| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
|
|
|
|||||||
|
|||||||||
Correspondence: Address correspondence to Courtney Van Houtven, HSR&D, 508 Fulton Street, Bldg. 16, Durham VAMC, Durham, NC 27705. E-mail: courtney.vanhoutven{at}duke.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Formal home care • Health care services use • Stratified analysis • Veterans
There has been mixed evidence on whether HHC reduces nursing home entry. The Channeling Experiment was a randomized control trial that increased availability of HHC in the community. In the initial analysis, researchers found either no effect or a small negative effect on the probability of nursing home entry (Christianson, 1988; Wooldridge & Schore, 1988). In a later reanalysis, researchers found that the more generous public HHC program significantly lowered the probability that an individual was in a nursing home (Pezzin, Kemper, & Reschovsky, 1996; Stabile, Laporte, & Coyte, 2006). The difference in results between these analyses is attributed in part to the consideration of the endogeneity of living arrangements. Among Medicaid recipients, more generous home care benefits led to less nursing home care and more formal home care (Ettner, 1994). In other studies, researchers have found that the subsidization of HHC services affects informal care but not nursing home use or other medical care (Hoerger, Picone, & Sloan, 1996).
Weissert and colleagues argued that HHC has been a disappointment by failing to improve outcomes despite large expenditures (Weissert & Hedrick, 1994; Weissert, Lesnick, Musliner, & Foley, 1997). Generally, home care does not reduce nursing home care unless it is targeted to the most disabled patients (Greene, 2005; Greene, Lovely, Miller, & Ondrich, 1995; Greene, Ondrich, & Laditka, 1998; Weissert, Chernew, & Hirth, 2001, 2003).
Patients referred for HHC in the VA system have been identified by clinicians and social workers as among the most disabled patients. They typically have limitations in basic activities of daily living, instrumental activities of daily living, or both that place them at risk for nursing home admission (Banaszak-Holl et al., 2004; Bharucha, Pandav, Shen, Dodge, & Ganguli, 2004; Holroyd-Leduc, Mehta, & Covinsky, 2004; Jette, Branch, Sleeper, Feldman, & Sullivan, 1992; Jette, Tennstedt, & Crawford, 1995; Nuotio, Tammela, Luukkaala, & Jylha, 2003). Thus, VA patients referred to HHC represent a group of individuals who—without HHC services—may be likely to incur substantial nursing home costs.
Despite recent expansions, access to community-based services is limited in the VA system, even for patients identified as being at risk for nursing home entry and referred to HHC. Strict eligibility requirements, supply constraints, mandated levels of institutional LTC at 1998 levels (Miller & Rosenheck, 2006), and a historical reliance on in-hospital and nursing home care over community-based care all limit access to community-based LTC. These factors, when combined, suggest that the current use of HHC in the VA system may greatly underestimate the demand for these services on the national level.
One possible result of excess demand for community-based LTC for veterans is that the VA system is incurring higher costs on other types of health care, including outpatient care, inpatient care, and nursing home care, leaving the system with fewer funds available to expand community-based LTC. Alternatively, VA patients denied HHC in the VA system may seek higher absolute VA care anyway, given their high need for care, or they may seek HHC from Medicare rather than from the VA, which could reduce costs to the VA. Understanding how VA HHC users seek other VA services in the near term is useful, given the recent major expansions of HHC in the VA system. The VA offers an interesting case because it covers LTC in the community and the home, and it is the only large, integrated, publicly financed health care system in the United States.
Using national VA and Medicare administrative data, we identified VA HHC users (cases) and VA nonusers of HHC (controls) matched on age and race. We then conducted a case-control study to (a) describe individual and organization characteristics associated with the receipt of HHC and (b) examine utilization across a broad spectrum of services. The factors constraining VA HHC access suggest that a retrospective study of HHC will be vulnerable to selection bias. Therefore, to minimize selection bias on observable characteristics, we used characteristics from the baseline year (2002) and a propensity-score analysis in order to examine differences in subsequent utilization (2003) between cases and controls (Abadie, Drukker, Herr, & Imbens, 2004; Imbens, 2003; Imbens & Angrist, 1994; Rosenbaum & Rubin, 1985; Rubin, 1997; Zhao, 2004). If the assumption that all variables related to both treatment and outcome were included in the propensity-score model holds, then simply comparing utilization differences in the matched sample will give us an unbiased estimate of the expected average effect of VA HHC on subsequent utilization (Landrum & Ayanian, 2001). It is important that crossover between the VA health care system and Medicare was considered to capture nearly all publicly funded utilization (less than 1% of patients had Medicaid). These findings can help inform policy makers about whether investments in HHC could be potentially cost saving compared with the current mix of institutional and community-based care supplied.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Data Sources
We extracted data from the following Austin Automation Center databases: Outpatient Care files (including the Visit and Event files), Inpatient Treatment files (including Inpatient Acute Care, Extended Care, Observation, and Non-VA Care files), Inpatient Census files (including Inpatient Acute Care and Extended Care files), Beneficiary Identification Records Locator Subsystem (BIRLS) file, and Fee Basis files for contract care.
We collected variables representative of demographics, health, eligibility, death, distance, and location that were expected to affect the likelihood of using HHC and other types of health care utilization. We primarily pulled the demographic, health, eligibility, and location of service variables from the VA Outpatient Visit file for the 2002 calendar year. We filled in any missing values by using information from the Inpatient Treatment file or the Medicare files. We created consistent rules for variable definition in the case of multiple observations per individual. Namely, if more than one race was reported in 2002, then we selected the modal value; we defined age as the patient's age on January 1, 2002; we defined marital status as the status at the last visit; and other insurance besides VA coverage was measured at the last visit. We garnered the distance characteristics and urban or rural specification based on patient zip code of residence from the Veterans Health Administration Office of Policy and Planning, Planning Systems Support Group data set (2002). Finally, we drew dates of death from The Beneficiary Identification Record Locator System (BIRLS) death file, a database containing vitals information on veterans who receive benefits from the VA system. Additionally, we cross-checked these against the Inpatient Treatment file. If the date of death varied between these two sources, then we used the BIRLS value as the accurate measure (Maynard & Chapko, 2004).
Medicare HHC
We considered a person to have Medicare HHC if they had one or more home health agency claims in the Medicare home health agency files in 2002. The requirement to receive Medicare HHC is stricter than that to receive VA HHC in that Medicare HHC recipients must be home bound and in need of skilled home care on an intermittent basis (www.medicare.gov/publications/pubs/pdf/10969.pdf). The VA requires that a physician refer a patient to receive services, using the criterion of being able to remain safely in the home, but these services can be skilled or unskilled. There are no time limits placed on VA HHC.
Outcomes
Utilization Outcomes
We considered a broad spectrum of VA utilization, including any inpatient, outpatient, nursing home, hospice care, home-based primary care (HBPC), respite, and adult day health care. We also examined four types of Medicare utilization: any inpatient, outpatient, skilled nursing facility, and hospice care.
VA Utilization
The Inpatient Treatment and Census files provided utilization information on inpatient, nursing home, respite, and hospice care. The Outpatient Care files provided information on outpatient visits as well as nursing home care, HBPC, and adult day health care. We used the Fee Basis files to define nursing home, respite, hospice, and adult day health care from contracted providers. Nursing home care is a composite measure, comprising any stay in a VA Medical Center nursing home care unit, a community nursing home, a domiciliary care home, or a community residential care home. Because key data were not available for 2003 in the separate HBPC database, we used the Outpatient Care file for measuring HBPC. This approach likely underestimates the number of HBPC users by around 16% (Kubal, Weaver, Guihan, Cowper, & Hynew, 2000).
Medicare Utilization
We used the following Medicare data files to capture Medicare utilization: The Inpatient Standard Analytic File, the Outpatient Standard Analytic file, the Skilled Nursing Facility file, and the Hospice file. We merged Medicare data with the VA data by using social security numbers that had been scrambled by use of a common algorithm (VIReC, 2006).
Data Analysis
We present descriptive statistics comparing HHC users and nonusers. Our main analysis used a matched propensity-score analysis to control for underlying differences between cases and controls and to minimize selection bias (Abadie et al., 2004). We calculated the average treatment effect of VA HHC on different types of utilization, and we used McNemar's test to test for differences in utilization between VA HHC users, or cases, and nonusers, or controls (Newgard, Hedges, Arthur, & Mullins, 2004). A negative treatment effect indicates a negative association between HHC and subsequent utilization. In addition, as a sensitivity analysis, we performed a stratified analysis using the entire sample of VA HHC users and nonusers based on quintiles of propensity scores to examine treatment effects. For the stratified analysis, we tested the HHC treatment effect on subsequent utilization by using Mantel–Haenszel chi-square tests. For both the matched and stratified propensity-score analysis by quintiles, we also calculated standardized differences of our overall sample (Austin, Grootendorst, Normand, & Anderson, 2007) to examine imbalances in variables between our groups. Generally, a standardized difference greater than 10% represents a meaningful imbalance between groups (Normand et al., 2001).
Propensity-Score Model
For the propensity-score model, we included variables representative of demographics, health, eligibility, death, distance, and location that were expected to affect a patient's chance of being assigned HHC as well as to affect our utilization outcomes.
The demographic variables that we included were age; race, measured as Black, White (referent category), other race, or unknown race; Hispanic (non-Hispanic is the referent category); and marital status, which we measured as never married, divorced, separated or widowed, and married (referent category). Marital status partially controls for informal care availability, and it is not usually available in claims studies.
We measured the health variables by using two approaches. We used the 20 Elixhauser comorbidity conditions (Elixhauser, Steiner, Harris, & Coffey, 1998) that were derived from the ninth edition of the International Statistical Classification of Diseases (ICD-9) diagnoses that occurred during VA inpatient or outpatient visits in 2002. The second approach we used was the individual diagnostic cost groups from the Diagnostic Cost Group (DCG) categorization of ICD-9 codes (DxCG, 2002). We eliminated categories that did not apply to our patients, such as female-specific conditions, mental retardation, or childhood conditions.
The eligibility variables included a Veteran's service connection status as well as indicator variables for whether or not a patient had Medicare, Medigap, Medicaid, or other private insurance, to help control for the likelihood of using non-VA services (no other insurance is the referent). In the VA system, eligibility for LTC services is restricted by a veteran's "service connectedness," which is defined by his or her military service history and circumstances surrounding any disabilities arising from military service (Miller & Rosenheck, 2006). The higher the score, the more eligible a person is for VA health care. As mandated in the MHBA, all veterans with service connection of 70% or more have a legal right to VA LTC services, as do all veterans who need LTC as a result of a service-related disability (Miller & Rosenheck, referencing Ileum, 2003). Additionally, service-connected status affects access to VA services more broadly.
The death variable, that is, whether a patient died in 2003 or not, controls for the fact that medical utilization often increases near death (Yang, Norton, & Stearns, 2003). Including a death indicator variable also helps control for the truncation of the utilization recall period for those who died in 2003. We are not concerned with point estimates of the propensity-score model; thus, the temporal inconsistency of including death in 2003 does not raise the usual concerns about the introduction of endogeneity bias.
We calculated the distance variable as straight-line miles from the population centroid of the patient's zip code to the nearest relevant VA Medical Center (McCarthy & Blow, 2004). Distance variables are used to control for supply constraints and travel burden to receive care, or the possibility that HHC providers may be located closer to a medical center, affecting access for rural elderly individuals.
Location variables account for access to care and practice style effects, such as patient's residence in an urban area. In addition, we included medical center indicator variables to account for site-specific differences in service provision.
For the propensity-score model, we fit a logistic regression model with VA HHC as the outcome and with demographic, health, eligibility, death, distance, and location as independent variables as described herein. The model includes an exhaustive list of independent variables expected to affect a patient's chance of being assigned HHC in order to help meet a key assumption of propensity-score analyses—ignorability of treatment assignment. That is, only when each control has a nonnegative chance of having become a case (Rubin, 1997) can the average causal treatment effect be estimated by: E[Y(1) – Y(0)]. We estimated predicted probabilities or propensity scores from this model for each subject.
The logistic regression model is as follows<--CO?1-->:
|
|
Propensity-Score Matching
We used the greedy matching algorithm that was developed in SAS (Greedy 5
1 Digit Match Macro) to create our matched-pair sample of cases and controls. The greedy algorithm uses the nearest available pair-matching method, which is also called nearest-neighbor matching (Bakas, 2001). If more than one match is found, a control is randomly selected from the group of matches. This macro first matches cases to controls on five digits of the propensity score, and then those that do not match on five digits are matched on four. This continues down to a one-digit match for those that remain unmatched. Subsequently, if no match is found for a case, then the case is removed from the analytic sample.
We also compared our main results from the propensity-score and stratified analyses to results from other specifications of the propensity-score model: using the Elixhauser Index in place of DCG indicator variables; excluding death in 2003 as a variable in the model to check whether our observed results are affected by the predictable upward spike in utilization that we would expect immediately preceding death; and excluding persons who died in 2003.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Patient characteristics appear in the two left-hand data columns of Table 1.<--CO?2--><--CO?3--> A higher proportion of users than nonusers were single, had high service connectedness, and had Medicare. Both groups had low levels of non-VA and Medicare supplement insurance. The users lived on average 4 miles further from the closest medical center than nonusers did, although more users than nonusers lived in an urban area. On the basis of a cutoff of 10% in standardized differences, we found that users were more likely than nonusers to have infectious diseases; ear, nose, throat, and mouth conditions; and postsurgical stays (See Table 1, column 3). Nearly 16% of the users also used Medicare HHC in 2002, compared with 4% of the nonusers (not shown in the table). Finally, there were markedly different death rates between home health users and nonusers in 2003 before matching, but this difference was eliminated after matching (standardized difference of –0.3).
|
|
For the matched propensity-score analysis, cases had higher utilization rates of VA health care services in 2003 than controls did (Table 2) for all categories except hospice and for all Medicare categories except outpatient and hospice. For the VA outpatient utilization we observed a 3% absolute difference in rates between cases and controls. The absolute difference in VA inpatient utilization between cases and controls was reduced from approximately 26% using the overall cohort to 12% using the matched propensity analysis. The absolute difference in VA HBPC use was 5.7% between cases and controls. Subsequent hospice and adult day health care use was slightly greater for cases versus controls.
Patterns for Medicare utilization remained similar for matched cases but the association was weaker after matching as a result of a sicker cohort of controls being matched. Cases were slightly more likely to have Medicare inpatient and outpatient utilization than controls were (Table 2, row headings under "Medicare"). Differences in Medicare outpatient care were not statistically significant across cases and controls (Table 2, data columns 3 and 4, bottom panel). Cases were likewise more likely to have inpatient and nursing home care from any source in 2003 compared with controls.
Medicare HHC
We drew our sample and performed matching based on VA HHC use, but we wanted to understand the association between Medicare HHC and subsequent utilization for our cohort to understand crossover effects. Medicare HHC was associated with higher subsequent VA inpatient and VA nursing home use by 2.7% and 2.9%, respectively. Medicare HHC is associated with higher subsequent Medicare use by much higher amounts—outpatient care by 30%, inpatient care by 26%, nursing home care by 12%, and hospice care by 5% (results not shown in a table). Clearly, use of Medicare HHC in 2002 is associated with a greater likelihood of having any Medicare utilization in 2003 in a matched sample of VA HHC users, but the effects on VA utilization were relatively slight.
Sensitivity Analyses
Using a stratified analysis based on quintiles of propensity scores, we found a pattern in utilization rates that was similar to that reported for the matched propensity analysis. The stratified analysis highlights that large differences in utilization persisted for cases and controls across the propensity score quintiles (Mantel–Haenszel chi-square test results not shown). However, the difference in VA inpatient utilization rates between cases and controls decreased by quintile, with the smallest differences in the fifth quintile (see Figure 1). The same general patterns applied to VA nursing home use, with a slightly more linear increase in nursing home use from the first to the fifth quintile of the propensity score (Figure 2). Looking across the quintiles generally, we found that patients with a low propensity for HHC had lower subsequent utilization, as we expected. We also note that, as expected, 28% of controls had propensity scores in the first quintile as compared with less than 1% of our cases, and 4% of controls had propensity scores in the fifth quintile as compared with 55% of our cases.
|
|
Finally, we examined the association of HHC with subsequent utilization for a subsample of patients who were alive for the full utilization recall period (the 2003 calendar year). The association between VA HHC and utilization are very similar for this subsample, although the utilization rates in both groups for inpatient care and nursing home care were slightly lower, as one would expect.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Despite these steps, as in any observational study, the study is still susceptible to bias from unobserved characteristics that differ between cases and controls. For example, marital status is a good proxy for informal care support, but we do not have information on other types of family caregivers, which could be a bias-inducing omitted variable if it is correlated with, for example, either HHC in 2002 or nursing home entry in 2003. As a result of data limitations, we also were not able to measure the intensity of the HHC received, which may have altered our conclusions. Additionally, measures of health status that drive the need for HHC may not have been captured in the model, particularly limitations in basic activities of daily living and instrumental activities of daily living. Improving on the Elixhauser Index (Elixhauser et al., 1998), we used the individual DCG severity score codes (Ash et al., 2000), which allowed us to better distinguish a patient's severity (DxCG, 2002). We also explicitly controlled for nearness to death in 2003. Furthermore, we controlled for differences in eligibility, and supplemental insurance coverage, which all may influence likelihood of the receipt of HHC.
Nevertheless, our study is the first that we know of to describe the health and utilization patterns of the full cohort of VA HHC users, offering important information about the intensity of subsequent service use among HHC recipients. In future research, we should consider making a direct comparison between users of HHC and VA nursing home care to better ascertain cost-saving potential for the VA system. It may be that only in cases of avoided nursing home use will the VA system realize cost savings from HHC use. Similarly, it also may be that VA HHC may be a cost-saving measure only when it is targeted to those individuals who are most at risk of nursing home care (Greene, 2005; Weissert et al., 2003). To best examine this, one would also have to consider the informal care network of the patient, given that informal care reduces the risk of nursing home care (Van Houtven & Norton, 2008).
From a wider policy perspective, patients receiving VA HHC also have spillover effects to Medicare (and vice versa). The crossover between the two systems was not small. Approximately 16% of VA HHC users also received Medicare home health, whereas 4% of VA HHC nonusers used Medicare HHC. Having Medicare HHC was associated with both higher VA and Medicare nursing home use and Medicare outpatient and inpatient use.
Overall, our results suggest that HHC expansion may not be cost saving to the VA health care system, at least in the very short term. In particular, receipt of VA HHC does not appear to decrease VA nursing home care in the following year. This result echoes other studies in the literature (Christianson, 1988; Pezzin et al., 1996; Weissert & Hedrick, 1994; Wooldridge & Schore, 1988), but our findings, unlike the Channeling Intervention and other randomized controlled trials on home care, are subject to bias from selection on unobserved characteristics. In particular, in our study, we are concerned that VA HHC users do not have the same health profile as the nonusers, given the differential death rates we saw in 2003 before matching, and the fact that our analysis does exclude a large proportion of individuals very near death (death rate in the nonmatched cohort was 0.17 vs around 0.10 in the matched cohort). Thus, more careful utilization and cost analyses using longitudinal data and more extensive matching by health are also warranted in the VA setting. This will help us better understand the cost-saving potential of the VA's HHC expansion over the long term in a more definitive way than was possible in this study.
It is important to point out that there has been evidence in the literature that increased publicly funded home care is correlated with improvements in the health of home care recipients (Stabile et al., 2006). Our focus on utilization does not address health gains to the patient from the receipt of HHC, not to mention to family caregivers, who often have negative health effects of caring for family members (see Schulz et al., 2001 for one example). Other evidence suggests that additional formal medical care services for the patients in the home can help reduce family caregiver burden and improve caregiver quality of life (Hughes et al., 2000). Therefore, a more complete understanding of the effectiveness of VA HHC in terms of quality-adjusted life years gained for the patient and the family caregiver is needed before any conclusions about cost effectiveness can be made in the VA health care system.
|
|
Footnotes |
|---|
Courtney Van Houtven thanks her mentors at the VA, Morris Weinberer, Eugene Oddone, and Elizabeth C. Clipp; and her UNC manuscript writing group, Marissa Domino, Michelle Mayer, Kristin Reiter, Sally Stearns, and Rebecca Wells, for their input. Desirée Hawkins provided word processing. Any mistakes or errors remain the authors'.
1 Center for Excellence in Health Services Research and Development in Primary Care, Durham VA Medical Center, Durham, NC.
2 Division of General Internal Medicine, Department of Medicine, Duke University, Durham, NC.
3 Department of Biostatistics, Duke University, Durham, NC.
Decision Editor: William J. McAuley, PhD
Received for publication October 18, 2007. Accepted for publication February 28, 2008.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| All GSA journals | Journals of Gerontology Series A: Biological Sciences and Medical Sciences | Journals of Gerontology Series B: Psychological Sciences and Social Sciences |
