Risk Factors of Skilled Nursing Facility Admissions and the Interrelation with Hospitalization and Amount of Informal Caregiving Received

Abstract

Background

The correlations between skilled nursing facility (SNF) admissions, number of hospitalizations, and informal caregiving hours received after adjusting for physical and cognitive function and sociodemographic covariates are not well understood.

Objective

To better understand risk factors for SNF admissions and the interrelation with hospitalizations and amount of informal caregiving received, this study applied a novel joint modeling analysis to simultaneously explore the correlation and shared information between the three outcomes.

Research design

Observational follow-up study.

Subjects

Data from 4836 older Americans included in the 2011–2015 rounds of the National Health and Aging Trends Study were linked with Centers for Medicare & Medicaid Services.

Measures

We jointly modeled SNF admission, hospital admissions, and informal caregiving hours received while accounting for possible risk factors. We addressed missing values by multiple imputation with chained equations.

Results

SNF admission evidenced a strong positive correlation with hospital admission and SNF admission evidenced a weak positive correlation with the informal caregiving hours received after adjustment for important risk factors. Non-Hispanic white race/ethnicity, living alone, not being Medicaid eligible, Alzheimer’s disease and related dementias diagnosis, activities of daily living disabilities, and frailty were associated with increased risk of SNF admissions and any/number of hospital admission. Lower educational level was also associated with the latter. Medicaid eligibility was the only factor not associated with any nor numbers of informal caregiving hours received.

Conclusions

Sociodemographic and health factors were important for predicting SNF admissions. After adjustment for important risk factors, SNF evidenced a strong positive correlation with the number of hospitalizations and a weak positive correlation with the hours of informal caregiving received.

INTRODUCTION

Nearly a quarter of Medicare beneficiaries are discharged directly to a skilled nursing facility (SNF) for post-acute care after an acute care hospitalization (1, 2). SNF admissions incur personal and societal costs (2, 3) and often lead to hospital readmissions or precede death (4, 5). Evidence of which care settings provide the best post-acute care for older adults is lacking (2, 6). A recent study showed that patients admitted to SNFs experienced lower readmission rates, but incurred higher Medicare costs, whereas functional status and mortality rates were comparable to patients discharged home with homecare (7). The decision of whether older adults are discharged to a SNF is often based on assessments of personal needs, cost, insurance coverage, preferences, and available support in the patients’ home (2, 6). Availability of informal caregivers may reduce SNF admissions after hospitalizations, but is also an indicator of needs because recipients of informal caregiving often have chronic conditions or disabilities (8).

To potentially reduce SNF admissions, it is imperative to simultaneously examine the interplay between SNF admissions, access to informal care, and number of hospitalizations, as well as their relationships to demographic and clinical characteristics of older adults. Previous studies have elucidated medical and societal risk factors of SNF admissions (6) and identified cognitive and physical functions as the strongest risk factors of SNF admissions (9, 10). Sociodemographic factors, such as socioeconomic position, race/ethnicity, and cohabitation status have also been shown to influence the probability of SNF admissions (9, 10). Cognitive and physical function express the individuals’ need for care in SNFs, whereas sociodemographic factors such as socioeconomic position, race/ethnicity, and cohabitation status may be related to health literacy and resources to seek care. Yet, the correlations between SNF admissions, hospitalizations, and informal caregiving received adjusted for all these potential risk factors have not been previously explored. To inform public health professionals and healthcare providers, it is necessary to explore risk factors of SNFs, as well as adjusted correlations between SNFs, number of hospitalizations, and number of informal caregiving hours received.

To better understand risk factors for SNF admissions and the interrelation with amount of hospitalization and informal caregiving hours received, this study applies a novel longitudinal joint modeling approach to simultaneously account for correlations and shared information between the three outcomes (11, 12, 13, 14, 15). The joint model enables examination of the conditional relationship between outcomes. Specifically, examining the subject-level relationship between SNF admissions, number of hospitalizations, and informal caregiving hours received after adjustment for sociodemographic information, as well as cognitive and physical functions. Furthermore, we explicitly address missing data using multiple imputation. This study investigates the relationship between functional and sociodemographic factors for SNF admission when modeled jointly with amount of hospitalization and informal caregiving hours received.

METHODS

Data source

The study was based on the 2011 through 2015 rounds of the National Health and Aging Trend Study (NHATS) linked with data from Center for Medicare and Medicaid (CMS). The NHATS cohort includes a nationally-representative sample of Medicare beneficiaries aged 65 years and older recruited in 2011 (baseline response rate of 71%) who were interviewed annually by trained research staff (16). The initial 2011 cohort included 8245 participants of which 7609 identified as community-dwelling or in a residential care facility. We excluded 120 participants with incomplete baseline information, 15 participants with missing values for informal caregiving hours received, and 2638 participants who were not fully covered by the Medicare fee-for-service during the first two rounds of the NHATS. Individuals who left the Medicare fee-for-service during the 3rd, 4th, or 5th round were censored at the last available annual interview. The final study population included 4836 participants who were Medicare fee-for-service enrollees for the duration of surveillance (Figure 1). The study was approved by the John Hopkins University Institutional Review Board (IRB) and Yale IRB (HIC# 1510016585).

Figure 1.

Flow chart of selection of study participants

Explanatory factors

Sociodemographic

Sociodemographic factors comprised baseline age, sex, race/ethnicity (non-Hispanic white (reference) and other), educational level (college and higher (reference), no school/<9 grade, 9–12 grade and high school, and vocational training), time-varying cohabitation status (living alone (reference), versus not living alone), and time-varying Medicare-Medicaid dual eligibility (binary indicator with no as the reference).

Clinical

Baseline self-rated health was based on a 5-point Likert scale ranging from excellent (1 point) to poor (5 point). Alzheimer’s disease and other related disorders or senile dementia (ADRD) was time-varying and defined based on the Chronic Condition Data Warehouse (CCW) algorithms that include a reference period of three years to accumulate ICD-9, Current Procedural Terminology (4th Edition), Healthcare Common Procedure Coding System (Level II) codes, and the number and types of claims. A description of the methodology to ascertain chronic conditions can be found at the CCW website (17). The time-varying comorbidity index was a count of 20 common chronic conditions (19) among older adults identified through the CCW algorithms (17, 20).

Function and Frailty

Longitudinal counts of activities of daily living (ADL) disabilities (0–6) included getting help with dressing, eating, bathing, toileting, transferring from bed, and getting around inside one’s home. Time-varying Fried Frailty Score was used to assess the criteria for exhaustion, low activity, weakness, slowness, and shrinking. The Fried Frailty score was categorized as robust by a score of 0 (reference), pre-frail by a score of 1–2, and frail by a score of 3–5 (18).

Outcomes

Three longitudinal outcome variables were considered: SNF admission, hospital admissions, and informal caregiving hours received. SNF and hospital admissions were accumulated from the Master Beneficiary Summary File including the Cost and Utilization and Chronic Conditions segments in Medicare claims data. These two outcomes were aggregated within time intervals that were aligned with the NHATS annual interviews. SNF admission was recorded as a dichotomous variable that indicated whether participants were admitted into SNF. Hospital admission was recorded as the number of hospitalizations within a given round. Hospital admission included all causes of hospitalizations. Informal caregiving hours received were computed from NHATS interviews and calculated as the sum of hours of all persons who, without receiving pay, helped with mobility, self-care activities, household activities, medical care-related activities, and/or driving and transportation during the month preceding the annual interview (21).

Statistical analyses

We applied a joint modeling approach to investigate the associations between the covariates and SNF admission, hospital admissions, and informal caregiving hours received, as well as correlation over time between the three latter variables. We fitted a multivariate multi-level generalized linear model with an unconstrained covariance matrix to generate dependence among the three outcomes (15). The first level of the model specified the measurements for the subject at the specific annual interview and it is nested within a second-level that specified variability across subjects. The three outcomes were assumed conditionally independent from each other given the subject-level intercepts. The joint model was formulated using hierarchical generalized linear models with different link functions for each outcome. We assumed that the subject-specific intercepts from the three outcome models followed a multivariate normal distribution with zero means and an unstructured variance-covariance matrix. This variance-covariance matrix captured the conditional subject-level correlations among the three outcomes.

For the separate outcome models, we applied the logit link function to model the log odds ratio of SNF admission status. We used a Negative-Binomial hurdle model (22) to describe the number of hospital admissions. A Bernoulli-lognormal two-part model was used to describe the number of informal caregiving hours received (23, 24). The hurdle model and two-part models were applied because the two outcomes comprised excessive number of zeros. The supplementary material provides additional details on the model assumption tests.

For the joint modeling, we used Bayesian inference and obtained the estimates using the Markov chain Monte Carlo (MCMC) algorithm implemented in the R package rjags (25). We set diffused normal prior distributions for coefficients estimates and the inverse Wishart prior distribution for the variance-covariance matrix of the second-level effects. The supplementary material provides additional details on the models.

We applied multiple imputation procedure to address the missing values in the covariates. This procedure replaced missing values with plausible values and resulted in five imputed datasets, which were analyzed separately. Final inferences were obtained using Rubin’s rules (26). Imputations were performed using the fully conditional specification method (27). Participants who were deceased and those who changed their insurance to Medicare Advantage plans were censored after their last available annual interview. Observations after censoring were not included in the imputation process. The data contain three incomplete variables: cohabitation status, ADL score, and the frailty score (Figure S1). For imputing cohabitation statuses and frailty scores, we implemented multilevel linear models on their corresponding latent Normal variables. This is equivalent to applying a multilevel Probit regression and a multilevel multinomial Probit regression, respectively (28, 29). Missing values in the ADL score were imputed using a multilevel linear model, where the imputations were truncated by 0 and 6 and rounded to closest integers. The multiple imputation procedure was implemented using the R package micemd (30). For comparison purposes, we also conducted the analyses only on complete cases.

RESULTS

Descriptive Results

Baseline characteristics for participants with and without SNF admissions, hospital admissions, and informal caregiving hours received during annual interviews in round 1–5 are presented in Table 1. In total, 16.8% had at least one SNF admission, 50.9% had at least one hospitalization, and 86.5% received at least one hour of informal caregiving. Beneficiaries who were admitted to a SNF had on average 1.5 times more informal caregiving hours (114.3 vs. 78.7) and larger mean ADL limitation scores (0.97 vs. 0.39) than those without a SNF admission. More beneficiaries with at least one SNF admission lived alone, were frail, and were diagnosed with ADRD, compared to beneficiaries without a SNF admission.

Table 1.

Baseline Characteristics of Study Population, %

Baseline characteristics	All	Skilled nursing facility admissions		Hospitalization admissions		Informal caregiving hours
	(N=4836)	No (n=4024)	Yes (n=812)	No (n=2376)	Yes (n=2460)	No (n=653)	Yes (n=4183)
Age
65–69 yrs	18.7	21.4	5.0	25.9	11.7	22.8	18.0
70–74 yrs	19.7	21.3	12.1	23.1	16.5	25.1	18.9
75–79 yrs	19.5	20.0	16.9	19.4	19.6	18.2	19.7
80–84 yrs	20.1	18.8	26.7	17.6	22.6	17.5	20.5
85–89 yrs	13.2	11.5	22.0	9.3	17.0	10.3	13.7
90 yrs	8.8	7.1	17.2	4.7	12.6	6.1	9.2
Sex
Male	43.0	44.0	38.1	43.6	42.4	40.0	43.5
Race/ethnicity
Non-Hispanic white	71.8	70.7	77.1	70.5	73.0	70.4	72.0
Cohabitation status
Living with a partner/spouse	66.8	69.4	54.2	70.2	63.6	41.3	70.8
Medicaid eligibility
Yes	15.3	15.2	16.1	12.8	17.8	11.8	15.9
Educational level
<9th grade education	12.6	12.3	14.0	10.8	14.3	8.0	13.3
9th-12th grade	13.2	12.7	15.4	11.9	14.4	13.6	13.2
High school/trade school	34.1	33.4	37.8	33.5	34.8	33.5	34.2
College	40.1	41.6	32.8	43.8	36.5	44.9	39.3
Alzheimer’s disease and related dementia
Yes	10.9	8.7	21.4	5.4	16.1	6.1	11.6
Activities of daily living limitations
Mean (SD)	0.49 (1.27)	0.39 (1.16)	0.97 (1.64)	0.24 (0.91)	0.73 (1.50)	0.21 (0.85)	0.54 (1.32)
Frailty score
Robust	30.9	34.7	12.1	41.4	20.8	40.4	29.4
Pre-frail	48.0	47.6	49.8	46.0	49.9	44.1	48.6
Frail	21.1	17.7	38.1	12.6	29.3	15.5	22.0
Self-rated health
Mean (SD)	2.86 (1.13)	2.78 (1.12)	3.26 (1.08)	2.58 (1.06)	3.13 (1.13)	2.67 (1.09)	2.89 (1.14)
Number of Comorbidities (0–15)
Mean (SD)	4.08 (2.76)	3.76 (2.64)	5.71 (2.79)	3.05 (2.34)	5.09 (2.77)	3.55 (2.66)	4.17 (2.77)

Joint modeling

Correlations between the outcomes

Table S1 shows that all three outcomes were correlated in the unadjusted correlation matrix of second-level effects from the joint model (Table S1). SNF evidenced a high positive correlation with hospital admission (binary indicator: 0.97, 95% confidence interval (CI): 0.96, 0.98; positive counts: 0.93, 95% CI: 0.88, 0.96). Receiving any informal caregiving hours evidenced a modest positive correlation with SNF admission (0.16, 95% CI: 0.09, 0.23), any hospital admission (0.21, 95% CI: 0.16, 0.27), and positive counts of hospital admissions (0.20, 95% CI: 0.09, 0.29). Hours of informal caregiving received evidenced a modest positive correlation with any SNF admission (0.23, 95% CI:0.17, 0.30), any hospital admission (0.29, 95% CI: 0.24, 0.34), and positive counts of hospital admissions (0.28, 95% CI: 0.18, 0.36).

In the adjusted correlation matrix, SNF admission remained strongly correlated with the binary indicator of any hospital admission (0.94, 95% CI: 0.92, 0.97) and positive counts of hospital admissions (0.81, 95% CI: 0.72, 0.92) (Table S2). In contrast to the unadjusted correlation estimates (Table S1), receiving any informal caregiving hours evidenced no significantly correlated with the other outcomes. The number of informal caregiving hours received evidenced a weak positive correlation with SNF admission (0.12, 95% CI: 0.01, 0.24).

Joint modeling estimates

Figure 2 shows the joint modeling results of SNF admission, hospital admissions, and informal caregiving hours received. Non-Hispanic white race/ethnicity was associated with any SNF admission (aOR: 1.56, 95% CI: 1.20, 2.05), hospital admission (aOR:1.16, 95% CI: 1.02;1.32), and informal caregiving hours received (aOR: 1.20, 95% CI:1.07;1.35); however, not with the positive counts of the latter two. Not living alone was associated with 40% decreased adjusted odds (aOR: 0.60, 95% CI: 0.48, 0.75) of any SNF admission and three times increased adjusted odds (aOR: 3.00, 95% CI: 2.71,3.32) and rates (aRR: 3.11, 95% CI: 2.88, 3.35) of informal caregiving hours received (if any). No significant association was identified between cohabitation status and hospital admission. Medicare-Medicaid dual eligibility was associated with decreased adjusted odds of SNF admission (aOR: 0.59, 95% CI: 0.44, 0.80) and decreased adjusted odds (aOR: 0.85, 95% CI: 0.72, 0.99) and rates (aRR: 0.80, 95% CI: 0.68, 0.95) of hospital admissions, but was not associated with informal caregiving hours received. Compared to college level education, lower education levels were associated with increased numbers of hospital admissions (<9^th grade aRR: 1.30, 95% CI: 1.07, 1.58; 9–12^th grade aRR: 1.22, 95% CI: 1.01, 1.47; high school/trade school aRR: 1.17, 95% CI: 1.01, 1.36) and informal caregiving hours received (<9^th grade aRR: 1.56, 95% CI: 1.37, 1.77; 9–12^th grade aRR:1.24, 95% CI: 1.10, 1.40; high school/trade school aRR: 1.13, 95% CI: 1.04, 1.24). Educational level was not associated with any SNF admission, hospital admissions, nor informal caregiving hours received.

Figure 2.

Joint analyses of 1) the binary indicators of any informal caregiving hours received, hospital admissions, and skilled nursing facility admission (left) and 2) the positive counts of hospital admissions and informal caregiving hours received (right). The vertical lines show the reference value. Results are from a hurdle model and a two-part model. For the number of hospital admissions, we applied a negative-binomial hurdle model and for the informal caregiving hours, we applied a Bernoulli-lognormal two-part model. The binary indicator signifies whether an admission or caregiving was received, while “positive counts” are calculated for those who had at least one admission or received caregiving.

ADRD was associated with any SNF admission (aOR: 1.43; 95% CI:1.12, 1.83), any hospital admission (aOR: 1.30, 95% CI:1.13, 1.50), and number of informal caregiving hours received (aRR: 1.26, 95% CI: 1.14, 1.40). ADRD was not associated with any informal caregiving hours received nor number of hospital admissions.

A one unit increase in ADL disabilities was associated with 34% increased adjusted odds of any SNF admission (aOR: 1.34, 95% CI:1.26, 1.43), 21% (aOR: 1.21, 95% CI: 1.17, 1.26) increased adjusted odds of any hospital admission, 8% (aRR: 1.08, 95% CI: 1.04,1.12) increased rate of hospital admissions, 11% (aOR: 1.11, 95% CI: 1.06, 1.16) increased odds of any informal caregiving hours received, and 22% (aRR: 1.22, 95% CI: 1.19, 1.25) increased rate of caregiving hours received. Frailty was associated with increased adjusted odds of all the outcomes.

DISCUSSION

We applied a joint model with multiple imputations to investigate the relationship between sociodemographic factors, as well as cognitive and functional abilities for SNF admission, hospital admissions, and informal caregiving hours received. SNF and hospital admissions evidenced a strong positive correlation over time after adjustment for sociodemographic factors, as well as ADRD, frailty, and functional disabilities. However, we only identified a weak positive correlation between SNF admission and informal caregiving hours received. All sociodemographic and health factors were identified as risk factors for SNF admission (except educational level), any or number of hospital admissions (except cohabitation status), and any or number of informal caregiving hours received (except Medicaid eligibility).

Sociodemographic factors

Being non-Hispanic white was associated with increased risk of any SNF admission, hospital admission, and informal caregiving hour received. Race/ethnic disparities in SNF admissions may be explained by differences in access to and quality of care facilities, family structures, social networks, and caregiving. Studies showed that African American older adults are less likely to endorse assisted living facilities and SNF admissions (31) and receive more assistance and support from their family (32, 43). We found that non-Hispanic whites evidenced greater likelihood of receiving informal caregiving, whereas there was no difference in the amount of informal caregiving received. We identified race/ethnic disparities in SNF admission even after accounting for socioeconomic measures, as well as concurrent outcomes of informal caregiving hours received and hospitalizations. African American older adults may avoid SNFs because they frequently are admitted to poor quality SNFs far from their community (34–37). Studies demonstrated that race/ethnic minority patients have greater hospital readmission rates than white patients after a SNF admission (38, 39). We only identified significant race/ethnic differences in any hospital admission and not in the number of hospitaliztions. Previous studies have shown that readmission rates decrease with higher SNFs’ quality; however, readmission rates are still significantly greater among Hispanics and African Americans at all quality levels (39).

Lower educational levels were only significantly associated with the number of hospital admissions and informal caregiving hours received among those who had a least one hospital admission and one informal hour of caregiving, respectively. Higher educational levels may increase access and resources, which enable patients to seek proper care at SNFs or may ensure resources to return home after a hospital admission. In a previous study, only less than a 9^th grade education was significantly associated with increased risk of SNF admissions compared to individuals with college or higher education (14, 40). Multivariate adjusted results from systematic reviews on educational level and SNF admissions were mixed (9, 10). This may be explained by differences in categorization of educational level, geographical variation, and cohort effects.

Medicare-Medicaid dual eligibility, as an indicator of low socioeconomic status, was significantly associated with lower risk of SNF admission and hospital admissions. A meta-analysis found no conditional association between Medicare-Medicaid dual eligibility and SNF admission (9). Yet, the meta-analysis showed that other indicators of socioeconomic status, such as house ownership and higher income, were protective of SNF admissions (9). Medicare-Medicaid dual eligibility determines which healthcare services are available to the beneficiary. Previous studies found that dual-eligible beneficiaries were more often admitted to SNFs with low nurse-to-patient ratio (41) and of low quality (33, 38); furthermore, they more often experience hospital readmissions (38, 41). A recent study showed that Medicare-Medicaid dual eligibility was the strongest predictor of being admitted to a low quality SNF (37). These factors may influence dual-eligible beneficiaries’ decision to seek care at SNF after a hospitalization. Furthermore, out-of-pocket payments have been shown to influence length of SNFs stays (43). While out-of-pocket payments may not restrict dual-eligible beneficiaries, SNFs receive higher payments for patients with private insurances. Thus, SNFs have incentives to avoid dual-eligible beneficiaries (41, 42). Dual-eligible beneficiaries often have fewer resources potentially making them less equipped to seek admission at high quality SNFs (42).

Cohabitation status was protective of SNF admission as shown in previous studies (9, 14). Yet, a systematic review found moderate to weak evidence of increased risk of SNF admissions among unmarried adults and inconsistent results for those living alone (10). Our analysis showed that beneficiaries who live alone less often received informal caregiving and a smaller amount of informal caregiving compared to cohabiting beneficiaries. This can be explained by previous findings that identified children and partners as the most frequent informal caregivers (14).

Health factors

We showed that cognitive and functional impairment were associated with all three outcomes. Functional impairments, frailty, and ADRD impact a person’s inabilities to self-care. Hospitalization indicates a medical necessity of illness or injury. While SNF admissions reflect a need for either post-acute care post-hospitalization or long-term care. Hours of informal caregiving received indicates both a need and access to care while living in the community. Thus, it is expected that ADRD, frailty, and functional impairment were significantly associated with all three outcomes. Our findings further support previous studies that identified functional impairments and ADRD as strong predictors of SNF admissions (9, 10). A recent review found that dementia patients are overrepresented in SNFs and among dementia patients, ADL limitations and caregiver burden are significant risk factors for SNF admission (44). This illustrates that the receipt of any informal caregiving is not significantly correlated with SNF care; however, the burden (measured in hours of informal care) on their caregivers is positively correlated with need for SNF care.

Strengths and limitations

This study is the first to jointly model three different types of outcomes by accounting for the within subject dependence amongst them and using multiple imputation for missing data. An advantage of the joint modeling is that it shows the conditional relationship between outcomes over time after adjustment for the covariates. The study includes a large sample of older adults based on a nationally-representative cohort with a high response rate (71%) at baseline and annual interviews over five years. This limits the risk of selection bias from analyses based only on individuals with complete information. Temporality of the covariates and outcomes also strengthens the results.

The study has a few limitations. Although we implemented advanced models that addressed the correlation amongst outcomes and temporality, we cannot draw causal conclusions given the observational nature of the study. There was no data on the length of the SNF stays. Our findings are also limited by investigating all cause hospitalizations. It may be that the correlations with SNF and informal caregivers, as well as the risk factors for hospitalizations, differ for elective compared to emergent hospital stays. Also, we only included information regarding informal caregivers even though formal paid caregivers may provide similar support. In the US, formal paid caregivers are infrequently covered by insurance plans and, thus, the effects would be more specific to those with sufficient wealth or specialized insurance to cover these costs. Nonetheless, future work should examine formal paid caregiving. Lastly, the population was restricted to those covered by Medicare fee-for-service, which may hinder generalizability to those with other insurance.

Implications

The study provides insight that could help identify high-risk persons for inclusion in future clinical interventions to prevent or delay SNF admissions among older beneficiaries in the US. We identified those at higher-risk by the following characteristics: non-Hispanic white, living alone, not being Medicaid eligible, frailty, and being cognitively or functionally impaired. This suggests that beneficiaries with these characteristics may need more support and guidance to live at home after a hospitalization. By applying the joint modeling approach with three outcomes in comparison to conventional single outcome models, we furthermore identified that SNF admission evidenced a strong positive correlation with hospitalizations, but SNF admission only evidenced a weak positive correlation with informal caregiving hours received after adjusting for important risk factors. Our findings suggest that the number of hospitalizations (medically-driven SNF care needs) evidenced a stronger positive correlation with SNF admission compared to informal caregiving hours received. This is an important and novel finding because SNF admission often is a consequence of a complex interplay between hospital discharge decisions considering the need, cost, and access to informal care.

In conclusion, we identified a strong positive correlation between SNF admission and hospital admissions, but not between SNF admission and informal caregiving hours received after taking important risk factors into account. Beneficiaries, who were non-Hispanic white, living alone, and not Medicare-Medicaid dual eligible and frail, as well as beneficiaries with ADRD and functional disabilities evidenced an increased risk of SNF admission. Apart from cohabitation status, these factors and lower levels of education were risk factors for any and/or number of hospital admissions. Medicaid eligibility was the only characteristic that was not associated with receipt or number of informal caregiving hours.