Transitional care outcomes in Veterans receiving post-acute care in a skilled nursing facility

Abstract

Background:

As the Veteran population ages, more Veterans are receiving post-acute care in skilled nursing facilities (SNFs). However, the outcomes of these transitions across VA and non-VA settings are unclear.

Objective:

To measure adverse outcomes in Veterans transitioning from hospital to SNF in VA and non-VA hospitals and SNFs.

Design:

Retrospective observational study using the 2012–2014 Residential History File, which concatenates VA, Medicare, and Medicaid data into longitudinal episodes of care for Veterans.

Setting:

VA- and non-VA hospitals, and SNFs in four categories: non-VA SNFs, VA-contracted SNFs, VA Community Living Centers (CLCs), and State Veterans Homes.

Participants:

Veterans age 65 or older who were acutely hospitalized and discharged to SNF; one transition was randomly selected per patient.

Measurements:

Adverse “transitional care” outcomes were a composite of hospital readmission, Emergency Department visit, or mortality within 7 days of hospital discharge.

Results:

More than 4 in 5 Veteran transitions (81.7%) occurred entirely outside the VA system. The overall 7-day outcome rate was 10.7% in the 388,339 Veterans included. Adverse outcomes were lowest in VA hospital-CLC transitions (7.5%, 95% CI 7.1–7.8%) and highest in non-VA hospital to VA-contracted nursing home transitions (17.5%, 95% CI 16.0–18.9%) in unadjusted analysis. In multivariate analyses adjusted for patient and hospital characteristics, VA hospitals had lower adverse outcome rates than non-VA hospitals (OR 0.80, 95% CI 0.74–0.86). In comparison to VA hospital-VA CLC transitions, non-VA hospital to VA-contracted nursing homes (OR 2.51, 95% CI 2.09–3.02) and non-VA hospital to CLC (OR 2.25, 95% CI 1.81–2.79) had the highest overall adverse outcome rates.

Conclusion:

Most Veteran hospital-SNF transitions occur outside the VA, though adverse transitional care outcomes are lowest inside the VA. These findings raise important questions about the VA’s role as a provider and payer of post-acute care in SNF.

BACKGROUND

The transition between the hospital and skilled nursing facility (SNF) has become one of the most active areas of healthcare payment reform in the United States, targeted by hospital and SNF value-based purchasing, bundled payments, and Accountable Care Organizations.^1–8 These reforms are intended to respond to the aging and growing comorbidity of the US population, reduced hospital lengths of stay, and reports of poor outcomes in SNF transitions.^9–12 Relatively little is known about SNF use and patient outcomes in the Veteran population, despite the fact that the VA has a similarly growing population of older Veterans – 52% of all male Veterans are age 65 or older, compared to 11.5% of men in the non-Veteran US population¹³ - and VA hospitals have similarly faced pressure to decrease length of stay.¹⁴ This may represent an opportunity for the nation’s largest integrated health system to improve outcomes during what is increasingly recognized outside the VA as one of the most perilous times in the care trajectory of older adults.^2,7

A major barrier to understanding Veteran post-acute care outcomes is tracking care provided across VA and non-VA settings. Veterans can utilize a variety of payers for care (for example, VA benefits to pay for an acute hospitalization, Medicare benefits for SNF, and Medicaid benefits for long-term nursing home care), which may lead to inaccurate estimates if only one of these payer sources is used to measure outcomes. The SNF options available to Veterans are also more varied than in the Medicare population. After hospitalization in a VA or non-VA hospital, Veterans can receive post-acute SNF care in VA-owned and operated nursing homes (called Community Living Centers, CLCs), State Veterans Homes (SVHs, which care for Veterans with significant VA financial support but are not otherwise integrated into the VA), community nursing homes the VA has established formal contracts with to provide nursing home care to Veterans (Contract Nursing Homes, CNHs), or Veterans can use their Medicare benefits to receive SNF care in non-VA facilities (non-VA SNFs). A new data source called the Residential History File concatenates VA-paid, Medicare fee-for-service, and Medicaid fee-for-service claims, allowing for the first time accurate measurement of longitudinal episodes of care for individual Veterans across these entities.¹⁵

Given the emphasis in Medicare on care transitions in current payment reforms involving SNF³, we sought to understand where Veterans are hospitalized and subsequently receive post-acute care in SNFs, and to measure short-term outcomes associated with these transitions. Based on known associations between care fragmentation and poor outcomes during care transitions, we hypothesize that the integration across VA hospitals and CLCs will be associated with lower adverse transitional care outcomes than other transitions across systems of care.

METHODS

Data source

We used the Residential History File (RHF) from January 1, 2012 to December 31, 2014.¹⁵ The RHF includes VA care, VA-paid care in the community, Medicare fee-for-service claims, and Medicaid fee-for-service claims for Veterans enrolled in the VA. Although the RHF does not include commercial payers, Medicare fee-for-service is the predominant payer for SNF care in the United States.¹⁶

Population description

We began with every Veteran enrolled in the VA who had an acute hospitalization followed by a stay in a non-VA SNF, CLC, SVH, or CNH (n=985,577 transitions). Swing bed stays, hospice care, and psychiatric or inpatient rehabilitation stays were excluded (n=132,757 excluded transitions). Behavioral health cohorts add additional complexity in terms of payment and care settings, and inpatient rehabilitation stays have different admission requirements than SNFs. We also excluded Veterans less than age 65 or older than 100 (n=87,558). After these exclusions, the remaining cohort was used to estimate the prevalence of transitions across hospital and SNF types.

To accurately measure outcomes, we further restricted the sample to hospitals that discharged at least one Veteran per week to any SNF type to avoid unstable estimates (n=154,662 Veteran transitions excluded, 3877 hospitals excluded). We then randomly selected one record per Veteran, so Veterans with multiple stays only contributed information to a single transition type. After these exclusions, we had 388,339 Veterans who had a single transition from hospital to SNF from 1502 hospitals. As a sensitivity analysis, we additionally excluded Veterans who were in a facility (non-VA SNF, CLC, CNH, or SVH) the day prior to/day of the index admission (n=80,596 Veterans excluded).

Predictor and outcome variables

Our primary outcome was a composite of death (not occurring in the context of hospice), Emergency Department (ED) visit, or rehospitalization within 7 days of discharge from the hospital. We define these as adverse transitional care outcomes because they are unlikely to have been anticipated within this time period, and 7-day outcomes are those most likely to be responsive to transitional care interventions.^17–19 Veterans could have more than one outcome during the 7-day period. We adjusted for predictors that have been linked to adverse transitional care outcomes in other studies.^11,17,20 Veteran demographic data included age, gender, race and ethnicity, income level from the 2015 American Community Survey (median income within the patient’s ZIP code), and comorbidities that constitute the Charlson-Deyo score (identified using ICD-9 diagnosis codes within 1 year of the initial hospital admission).²¹ We adjusted for each comorbidity individually. We also identified the number of hospitalizations that Veterans had in the year prior to the index admission, and characteristics of their index admission (length of stay, and whether the hospital stay included an Intensive Care Unit (ICU) stay).

While Minimum Data Set (MDS) admission information can be useful in predicting adverse outcomes²², SNFs have up to two weeks to collect this information. We found a more than 10% rate of missing data in this cohort given the short time-frame after SNF admission, and did not include MDS variables in the analysis. We conducted a sensitivity analysis including MDS variables found to be risk factors for hospital readmission from SNF (Barthel Index, presence of indwelling catheter, heart failure diagnosis, dementia)¹¹ and inclusion of these variables did not materially change our results.

We also adjusted for hospital characteristics between VA and non-VA hospitals (includes stays paid by Medicare, Medicaid, and the VA for “fee-basis” care), including number of beds, medical school/residency affiliation, urban/rural location, ownership, and whether the hospital was capable of performing organ transplants (a marker of its patient population and relative complexity of services offered). Non-VA hospital characteristics urban/rural status was identified from the Medicare Provider of Services file. VA hospital characteristics were taken from three sources. Urban/rural status was taken from the Medicare Provider of Services file. Number of operating beds was taken from VHA Support Services Center (VSSC), which maintains accurate operational data across VA facilities. Residency affiliation and organ transplant information was taken from the VA Office of Productivity, Efficiency, and Staffing (OPES) which maintains up-to-date information on the complexity of care a VA facility can provide, including these elements.

Statistical analysis

To estimate the prevalence of transitions types, we categorized transitions as originating from VA or non-VA hospitals to each of the four SNF types (non-VA SNF, CLC, CNH, SVH) for eight possible transition types using the sample selected above (n=765,262). We used our more restrictive sample when describing these populations and their transitional care outcomes (n=388,339). We used two logistic regression models to adjust for patient and hospital characteristics and used hospital random effects to account for clustering of patients within hospitals and SNFs. The first model compared outcomes for VA and non-VA hospitals; the second model compared outcomes across all 8 transition types. We adjusted the 95% confidence intervals for multiple comparisons using the Bonferroni adjustment. This study was approved by the Colorado Multiple IRB (COMIRB) and Corporal Crescenz VA Medical Center IRB.

RESULTS

Our prevalence sample included 765,262 Veteran transitions from 5379 hospitals. More than 4 in 5 (81.7%) transitions occurred between non-VA hospitals and non-VA SNFs; only 14.9% of Veteran transitions to SNF began at a VA hospital (Figure 1).

Figure 1. Prevalence of Veteran transitions across VA and non-VA hospitals.

VA hospitals and VA-affiliated nursing homes are listed in blue boxes, non-VA hospitals and SNFs are in green boxes. Transitions outside the VA are displayed by the green arrow, transitions that span VA and non-VA settings are displayed by red arrows, and transitions within the VA setting by blue arrows. The percentages listed next to each arrow are the proportion of all Veteran transitions to SNF accounted for by each of the 8 possible transitions.

Our sample analyzing outcomes of these transitions included 388,339 Veterans and 1502 hospitals. The largest differences in Veteran populations hospitalized at VA and non-VA hospitals were income levels (Veterans at VA hospitals were 10% more likely to have an income below $40,000 annually) and length of stay (Veterans at VA hospitals were 20% more likely to have an index length of stay of more than 3 days -Table 1). These characteristics also varied across the eight different transition types (Supplementary Table S1). VA hospitals in our sample tended to be smaller and more commonly affiliated with a medical school; all hospitals in our sample were predominantly urban (Table 2).

Table 1 –

characteristics of Veterans in VA and non-VA hospitals

Veteran characteristic	VA hospital (n=70,449)	Non-VA hospital (n=317,890)	Total cohort (n=388,339)	Absolute difference (VA-non-VA)
Age (mean, standard deviation)	78.4 (8.8)	82.5 (7.9)	81.8 (8.2)	−4.1
Male	97.4	90.6	91.8	6.8
White	80.7	90.4	88.6	−9.6
Black	15.8	7.9	9.3	8.0
Hispanic	3.9	1.2	1.7	2.7
Income
<$40,000/year	27.3	17.0	18.9	10.3
$40,000–49,999/year	25.8	23.8	24.1	2.0
$50,000–69,999/year	30.3	33.3	32.8	−3.0
>$70,000/year	16.6	25.9	24.2	−9.3
Congestive heart failure	31.4	39.6	38.1	−8.2
COPD	37.3	40.3	39.9	−3.1
Cerebrovascular disease	23.6	23.7	23.7	0
Dementia	13.3	16.5	15.9	−3.2
Diabetes	44.2	38.9	39.9	5.3
Diabetes with complications	16.9	11.6	12.5	5.4
Hemiplegia	4.6	4.4	4.4	0.2
AIDS	0.4	0.1	0.1	0.3
Moderate liver disease	6.4	4.2	4.6	2.2
Severe liver disease	1.7	1.0	1.1	0.8
Malignant tumor	22.0	17.2	18.1	4.8
Metastatic malignancy	5.1	4.8	4.9	0.3
Myocardial infarction	12.2	20.7	19.1	−8.5
Peptic ulcer disease	2.9	3.7	3.5	−0.7
Peripheral vascular disease	21.0	22.5	22.2	−0.5
Chronic kidney disease	32.4	38.7	37.6	−6.3
Connective tissue disease	2.8	3.8	3.6	1.0
Intensive care unit stay	22.9	32.3	30.6	−9.4
Index length of stay >3 days	83.5	63.1	66.8	20.3
Hospitalizations, prior year
0	45.0	49.1	48.4	−4.1
1	24.6	26.2	25.9	−1.6
>1	30.5	24.7	25.8	5.7

All values are percentages unless otherwise noted. Comorbidities are identified using ICD-9 codes. COPD = chronic obstructive pulmonary disease; AIDS = acquire immunodeficiency syndrome. Given the large sample size, statistical significance is unlikely to connote a meaningful difference, thus absolute differences of 10% or more are highlighted in bold.

Table 2 –

Hospital characteristics of VA and non-VA hospitals

	VA hospital (n=122)	Non-VA hospital (n=1,380)	Total (n=1,502)
Number of beds
<50	22.1	0.4	2.1
50–100	19.7	4.5	5.7
101–200	40.2	21.0	22.6
>200	18.0	74.1	69.6
Medical school affiliation	91.8	45.7	49.4
Urban	86.9	86.7	86.8
Transplant available	11.5	21.9	21.0
Ownership
Private for profit	-	13.9	12.8
Private not for profit	-	60.2	55.3
Public	100	11.7	18.9
Other	-	14.2	13.1

All values listed are percentages.

Across all transitions, 10.7% of Veterans experienced at least one 7-day adverse transitional care outcome (rehospitalization: 5.4%, ED visit: 5.3%, death: 2.5%). VA hospital to CLC transitions had the lowest rates in unadjusted analyses (7.5%, 95% CI 7.1–7.8%) while non-VA hospital to CNH had the highest adverse event rates (17.5%, 95% CI 16.0–18.9%, Table 3). After adjustment for patient and hospital characteristics, VA hospitals had lower rates of 7-day adverse events (OR 0.80, 95% CI 0.74–0.86) than non-VA hospitals across all transitions. Similarly, all transitions had higher adverse event rates when compared to VA hospital-CLC transitions, with non-VA hospital to CNH (OR 2.51, 95% CI 2.09–3.02) and non-VA hospital to CLC (OR 2.25, 95% CI 1.81–2.80) having the highest adjusted odds of 7-day adverse event (Table 3). In fact, every transition starting at a non-VA hospital had higher odds of the composite outcome than transitions starting at a VA hospital, with the exception of VA hospital-CNH transitions.

Table 3 –

Unadjusted and adjusted outcomes by transition type

Transition type	Unadjusted outcome % (95% CI)	Adjusted outcome OR (95% CI)	Adjusted outcome OR (95% CI)
Non-VA hospital		Reference
to CLC	15.3 (13.7–17.0)		2.25 (1.81–2.79)
to CNH	17.5 (16.0–18.9)		2.51 (2.09–3.02)
to non-VA SNF	10.9 (10.8–11.0)		1.58 (1.41–1.78)
to SVH	10.5 (9.7–11.3)		1.43 (1.21–1.70)
VA hospital		0.80 (0.74–0.86)
to CLC	7.5 (7.1–7.8)		Reference
to CNH	12.3 (11.6–13.1)		1.72 (1.53–1.95)
to non-VA SNF	9.8 (9.5–10.1)		1.32 (1.21–1.44)
to SVH	9.3 (8.1–10.4)		1.31 (1.07–1.62)

OR = odds ratio, CI = confidence interval. CLC= Community Living Center, CNH = contract nursing home, SNF = skilled nursing facility, SVH = State Veterans Home. The middle column compares VA to non-VA hospitals on the primary outcome; the right column compares transitions to SNFs of different types on the primary outcome.

When the composite outcome was decomposed to measure the individual components of the composite primary outcome, it seemed the differences in the composite outcome were mostly driven by differences in ED visit rates and to a lesser extent mortality; rehospitalization rates were not significantly different across transition types (Table 4). A sensitivity analysis excluding Veterans who were admitted from a facility resulted in similar findings (Supplementary Table S2).

Table 4 –

adjusted outcomes by individual components of the composite primary outcome

	Mortality OR (95% CI)	ED visit OR (95% CI)	Readmission OR (95% CI)	Composite OR (95% CI)
Non-VA hospital	-	-	-	-
VA hospital	0.69 (0.62–0.76)	0.94 (0.85–1.04)	0.97 (0.89–1.06)	0.80 (0.74–0.86)
Non-VA hospital
to CLC	1.27 (0.77–2.09)	8.51 (6.34–11.43)	1.25 (0.94–1.67)	2.25 (1.81–2.79)
to CNH	3.08 (2.25–4.21)	8.96 (6.89–11.66)	1.13 (0.88–1.46)	2.51 (2.09–3.02)
to non-VA SNF	1.57 (1.30–1.89)	4.55 (3.72–5.55)	1.00 (0.87–1.15)	1.58 (1.41–1.78)
to SVH	0.99 (0.72–1.37)	4.99 (3.87–6.43)	0.89 (0.71–1.11)	1.43 (1.21–1.70)
VA hospital
to CLC	Reference	Reference	Reference	Reference
to CNH	1.72 (1.36–2.18)	7.14 (5.87–8.68)	1.14 (0.97–1.33)	1.72 (1.53–1.95)
to non-VA SNF	0.99 (0.82–1.18)	5.80 (4.89–6.86)	0.92 (0.83–1.03)	1.32 (1.21–1.44)
to SVH	0.89 (0.55–1.44)	6.08 (4.65–7.96)	0.80 (0.59–1.08)	1.31 (1.07–1.62)

The fourth column is the same as the third column in Table 3 (composite 7-day adverse outcome rate). The first three columns analyze the individual components of the primary outcome adjusted for hospital and patient characteristics and using hospital fixed effects models.

DISCUSSION

In the largest analysis to date of Veteran post-acute care outcomes, we found more than 1 in 10 Veterans suffered an adverse transitional care outcome within 7 days of discharge to SNF. Transitions from VA hospitals were associated with 20% lower odds of an adverse event compared to transitions from non-VA hospitals. VA hospital to CLC transitions - occurring entirely within the VA system - had the lowest rate of adverse events. Compared to these transitions, transitions from non-VA hospitals to CLCs or CNHs had more than two times the odds of emergency department visit, readmission, or death during the week following hospital discharge. The largest differences across transition types were in ED visits and to a lesser extent mortality; rehospitalizations were not significantly different.

We are unaware of prior data investigating Veteran outcomes in post-acute care, and direct comparisons of Veteran and Medicare beneficiary outcomes in SNF, particularly around care transitions, are similarly lacking. One prior report identified timing of readmissions from SNF among a nationally representative sample of Medicare beneficiaries, observing 22.8% of beneficiaries were readmitted overall, and 27.5% of these readmissions occurred within 7 days (a 7-day event rate of 6.3%).¹¹ This is higher than the composite 5.4% rehospitalization rate among Veterans in our study, though this rate was quite variable across transition types.

Growing evidence suggests early adverse events (within 7 days) may be more responsive to changes in transitional care processes.^17–19,23 These early events may also be important as a quality target since they are correlated with long-term outcomes. Medicare beneficiaries readmitted from SNF are nearly four times as likely to have died by 100 days post-hospital discharge than patients who were not readmitted even after risk adjustment.¹¹ They also have taken on particular significance for hospitals and SNFs participating in a growing number of payment reforms that target these outcomes, such as Accountable Care Organizations, Bundled Payments for Care Improvement, and SNF value-based purchasing.

These findings open new questions about why some transitions seem to be more associated with early adverse events than others. Variability in hospital transitional care processes²⁴, factors such as linkages between hospital and SNF providers^25,26, and SNF staffing and turnover levels could play important roles.^27,28 Discharge decision-making regarding post-acute care has received increased attention as a deficit that could contribute to adverse outcomes.^29–32 Identification and dissemination of best practices in reducing adverse outcomes in the transition to SNF is sorely needed. Most published interventions to date in this regard have relied on resource-intensive interventions that have been difficult to implement broadly.^33–36

The variety of settings for SNF post-acute care available to Veterans provides a rich potential resource to evaluate the role of processes of care and other modifiable factors that differ between VA and non-VA settings in SNF outcomes. These evaluations may be impactful because the VA, as both payer and provider of care, has the ability to alter its SNF provider “menu” to achieve better outcomes and lower costs (higher value) in SNF. For example, our analysis suggests transitions within the VA environment (which includes a shared electronic medical record and employment) are associated with fewer adverse events, and this is strikingly different for non-VA hospital to CLC transitions, suggesting differences in the transition process may partly explain the difference in outcomes. However, we note that though we adjusted for patient and hospital characteristics used in prior studies to compare VA and non-VA settings and different SNF types^37,38, we could not fully account for unmeasured differences between VA and non-VA facilities. Nevertheless, these findings are informative as VA considers more widespread use of non-VA facilities to improve access. In particular, outcomes at CNHs were worse in transitions from both VA and non-VA hospitals; it is unclear if CNHs willing to contract with the VA to provide nursing home care have similar quality to other nursing homes. To our knowledge, differences in these facilities have received limited examination in the published literature.³⁹

Non-VA hospitals and SNFs may also benefit from research identifying why these differences in outcomes exist. For example, VA CLCs represent an example of vertically integrated SNFs – something many health systems may be considering to improve SNF value in the setting of bundled payments and Accountable Care Organizations, though evidence so far is lacking that this improves longer-term outcomes.⁴⁰ Identifying and implementing specific aspects of VA hospital-CLC transitions may improve SNF transitional care outcomes across VA and non-VA care settings. Of note, we did not assess to what extent private SNFs in our sample were part of non-VA integrated hospital-SNF networks. Although these networks are still uncommon even among Accountable Care Organizations,^5,41 they may have played a role in lowering adverse event rates in non-VA facilities.

Our results also highlight the need for information transparency about post-acute care quality for Veterans and hospitals treating Veterans. Many SVHs and all CLCs are not listed on Nursing Home Compare, while the VA’s version of Nursing Home Compare only includes CNHs and CLCs. There isn’t a unified resource where Veterans or hospitals (VA or non-VA) could compare all SNF options for Veterans. It is important to establish whether short- and long-term outcomes in SNF are linked to quality metrics collected by the VA, such as Strategic Analytics for Improvement and Leaning (SAIL) metrics⁴², and consider a unified data source that would allow more meaningful comparisons of SNFs across VA and non-VA settings.

Strengths of this analysis include the 100% sample of all Veterans enrolled in the VA who were discharged from a hospital to SNF and the ability to link VA, fee-basis, Medicare fee-for-service, and Medicaid data to create longitudinal episodes of care and accurately assess outcomes. However, we were unable to adjust for potentially important reasons Veterans would be discharged to one type of SNF or another. For example, functional status, cognition, linkage between hospitals and SNFs, proximity to different facilities, patient and caregiver preferences, and the intensity of a particular Veteran’s prior relationship with the VA are all important potential confounders. We did not use data from the nursing home stay to help with risk adjustment because of missing data. However, sensitivity analyses that included the available MDS data produced similar results. Since there is no standard definition for short-term stays in VA administrative data, we included all Veterans discharged to SNF. A sensitivity analysis that excluded Veterans who resided in a facility before admission did not meaningfully change our results. We did not include transitions covered by commercial payers or Medicare Advantage and our findings may not be generalizable to those populations. Finally, the population of patients utilizing VA vs non-VA facilities differs as illustrated in Table 1 and since we may not fully account for these differences in risk adjustment, unmeasured confounding remains a significant concern.

As with many systems that increasingly bear risk for post-acute care outcomes of their patients, the VA faces an increasingly older, frail, comorbid population who may require more post-discharge supports. Post-acute outcomes among Veterans discharged to SNFs for post-acute care are highly variable across facility types. Identifying specific factors driving these differences may inform efforts to increase the value care in this setting and improve short- and long-term outcomes in this vulnerable population.