| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
|
|
|
|||||||
|
|||||||||
Correspondence: Address correspondence to Bruce Stuart, PhD, Professor and Director, The Peter Lamy Center on Drug Therapy and Aging, Department of Pharmaceutical Health Services Research, University of Maryland Baltimore, 515 W. Lombard Street, Suite 157, Baltimore, MD 21201. E-mail: bstuart{at}rx.umaryland.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Dementia • Alzheimer's disease • Medicare • Nursing home • Health expenditures • Medicare qualified stay
Determining what is appropriate medical treatment with persons with dementia is a challenge from both a clinical and an economic perspective. The issue goes well beyond what direct services these individuals receive (or do not) to treat their dementia, into the question of treatment for other comorbidities that would be routinely recommended for patients without dementia. It would seem straightforward enough to compile accurate accounts of the services that individuals with dementia actually receive and then to compare that to those without the condition. However, that task turns out to be more complex, as the considerable literature on the utilization and cost of treating persons with dementia attests. Part of the problem has to do with diagnosing dementia and coding services for the condition, but a more fundamental factor is selecting appropriate comparison groups and settings for the analysis.
Since the late 1980s, more than a dozen studies have examined the cost of treating persons with Alzheimer's disease and related dementias (Bynum et al., 2004; Coughlin & Liu, 1989; Ernst & Hay, 1994; Fillenbaum, Heyman, Peterson, Pieper, & Weiman, 2001; Fillit, Hill, & Futterman, 2002; Gutterman, Markowitz, Lewis, & Fillit, 1999; Hay & Ernst, 1987; Hill et al., 2002; Kane & Atherly, 2000; Leon, Cheng, & Neuman, 1998; Martin, Ricci, Kotzan, Lang, & Menzin, 2000; Menzin, Lang, Friedman, Neuman, & Cummings, 1999; Newcomer, Clay, Luxenberg, & Miller, 1999; O'Brien & Caro, 2001; Ostbye & Crosse, 1994; Rice et al., 1993; Richards, Shepherd, Crismon, Snyder, & Jermain, 2000; Taylor, Schenkman, Zhou, & Sloan, 2001; Taylor & Sloan, 2000; Weiner, Powe, Weller, Shaffer, & Anderson, 1998; and Welch, Walsh, & Larson, 1992). This work is important in understanding the impact of dementia on resource utilization in the health care sector, particularly for programs focused on the aged. However, the current literature is limited in several important respects. Most studies are restricted to community-dwelling populations whereas a significant share of the cost of treating individuals with dementia occurs in nursing homes and other long-term-care facilities. The community-based studies have consistently found that persons with dementia have greater health care utilization and expenditures than those without dementia; the same may not hold true in nursing facilities with highly disabled populations.
Studies by Rice and colleagues (1993), Welch and colleagues (1992), and Leon and colleagues (1998) examined nursing home costs for patients with dementia but did not compare their costs to nursing home residents without dementia. Lack of comparative data makes it difficult to ascertain how much cost burden is due to dementia. Two recent studies by Kane and Atherly (2000) and O'Brien and Caro (2001) compared nursing home costs between those with and without dementia but reported conflicting results. Kane and Atherly used 1991–1995 Medicare Current Beneficiary Survey (MCBS) data and found that residents with dementia generated $2,470 less in annual Medicare Part A and B claims compared to those with dementia. The O'Brien and Caro study analyzed 1997 Massachusetts Medicaid data and found that, on average, nursing home residents with dementia cost $3,865 more compared to those without dementia. Although Medicare and Medicaid cover different services, it is not obvious why the dementia-related cost differentials should have opposite signs.
Prior work by the investigators (Burton et al., 2001) found that utilization rates for hospital and physician services were lower for a cohort of Maryland nursing home residents diagnosed with dementia upon admission compared to those with no evidence of dementia. Assuming that utilization differentials translate into cost differences, that would imply it is less expensive to treat residents with dementia in the nursing home setting, which is consistent with the Kane and Atherly (2000) findings. There are several possible reasons that might explain the direction of this relationship. One possibility is that patients admitted with dementia have fewer comorbid conditions requiring expensive medical interventions compared to individuals with no dementia. This interpretation is consistent with research conducted by Murtaugh and Freiman (1995) and Fried and Mor (1997); both studies found that nursing home residents with dementia were significantly less likely to be hospitalized compared to those with no dementia. Another possible reason for lower utilization rates among nursing home residents with dementia is they are treated less aggressively for other medical conditions, possibly as a result of advance directives. For example, Magaziner and colleagues (1991) found that for nursing home-acquired infections, those with dementia were less likely to receive an evaluation meeting minimal diagnostic criteria established by an expert panel. Yet a third possibility is that the behavioral manifestations of dementia mask underlying morbidity that then goes untreated. The literature offers no conclusive evidence about which interpretations are correct.
Published comparisons of utilization and cost differences between nursing home residents with and without dementia suffer from an additional shortcoming relating to ascertainment of the disease. There is no consistent diagnostic standard applied in this literature, and it is therefore difficult to make meaningful comparisons across studies. Most of the smaller comparative studies use mental status tests (Leon et al., 1998; Hu, Huang, & Cartwright, 1986). But low functioning on mental status tests is only one aspect of dementia diagnosis (Magaziner et al., 1996). Various investigators, including O'Brien and Caro (2001), identified dementia cases using diagnosis codes derived from medical charts. In addition to the problem of inconsistent coding of dementia across charts, it has been shown that chart-based methods underestimate the true prevalence of dementia in nursing homes by as much as 30% (Magaziner et al., 1996; Magaziner et al., 2000). The Kane and Atherly (2000) study identified dementia cases based on proxy reports of dementia, which lack standardized diagnostic criteria and raise the possibility of recall bias.
Purpose
This article extends the research on cost of treating nursing home residents with and without dementia in several important respects. First, we gathered Medicare expenditure data for our study cohort both pre- and postadmission. This before-and-after feature of the study design enabled us to control for differences between residents with and without dementia prior to admission and thereby helped isolate cost differences postadmission attributable to dementia status alone. Second, using a longitudinal design we could determine whether observed cost differentials between residents with dementia and no dementia at admission diverged or converged over time. Third, we had access to a more extensive set of control variables than available to other researchers. These variables permitted us to discern whether measured cost differences were associated with demographic factors, payment source at admission, and characteristics of the nursing facility, as well as patient comorbidities. Unlike previous research, we utilized a comorbidity measure designed specifically to predict Medicare costs. Fourth, we analyzed cost factors by type of service, thereby making it possible to determine where and (by extension) why costs diverge. Last, but not least, we used an expert panel for dementia case ascertainment in accordance with criteria developed by the American Psychiatric Association.
Before describing the study methods it will be helpful to briefly review the role of Medicare in paying for services to nursing home residents. About 91% of all nursing home residents nationwide are enrolled in Medicare (Doshi, Shafer, & Briesacher, 2005), and they retain the same rights to Part A and B services as community-dwelling beneficiaries. Medicare is the primary payer for physician visits, inpatient hospital care, diagnostic tests, and various other services for these residents. However, Medicare payment for nursing home care per se is limited to Medicare qualified stays (MQS), which are defined in statute as being short (less than 100 days, typically a fraction of that) and for the purpose of rehabilitation rather than custodianship. Moreover they must immediately follow an inpatient hospital stay of at least 3 days. In 1996, payments by Medicare for MQS stays represented approximately 19% of all nursing home revenues compared to 44% from Medicaid, 30% from self-payment, and the rest from insurance (Rhoades & Sommers, 2001). However, MQS only pays full charges for 20 days, after which time the resident is responsible for a daily copayment ($109.50 in 2005).
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
In all, 2,285 individuals were enrolled in the study. Medicare records could be matched for 2,074, and of these, 1,276 (representing the final analytical sample for this study) were Medicare eligible for 12 months prior to their index nursing facility admission and had complete data on all key variables used in the statistical analysis. (We tested the impact of this inclusion criterion by comparing postadmission Medicare costs by dementia status for the final sample of 1,276 with the larger sample of 2,074 and found no statistically significant differences.)
Resident Assessment on Admission
We collected baseline data from multiple sources including structured interviews by trained lay interviewers with residents, nursing staff, and significant others. Information collected included age, gender, race, education, marital status, and functional status. Dementia status was determined in accordance with Diagnostic and Statistical Manual III-R criteria (American Psychiatric Association, 1987) by an expert panel of geriatric psychiatrists, neurologists, and a geriatrician. A detailed description of the dementia ascertainment methodology is available elsewhere (Magaziner et al., 1996). In the overall study population (N = 2,285), we designated 48.2% as having dementia (Magaziner et al., 2000).
Medicare Expenditure Data
We obtained Medicare bill records between 1992 and 1997 for all study participants. We aggregated final paid claim amounts into 5 categories: inpatient hospital, MQS in skilled nursing facilities, physician or supplier, other services (hospital outpatient, home health, durable medical equipment, and hospice), and total Medicare payments. We counted only payments made by Medicare to health care providers and did not include beneficiary cost-sharing amounts. We used the date of service information on each bill record to group payments into 30-day "months" indexed to the resident admission date. According to the study inclusion criteria, each resident had 12 months of preadmission Medicare observations and up to 24 observations after admission. Thus, "months" prior to the admission were coded –12 to –1, and "months" after admission were coded 1 to 24. Each "month" observation also carried an indicator representing the calendar month and year of the observation (operationally, the month with the most observation days or the earlier month when equal observation day). We used the calendar indicator to inflate all cost observations to constant December 1997 dollars using the Consumer Price Index.
We used Medicare claims data to stratify the study sample into residents admitted into the nursing facility with or without a MQS. All study participants with an MQS claim with a start date within one week of the nursing home-admission date were considered to be admitted with a MQS.
We also used the Medicare-claims data to risk adjust each individual in the study based on the Diagnostic Cost Group/Hierarchical Coexisting Condition (DCG/HCC) risk adjuster developed by the Centers for Medicare and Medicaid Services (CMS). The DCG/HCC is the basis for the "selected significant disease model" that CMS has used to set capitation rates for Medicare HMOs since January 2004 (Pope et al., 2004). Briefly, the model requires a year of Medicare inpatient, outpatient, and physician payment records to create indicators for the presence of 189 medical conditions. The version of the DCG/HCC model we used in the present study applied previously calibrated weights to 101 of these conditions to create a summary score of the patient's expected Medicare expenditure under Parts A and B for the next year.
Our application of the DCG/HCC model used the 12 months of preadmission Medicare claims to create a predicted value for Medicare expenditures in the immediate postadmission year for each study participant. We did not expect that the prediction would hold true because every one of the individuals was admitted to a nursing facility, most after an expensive medical event. However, the DCG/HCC predictions served as an effective way to control for individual-level comorbidities that might mask the independent effect of dementia on residents' Medicare expenditures. Because our interest was in finding the marginal contribution of dementia to residents' cost patterns, we explicitly excluded the HCC category that represents dementias (HCC 49).
Statistical Analysis
We have presented charts showing mean constant-dollar Medicare PPM expenditures for residents admitted with and without dementia for the 12-month preadmission and up to 24-month postadmission periods for the entire sample and stratified by MQS status. By our inclusion criteria, all sample residents contributed to every PPM measure up to the admission month (not all residents had Medicare expenditures in each month but all were included in the denominators for these months). Beginning in the month after admission, the number of residents contributing cost information gradually declined because of discharge and death to the point that just 389 residents remained after 630 days. We have also reported the mean PPM spending postadmission by dementia status and type of Medicare-covered service for the total sample and 2 subsamples during their entire stays.
Our multivariate analysis used a sequential model development framework beginning with a simple model including a single indicator for dementia status (Model 1) and progressing in stages to a comprehensive model with all covariates (Model 7). This forward-selection approach permitted us to establish which domains were most important in explaining the difference in Medicare costs between residents with and without dementia. The full model is:
|
|
We estimated these models using generalized estimating equations (GEE) in STATA v. 8 (StataCorp, 2003). This STATA procedure (xtgee) adjusts the variances for correlations of repeated measures within individuals and within nursing homes. It uses a theoretical bootstrap method for correcting the standard errors of the regression coefficients and can be applied in regression analyses using many distributions. In our samples, the costs (dependent variable) were all right skewed and included numerous zero values (primarily in the inpatient hospital and skilled nursing facility service categories). We chose a gamma distribution with a log link to model the cost distributions as this approach has gained acceptance and is now preferred to the more commonly used method of log transformation (Veazie, Manning, & Kane, 2003).
We have presented all of the multivariate results in the form of "cost ratios," defined as the adjusted mean Medicare cost for residents with dementia divided by the adjusted mean Medicare cost for residents without dementia (thus a cost ratio below 1.0 indicates that residents with dementia incur lower Medicare costs than those without dementia). Most of the analyses focused on cost comparisons at the PPM level, but we also conducted analyses for sequential 90-day periods postadmission. In one version, we stratified the stay into distinct 90-day periods and replicated the same Model 7 regression for each stratum. In another model we used GEE to examine interactions between dementia status, time (in 90-day increments), and MQS status.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
. Figure 1 depicts Medicare PPM costs for the entire sample pre- and postnursing home admission. The PPM costs were generally higher for the nondementia group, albeit the only significant differences were at months –3, –2, –1, 1, and 8. However, the most telling feature of these time paths was the gradual increase in spending beginning in month –7 and peaking in the month immediately preceding admission to the nursing home. This pattern was expected and reflects the accumulation of health problems that eventually leads to admission.
|
|
|
The characteristics of study participants are shown in Table 1 for the full sample and stratified according to MQS status. In general, residents with dementia were older, more likely male, less likely Caucasian, less educated, more likely married, more ADL impaired, more chairfast but less bedfast, and predicted to incur lower Medicare expenditures than residents with no dementia. Residents with dementia were also much more likely to have Medicaid eligibility upon admission, stay longer in the nursing home, and have a lower likelihood of being discharged home or dying in the first 90 days of the nursing home stay. These comparisons were consistent across the subgroups by MQS status.
|
|
|
|
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Longitudinal analysis showed that Medicare costs for both groups peaked in the month immediately preceding admission and then dropped within 3 months of admission to roughly the same level as before admission. The time plots for residents with and without dementia showed no evidence of convergence over time. It is worth noting, however, that the higher intensity of treatment for the nondemented is balanced by the much longer average stay (nearly double) for residents with dementia. In fact, if one were to compare total Medicare spending over the entire nursing home stay, average costs would be higher for the resident group with dementia.
We then investigated the effect of stratifying the sample according to whether the admission qualified for skilled nursing facility payment under Medicare Part A. This stratification produced enlightening results: For those admitted with a MQS, overall Medicare costs were indistinguishable between the dementia and nondementia groups (cost ratio =.98), whereas for those without a MQS at admission, the cost ratio was.63 (p <.001). In other words, virtually all the dementia cost differential originated from those admitted without a MQS. We found some evidence that the MQS effect diminished for those residents remaining in the nursing facility after their MQS eligibility expired (the cost ratios for initial MQS residents fell to between.41 and.48 during the period 271–450 days from admission [p <.05]), but sample attrition precluded further investigation of this relationship.
Given the structure of our sample, these results are not strictly comparable to other studies. Our finding that total Medicare costs are higher for nursing home residents without dementia is similar to Kane and Atherly (2000) during an equivalent time frame (1991–1995). They found that the biggest Medicare cost differences were associated with probability of hospitalization and other Part A services and the level of physician spending. Although we did not estimate two-part spending models (part 1 being the probability of any spending and part 2 being the level of spending for those with any spending), we found that residents with dementia had significantly lower Medicare spending for both types of service after admission. Our finding that residents with dementia also had lower rates of Medicare spending prior to admission might be considered inconsistent with prior studies showing that patients with dementia are more costly in community settings. However, as our entire study sample eventually entered a nursing facility, the observations of spending during their time in the community are heavily influenced by health events leading up to admission; and the events suffered by the group without dementia appear to be more costly to Medicare than dementia per se (also, it is important to note that because we ascertained dementia at nursing home admission, it is not certain that all individuals in this group actually experienced problems with dementia prior to that).
Our findings regarding preadmission Medicare-spending patterns are important primarily because they permit us to adjust for non-dementia-related influences on postadmission spending. To our knowledge, this feature of our study design is unique to the literature on treatment of dementia in long-term-care settings. Assuming that the resident populations (dementia vs nondementia) maintain the same relative costliness pre- and postadmission, then the application of the HCC risk adjuster (sans adjustment for dementia) will generate an estimate of the Medicare cost of treating dementia net of all other comorbidities. If we accept this assumption, then our findings imply that the net cost of Medicare treatment for dementia in nursing homes is actually negative. That is not as far fetched as it sounds. Studies leading to the development of the Medicare skilled nursing facility prospective payment system (PPS) have shown that the diagnosis of dementia is only important in predicting costs among residents if they are receiving no rehabilitation and/or have no other medical conditions (Fries, Mehr, Schneider, Foley, & Burke, 1993). In the introduction to this article, we suggested two other possible reasons why the presence of dementia might generate negative costs: a failure to pursue aggressive medical treatment for comorbid conditions or dementia masking other underlying comorbidity. Our study was not designed to isolate these potential explanations, but we cannot rule them out either. A final possibility is that the HCC underadjusts for true differences in comorbidity between the two groups. There is evidence (Pope at al., 2004) that the HCC tends to underpredict future Medicare costs for high-cost beneficiaries (such as those in our sample), but that should not bias the measured difference in costs between residents with and without dementia.
We cannot rule out the possibility that some of the measured difference in costs between the two types of residents was due to Medicare carrier reactions to the government's attempts to reign in fraud and abuse in Medicare spending for mental health services in nursing facilities in the mid-1990s. The effect, if any, is likely to be small. The push for carriers to tighten up reimbursement procedures under the aegis of "Operation Restore Trust" did not gain steam until after publication of the Inspector General's report in May 1996 (DHHS, 1996), and most of our data collection (1992–1997) preceded that date. Moreover, Maryland was not one of the states targeted for "Operation Restore Trust." That is not to say, however, that the Medicare carrier serving Maryland at the time may not have denied care it believed unnecessary for persons diagnosed with dementia.
The findings from this study should be interpreted in the light of other limitations. We analyzed only Medicare costs and did not consider Medicaid or private payments. Although Medicare is the primary payer for both MQS admissions and all hospital and physician services provided to Medicare-eligible residents, the program does not cover routine nursing home expenses nor prescription medications (until 2006). We have no reason to believe that consideration of routine nursing home expenses or palliative remedies to treat symptoms of the disease would negate or reverse our finding that residents with dementia cost less to treat. It is true that patients with dementia often have concurrent psychiatric symptoms that require drug therapy, but those without dementia also have high medication costs (which we did not tally either), so this exclusion should have little net effect on our findings. The small size of our study sample precluded analysis of the effect of dementia severity on the relative costliness of treating the disease. If there is a cost gradient, that might lead to different conclusions regarding relative treatment costs for mild and very severely demented patients.
A more serious limitation is that our cost comparisons are strictly generalizable only to individuals newly admitted to nursing homes in the state of Maryland between 1992 and 1995 (with data collection through 1997 for the most recent study entrants). As demonstrated elsewhere (Magaziner et al., 2000), the characteristics of the study sample of Maryland nursing home residents were very similar to those admitted to nursing homes nationally during the same period. However, the long-term-care market has changed dramatically since 1997, with prospective payments systems adopted by Medicare and numerous state Medicaid programs and explosive growth in alternative living arrangements for elders needing various levels of assistance. It is not clear whether these developments would lead to different results were our study replicated today, but it does suggest that our findings be interpreted in light of them.
A final limitation is that we lacked information on whether residents had private or public Medicare supplements (other than Medicaid). These supplements reduce the burden of Medicare cost sharing, and, thus, persons with them might be expected to have higher Medicare spending as a result. If, as Kane and Atherly (2000) found, persons with dementia are much less likely to have supplemental insurance compared to those with no dementia, then this omission could bias our results. It is worth noting, however, that the Kane and Atherly study (and the entire literature on insurance-induced demand, for that matter) has been conducted on samples of community-dwelling individuals. Given the regulations regarding provision of medical care in nursing homes, it is not at all certain that economic incentives could even find expression in these facilities.
These limitations notwithstanding, our study incorporated design elements that should help guide future research in this area. First, our use of an expert panel to identify dementia cases based on Diagnostic and Statistical Manual III-R criteria represents a significant improvement over previous studies that used proxy reports (Kane & Atherly, 2000), medical charts (O'Brien & Caro, 2001), or diagnoses from medical claims (Fillit, Geldmacher, et al., 2002) alone. Second, having preadmission Medicare cost information permitted us to employ a quasi-experimental, pre–post comparison series design that is inherently superior to simple cross-sectional comparisons. Third, our focus on incident nursing home residents, while unrepresentative of the entire facility population, permitted analysis of trends in cost ratios relative to date of admission that would not have been possible with other sample configurations. Finally, and perhaps most important of all, was our decision to stratify nursing home residents admitted with and without a MQS. In our study sample, a majority of dementia patients were admitted without a MQS (354 vs 286), but the rates were reversed for the nondementia group (215 vs 421). Only through stratification on MQS status did it become clear that the sample-wide finding of somewhat lower Medicare costs for residents with dementia was driven entirely by the subset admitted without a MQS (for whom the cost differential was markedly greater). Had we failed to stratify the sample, we would have missed this important distinction. This finding should spur future work on the cost of treating dementia residents admitted without qualified stays or whose duration of residence extends beyond the Medicare qualifying period.
|
Footnotes |
|---|
1 Department of Pharmaceutical Health Services Research, University of Maryland Baltimore.
2 Department of Epidemiology & Preventive Medicine, School of Medicine, University of Maryland Baltimore.
3 Health Research and Education Trust, Washington, DC.
4 Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD
5 School of Social Work, University of North Carolina at Chapel Hill.
Decision Editor: Linda S. Noelker, PhD
Received for publication October 14, 2004. Accepted for publication February 21, 2005.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Zimmerman, A. L. Gruber-Baldini, J. R. Hebel, L. Burton, K. Boockvar, G. Taler, C. C. Quinn, and J. Magaziner Nursing Home Characteristics Related to Medicare Costs for Residents With and Without Dementia American Journal of Alzheimer's Disease and Other Dementias, March 1, 2008; 23(1): 57 - 65. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| 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 |
