Abstract
Objective
To compare the outcomes of postacute care between home health (HH) and skilled nursing facilities (SNFs) following hospitalization among Medicare beneficiaries with a diagnosis of dementia.
Data Sources
100% MedPAR data, Minimum Data Set, and Outcome and Assessment Information Set assessment data from January 1, 2015 to December 31, 2016.
Study Design
Retrospective cohort analysis using an instrumental variable design to compare outcomes (30‐day readmission and mortality, 100‐day mortality) of HH versus SNF following acute hospitalization. We used the differential distance between patients' home and the closest HH agency and SNF to instrument for nonrandom allocation of patients.
Data Collection/Extraction Methods
We identified hospital discharges followed by SNF and HH stays for Medicare fee‐for‐service beneficiaries with dementia. We excluded beneficiaries younger than age 65, admitted to the hospital from a nursing home, or enrolled in hospice. We identified dementia using validated diagnostic codes with a 3‐year look‐back.
Principal Findings
Our sample included 977,946 beneficiaries with a diagnosis of dementia; 297,732 (30.4%) received HH, while 680,214 (69.6%) went to SNF. Overall, 16.8% were readmitted to the hospital and 6.1% died within 30 days, while 15.4% died within 100 days of hospital discharge. In the instrumental variable analysis, there were no differences in any outcome between the two postacute care settings.
Conclusions
Medicare beneficiaries with a diagnosis of dementia receiving postacute care in HH or SNF experienced similar rates of readmission and mortality across settings. This finding raises important questions about current postacute care referral patterns, given 7 in 10 patients with a diagnosis of dementia in our sample were discharged to SNF.
Keywords: dementia, health policy, home health, postacute care, skilled nursing facility
What is known on this topic
The prevalence of dementia and of postacute care use is rising in the United States.
No studies have compared outcomes of different postacute care settings for patients with dementia even though they vary significantly on cost and services provided.
What this study adds
Medicare beneficiaries with a diagnosis of dementia receiving HH care had similar rates of hospital readmission and mortality to those receiving care in a SNF.
This finding raises questions about the appropriateness of postacute care referral patterns in beneficiaries without a diagnosis of dementia (where 50% of postacute care use is in SNF) and in beneficiaries with a diagnosis of dementia (where 70% of postacute care use in this sample is in SNF).
1. INTRODUCTION
Currently, more than 40% of Medicare beneficiaries receive postacute care in a skilled nursing facility (SNF) or from a home health (HH) agency after hospital discharge, for which Medicare pays more than $60 billion annually. 1 As a result, postacute care has received national policy attention as one of the most rapidly rising areas of Medicare costs. 2 , 3 , 4 It is also one of the most vulnerable times in the life of a Medicare beneficiary: approximately one in four beneficiaries are readmitted to the hospital from postacute care settings, and slightly more than half of the beneficiaries have returned to the community from SNF by 100 days after hospital discharge. 5 , 6 , 7
Over 3.1 million Medicare beneficiaries currently carry a diagnosis of dementia, and this number is expected to increase to more than 14 million by 2050. 8 These patients, like many Medicare beneficiaries, often receive postacute care after a hospitalization to help them recuperate and rehabilitate. 9 , 10 , 11 Results from the few published studies of the outcomes of patients with dementia in the SNF setting show mixed results regarding whether patients with dementia enjoy similar benefits from SNF compared to patients without dementia, 12 or whether treatment in SNF could represent a “burdensome” transition followed by long‐term institutionalization or death. 13 , 14
Individuals with dementia represent a particularly vulnerable population whose postacute care needs may differ from other patients. Despite the fact that both postacute care use and dementia are highly prevalent in the Medicare population, very few studies have compared the outcomes of different postacute care settings for patients with dementia, leaving patients, caregivers, and clinicians uncertain how to make optimal decisions around hospital discharge. Recent studies suggest the trade‐offs between these two settings could be large: for example, Medicare beneficiaries discharged to SNF experience a 30‐day readmission reduction of 5.6 percentage points compared to HH. 15
We sought to compare outcomes between SNF and HH postacute care for Medicare beneficiaries with dementia, to help inform discharge decision making and payment incentives regarding postacute care use.
2. METHODS
This was a retrospective, observational study using national Medicare data from 2015 and 2016 from the 100% MedPAR, Minimum Data Set (MDS) 3.0, Outcome and Assessment Information Set (OASIS), Hospice, and the Medicare Master Beneficiary Summary Files (MBSF; including Chronic Conditions segment). These data were the most up‐to‐date available at the time of project start. Further, when we evaluated trends in postacute care referral rates and outcomes in Medicare beneficiaries with dementia over a 5‐year time period (2012–2016), we found these were stable (manuscript under review).
2.1. Population description
We identified Medicare fee‐for‐service beneficiaries discharged from the hospital between January 1, 2015 and December 31, 2016 (identified in MedPAR) with a HH or SNF stay starting within 3 days of hospital discharge (identified in OASIS or MDS, respectively). We included individuals with a diagnosis of dementia present in claims or assessment data in the 3 years prior to the index hospitalization. We excluded Medicare beneficiaries who were younger than age 65, were in a nursing home within 30 days of hospital admission, discharged to a swing bed or hospice, or who had both SNF and HH admission assessments on the same day following hospital discharge. To avoid the inclusion of multiple hospitalizations and postacute care stays of single patients, we included only the first hospitalization followed by postacute care stay for each beneficiary (Appendix Figure 1, flowchart with numbers of beneficiaries included or excluded at each step).
2.2. Predictor variables
2.2.1. Dementia diagnosis
We used a 3‐year look‐back for validated diagnostic codes for dementia used as part of the chronic conditions segment of the MBSF. The MBSF identifies “Alzheimer's Disease” and “Alzheimer's disease and related disorders or senile dementia” using International Classification of Diseases (ICD)‐9/10, current procedural terminology, and Healthcare Common Procedure Coding System codes from the prior 3 years on any inpatient, SNF, HH, hospital outpatient, or carrier claims.
2.2.2. Beneficiary characteristics
We also identified the age, sex, race, and ethnicity of beneficiaries, dual‐eligibility for Medicare‐Medicaid enrollment, comorbidities (measured by Elixhauser comorbidity score with 1‐year look‐back), and functional status at the start of the postacute care episode (measured as Activity of Daily Living deficiencies using OASIS AND MDS items). We identified how many of six Activity of Daily Living (ADL) dependencies (defined as “needing assistance”) the beneficiary had at the start of their postacute care episode related to bathing, dressing, feeding, transferring, toileting, and continence. We calculated the length of stay for the index hospitalization and identified whether that hospital stay included an intensive care unit admission. We calculated the number of hospitalizations, SNF stays, and HH episodes the beneficiary had in the prior year.
2.2.3. Locations and distance
We identified the centroid of the zip code of beneficiary location, and then identified all hospital, HH agency, and SNF locations using the Medicare Provider of Services file. We first classified HH and SNFs into hospital service areas as defined by the Dartmouth Atlas, but found a significant number without a HH or SNF. Thus, we then classified beneficiaries, HH agencies, and SNFs into hospital referral regions (HRRs).
2.2.4. Outcomes
We calculated the rate of unplanned hospital readmissions within 30 days of hospital discharge using Centers for Medicare and Medicaid Services planned readmission algorithm (v4.0), accounting for censoring by adjusting for number of days alive in the 30‐day period as our primary outcome. We also calculated 30‐day and 100‐day mortality rates.
2.3. Statistical analysis
We first described characteristics of beneficiaries with a diagnosis of dementia discharged to HH and SNF and their outcomes. Beneficiaries with dementia are not randomly allocated to the type of postacute care, and this nonrandom allocation introduces bias when comparing outcomes across different types of postacute care. We sought to address this by first conducting multivariable adjustment for beneficiary characteristics using logistic regression. We included fixed effects for year and hospital, and adjusted for clustering using Huber–White sandwich estimators to compare outcomes between HH and SNF.
However, as prior research has demonstrated, there may be significant residual confounding from patient selection on unobservable characteristics using this approach, 15 we also used an instrumental variable design. Instrumental variable analyses approximate random assignment of patients to treatment groups based on an instrumental variable, which strongly influences type of treatment received but is not linked to the outcome or other factors associated with the outcome, and is randomly distributed across the population of interest. As distance to postacute care has been shown to be a strong determinant of postacute care choice, 15 , 16 , 17 we used the difference in the distance between the centroid of the zip code of the patient's residence and the nearest HH agency and SNF within their HRR. We evaluated a continuous measure of differential distance and found it did not meet the monotonicity assumption of the instrumental variable model. Thus, consistent with prior work, we dichotomized differential distance between the HH agency and SNF to equal 1 if the beneficiary lives closer to HH agency than SNF, and zero if they live closer to SNF (Appendix Table 1a). 15
We tested several other assumptions of our model. First, we evaluated the strength of our instrument in predicting treatment (HH vs. SNF) using an F‐statistic (F = 31.1) in the overall cohort. The F‐statistic measures how strongly differential distance predicts treatment choice. It is based on the fact that the probability of receiving care from a HH agency is 1.7 percentage points higher among patients who live closer to a HH agency than to an SNF. F‐Statistics greater than 10 are generally considered strong. Second, we examined the strength of the instrument in cases where the instrument should not influence the treatment received (e.g., in patients whose home zip code was very far from the hospital of their index admission—a “vacationer”). One of the reasons this falsification test is useful is to address concerns about the referral patterns of the distant hospital to postacute care affecting treatment. It is important to clarify that as our instrument measures the differential distance between the centroid of the patient's home zip code and the nearest SNF and HH to their home, the local supply or referral patterns at a distant hospital should not influence postacute care choice. Similarly, the falsification test helps evaluate whether factors related to the neighborhood where patients live (rather than the differential distance to HH vs. SNF) is driving treatment choice. As expected, we found that the instrument did not influence postacute care choice in this situation (Appendix Table 1b). Third, we examined the balance of measured confounders; demonstrating good balance on measured confounders may increase confidence that unmeasured confounders may be similarly balanced. We found most measured covariates were balanced with the exception of race/ethnicity, and adjusted for residual imbalances in our IV model. Fourth, we calculated the proportion of “compliers” in our sample: those to whom our results more clearly apply, and compared the “complier” or “marginal” group to our overall sample to provide insights into how their characteristics may be similar or different. 18
We conducted 2‐stage least‐squares regressions (the first stage predicted likelihood of discharge with HH based on the value of the instrument; the second estimated the association between predicted admission to HH from the first stage and the outcome of interest). These models adjusted for beneficiary characteristics listed above, included fixed effects for hospital, year, and diagnosis‐related group (DRG), and adjusted for clustering at the hospital level using Huber–White sandwich estimators.
2.3.1. Sensitivity analyses
We first created a composite outcome of readmission or death within 30 days given competing risks of these outcomes and repeated our main analysis. To ensure we were evaluating a previously community‐dwelling cohort, we tested excluding nursing home residence in the year prior to the index admission. As postacute care treatment may vary significantly by diagnosis, in addition to including DRG fixed effects, we calculated the top 10 most common DRGs listed as the principal hospital discharge code and compared them in the “complier” group to the overall sample to compare the prevalence of these conditions. The study was considered exempt by the University of Pennsylvania IRB.
3. RESULTS
Of 4,870,866 beneficiaries discharged to SNF or HH during our study period, our sample included 977,946 fee‐for‐service beneficiaries with dementia. In this group, 297,732 (30.4%) received HH after hospital discharge, while 680,214 went to SNF (69.6%). Beneficiaries in both groups were older (mean age 83 years), more likely to be female (63%) and white (85%); 19.8% were dual‐eligible (Table 1). Functional impairment was significant in this cohort; only 6% of beneficiaries had fewer than four ADL dependencies at the start of their postacute care episode. Beneficiaries discharged to SNF generally had more dependencies, and in particular, needed more assistance with eating (HH: 12.6%, SNF: 32.6%).
TABLE 1.
Characteristics of study cohort stratified by discharge to HH or SNF
| Characteristic | Total | HH | SNF |
|---|---|---|---|
| N = 977,946 (100%) | N = 297,732 (30.4%) | N = 680,214 (69.6%) | |
| Age, mean, years (SD) | 83.6 (8.2) | 82.7 (8.3) | 84.0 (8.1) |
| Male (%) | 362,495 (37.1%) | 114,494 (38.5%) | 248,001 (36.5%) |
| Race | |||
| White | 832,050 (85.1%) | 244,908 (82.3%) | 587,142 (86.3%) |
| Black | 95,079 (9.7%) | 33,302 (11.2%) | 61,777 (9.1%) |
| Asian | 15,465 (1.6%) | 6032 (2.0%) | 9433 (1.4%) |
| Other | 13,746 (1.4%) | 4708 (1.6%) | 9038 (1.3%) |
| Hispanic ethnicity | 18,351 (1.9%) | 7550 (2.5%) | 10,801 (1.6%) |
| Dual‐eligible | 193,288 (19.8%) | 58,420 (19.6%) | 134,868 (19.8%) |
| Elixhauser comorbidity score (SD) | 5.0 (3.0–7.0) | 5.0 (3.0–7.0) | 5.0 (3.0–7.0) |
| Prior hospitalizations, mean (SD) | 0.7 (1.2) | 0.9 (1.4) | 0.7 (1.2) |
| Prior SNF stays, mean (SD) | 0.4 (1.1) | 0.3 (1.0) | 0.4 (1.1) |
| Prior HH episodes, mean (SD) | 0.7 (1.2) | 0.7 (1.3) | 0.7 (1.2) |
| Hospital LOS, days (SD) | 6.4 (6.0) | 5.5 (5.1) | 6.8 (6.3) |
| ICU stay (%) | 26.0 (43.9) | 24.4 (42.9) | 26.7 (44.2) |
| Functional impairment at start of postacute care episode | |||
| Bathing | 891,171 (91.1%) | 266,474 (89.5%) | 624,697 (91.8%) |
| Dressing | 900,353 (92.1%) | 283,084 (95.1%) | 617,269 (90.7%) |
| Toileting | 898,228 (91.8%) | 281,360 (94.5%) | 616,868 (90.7%) |
| Transferring | 893,647 (91.4%) | 281,390 (94.5%) | 612,257 (90.0%) |
| Eating | 259,473 (26.5%) | 37,661 (12.6%) | 221,812 (32.6%) |
| Continence | 704,267 (72.0%) | 206,238 (69.3%) | 498,029 (73.2%) |
| Number of ADL dependencies at start of postacute care episode | |||
| 0 | 4075 (0.4%) | 37 (0.0%) | 4038 (0.6%) |
| 1 | 10,308 (1.1%) | 2837 (1.0%) | 7471 (1.1%) |
| 2 | 14,734 (1.5%) | 6087 (2.0%) | 8647 (1.3%) |
| 3 | 29,787 (3.0%) | 17,287 (5.8%) | 12,500 (1.8%) |
| 4 | 189,163 (19.3%) | 80,706 (27.1%) | 108,457 (15.9%) |
| 5 | 451,224 (46.1%) | 152,903 (51.4%) | 298,321 (43.9%) |
| 6 | 234,205 (23.9%) | 33,666 (11.3%) | 200,539 (29.5%) |
| Missing | 44,450 (4.5%) | 4209 (1.4%) | 40,241 (5.9%) |
| 30‐day readmission | 164,397 (16.8%) | 50,825 (17.1%) | 113,572 (16.7%) |
| 30‐day mortality | 59,521 (6.1%) | 9204 (3.1%) | 50,317 (7.4%) |
| 100‐day mortality | 150,954 (15.4%) | 29,263 (9.8%) | 121,691 (17.9%) |
| 30‐day mortality or readmit | 198,071 (20.3%) | 55,213 (18.5%) | 142,858 (21.0%) |
Abbreviations: ADL, activity of daily living; HH, home health; ICU, intensive care unit; LOS, length of stay; SD, standard deviation; SNF, skilled nursing facility.
Overall, 164,397(16.8%) beneficiaries were readmitted to the hospital and 59,521 (6.1%) died within 30 days, while 150,954 (15.4%) died within 100 days of hospital discharge. In unadjusted analyses, beneficiaries discharged to HH had higher rates of 30‐day unplanned readmission (HH: 17.1%, SNF: 16.7%), but those discharge to SNF had higher mortality rates (30 days: HH 3.1%, SNF 7.4%; 100 days: HH 9.8%, SNF 17.9%). Those discharged to SNF had a higher rate of a composite 30‐day outcome of death or hospital readmission (HH: 18.5%, SNF: 21.0%).
After risk adjustment, patients discharged to HH had similar 30‐day readmission rates (OR 1.01, 95% CI 1.00–1.02) but much lower odds of 30‐ and 100‐day mortality (OR 0.42, 95% CI 0.41–0.43 for 30‐day mortality, OR 0.51, 95% CI 0.51–0.52 for 100‐day mortality). Similarly, patients discharged to HH had reduced odds of a composite 30‐day outcome of readmission or death (OR 0.85, 95% CI 0.84–0.86).
Characteristics of the two groups were similar when stratified by the instrument (Table 2). In the instrumental variable analyses, there were no statistically significant differences in any outcome between HH and SNF. This included the composite outcome for mortality or readmission within 30 days (Table 3). Findings from sensitivity analyses were similar to the primary analysis (Table 3 and Appendix Table 2).
TABLE 2.
Characteristics of study cohort, stratified by instrumental variable
| Characteristic | Closer to SNF | Closer to HH |
|---|---|---|
| N = 874,265 (89.4%) | N = 103,399 (10.6%) | |
| Age, mean, years (SD) | 83.6 (8.2) | 83.5 (8.3) |
| Male (%) | 323,976 (37.1%) | 38,405 (37.1%) |
| Race | ||
| White | 748,106 (85.6%) | 83,719 (81.0%) |
| Black | 84,276 (9.6%) | 10,785 (10.4%) |
| Asian | 12,561 (1.4%) | 2892 (2.8%) |
| Other | 11,843 (1.4%) | 1893 (1.8%) |
| Hispanic ethnicity | 14,652 (1.7%) | 3687 (3.6%) |
| Dual‐eligible | 170,470 (19.5%) | 22,770 (22.0%) |
| Elixhauser comorbidity score (SD) | 5.0 (3.0–7.0) | 5.0 (3.0–7.0) |
| Prior hospitalizations, mean (SD) | 0.7 (1.3) | 0.8 (1.4) |
| Prior SNF stays, mean (SD) | 0.4 (1.1) | 0.3 (1.0) |
| Prior HH episodes, mean (SD) | 0.7 (1.2) | 0.8 (1.2) |
| Hospital LOS, days (SD) | 6.4 (6.0) | 6.5 (6.1) |
| ICU stay, % (SD) | 25.8 (43.8) | 27.4 (44.6) |
| Functional impairment at start of postacute care episode | ||
| Bathing | 796,517 (91.1%) | 94,400 (91.3%) |
| Dressing | 804,214 (92.0%) | 95,882 (92.7%) |
| Toileting | 802,463 (91.8%) | 95,506 (92.4%) |
| Transferring | 798,056 (91.3%) | 95,334 (92.2%) |
| Eating | 229,869 (26.3%) | 29,510 (28.5%) |
| Continence | 628,822 (71.9%) | 75,245 (72.8%) |
| Number of ADL dependencies at start of post‐acute care episode | ||
| 0 | 3718 (0.4%) | 356 (0.3%) |
| 1 | 9442 (1.1%) | 863 (0.8%) |
| 2 | 13,414 (1.5%) | 1316 (1.3%) |
| 3 | 26,839 (3.1%) | 2943 (2.8%) |
| 4 | 169,504 (19.4%) | 19,609 (19.0%) |
| 5 | 404,036 (46.2%) | 47,065 (45.5%) |
| 6 | 207,493 (23.7%) | 26,632 (25.8%) |
| Missing | 39,819 (4.6%) | 4615 (4.5%) |
Abbreviations: ADL, activity of daily living; HH, home health; ICU, intensive care unit; LOS, length of stay; SD, standard deviation; SNF, skilled nursing facility.
TABLE 3.
Outcomes of instrumental variable analysis of HH versus SNF
| Instrumental regression of all patients with dementia | |||
|---|---|---|---|
| Number of observations | Difference (SE) | p‐Value | |
| 30‐day readmission | 969,638 | 0.12 (0.14) | 0.41 |
| 30‐day mortality | 969,638 | −0.03 (0.08) | 0.70 |
| 100‐day mortality | 969,638 | 0.12 (0.13) | 0.36 |
| Mortality or readmission within 30 days | 969,638 | 0.09 (0.15) | 0.55 |
Abbreviations: HH, home health; SE, standard error; SNF, skilled nursing facility.
We estimated 30.5% of the total population are “compliers” or “marginal” patients. We describe characteristics of the “compliers” or “marginal” patients compared to the total cohort in Appendix Table 3. Compared to the overall cohort, compliers were more likely to be dual‐eligible (29.8% vs. 19.8%) with small differences in other demographic characteristics or DRGs.
4. DISCUSSION
In a sample of nearly 1 million Medicare beneficiaries discharged to postacute care following hospitalization, outcomes associated with the use of HH services and SNF were equivalent in terms of unplanned readmissions and 30‐ and 100‐day mortality. These are important findings given more than 7 in 10 Medicare beneficiaries with dementia are currently discharged to SNF for postacute care at a much higher cost than HH, and the prevalence of both dementia and postacute care use are increasing in the United States.
Prior evaluations of postacute care use in patients with dementia have suggested SNF may be overutilized in this population, particularly among patients with moderate to severe dementia who may be “rehabbed to death.” 13 , 14 , 19 , 20 Even Medicare beneficiaries with severe dementia living in nursing homes are often sent to SNF for postacute care after a hospitalization. 20 This commonly results in multiple care transitions across hospitals and nursing facilities, particularly near the end of life. 11 , 19 Patients, families, and clinicians may not have sufficient supports at home, or use SNF as a bridge to long‐term care. 14 , 21 , 22 However, few prior studies have compared SNF and HH outcomes in Medicare beneficiaries with dementia, and the equivalent outcomes found in our analysis was surprising, particularly given studies with strong causal inference designs suggest SNFs reduce readmissions compared to HH in Medicare beneficiaries overall. 15 , 23
There are several additional potential explanations for this finding. First, patients with dementia may benefit more from a supportive home environment than a transition from one unfamiliar facility to another. Second, the more medical and rehabilitative focus of many SNFs may take precedence over a focus on high‐quality dementia care. In contrast, HH services can co‐exist with home and community‐based services to support patients with dementia in the community at the direction of the patient and their family. However, we are unable to capture all the relevant outcomes of this decision. For example, caregiver burden is an important outcome that may be quite different between home and SNF care for patients with dementia, but is not captured in our datasets. 24 , 25 , 26
Taken together with prior research, these findings suggest that current practice could lead to overutilization of SNF for postacute care in Medicare beneficiaries with dementia. However, for home care to become more prevalent following hospital discharge, several barriers would have to overcome. First, more support for caregivers—who are frequently those bearing the “cost” of care at home— is clearly needed. Frequently, patients need both medical supports provided by a HH agency (paid by Medicare), but also home and community‐based long‐term services and supports (usually paid by Medicaid) for more round‐the‐clock support for activities of daily living. Novel treatment paradigms such as Rehabilitation at Home (building on Hospital at Home models) hold promise, 27 as does new flexibility in the ability of Medicare Advantage plans to pay for home and community‐based services in addition to postacute care. 28 Second, decision making about postacute care in the hospital is of low quality, and interventions to help patients and families understand trade‐offs of HH and SNF are important to develop and test. 22 , 29 Third, improved advanced care planning around the hospital to postacute care transition is needed to avoid postacute care stays that are more consistent with harm than benefit. 13 Risk prediction scores have been developed to identify patients at high risk for adverse outcomes in SNF and HH, and several advanced care planning interventions in SNF are being tested in trials. 30 , 31 , 32 , 33 , 34 However, these efforts may be occurring too late in the care trajectory (e.g., once the patient is already in SNF).
Strengths of the analysis include the use of a large dataset with robust identification of hospitalizations, postacute care stays, and outcomes. We were able to incorporate important exclusions such as use of hospice, which could confound our mortality findings, as well as data from prior hospital, and postacute care use and physical function on postacute care admission. Although dementia diagnoses may underestimate the true prevalence of dementia, we combined multiple data sources from the inpatient, outpatient, and postacute care settings to robustly identify Medicare beneficiaries with this diagnosis. Our analysis using traditional multivariable risk adjustment suggests unmeasured confounding is a significant concern, and the instrumental variable analysis is one way to potentially better address this confounding. In particular, instrumental variable methods may better address important unmeasured factors in our data, such as caregiver support. While caregiver support may play a large role in hospital discharge decision making, the presence of a caregiver is unlikely to confound our findings unless caregiver support is distributed in such a way that it is correlated with differential distance—for example, if patients who live closer to SNF than HH are more or less likely to have caregiver support. Although we are not aware of any data that speak to this, it does not seem plausible that caregiver availability and differential distance to postacute care provider type are strongly correlated. However, we are unable to directly test for the relationship between any unobserved factors and the instrument, a limitation of this analytic approach.
It is important to note that our results apply to the “marginal” patient in our analysis (those for whom the decision about HH and SNF might be strongly influenced by location). While we estimate this group constitutes 30.5% of the overall population, this suggests caution should be applied when seeking to apply these findings to broader populations of patients with dementia. The change between ICD‐9 and ICD‐10 codes has led to changes in the calculated prevalence of dementia, introducing a potential concern with our identification strategy. 35 However, this is unlikely to be related to our outcomes of interest. The inclusion of year fixed effects helps to address this concern, but it remains a limitation of this analysis. We were unable to complete subanalyses within specific DRGs due to small sample size. Our results only apply to Medicare fee‐for‐service beneficiaries given we did not have Medicare Advantage claims data that would allow us to mirror our identification strategy in that population.
Unlike SNF, the business address of a HH agency is not where care is delivered, raising the question of why distance to the business address of the agency would predict treatment choice. First, qualitative studies demonstrate that distance to a HH provider is an important criterion for patients and caregivers when choosing among agencies. 36 In fact, location of HH agency is the main criterion hospital discharge planners use to create a list for patients to choose from prior to discharge, and is how HH compare is designed (entering the patient's address to find nearby home care agencies). Second, the business office is likely to be proximate to where care is delivered. As most referrals are driven by the location of the business address, not providing care in that area would not be in the economic interest of the home care agency. The business address may also serve relevant administrative functions that make it important for it to be proximate to the areas of care delivery by HH agency staff. For example, HH supplies would be delivered to the business address, making it unlikely the business address would be far from where care is delivered. New referrals, billing questions, and other administrative functions are also consolidated at the business address, potentially requiring HH staff to travel to the office. Empirically, prior studies and our own have demonstrated differential distance strongly predicts treatment choice. 15 Thus, although our analysis is limited, as we are unable to measure the location of where care is delivered by the home care agency, the differential distance between a beneficiary's home and HH versus SNF can be thought of as a proxy for the local supply of HH and SNF.
Another important concern is whether the patient is actually choosing between SNF and HH, or choosing within just one group. The decision making process regarding the choice of HH, and between SNF and HH, is understudied, and this is an important potential limitation. However, existing literature suggests the decision making process about postacute care is iterative, with considerable uncertainty about which type of postacute care may be better for patients 22 , 36 , 37 , 38 , 39 —leading to substantial variability in use nationally. 40 It is therefore plausible for a sizeable subpopulation of patients that the “correct” postdischarge supports are not clear, and the local supply of SNF or HH could influence decision making.
As the older adult population of the United States grows, important questions remain about how best to provide postacute care to meet the needs of patients with dementia and their caregivers that maximizes value. Our findings contribute to an important national dialog regarding how to better support patients and caregivers following hospital discharge to home settings, since HH may provide similar outcomes at lower cost following hospitalization for Medicare fee‐for‐service beneficiaries with a diagnosis of dementia.
Supporting information
Appendix S1 Supporting information
ACKNOWLEDGMENTS
The study was funded by the University of Pennsylvania Institute on Aging. The Hospital Service Area and Hospital Referral Region were obtained from Dartmouth Atlas Data website, which was funded by the Robert Wood Johnson Foundation, The Dartmouth Clinical and Translational Science Institute, under award number UL1TR001086 from the National Center for Advancing Translational Sciences (NCATS) of the National Institutes of Health (NIH), and in part, by the National Institute of Aging, under award number U01 AG046830.
Burke RE, Xu Y, Ritter AZ, Werner RM. Postacute care outcomes in home health or skilled nursing facilities in patients with a diagnosis of dementia. Health Serv Res. 2022;57(3):497-504. 10.1111/1475-6773.13855
Funding information National Institute of Aging, Grant/Award Number: U01 AG046830; National Center for Advancing Translational Sciences, Grant/Award Number: UL1TR001086; Robert Wood Johnson Foundation, The Dartmouth Clinical and Translational Science Institute; University of Pennsylvania Institute on Aging
REFERENCES
- 1. June 15th Data Book: Health Care Spending and the Medicare Program. Accessed August 5, 2015. http://www.medpac.gov/-documents-/data-book
- 2. Mechanic R. Post‐acute care—the next frontier for controlling Medicare spending. N Engl J Med. 2014;370(8):692‐694. 10.1056/NEJMp1315607 [DOI] [PubMed] [Google Scholar]
- 3. Ackerly DC, Grabowski DC. Post‐acute care reform—beyond the ACA. N Engl J Med. 2014;370(8):689‐691. 10.1056/NEJMp1315350 [DOI] [PubMed] [Google Scholar]
- 4. Chandra A, Dalton MA, Holmes J. Large increases in spending on postacute care in Medicare point to the potential for cost savings in these settings. Health Aff. 2013;32(5):864‐872. 10.1377/hlthaff.2012.1262 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Mor V, Intrator O, Feng Z, Grabowski DC. The revolving door of rehospitalization from skilled nursing facilities. Health Aff. 2010;29(1):57‐64. 10.1377/hlthaff.2009.0629 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Burke RE, Whitfield EA, Hittle D, et al. Hospital readmission from post‐acute care facilities: risk factors, timing, and outcomes. J Am Med Dir Assoc. 2016;17(3):249‐255. 10.1016/j.jamda.2015.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Kramer A, Lin M, Fish R, Min S. Refinement of Community Discharge, Potentially Avoidable Readmission, and Functional Outcome SNF Quality Measures, for Fiscal Years 2011, 2012, and 2013. Accessed December 4, 2017. http://www.medpac.gov/docs/default-source/contractor-reports/refinement-of-community-discharge-potentially-avoidable-readmission-and-functional-outcome-snf-quali.pd
- 8. Goodman RA, Lochner KA, Thambisetty M, Wingo TS, Posner SF, Ling SM. Prevalence of dementia subtypes in United States Medicare fee‐for‐service beneficiaries, 2011‐2013. Alzheimers Dement. 2017;13(1):28‐37. 10.1016/j.jalz.2016.04.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Burke RE, Juarez‐Colunga E, Levy C, Prochazka AV, Coleman EA, Ginde AA. Rise of post‐acute care facilities as a discharge destination of US hospitalizations. JAMA Intern Med. 2015;175(2):295‐296. 10.1001/jamainternmed.2014.6383 [DOI] [PubMed] [Google Scholar]
- 10. Werner RM, Konetzka RT. Trends in post‐acute care use among Medicare beneficiaries: 2000 to 2015. JAMA. 2018;319(15):1616‐1617. 10.1001/jama.2018.2408 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Callahan CM, Tu W, Unroe KT, LaMantia MA, Stump TE, Clark DO. Transitions in care in a nationally representative sample of older Americans with dementia. J Am Geriatr Soc. 2015;63(8):1495‐1502. 10.1111/jgs.13540 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Allen J, Koziak A, Buddingh S, Liang J, Buckingham J, Beaupre LA. Rehabilitation in patients with dementia following hip fracture: a systematic review. Physiother Can. 2012;64(2):190‐201. 10.3138/ptc.2011-06BH [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Flint LA, David DJ, Smith AK. Rehabbed to death. N Engl J Med. 2019;380(5):408‐409. 10.1056/NEJMp1809354 [DOI] [PubMed] [Google Scholar]
- 14. Aragon K, Covinsky K, Miao Y, Boscardin WJ, Flint L, Smith AK. Use of the Medicare posthospitalization skilled nursing benefit in the last 6 months of life. Arch Intern Med. 2012;172(20):1573‐1579. 10.1001/archinternmed.2012.4451 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Werner RM, Coe NB, Qi M, Konetzka RT. Patient outcomes after hospital discharge to home with home health care vs to a skilled nursing facility. JAMA Intern Med. 2019;179(5):617‐623. 10.1001/jamainternmed.2018.7998 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Rahman M, Foster AD, Grabowski DC, Zinn JS, Mor V. Effect of hospital‐SNF referral linkages on rehospitalization. Health Serv Res. 2013;48(6 pt 1):1898‐1919. 10.1111/1475-6773.12112 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Grabowski DC, Feng Z, Hirth R, Rahman M, Mor V. Effect of nursing home ownership on the quality of post‐acute care: an instrumental variables approach. J Health Econ. 2013;32(1):12‐21. 10.1016/j.jhealeco.2012.08.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Baiocchi M, Cheng J, Small DS. Instrumental variable methods for causal inference. Stat Med. 2014;33(13):2297‐2340. 10.1002/sim.6128 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Lage DE, Caudry DJ, Ackerly DC, Keating NL, Grabowski DC. The care continuum for hospitalized Medicare beneficiaries near death. Ann Intern Med. 2018;168(10):748‐750. 10.7326/M17-2651 [DOI] [PubMed] [Google Scholar]
- 20. Givens JL, Mitchell SL, Kuo S, Gozalo P, Mor V, Teno J. Skilled nursing facility admissions of nursing home residents with advanced dementia. J Am Geriatr Soc. 2013;61(10):1645‐1650. 10.1111/jgs.12476 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Middleton A, Li S, Kuo Y‐F, Ottenbacher KJ, Goodwin JS. New institutionalization in long‐term care after hospital discharge to skilled nursing facility. J Am Geriatr Soc. 2018;66(1):56‐63. 10.1111/jgs.15131 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Burke RE, Lawrence E, Ladebue A, et al. How hospital clinicians select patients for skilled nursing facilities. J Am Geriatr Soc. 2017;65(11):2466‐2472. 10.1111/jgs.14954 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Rose L. The effects of skilled nursing facility care: regression discontinuity evidence from Medicare. Am J Health Econ. 2020;6(1):39‐71. [Google Scholar]
- 24. Reckrey JM, Bollens‐Lund E, Husain M, Ornstein KA, Kelley AS. Family caregiving for those with and without dementia in the last 10 years of life. JAMA Intern Med. 2021;181(2):278. 10.1001/jamainternmed.2020.4012 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Vick JB, Ornstein KA, Szanton SL, Dy SM, Wolff JL. Does caregiving strain increase as patients with and without dementia approach the end of life? J Pain Symptom Manage. 2019;57(2):199‐208.e2. 10.1016/j.jpainsymman.2018.11.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Kasper JD, Freedman VA, Spillman BC, Wolff JL. The disproportionate impact of dementia on family and unpaid caregiving to older adults. Health Aff. 2015;34(10):1642‐1649. 10.1377/hlthaff.2015.0536 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Augustine MR, Davenport C, Ornstein KA, et al. Implementation of post‐acute rehabilitation at home: a skilled nursing facility‐substitutive model. J Am Geriatr Soc. 2020;68(7):1584‐1593. 10.1111/jgs.16474 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Thomas KS, Durfey SNM, Gadbois EA, et al. Perspectives of Medicare advantage plan representatives on addressing social determinants of health in response to the CHRONIC care act. JAMA Netw Open. 2019;2(7):e196923. 10.1001/jamanetworkopen.2019.6923 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Burke RE, Jones J, Lawrence E, et al. Evaluating the quality of patient decision‐making regarding post‐acute care. J Gen Intern Med. 2018;33(5):678‐684. 10.1007/s11606-017-4298-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Burke RE, Hess E, Barón AE, Levy C, Donzé JD. Predicting potential adverse events during a skilled nursing facility stay: a skilled nursing facility prognosis score. J Am Geriatr Soc. 2018;66(5):930‐936. 10.1111/jgs.15324 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Burke RE, Canamucio A, Medvedeva E, Manges KA, Ersek M. External validation of the skilled nursing facility prognosis score for predicting mortality, hospital readmission, and community discharge in veterans. J Am Geriatr Soc. 2020;68(9):2090‐2094. 10.1111/jgs.16650 [DOI] [PubMed] [Google Scholar]
- 32. Loomer L, Ogarek JA, Mitchell SL, et al. Impact of an advance care planning video intervention on care of short‐stay nursing home patients. J Am Geriatr Soc. 2021;69(3):735‐743. 10.1111/jgs.16918 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Sales AE, Ersek M, Intrator OK, et al. Implementing goals of care conversations with veterans in VA long‐term care setting: a mixed methods protocol. Implement Sci IS. 2016;11(1):132. 10.1186/s13012-016-0497-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Jones CD, Falvey J, Hess E, et al. Predicting hospital readmissions from home healthcare in Medicare beneficiaries. J Am Geriatr Soc. 2019;67(12):2505‐2510. 10.1111/jgs.16153 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Yoon J, Chow A. Comparing chronic condition rates using ICD‐9 and ICD‐10 in VA patients FY2014‐2016. BMC Health Serv Res. 2017;17(1):572. 10.1186/s12913-017-2504-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Baier RR, Wysocki A, Gravenstein S, Cooper E, Mor V, Clark M. A qualitative study of choosing home health care after hospitalization: the unintended consequences of “patient choice” requirements. J Gen Intern Med. 2015;30(5):634‐640. 10.1007/s11606-014-3164-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Ayele R, Jones J, Ladebue A, et al. Perceived costs of care influence post‐acute care choices by clinicians, patients, and caregivers. J Am Geriatr Soc. 2019;67(4):703‐710. 10.1111/jgs.15768 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Gadbois EA, Tyler DA, Mor V. Selecting a skilled nursing facility for postacute care: individual and family perspectives. J Am Geriatr Soc. 2017;65(11):2459‐2465. 10.1111/jgs.14988 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Manges KA, Ayele R, Leonard C, Lee M, Galenbeck E, Burke RE. Differences in transitional care processes among high‐performing and low‐performing hospital‐SNF pairs: a rapid ethnographic approach. BMJ Qual Saf. 2021;30:648‐657. 10.1136/bmjqs-2020-011204 [DOI] [PubMed] [Google Scholar]
- 40. Newhouse JP, Garber AM. Geographic variation in health care spending in the United States: insights from an Institute of Medicine report. J Am Med Assoc. 2013;310(12):1227‐1228. 10.1001/jama.2013.278139 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix S1 Supporting information