjusted Modified” includes interactions between the functional limitation exposure and (A) comorbid classifications and (B) race/ethnicity. Source: Public use data files from the Medical Expenditure Panel Survey (MEPS) for 2008-2022, excluding 2020 to avoid direct confounding with COVID data collection.

Figure 2.

Post-hoc marginal estimates of the association of physical limitations and health expenditures/use and their 95% confidence intervals. Estimates are based on 2-part models (for expenditures) and negative binomial regression (for office visits, emergency department [ED], and inpatient hospitalizations). LCC = low prevalence of chronic conditions; NHW = non-Hispanic White Adults; NHB = non-Hispanic Black Adults; H = Hispanic adults. The partially adjusted (“P. Adjusted”) model includes physical limitations, comorbidity classifications, and race/ethnicity. The “Adjusted” model includes predisposing (age, sex, marital status, region of residence), and enabling factors (insurance coverage, employment, education, income (poverty to income ratio) (we estimate an additional model with predisposing factors only, not presented in this figure, but shown as M2 results in supplementary tables). The “F. Adjusted” additionally adds the SF-12 mental score and a linear trend for time. Finally, the “Adjusted Modified” includes interactions between the functional limitations exposure and (A) comorbid classifications and (B) race/ethnicity. Source: Public use data files from the Medical Expenditure Panel Survey (MEPS) for 2008-2022, excluding 2020 to avoid direct confounding with COVID data collection.

Results

Descriptive characteristics

An unweighted yearly average n = 4514 adults ages 50-64 (total n = 63,191) are included in the analyses (the weighted yearly equivalent of N = 55,624,105 adults). Average age was 56.8 years, 51.4% were female, 40.8% had a high school education or less, and 22.3% met criteria for poor/near poor/low household income (<200% of the federal poverty level). About 1 in 10 individuals reported being uninsured, and 16.3% did not have a usual source of care. The average count of chronic conditions was 1.25, and the average MCS was 51.48 (range: 0-100). NHB and Hispanic middle-aged adults were less likely to report a college degree or more, and less likely to meet criteria for high household income (≥400% of FPL). Both groups were more likely to be uninsured (NHB: 11.9%; Hispanic: 22.0%), but, particularly, Hispanic adults reported having public coverage only (NHB: 22.8%; Hispanic: 19.3%) and to report not having a usual source of care (NHB: 18.8%; Hispanic: 26.9%). The NHB group had a higher count of chronic conditions compared to the other groups. See Supplementary Table 1 for the complete results.

Comorbid phenotypes and limitations

We found that a 3-class solution indicated the best fit to the data and the assigned class labels, based on phenotype comorbidities profiles, were (1) low prevalence of chronic conditions (LCC) (63.4%), (2) hypertension/arthritis/joint pain, henceforth “hypertensive arthritis,” (29.3%), and (3) complex cardiovascular (C-CVD) (7.3%). The prevalence rates of the individual conditions within each phenotype are depicted in Supplementary Figure 1 (see online supplementary material for a color version of this figure). It is critical to note that these labels are “suggestive” and based on substantive interpretation of the comorbidity profiles within the generated classes. The C-CVD phenotype included high prevalence of high blood pressure, arthritis, CHD, angina, MI, and other heart diseases, as well as an elevated rate of strokes (twice the rate in the hypertensive arthritis group). The prevalence of diabetes, depression, and cancer was largely similar between the hypertensive arthritis and C-CVD phenotypes. NHBs were less likely to meet criteria for classification in the LCC group (52.3%) and more likely to be classified in the hypertensive arthritis (40.2%) phenotype compared to both NHW (65.2% and 6.9%, respectively) and Hispanic adults (67.6% and 6.9%, respectively). All 3 race/ethnic groups had largely equal rates of C-CVD (6.85%, 7.5%, and 6.9% for NHW, NHB, and Hispanic adults, respectively).

Overall, 16.5% of middle-aged adults reported having physical limitations, and 3.3% required help with I/ADLs. Rates of physical limitations and I/ADLs varied across the comorbid phenotypes. A third (31.9%) and 1-in-4 (39.4%) middle-aged adults in the hypertensive arthritis groups, respectively, had prevalent physical limitations (vs 7.0% of LCC middle-aged adults), and 6.2% and 10.2%, respectively, required help with I/ADLs (vs 1.2% in the LCC phenotype). See Supplementary Table 2 for the complete results.

Healthcare expenditures

In cross-adjusted models, physical limitations doubled the odds ratios (odds ratio [OR] = 2.06, 95% CI = [1.72; 2.43]) of having any healthcare spending, hypertensive arthritis classification increased the odds 5-fold (4.70 [4.16; 5.31]), and those in the complex cardiovascular phenotype had 5.98 [4.45; 8.04] times higher odds ratios for having any spending (vs those in the LCC group). Adjusting for physical limitations and comorbid classification, both NHBs and Hispanic adults had significantly lower odds of spending compared to NHWs. Among those with any health spending, individuals with physical limitations, controlling for comorbid classification and race/ethnicity, spent $10,008 more than those with no physical limitations. Individuals meeting criteria for hypertension, arthritis, and C-CVD spent, on average, $5,186 and $12,018 more on healthcare compared to those in the LCC group. Finally, NHB and Hispanic middle-aged adults spent $1,540 and $2,347 less, respectively, relative to NHWs.

We found that adjusting for predisposing and enabling factors (M3) and mental health (M4) did not significantly change the associations (Supplementary Table 3). Furthermore, tests of interactions by physical limitations for comorbid phenotypes showed no evidence for changes in the propensity to spend or in spending levels in groups with comorbidities in the presence of physical limitations. Tests of interactions by physical limitations for race/ethnic groups showed higher odds ratios for spending among NHBs in the presence of physical limitations, but no changes in the level of spending.

Similar but more pronounced trends emerged from cross-adjusted I/ADL models. Individuals with I/ADLs had 6.4 times higher odds ratios for having any spending relative to those without those needs, and spent $20,701 more on healthcare. Adjustments for predisposing, enabling, and health factors had similar attenuating effects as described with physical limitations. However, I/ADLs did not significantly modify the associations between either comorbid phenotypes or race/ethnicity with respect to propensity to spend or average spending. See Table 1, Supplementary Table 3, and Figures 1 and 2 for the complete results.

Office-based visits

In cross-adjusted models, individuals with physical limitations (incident rate ratios [IRR] = 1.96, 95% CI = [1.87; 2.04]; p < .01; vs no limitations), those in the hypertensive arthritis group (IRR = 1.56, 95% CI = [1.50; 1.63]; p < .01; vs LCC), and those in the C-CVD phenotype (IRR = 1.57, 95% CI = [1.47; 1.67]; p < .01; vs LCC) have higher average number of office-based visits. Both NHBs (IRR = 0.71, 95% CI = [0.66; 0.76]; p < .01) and Hispanic adults (IRR = 0.67, 95% CI = [0.63;0.72]; p < .01) have lower average estimates for office-based visits (relative to NHWs). The associations remained robust to adjustment for predisposing and enabling factors, and additional adjustment for mental health had minimal impact on the estimates.

We found similar results when I/ADLs were tested as the exposure. In cross-adjusted models, I/ADLs increase the relative risk ratios for office-based visits (IRR = 2.26; 95% CI = [2.06; 2.49]; p < .01), and these associations were attenuated by adjustment to enabling factors (IRR = 1.83; 95% CI = [1.66; 2.02]; p < .01) and mental health status (IRR = 1.74; 95% CI = [1.57; 1.92]; p < .01).

We found evidence for significant modifications of both comorbid phenotype classifications and race/ethnicity by physical limitations and I/ADLs. Individuals in both the hypertensive arthritis and C-CVD groups with either of these 2 limitations had lower IRRs, whereas NHB and Hispanic adults with either of these functional limitations had higher IRRs. See Table 2, Supplementary Table 4, and Figures 1 and 2 for the complete results.

ED use and inpatient hospitalizations

In covaried models (including comorbidities, race/ethnicity), physical limitations (IRR = 2.18, 95% CI = [2.03; 2.35]; p < .01) and I/ADLs (IRR = 2.70; 95% CI = [2.42; 3.01]; p < .01) were associated with higher IRRs of ED services use. Hypertensive arthritis and complex cardiovascular phenotype classifications (vs LCC) increased the IRRs of ED use by more than 2-fold and between 3.3 (adjusting for physical limitations) and 4-fold (adjusting for I/ADLs) for these groups, respectively. NHBs, but not Hispanic adults, also had higher relative risks of ED use (IRR = 1.35, 95% CI [1.26; 1.46]; p < .01) compared to NHWs. Adjusting for predisposing and enabling factors decreased the IRRs for physical limitations and I/ADLs by nearly 25% and 34%, respectively. Adjustment for subjective mental health assessment only slightly reduced these IRRs. We found similar reductions in the IRRs (around 20% and 30% for hypertensive arthritis and C-CVD) for the comorbid phenotypes, through adjustment for predisposing, enabling, and subjective mental health factors. The IRRs for NHBs remained largely stable (a 10% reduction) through covariables adjustment (IRR = 1.22, 95% CI = [1.13; 1.32], p < .01). In models including interactions, we found support for changes in the associations between comorbid phenotype classification and ED use by physical limitations but not I/ADLs. Neither physical limitations nor I/ADLs consistently modified the associations between race/ethnicity and ED use.

Finally, similar findings regarding physical limitations and I/ADLs emerged when modeling inpatient hospitalizations. The IRRs for C-CVD were especially pronounced (IRR = 5.88, 95% CI = [5.26; 6.59]; p < .01 and IRR = 7.12, 95% CI = [6.41; 7.92]; p < .01) in models cross-adjusting for physical limitations and I/ADLs, respectively. NHBs did not differ in their inpatient hospitalization patterns compared to NHWs, but Hispanic adults had lower incidence rate ratios (IRR = 0.83 95% CI = [0.74; 0.94]; p < .01 and IRR = 0.77, 95% CI = [0.69; 0.86], p < .01), cross-adjusting for physical limitations and I/ADLs, respectively. As with ED use, the associations between the limitations exposures and phenotype classification were attenuated, but only partially so, through predisposing, enabling, and mental health needs factors. The IRRs for Hispanic adults were not substantively affected by covariable adjustment. Both physical limitations and I/ADLs downward modified the IRRs for both hypertensive arthritis and C-CVD. I/ADLs downward modified the IRRs for inpatient hospitalizations among NHBs and upward modified the IRRs for Hispanic adults. See Table 2, Supplementary Tables 5 and 6, and Figures 1 and 2 for the complete results.

Discussion

We used pooled cross-sectional data from middle-aged adults (ages 50-64), spanning more than a decade (2008-2022, excluding 2020), from the MEPS, a large nationally representative survey of the non-institutionalized United States population, to generate data-driven classifications of comorbidity phenotypes. We then examined their associations with health services outcomes in the context of vulnerabilities introduced by functional limitations and race/ethnicity. We report 4 main findings: (1) Only two-thirds (63.5%) of middle-aged adults met criteria for classification into a LCC group, and NHBs were particularly more likely to be in the hypertensive arthritis group; (2) Individuals with severe cardiovascular morbidity had elevated levels of health spending and healthcare use, including problematically high levels of ED and inpatient hospitalizations, despite higher use of office-based visits; (3) The effects of race/ethnicity on health spending and services use were only found for overall health spending and use of office-based services; with NHBs and Hispanic adults having both lower levels of overall health spending and less use of office-based clinical settings; and finally, (4) Health services use patterns in middle-age was less influenced by classic SDOH and potentially more driven by structural factors that govern health provision.

Three comorbidity phenotypes emerged: First, a select group of middle-aged adults (7%) satisfied criteria for severe levels of morbidity driven by cardiovascular complications; Second, nearly a third of the target population were characterized by having hypertensive arthritis; and third, nearly 64% of participants were classified as LCC. Our results suggest that, annually, nearly 3.9 million non-institutionalized middle-aged individuals (7%) in the United States have substantially high health needs and markedly elevated risks for concomitant functional limitations (5- and 8.5-fold higher likelihood of physical and I/ADL limitations relative to the LCC group) that are further complicated by vulnerabilities due to SDOH. Previous studies examining comorbidities have reported a varied number of classifications ranging from 3 to 8. The proportion of LCC groups found varied depending on the age of the considered population (eg, near 50% in middle-aged and older adults)^2,9 but was pronouncedly lower (16% to 33%) when focusing on older adults.^10,11 We found no differences in morbidity classifications between Hispanic adults and NHWs, but NHBs had particularly elevated rates of hypertensive arthritis classification and were least likely to satisfy criteria for the LCC phenotype. This finding aligns with past cross-sectional and longitudinal work indicating that NHBs, especially given their elevated rates of hypertension relative to other groups, are more likely than NHWs to have multisystem morbidities.^2,12,13 Continued efforts to increase awareness, monitor, and treat hypertension among NHBs can reduce disease burden and associated costs.

Second, individuals with severe levels of cardiovascular morbidity had elevated health spending and healthcare use patterns, including high rates of ED visits and inpatient hospitalizations. These findings align with established arguments that healthcare consumption and costs are primarily driven by a small fraction of patients with a higher propensity for using expensive healthcare resources that account for a substantial portion of national health spending. Yet, nearly 3-in-10 middle-aged adults with intersecting conditions of hypertension and arthritis showed a similar but less pronounced trend. This was observed in individuals who also demonstrated increased health spending and a higher propensity for ED use and inpatient hospitalizations, even in the presence of more frequent office-based visits. Previous research shows that comorbidity increases health expenditures, and documents that conditions, such as arthritis^14,15 and cardiovascular disease^16,17 are most expensive, and that clusters of cardiovascular risk and disease and arthritis are linked to the highest annual medical expenditures.¹⁸ As such, our findings suggest considerable opportunities for improving care provided in a substantial portion of the middle-aged population. Similarly, we found that conditions that cause functional limitations, particularly those affecting I/ADLs, substantially increased health spending levels. This increase persisted even after adjusting for comorbidities. These conditions were associated with higher levels of ED use and inpatient hospitalizations, despite more frequent office-based visits. Healthcare costs associated with functional limitations constitute 36% of national health expenditures, are mostly borne by public insurers, and have increased substantially over time¹⁹ Critically, studies suggest that people with functional limitations are more likely to delay or forgo necessary healthcare,²⁰ and have higher overall “unmet healthcare needs,” including access to prescription drugs.²¹ Our findings suggest that by middle age, individuals bear high healthcare costs due to the worsening of existing conditions and primary preventative care becoming less effective.

We also found that both measures of functional limitations accentuated the effects of the comorbid phenotypes on the use of office-based visits. However, they surprisingly downward modified ED and inpatient hospitalization use, and had minimal modifying effects on healthcare spending. This finding was unexpected, as existing literature consistently associates increased use of hospital and healthcare services with both comorbidities and functional limitations, independently,^11,22–24 and points to a combined effect of multimorbidities and functional limitations on an increase in use of hospital facilities.^22,25 Most of this published work has focused on older adults, and as such, use patterns due to age might have contributed to our differing findings.

Third, we showed that the links between race/ethnicity and health spending were mixed. NHBs and Hispanic adults had lower levels of overall health spending compared to NHWs and were less likely to use office-based clinical settings. This aligns with longstanding research indicating that racial/ethnic minorities, over the life course, have lower medical spending²⁶ and preventive care utilization compared to NHWs.^27,28 Moreover, estimates for specific diseases show similar patterns. For instance, studies have found that in adults with arthritis, annual health expenditures are lower for racial and ethnic minorities relative to NHWs.^14,29 Existing studies also suggest that NHWs spend more on outpatient or ambulatory care compared to minorities^30,31 and that racial/ethnic minorities spend more on emergency healthcare.^29,30 Racial/ethnic minorities often report greater difficulties accessing medical care, which could explain their lower outpatient care utilization. Previous studies have also reported higher ED utilization among Blacks and Hispanic adults³² and higher rates of hospitalizations specifically for high-need race/ethnic groups.³³ In our study, we find that only middle-aged NHBs had a higher propensity for ED use, while Hispanics were less likely to be hospitalized. However, physical limitations and I/ADLs increased the levels of office-based visits for both NHBs and NHWs, while I/ADLs also increased the use of inpatient hospitalizations among Hispanic adults. Interestingly, no other significant changes were observed in our results, suggesting a different dynamic for health services use in middle-aged populations, particularly after adjusting for differences in health needs requirements due to complex comorbidities.

Finally, we found that predisposing factors had minimal explanatory effects in our target population. Although enabling factors (insurance, employment, income, and education) did reduce some of the effects noted above, these reductions were also small. These results were surprising, as extensive literature has documented that SDOH, including predisposing health risks and enabling health and healthcare factors, play a significant role in explaining disparities among at-risk groups, such as individuals with functional limitations^34,35 and racial/ethnic groups. The limited attenuation through insurance, in particular, raises several questions. Uninsurance and underinsurance, significant barriers to healthcare access, are higher among individuals with functional limitations,^36,37 middle-income families, and Hispanic adults.^37–40 These findings suggest that the differences in health services use and spending at the intersection of severe morbidity, functional limitations, and race/ethnicity are less influenced by standard social drivers of disparities. Instead, they appear to be more strongly influenced by structural factors governing health provision for high-need patient populations.

Limitations

A few limitations of our study should be noted. The MEPS relies on self-reported data and is subject to the inherent limitations of survey data. Although generally accurate, the level of morbidity can affect the precision and specificity of reporting healthcare services compared to administrative data. Because MEPS is a nationally representative study, its design does not allow for generalization to specific populations, such as those with functional limitations, particularly given the exclusion of institutionalized individuals. Additionally, as MEPS provides cross-sectional data, we could not establish causality or mechanistic pathways linking race/ethnicity, comorbidities, and limitations. Also, MEPS does not include any biological markers that may allow us to measure biological aging. We also did not distinguish between primary and secondary functional limitations and treated functional limitations and comorbidity as co-occurring without differentiating based on the timing of onset. In terms of comorbidities, we limited our analyses to the 11 conditions recognized by AHRQ as priority conditions, and as such, accounting for a larger set of conditions could yield different health profiles and is worth exploring with more detailed data such as claims data and electronic health records. Furthermore, health disparities for individuals with functional limitations can include a lack of transportation, risk of social isolation, higher prevalence of mental health issues (eg, depression and anxiety), and economic precariousness, among other factors. We did not account for these specific measures in our analyses. Finally, we used yearly pooled cross-sectional national data and did not capture any changes in the macro political and economic environment. Since publicly available MEPS does not allow identification of U.S. states, we were unable to control for individuals living in states that have accepted the Medicaid changes provided by the ACA. Future research can explore these aspects as insurance and economic policy have the potential to influence health services use and spending at the intersection of comorbidity and functional limitations.

Conclusion

In our analysis of a nationally representative sample, we identified that co-occurring comorbidities and functional limitations in midlife significantly impact the utilization of healthcare resources and healthcare spending among U.S. racial/ethnic minorities. Targeted interventions that focus on these intersectional risks, especially those tailored for individuals with co-occurring complex comorbidities and functional limitations, can yield significant efficiencies in the healthcare system and potentially decrease the burden of disease among those impacted.

Supplementary material

Supplementary data are available at Innovation in Aging online.

Funding

None declared.

Conflict of interest

None declared.

Data Availability

This study was not preregistered. This study uses publicly available data from the Medical Expenditures Panel Survey (MEPS). The data and more details are available at https://meps.ahrq.gov/mepsweb/.