Abstract
Objectives:
We examined whether previously identified relationships between sociodemographic factors and caregivers for skilled nursing facility (SNF) admissions are replicated when jointly accounting for longitudinal change in functional limitations. We further explored the impact of caregivers by investigating the relationship between caregiver’s characteristics and SNF admissions.
Design:
Longitudinal follow-up study.
Setting:
The United States of America.
Participants:
In total, 3875 older Americans from the 2011–2015 rounds of the National Health and Aging Trends Study linked with Centers for Medicare and Medicaid Services.
Measures:
Sociodemographic factors and caregiver’s characteristics were used to predict change in functional limitations from baseline and time to first SNF admission using a joint modeling approach.
Results:
In total, 11.3% of the study population had a SNF admission during follow-up. For sociodemographic factors, non-Hispanic white, <9th grade education, and having at least 1 caregiver were associated with higher hazards of SNF admission than other race/ethnicity, college or higher education, and no caregiver, respectively. In contrast, living with a partner or living with others was associated with lower hazard of SNF admissions. For characteristics of caregivers, medical-supportive caregiver was associated with increased hazard of SNF admissions, whereas partner caregiver was protective of SNF admissions. Jointly modeling SNF admissions and change in functional limitations resulted in greater precision of effect estimates than modeling these outcomes separately.
Conclusions and implications:
The study provides insight that can help identify high-risk populations for future interventions to prevent or delay SNF admissions. The relation between caregivers and SNF admissions depended on caregiver’s characteristics. Future work should focus on providing help to those without a partner caregiver or needing help managing their health to ensure independent living and improve the well-being of older adults. Precision increased when jointly modeling the SNF admission with change in functional limitations.
Keywords: Skilled nursing facility admissions, functional limitations, joint modeling
Skilled nursing facility (SNF) admissions are costly for the individual and the society.1 SNFs provide post-acute skilled care to older adults who need care before a potential return to their community after a hospitalization.2 However, a substantial proportion of patients admitted to a SNF experience readmissions within 30 days.2,3 Negative consequences of SNF admissions for older adults’ well-being have prompted a large number of studies to investigate predictors of SNF admissions to prevent or delay older adults going to these facilities.4 An abundance of studies have shown that functional limitations are one of the main risk factors of SNF admissions.4,5 Further, sociodemographic characteristics, such as low income, non-Hispanic whites, living alone, and having a caregiver are shown to be risk factors for SNF admissions and functional limitations.4,6–8 The previously detected associations between sociodemographic factors and caregivers and SNF admission may be explained by the high correlation between functional limitations and SNF admission. In particular, the relationship between number of caregivers and SNF admissions could be explained by health status. Characteristics of both the caregiver and the recipient have been shown to drive the caregivers’ decision of long-term SNF admissions for American patients with dementia.9 Caregivers may strive to keep the older adults in the community and to facilitate SNF admissions for serious health problems. The dominance of these counteracting mechanisms may depend on the relationship between the caregiver and the recipient. To investigate potential mechanisms behind the relationship between caregivers and SNF admissions, characteristics of caregivers should be explored. Furthermore, to account for the correlation between functional limitations and SNF admissions, the 2 outcomes should be modeled jointly when estimating the effects of sociodemographic factors and caregivers. Joint modeling is a powerful method to account for correlation and shared information between longitudinal outcomes and time-to-event outcomes.10 This approach leads to more efficient and accurate estimation of effects compared with models that solely include time-varying covariate.10
Our objectives were 2-fold. First, we investigated whether previously detected sociodemographic factors and caregivers as predictors of SNF admissions are replicated when modeled jointly with longitudinal change in functional limitations. Second, we disentangled the previously detected relationship between caregivers and SNF admissions by exploring the relationship between different characteristics of caregivers and SNF admissions.
Methods
Data Source
We used data from 2011–2015 rounds of the National Health and Aging Trends Study (NHATS) linked with Centers for Medicare and Medicaid Services data. In brief, the NHATS cohort includes a nationally representative sample of Medicare beneficiaries who were ≥65 years old.11 The study protocol was approved by the Johns Hopkins University Institutional Review Board (IRB) and the Yale IRB (HIC# 1510016585).
Variables
The main exposures included sociodemographic factors and characteristics of caregivers. The sociodemographic factors included being female (reference), race/ethnicity (race/ethnicity other than non-Hispanic whites as reference), cohabitation status, and educational level. Cohabitation status was based on the previous round categorized as living alone (reference), living with a partner, living with others, and living with a partner and others. Educational level was categorized as ≥college (reference), <9th grade education, 9th-12th grade or high school, and vocational training. Caregivers were identified for each participant as a person who had helped with mobility, self-care activities, household activities, medical care-related activities, or driving and transportation.12 Number of caregivers was categorized as 0 (reference), 1–2, and ≥3 caregivers. For those with a caregiver, we investigated the following caregiver characteristics: (1) paid (no/yes), (2) medical-supportive who helped with physician contact and prescription medicine (no/yes),13 (3) partner (no/yes), (4) child (no/yes), (5) neighbor/friend (no/yes), and (6) overall family (no/yes). Number of different types of caregiver sources (partner, child, other family, neighbor/friend, professionals, and others) was categorized as 1 (reference), 2–3, and ≥4 sources (Appendix: Supplementary Table S1).
The covariates included age, dementia status, self-rated health, and a morbidity index. Age was categorized as 65–69 years (reference), with 5-year increments to ≥90 years. Dementia status was dichotomized as normal (reference) or living with dementia if the individual fulfilled any of the following 3 validated NHATS criteria for dementia14: self-reported, physician-diagnosed dementia; probable dementia classification scores (score ≤ 2/8 items) from the validated proxy-report;15 and an impairment classification score from a cognitive test battery. Overall self-rated health was measured on a 5-point Likert scale ranging from excellent to poor. The morbidity index was based on self-report of the following 9 conditions: heart attack, heart disease, high blood pressure, arthritis, osteoporosis, diabetes mellitus, lung disease, stroke, and cancer.
Change in functional limitations and first SNF admission were the main outcomes. Change in functional limitations was measured on a scale developed and validated by Gill and Williams,16 which ranges from 7 (full function) to 28 (no function). The scale is based on the ability to perform 4 self-care activities (eating, washing oneself, toileting, and dressing) and 3 mobility activities (going outside, getting around inside, and getting out of bed). We calculated the change in the functional score from baseline (2011) for each follow-up round. First SNF admission was based on an algorithm developed and validated by Yun et al using Centers for Medicare and Medicaid Services data to identify SNF admissions17 (Appendix: Supplementary Table S2). SNF admissions were included as a combination of short- and long-term stays in the main analyses. Separate analyses were conducted for the short- (<100 days) and long-term (≥100 days) stays.
Statistical Analyses
Descriptive statistics were conducted using cross tabulation by SNF admission status. We applied a joint modeling approach using 3 steps for the primary analysis. The joint modeling approach accounts for existing correlation between a survival outcome and a longitudinal outcome; here, the correlation between functional limitations and time to first SNF admission. Older adults who are functionally disabled are more likely to be admitted to a SNF, and this correlation may provide important clinical insight into the health trajectory of older adults. Further, joint modeling approach can correct for random measurement error in the longitudinal outcome. In the joint model, a linear mixed effects model was used to model functional limitations and a Cox proportional hazards model was used to model time to first SNF. We chose the semiparametric Cox proportional hazards model because of its less restrictive assumptions as opposed to parametric survival models such as Weibull. The following modeling steps were followed. First, we used a sociodemographic linear mixed effect model with random intercepts and slope to analyze the associations of the following characteristics with change in functional limitations: sex, race/ethnicity, educational level, cohabitation status, age, time, dementia, self-rated health, and morbidity. Next, we applied Cox proportional hazard model to analyze the associations of the characteristics in the sociodemographic model with time to first SNF admission. Finally, we jointly modeled change in functional limitations and time to first SNF admission as outcomes and sociodemographic factors as predictors using estimates from the separate models for each outcome as starting values. We repeated these steps in the analysis restricted to individuals with a caregiver, focusing on the following caregiver characteristics: paid caregiver, medical-supportive caregiver, partner caregiver, child caregiver, neighbor/friend caregiver, any family caregiver, and number of caregiver sources. The statistical analysis was implemented using JM package in R statistical software (The R Foundation, Rotterdam, The Netherlands).18 JM package is a user-friendly joint modeling tool, where the user simply specifies the analytic data, longitudinal outcome, survival outcome, and a set of predictors. For all Cox regression analyses, the available data were weighted with the inverse probability of not being missing (n = 1583) between rounds 1 and 2 (Figure 1). Missing values were due to death and loss to follow-up. These weights were estimated as inverse probabilities from a logistic regression with age, sex, race/ethnicity, educational level, cohabitation status, number of caregivers, dementia, and functional limitations as the covariates and missing indicator between rounds 1 and 2 as the response variable. For sensitivity analyses, we fitted 2 separate Cox regression models for the sociodemographic and caregiver model with physical limitations as an additional covariate. Further sensitivity analyses involved joint modeling of change in functional limitations with short- or long-term SNF stays.
Fig. 1.
Flow chart of study selection.
Results
Descriptive Results
The baseline characteristics for those with and without a SNF admission during follow-up are presented in Table 1. In total, 11.3% of the study population had a SNF admission during follow-up. The mean change in functional limitations from baseline was 3 times greater among those who had a SNF admission compared with those who did not. In total, 92.4% of the study population had a caregiver at some point during follow-up.
Table 1.
Baseline Characteristics of Study Population, n (%)
| Baseline characteristics | All (N = 3875) | SNF | |
|---|---|---|---|
| No (n = 3439) | Yes (n = 436) | ||
| Age | |||
| 65–69 y | 784 (20.2) | 755 (22.0) | 29 (6.6) |
| 70–74 y | 790 (20.4) | 735 (21.4) | 55 (12.6) |
| 75–79 y | 783 (20.2) | 696 (20.2) | 87 (20.0) |
| 80–84 y | 767 (19.8) | 653 (19.0) | 114 (26.2) |
| 85–90 y | 462 (11.9) | 370 (10.7) | 92 (21.1) |
| 90+ y | 289 (7.5) | 230 (6.7) | 59 (13.5) |
| Sex | |||
| Female | 2196 (56.7) | 1911 (55.6) | 285 (65.4) |
| Male | 1679 (43.3) | 1528 (44.4) | 151 (34.6) |
| Race/ethnicity | |||
| Other than non-Hispanic whites | 1088 (28.1) | 978 (28.4) | 110 (25.2) |
| Non-Hispanic whites | 2787 (71.9) | 2461 (71.6) | 326 (74.8) |
| Educational level | |||
| College or higher | 453 (11.7) | 400 (11.6) | 53 (12.1) |
| <9th grade education | 1515 (39.1) | 1308 (38.0) | 207 (47.5) |
| 9th-12th grade and high school | 280 (7.2) | 250 (7.3) | 30 (6.9) |
| Vocational training | 1627 (42.0) | 1481 (43.1) | 146 (33.5) |
| Cohabitation status | |||
| Living alone | 1288 (33.3) | 1078 (31.3) | 210 (48.2) |
| Living with a partner | 1616 (41.7) | 1502 (43.7) | 114 (26.1) |
| Living with others | 342 (8.8) | 316 (9.2) | 26 (6.0) |
| Living with partner and others | 629 (16.2) | 543 (15.8) | 86 (19.7) |
| Number of caregivers | |||
| 0 | 590 (15.2) | 536 (15.6) | 54 (12.4) |
| 1–2 | 2710 (69.9) | 2430 (70.6) | 280 (64.2) |
| 3+ | 575 (14.9) | 473 (13.8) | 102 (23.4) |
| Cognition | |||
| Normal | 3008 (77.6) | 2717 (79.0) | 291 (66.7) |
| Dementia | 867 (22.4) | 722 (21.0) | 145 (33.3) |
| Change in functional limitations | |||
| Mean (SD) range −16;18 | 0.35 (2.75) | 0.28 (2.62) | 0.95 (3.59) |
| Self-rated health | |||
| Mean (SD) range 0;5 | 2.77 (1.10) | 2.74(1.10) | 3.02 (1.09) |
| Comorbidities | |||
| Mean (SD) range 0;10 | 2.50(1.55) | 2.45 (1.53) | 2.89 (1.65) |
| Characteristics of Caregivers for Those with Caregivers* | |||
| All (N = 3580) | No (n = 3172) | Yes (n = 408) | |
| Paid caregiver | |||
| No | 3100 (86.6) | 2775 (87.5) | 325 (79.7) |
| Yes | 480 (13.4) | 397 (12.5) | 83 (20.3) |
| Medical supportive caregiver | |||
| No | 2020 (56.4) | 1867 (58.9) | 153 (37.5) |
| Yes | 1560 (43.6) | 1305 (41.1) | 255 (62.5) |
| Partner caregiver | |||
| No | 1615 (45.1) | 1354 (42.7) | 261 (64.0) |
| Yes | 1965 (54.9) | 1818 (57.3) | 147 (36.0) |
| Child caregiver | |||
| No | 332 (9.3) | 293 (9.2) | 39 (9.6) |
| Yes | 3248 (90.7) | 2879 (90.8) | 369 (90.4) |
| Neighbor/friend caregiver | |||
| No | 2546 (71.1) | 2264 (71.4) | 282 (69.1) |
| Yes | 1034 (28.9) | 908 (26.6) | 126 (30.9) |
| Any family caregiver | |||
| No | 63 (1.8) | 57 (1.8) | 6 (1.5) |
| Yes | 3517 (98.2) | 3115 (98.2) | 402 (98.5) |
| Number of caregiver sources | |||
| 1 | 434 (12.1) | 370 (11.7) | 64 (15.7) |
| 2–3 | 2756 (77.0) | 2453 (77.3) | 303 (74.3) |
| 4+ | 309 (10.9) | 349 (11.0) | 41 (10.0) |
Among those who have a caregiver at any time point during follow-up.
Joint Modeling of Functional Limitations and SNF Admissions
For the sociodemographic model (Table 2), only male sex significantly predicted improvement in functional limitation. For SNF admissions, non-Hispanic whites, <9th grade education, and more caregivers were associated with increased hazards of SNF admission compared with their respective references. In contrast, living with a partner or living with others was associated with lower hazard of SNF admission. One point worsening in change of functional limitations was associated with 26% higher hazard of SNF admission. The correlation between the person-specific intercept and person-specific slope was −0.12.
Table 2.
Covariates Adjusted Association Between Sociodemographic Factors and Caregivers and the Outcomes Change in Functional Limitations and SNF Admissions From Joint Modeling
| Exposure variables | Change in Functional Limitations Beta (95% CI)* | SNF Admission HR (95% CI)† |
|---|---|---|
| Sociodemographic Model‡ | ||
| Sex | ||
| Male vs female | −0.17 (−0.34; −0.00) | 1.02 (0.82; 1.26) |
| Race/ethnicity | ||
| Non-Hispanic whites vs others | 0.13 (−0.06; 0.32) | 1.31 (1.03; 1.65) |
| Education (reference: college or higher) | ||
| <9th grade education | 0.03 (−0.25; 0.31) | 1.57 (1.14; 2.17) |
| 9th-12 th grade and high school | −0.12 (−0.51; 0.27) | 1.19(0.75; 1.90) |
| Vocational training | 0.04 (−0.24; 0.33) | 0.95 (0.68; 1.33) |
| Cohabitation status (reference: living alone) | ||
| Living with a partner | 0.06 (−0.12; 0.24) | 0.42 (0.32; 0.54) |
| Living with others | 0.09 (−0.18; 0.36) | 0.46 (0.30; 0.70) |
| Living with partner and others | −0.06 (−0.26; 0.14) | 0.80 (0.61; 1.04) |
| Number of caregiver (reference: 0) | ||
| 1–2 | −0.04 (−0.21; 0.12) | 1.53 (1.13; 2.08) |
| 3+ | −0.04 (−0.25; 0.16) | 2.27 (1.61; 3.21) |
| Change in functional limitations‡ | ||
| One unit change | 1.26(1.22; 1.30) | |
| Caregiver Model4§ | ||
| Paid caregiver | ||
| Yes vs no | 0.02 (−0.16; 0.20) | 1.30 (0.99; 1.69) |
| Medical-supportive caregiver | ||
| Yes vs no | 1.06 (0.92; 1.20) | 1.95 (1.57; 2.43) |
| Partner caregiver | ||
| Yes vs no | −0.23 (−0.72; 0.25) | 0.48 (0.38; 0.61) |
| Child caregiver | ||
| Yes vs no | −0.08 (−0.40; 0.25) | 0.91 (0.63; 1.31) |
| Neighbor/friend caregiver | ||
| Yes vs. no | −0.09 (−0.28; 0.09) | 0.93 (0.72; 1.19) |
| Any family caregiver | ||
| Yes vs no | −0.15 (−0.85; 0.55) | 1.15 (0.49; 2.72) |
| Number of caregiver sources (reference: 1) | ||
| 2–3 | −0.26 (−0.02; 0.53) | 1.08 (0.78; 1.48) |
| 4+ | 0.17 (−0.19; 0.53) | 1.16 (0.69; 1.95) |
| Change in functional limitations‖ | ||
| One unit change | 1.20 (1.15;1.24) | |
Note. Bold values are statistically significant (P < .05).
CI, confidence interval.
Based on estimates from linear mixed effect model with random intercept and slope in joint model.
Based on estimates from Cox regression model in the joint model.
All exposures included in the same model + age, time, dementia, self-rated health, and comorbidities.
All exposures included in the same model + age, time, dementia, self-rated health, comorbidities, cohabitation, education, race/ethnicity, and sex.
This is a parameter estimated from the joint model.
For the caregiver model (Table 2), having a medical-supportive caregiver was significantly associated with worsening in functional limitations and was associated with increased hazard of SNF admission, whereas having a partner caregiver was protective of SNF admission. One point worsening in functional limitations was associated with 20% higher hazard of SNF admissions. The correlation between the person-specific intercept and person-specific slope was −0.21.
Sensitivity Analyses
The estimates from the separate Cox regression models are presented in Supplementary Table S3. Overall, the estimates from the joint models were more precise and greater in absolute value than the estimates from the separate analyses. The estimates from the joint models for first short- (n = 344) and long-term (n = 150) SNF admissions are presented in Supplementary Tables S4 and S5, respectively. Comparing the results for short-term SNF admission with the main results, the estimate for race/ethnicity became slightly greater in absolute value, whereas the estimates for cohabitations status, number of caregivers, and medical-supportive caregivers became slightly smaller. In the analyses for long-term SNF, the estimate for race/ethnicity was not significant, whereas the estimates for cohabitation status, number of caregivers, paid caregiver, medical-supportive caregiver, partner caregiver, and neighbor/friend caregiver became greater in absolute value.
Discussion
We applied a joint modeling approach to investigate the relationship between sociodemographic factors and characteristics of caregivers for SNF admissions accounting for repeated measures of change in functional limitations among older adults. The joint modeling analyses provided larger point estimates with higher precision than the corresponding estimates from separate Cox regression analyses.
Sociodemographic Factors and SNF Admission
For sex, similar to the present study, a meta-analysis of American studies found no association; whereas a systematic review of studies from developed countries reported inconsistent evidence.4,5 No sex difference in SNF admissions could be explained by the paradox that female individuals live longer than male individuals despite experiencing worse health.19 Thus, female individuals may experience worse health, while simultaneously, being less affected by it. These counteracting mechanisms may lead to no sex difference in SNF admissions. Furthermore, female individuals report poorer satisfaction with health and nursing care than male individuals,20–22 which may cause reluctance toward SNF admissions despite poorer health among female individuals. Moreover, extended social networks among female than male individuals23 and differences in social roles could also be explanations.
Similar to our results for race/ethnicity, the previous mentioned meta-analysis and systematic review found that white Americans were associated with more SNF admissions.4,5 These findings could be explained by race/ethnical differences in access to and quality of care facilities, family structures, social networks, and caregiving. The American history of race/ethnical segregation may have caused continued underrepresentation of African American and Hispanic older adults in assisted living facilities and nursing homes in the United States.24–27 Today, African American older adults experience placement at SNFs farther away from their community and with poor quality of care in comparison to white older adults.25,27,28 This may explain why African American older adults are less likely to endorse assisted living facilities and nursing home admissions.29 Instead, older African Americans receive more assistance and support from their family members.30,31 Also, African American family caregivers of older adults with dementia experience fewer depressive symptoms and are less likely to seek SNF admissions compared with their white counterparts.9,32
For educational level, we found increased hazards of SNF admissions among those with <9th grade education in contrast to the meta-analysis that found no influence of educational level and the systematic review that found mixed results.4,5 Mixed results may be explained by differences in categorization of educational level, geographic variation, and cohort effects. In the American meta-analysis, house ownership and higher levels of income were found to be protective of SNF admissions,4 which shows the importance of socioeconomic resources.
For cohabitation status, the present study and the meta-analysis found that having a partner or cohabiting was protective of SNF admissions, whereas the systematic review found moderate to weak evidence of increased risk of SNF admissions among unmarried adults and inconsistent results for those living alone.4,5 Different mechanisms in various cultural settings may explain the inconsistencies in the results.
Caregivers and SNF Admission
In the meta-analysis, the sizes of the associations of available informal caregiver and formal help for SNF admissions4 were comparable to those from our separate Cox regression analysis, whereas the results from the joint modeling analysis yielded larger estimates. Thus, sharing information on concurrent changes in physical limitations strengthens the relationship between caregivers and SNF admissions. This contradicts with a hypothesis that more caregivers indicate complex healthcare needs. Instead, these findings suggest that caregivers initiate SNF admissions when health problems become severe. Contrary to previous findings, 2 American studies found a protective effect of informal care on SNF admission.33,34 These differences may be explained by specific definitions of child caregivers in these 2 studies.
Variation in findings based on characteristics of caregivers could imply differences in mechanisms behind the relationship between various types of caregivers and SNF admissions. First, the strong protective effect of partner caregiver in our study is supported by 2 studies that showed caregiver co-residence was protective of SNF admissions; co-residence with adult offspring was most protective.35,36 One meta-analysis of patients with dementia showed that closeness of the relationship between caregiver and recipient was protective of SNF admissions,37 whereas another systematic review of patients with dementia did not find an effect of caregiver relationship.38 Partner caregiver may be of great importance because they are more persistent and provide more care, thus, they may strive more to keep the person in the community compared with children caregivers.39
A medical-supportive caregiver may act as a predictor of SNF admission because they, because of their medical insight, potentially advocate more for a SNF admission if the older adults’ healthcare needs are unmet. Also, this association could be explained by residual confounding based on health status and care need. For the other characteristics, the meta-analysis of patients with dementia showed that help with activities of daily living from family was associated with decreased risk of SNF admissions, whereas paid help and amount of family help and support were not.37 No association between overall family caregiver and SNF admission in the present study could be due to lack of variation because the majority had a family caregiver and these caregivers may be more distant compared with a partner care-giver. Also, family caregivers have been identified to either provide supportive caregiving or conflicting caregiving,40 which demonstrates that family caregivers can provide very different types of caregiving depending on their relationship with the care recipient.
The findings support that the relationship between caregivers and SNF admissions depends on the context and characteristics of the caregiver-recipient relationship.
Short- vs Long-term SNF Admissions
The results deviated slightly between short- and long-term SNF admissions. Cohabitation status, number of caregivers, and almost all characteristics of caregivers were more strongly associated with long-term SNF admissions. Caregiver characteristics may be more important for long-term compared with short-term SNF admissions because of different impacts of social network, poor health, and co-payment on short- and long-term SNF admissions. Long-term SNF admissions could reflect a more permanent need for care and help where geographically close relations and caregivers are more essential for the ability to stay in the community. This is supported by a previous study, which found that geographic closeness of children was protective of long- but not short-term SNF admissions.33 Number of caregivers and medical-supportive caregivers may be strongly associated with long-term SNF admissions because long-term SNF admissions reflect poorer health and more permanent care needs. This is supported by our findings that dementia was associated with greater hazards for long-term compared with short-term SNF admissions (results not shown). Finally, a stronger negative association between paid caregiver and long-term SNF admissions could be explained by the ability to pay for both of these services. Medicare coverage of SNF costs decreases when admissions extend >20 days.41
Difference in short- vs long-term SNF admissions was detected for race/ethnicity in that non-Hispanic white older adults were associated with greater hazards of short-term but not long-term SNF admissions. As described above, this may also be explained by long-term SNF admissions depending more on poor health and permanent care needs. Reluctance toward SNF admissions may, thus, become of less importance. Older African Americans’ greater resistance toward assisted living facilities and nursing home admissions29 may limit short-term admissions more than long-term admissions because the latter is more dependent on permanent care need. In addition, the tendency that African Americans more often are placed at SNFs with poorer quality of care and lower degree of successful discharges to the community25 may also increase the risk of an SNF admission turning into a long-term admission.
Strengths and Weaknesses
This study has several strengths. A major strength is the improved precision and validity of the estimates by jointly modeling SNF admissions and change in functional limitations. The study includes a large sample size of 3875 older adults based on a population with a high response rate (71%) at baseline and yearly follow-up over a 5-year period. Both outcomes included in the study are validated.16,17 We used inverse probability weights to account for those excluded because of missing follow-up information (Figure 1), and the method uses full likelihood-based method to account for missing data in the outcomes.
The study has a few limitations. We cannot draw causal conclusion given the observational nature of the study, although there is temporality and we applied advanced methods to assess the associations. Despite jointly modeling of change in functional limitations and SNF admission and including a range of measures of health status, there might be residual confounding. Another issue is that restricting the population to those covered by Medicare fee-for-service may hinder generalizability.
Conclusions and Implications
This study showed that a joint modeling approach provided greater and more precise estimates than the separate Cox regression analyses of sociodemographic factors and caregivers for SNF admissions. The study provided insight that could help identify high-risk populations for future interventions to prevent or delay SNF admissions. High-risk populations are identified by the following characteristics: non-Hispanic white, living alone, <9th grade education, and more caregivers. Among those with a caregiver, caregiver characteristics are shown to be important for SNF admissions. Older adults who have a partner caregiver are protected against a SNF admission, whereas those who have a medical-supportive caregiver are more likely to get a SNF admission. Future interventions should focus on helping those without a partner caregiver or those needing help and medical support to ensure independent living. To promote independent living, it is important to monitor functional limitations for older adults.
Supplementary Material
Acknowledgments
Terese Sara Høj Jørgensen was supported by the Social Inequalities in Ageing (SIA) project, funded by NordForsk, project no. 74637 and the Faculty of Health Sciences and Center for Healthy Aging, University of Copenhagen. The National Health and Aging Trends Study (NHATS) is sponsored by the National Institute on Aging (grant number NIA U01AG032947) through a cooperative agreement with the Johns Hopkins Bloomberg School of Public Health. Heather Allore, Janet L. MacNeil Vroomen, Brent Vander Wyk, and George O. Agogo were partially supported by the National Institute on Aging (R01 AG047891-01A1, P50AG047270, and a P30AG021342-14S, and the Yale Claude D. Pepper Older Americans Independence Center P30AG021342). Janet L. MacNeil Vroomen was additionally supported by a James Hudson Brown-Alexander Brown Coxe fellowship.
Footnotes
The authors declare no conflicts of interest.
The study provides insight that can help identify high-risk populations for future interventions to prevent or delay skilled nursing facility admissions among older adults in the United States.
References
- 1.The U.S. Centers for Medicare & Medicaid Services: Downloads NHE Tables. NHE Fact Sheet 2018. Available at: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/NationalHealthExpendData/NHE-Fact-Sheet.html. Accessed May 16, 2018.
- 2.Mor V, Intrator O, Feng Z, et al. The revolving door of rehospitalization from skilled nursing facilities. Health Aff (Millwood) 2010;29:57–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Buurman BM, Han L, Murphy TE, et al. Trajectories of disability among older persons before and after a hospitalization leading to a skilled nursing facility admission. J Am Med Dir Assoc 2016;17:225–231. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Gaugler JE, Duval S, Anderson KA, et al. Predicting nursing home admission in the U.S: A meta-analysis. BMC Geriatr 2007;7:13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Luppa M, Luck T, Weyerer S, et al. Prediction of institutionalization in the elderly. A systematic review. Age Ageing 2010;39:31–38. [DOI] [PubMed] [Google Scholar]
- 6.Kingston A, Davies K, Collerton J, et al. The enduring effect of education-socioeconomic differences in disability trajectories from age 85 years in the Newcastle 85+ Study. Arch Gerontol Geriatr 2015;60:405–411. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Clouston SA, Lawlor A, Verdery AM. The role of partnership status on late-life physical function. Can J Aging 2014;33:413–425. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Jacob ME, Marron MM, Boudreau RM, et al. Age, race, and gender factors in incident disability. J Gerontol A Biol Sci Med Sci 2018;73:194–197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Buhr GT, Kuchibhatla M, Clipp EC. Caregivers’ reasons for nursing home placement: Clues for improving discussions with families prior to the transition. Gerontologist 2006;46:52–61. [DOI] [PubMed] [Google Scholar]
- 10.Ibrahim JG, Chu H, Chen LM. Basic concepts and methods for joint models of longitudinal and survival data. J Clin Oncol 2010;28:2796–2801. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Kasper JD, Freedman VA. Findings from the 1st round of the National Health and Aging Trends Study (NHATS): Introduction to a special issue. J Gerontol B Psychol Sci Soc Sci 2014;69:S1–S7. [DOI] [PubMed] [Google Scholar]
- 12.Kasper JD, Freedman VA. National Health and Aging Trends Study User Guide: Rounds 1–6 Final Release. Baltimore: John Hopkins University School of Public Health; 2017. p. 30–42. [Google Scholar]
- 13.Wolff JL, Spillman B. Older adults receiving assistance with physician visits and prescribed medications and their family caregivers: Prevalence, characteristics, and hours of care. J Gerontol B Psychol Sci Soc Sci 2014;69:S65–S72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Kasper JDF, Spillman BC. Classification of Persons by Dementia Status in the National Health and Aging Trends Study Technical Paper #5. Baltimore: Johns Hopkins University School of Public Health; 2013. [Google Scholar]
- 15.Galvin JE, Roe CM, Powlishta KK, et al. The AD8: A brief informant interview to detect dementia. Neurology 2005;65:559–564. [DOI] [PubMed] [Google Scholar]
- 16.Gill TM, Williams CS. Development and validation of a functional outcome measure in the National Health and Aging Trends Study. J Gerontol Ser A Biol Sci Med Sci 2018;73:1111–1118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Yun H, Kilgore ML, Curtis JR, et al. Identifying types of nursing facility stays using Medicare claims data: An algorithm and validation. Health Serv Outcomes Res Method 2010;10:100–110. [Google Scholar]
- 18.Rizopoulos D. Joint Models for Longitudinal and Time-to-Event Data: With Applications in R. New York, USA: Chapman & Hall/CRC Biostatistics Series; 2012. p. 51–97. [Google Scholar]
- 19.Oksuzyan A, Juel K, Vaupel JW, et al. Men: Good health and high mortality. Sex differences in health and aging. Aging Clin Exp Res 2008;20:91–102. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Teunissen TAM, Rotink ME, Lagro-Janssen ALM. Gender differences in quality of care experiences during hospital stay: A contribution to patient-centered healthcare for both men and women. Patient Educ Couns 2016;99: 631–637. [DOI] [PubMed] [Google Scholar]
- 21.Elliott MN, Lehrman WG, Beckett MK, et al. Gender differences in patients’ perceptions of inpatient care. Health Serv Res 2012;47:1482–1501. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Foss C. Gender bias in nursing care? Gender-related differences in patient satisfaction with the quality of nursing care. Scand J Caring Sci 2002;16: 19–26. [DOI] [PubMed] [Google Scholar]
- 23.Fuhrer R, Stansfeld SA. How gender affects patterns of social relations and their impact on health: A comparison of one or multiple sources of support from “close persons”. Soc Sci Med 2002;54:811–825. [DOI] [PubMed] [Google Scholar]
- 24.Feng Z, Fennell ML, Tyler DA, et al. The Care Span: Growth of racial and ethnic minorities in US nursing homes driven by demographics and possible disparities in options. Health Aff (Millwood) 2011;30:1358–1365. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Rivera-Hernandez M, Rahman M, Mukamel DB, et al. Quality of post-acute care in skilled nursing facilities that disproportionately serve black and Hispanic patients. J Gerontol A Biol Sci Med Sci 2018;25:4985194. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Fabius CD, Thomas KS. Examining black-white disparities among Medicare beneficiaries in assisted living settings in 2014. J Am Med Dir Assoc 2018;14: 30547–30554. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Smith DB, Feng Z, Fennell ML, et al. Separate and unequal: Racial segregation and disparities in quality across U.S. nursing homes. Health Aff (Millwood) 2007;26:1448–1458. [DOI] [PubMed] [Google Scholar]
- 28.Rahman M, Foster AD. Racial segregation and quality of care disparity in US nursing homes. J Health Econ 2015;39:1–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Wolff JL, Kasper JD, Shore AD. Long-term care preferences among older adults: A moving target? J Aging Soc Policy 2008;20:182–200. [DOI] [PubMed] [Google Scholar]
- 30.Taylor RJ, Chatters LM, Woodward AT, et al. Racial and ethnic differences in extended family, friendship, fictive kin and congregational informal support networks. Fam Relat 2013;62:609–624. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Peek MK, Coward RT, Peek CW. Race, aging, and care: Can differences in family and household structure account for race variations in informal care? Res Aging 2000;22:117–142. [Google Scholar]
- 32.Skarupski KA, McCann JJ, Bienias JL, et al. Race differences in emotional adaptation of family caregivers. Aging Ment Health 2009;13:715–724. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Charles KK, Sevak P. Can family caregiving substitute for nursing home care? J Health Econ 2005;24:1174–1190. [DOI] [PubMed] [Google Scholar]
- 34.Lo Sasso AT, Johnson RW. Does informal care from adult children reduce nursing home admissions for the elderly? Inquiry 2002;39:279–297. [DOI] [PubMed] [Google Scholar]
- 35.Betini RSD, Hirdes JP, Lero DS, et al. A longitudinal study looking at and beyond care recipient health as a predictor of long term care home admission. BMC Health Serv Res 2017;17:709. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.McCann M, Donnelly M, O’Reilly D. Living arrangements, relationship to people in the household and admission to care homes for older people. Age Ageing 2011;40:358–363. [DOI] [PubMed] [Google Scholar]
- 37.Toot S, Swinson T, Devine M, et al. Causes of nursing home placement for older people with dementia: A systematic review and meta-analysis. Int Psychogeriatr 2017;29:195–208. [DOI] [PubMed] [Google Scholar]
- 38.Gaugler JE, Yu F, Krichbaum K, et al. Predictors of nursing home admission for persons with dementia. Med Care 2009;47:191–198. [DOI] [PubMed] [Google Scholar]
- 39.Pinquart M, Sorensen S. Spouses, adult children, and children-in-law as caregivers of older adults: A meta-analytic comparison. Psychol Aging 2011;26: 1–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Riffin C, Van Ness PH, Iannone L, et al. Patient and caregiver perspectives on managing multiple health conditions. J Am Geriatr Soc 2018;66: 1992–1997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Medicare.gov. Skilled nursing facility (SNF) care. 2018. Available at: https://www.medicare.gov/coverage/skilled-nursing-facility-care.html; 2018. Accessed May 16, 2018.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.