| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Duke University, Durham, North Carolina
b University of Colorado, Denver
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Methods. Participants were drawn from community respondents to the 1994 National Long Term Care Survey. The authors compared total cost of caring for persons without ADRD with that of those who had moderate and severe ADRD. Using regression analysis, the author identified the adjusted effect of ADRD, limitations in activities of daily living (ADLs), and common comorbidities on total costs.
Results. Persons with severe ADRD had higher median total costs ($10,234) than did persons with moderate ADRD ($4,318) and those without ADRD ($2,268, p < .001). However, disability measured by ADL limitations was a more important predictor of total cost than was ADRD status in both stratified and multivariate analyses. Comorbidities such as heart attack, stroke, and chronic obstructive pulmonary disease also increased costs. Severe ADRD was an effect modifier for ADL limitations, increasing the positive impact of disability on total costs among persons with severe ADRD, but not for comorbidities.
Discussion. Disability, severe ADRD, and comorbidity all had independent effects that increased total costs. Thus, any risk adjustment procedure should account for disability and comorbidity and not just ADRD status.
ALZHEIMER'S disease and related dementias (ADRD) has been estimated to be the third most expensive disease to treat in the United States; causes substantial disability to afflicted persons; and has a major impact on family members, who often provide a substantial amount of informal care (Ernst and Hay 1994
; Meek, McKeithan, and Schumock 1998; Small et al. 1997). The cost of ADRD includes the cost of formal personal health services provided in a variety of settings—nursing homes, hospitals, physicians' offices, and in homes—as well as informal care provided by families and paid caregivers (Coughlin and Liu 1989; Ernst and Hay 1994; Hay and Ernst 1987; Hu, Huang, and Cartwright 1986; Huang, Cartwright, and Hu 1988; Max 1993; Meek et al. 1998; Ostbye and Crosse 1994; Rice et al. 1993; Stommel, Collins, and Given 1994; Weinberger et al. 1994). Prescription drugs and assistive devices such as durable medical equipment are also factors. Intangible costs such as reductions in quality of life or reduced life expectancy are difficult to measure. The effect of ADRD on the cost of caring for elderly persons is important because of the increasing numbers of persons diagnosed with ADRD in the United States (U.S. General Accounting Office 1997).
Estimates of the prevalence of ADRD vary from 2.7% to 11.2% of the population aged 65 and older across U.S. studies (Bachman et al. 1993; Beard, Offord, and Kurland 1991; Evans et al. 1990; R. I. Pfeiffer, Afifi, and Chance 1987), and a recent U.S. General Accounting Office 1997 meta-analysis placed the figure at 5.7% of the population aged 65 and older. Study populations and methods of diagnosis differed across studies, contributing to the uncertainty, but the prevalence of dementia has even been found to differ within the same study population when multiple diagnostic criteria are used (Erkinjunti et al. 1997). However, there is consensus that ADRD prevalence rises with age, making it of great concern as the population of the United States continues to age (U.S. General Accounting Office 1997).
Ernst and Hay 1997 reviewed the literature on the cost of ADRD and found that estimates varied by a factor of 4 across studies. They pointed out that drawing conclusive cost estimates from the existing literature is made difficult by the localized nature of many samples, differences in controls for severity of ADRD, lack of control for other important patient characteristics, and uncertainty regarding the period of data collection in some studies. The clearest findings from the cost literature are as follows: (a) Caregiver time is a large portion of the cost of caring for persons with ADRD in community settings; (b) costs rise with increased dementia and limitations in activities of daily living (ADLs); and (c) the cost and distribution of cost among formal and informal care categories varies according to residential location. Demented individuals living in nursing homes tend to receive more paid caregiving and to incur higher total costs, whereas those living in community settings tend to cost less overall but receive a larger proportion of their care from informal, unpaid caregivers.
Recently, Leon, Cheng, and Neumann 1998 found that monthly total costs (formal and caregiving services) of caring for a person with ADRD were lower for community residents and that increased disability was an important predictor of higher cost, in a sample that included persons from several states. Five studies have focused on the cost to the Medicare program of caring for persons with ADRD by studying Medicare claims records (Cano, Hennessey, Warren, and Lubitz 1997; Gutterman, Markowitz, Lewis, and Fillit 1999; Newcomer, Clay, Luxenberg, and Miller 1999; Taylor and Sloan 2000; Weiner, Powe, Weller, Shaffer, and Anderson 1998). These studies showed that persons with ADRD use more Medicare-financed care compared with controls without the disease and suggested that prior research based on localized samples that did not have administrative claims data probably underestimated the magnitude of such care. Comparison of these studies is also made difficult because of different controls used across studies, and there is still much that is unclear about cost of care and ADRD. Ernst and Hay 1997 pointed out in their review of the literature that "despite the studies that have been done, there is clearly room for new work on AD costs, ideally with national instead of small and geographically limited samples" (p. 139).
Our study was designed to provide more information on the total cost of caring for elderly persons living in community settings, including the distribution of cost among different types of care, by using a national sample that included persons over the age of 65 who lived in the community. Persons living in nursing homes were excluded from our analysis because the major cost driver of such persons' care is being in a nursing home per se, and not the mixture of disabilities and/or diseases that they have. There were two main goals of our study: (a) to determine the relative impact of ADRD, disability as measured by ADL limitations, and comorbidities on the total cost of caring for elderly persons living in the community, and (b) to determine whether the presence of ADRD serves as an effect modifier of the relationship between disability and comorbidity on total costs. These analyses are relevant to current policy discussions regarding the incremental burden on society of a rise in the prevalence of ADRD and help answer questions regarding whether ADRD per se or comorbidity and disability are better variables that might be used as risk adjustors for policy purposes.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Method of Diagnosing ADRD
We defined persons as having ADRD if they had low cognition as measured by the Short Portable Mental Status Questionnaire (SPMSQ; Pfeiffer 1975). The SPMSQ "is intended to offer a rapid screen for cognitive deficit in the community-dwelling elderly" (Fillenbaum 1988, p. 120). Researchers have used a variety of cutoff points with the SPMSQ to define persons with impaired cognition, and we used a variant of what was suggested by Fillenbaum without defining different standards for persons of different races or educational levels because of a relatively homogenous study sample in terms of race and because of incomplete data for education. We placed all respondents into three groups on the basis of their answers to the SPMSQ. Persons who got zero to three items correct on the SPMSQ were classified as having severe ADRD. Those who got four to six items correct were classified as having moderate ADRD, and those who had seven or more items correct were classified as not having ADRD. Persons who responded via a proxy were assigned to the severe ADRD group if the reason for a proxy response was noted as being due to cognitive limitations; when a proxy was due to other reasons, the person was assigned to the "none" category. Descriptive analyses were stratified by this variable, and ADRD status was controlled for in multivariate models directly as well as interacted with ADL limitations and the presence of four serious health conditions.
Defining the Cost of ADRD
We took a broad perspective of cost and included costs that were paid for by the Medicare program, costs paid out of pocket by families for items such as caregiving and prescription drugs, and costs that were not reimbursed such as caregiving provided by family members. Intangible costs such as pain and suffering associated with disability were not measured in this study, but we did quantify levels of disability to provide context for cost figures. The period for which we measured costs was the 365 days prior to the 1994 NLTCS interview. For Medicare-financed costs we identified the services that were provided to an individual for the 365 days prior to the 1994 NLTCS according to Medicare claims. For other cost items, we annualized costs that were measured in terms of utilization in the month or week prior to the 1994 interview. For example, respondents reported the number of hours of assistance for ADL limitations that they received in the week prior to the 1994 NLTCS. We multiplied this figure by 52 to obtain an estimate of the number of hours of ADL assistance that a respondent received in the year prior to the 1994 NLTCS, which we then multiplied by an hourly wage rate.
Nursing home costs.
Given that a person answered the 1994 community NLTCS survey, it was rare that they reported having been in a nursing home in the previous 365 days. Only 3 persons reported having spent time in a nursing home in the year prior to the 1994 NLTCS. For this reason, we did not include nursing home costs in our total costs estimates based on self-reports. Our Medicare costs noted previously did include Medicare-financed stays in skilled nursing facilities.
ADL helper cost.
For community respondents, we used the self-reported number of hours of ADL help an individual reported receiving for the week prior to the 1994 NLTCS. We multiplied this by 52 to devise an annual estimate of hours of ADL help received. We then multiplied these hours by the $6.95/hr figure used by Leon and colleagues 1998 to estimate the cost of caregiving for persons with ADRD. When the number of hours of ADL support was missing, we assigned zero ADL help hours if the persons had no ADL limitations, because in such a case they had missing ADL help hours because of the skip pattern of the database. If they had at least one ADL limitation, we imputed the ADL help hours received if they had a missing value for this item as follows. For respondents with valid responses for ADL hours, we estimated a median regression model with number of ADL hours in the previous week as the dependent variable. The explanatory variables used were number of ADL limitations, male gender, and married. Using the coefficients obtained from estimating this model on persons with valid responses, we then predicted the value for number of ADL hours in the previous week for persons with missing ADL hours. The predicted values ranged from 2 hr/week to 13.5 hr/week. We then multiplied the predicted hours for persons with missing values by 52 to develop an estimate of the number of ADL support hours received in the year prior to the 1994 NLTCS. A total of 1,117 persons had their ADL hours of help imputed in this manner.
Assuming that the ADL hours from the week prior to the survey held constant for the entire 365-day period may have resulted in an overestimation of the mean and median ADL helper cost and, therefore, the total annual cost of ADRD. We chose this approach over projecting forward (e.g., assuming costs in the week prior to the survey held constant for the 365 days after the interview) because it ensured that we measured costs for the same period of time for each person, whereas going forward would have meant differential cost follow-up periods because of mortality. In multivariate analyses, the assumption we made had no effect in determining the relative importance of explanatory variables in explaining total cost, because the annualization approach we used simply multiplied a weekly cost by a constant for each person.
To provide information regarding the sensitivity of our findings to these imputation methods, our data show results when all cases (including those with imputed ADL hours) were used and when only cases in which no imputation was necessary were used. The overall multivariate results were robust and did not differ when cases with imputed ADL helper hours were used.
Medicare cost.
The amount actually paid by Medicare for all services provided by Medicare for the 365 days prior to the 1994 interview date was measured for all respondents.
IADL helper cost.
The approach was identical to that used for ADL helper cost, including the imputation of missing values for IADL helper hours among persons with some limitations after those with no IADL limitations had their IADL helper hours set to zero. We multiplied the number of hours reported for the week prior to the 1994 interview by 52 to get annual hours. The hourly price of $6.08 for IADL helper hours was used following Leon and colleagues 1998. Some 331 persons had their IADL hours of help imputed on the basis of their reported limitations in IADLs with the same regression model approach that was used for missing ADL hours and the same explanatory model. The predicted hours of IADL support ranged from 4 to 25 hr/week.
Prescription costs.
Respondents were asked how many prescription drugs they had taken during the past month. We multiplied this number by 12 to obtain the annual number of prescription medications being taken. We used the average cost per prescription medication used by Ostbye and Crosse 1994 to obtain annual prescription drug costs.
Controlling for Disability and Comorbidity
A distinction should be made between the cost of caring for a person with a particular condition, such as ADRD, and the incremental cost that a condition such as ADRD adds to the total cost of caring for an individual above and beyond the otherwise expected cost. We wanted to identify any additional cost associated with having ADRD after accounting for disability and comorbidities. If comorbidity and disability are not separately accounted for, then it is likely that the cost effect of ADRD will be biased. Therefore, we controlled for these factors in our multivariate models in addition to a limited number of demographic variables.
Disability.
Disability was measured by the number of limitations in the following six ADLs: dressing, bathing, eating, getting in and out of bed, moving around inside the home, and using the toilet. A limitation meant that a person either needed help doing the activity or used equipment to help them do the activity. We used number of ADL limitations to stratify the sample for descriptive purposes and included number of ADL limitations as a linear control variable, on the basis of descriptive findings that hours of help and overall cost rise with increasing number of ADL limitations. We also used the number of limitations in IADLs and behavioral problems, such as getting lost, forgetting to do important things, and losing one's temper frequently, to characterize the sample and illustrate differences among the three ADRD groups (severe, moderate, and none). We did not control for limitations in IADLs in multivariate models because of their high degree of collinearity with limitations in ADL.
Comorbiditiy.
The presence of certain diseases has been found in past work to be important in explaining variations in costs of health care generally (Bass, Looman, and Ehrlich 1992; Ellis, Pope, and Iezzoni 1996; Gruenberg, Kaganova, and Hornbrook 1996), and costs for caring for persons with ADRD specifically (Gutterman et al. 1999). We used self-reported information from the 1994 NLTCS on the presence of four common conditions both to control for severity of illness and to investigate whether having ADRD altered the total costs of caring for an individual with a particular condition given the presence of ADRD. The presence of the following four conditions was measured at the 1994 NLTCS: heart attack, cerebrovascular event/stroke, cancer, and chronic obstructive pulmonary disease (COPD). For heart attack and stroke, the survey asked whether either had occurred during the past year, so a person answering yes experienced this adverse health event during the period for which we measured total costs. For cancer and COPD, the survey asked whether these chronic conditions were present at the time of the survey, so the onset of the disease was not clear if they reported that they had the condition. We included binary variables equal to 1 if a respondent self-reported a condition, zero otherwise. We controlled for these four conditions separately as well as interacted them with severe ADRD to determine whether the presence of severe ADRD altered the total cost of caring for a person with one of these diseases compared with the costs of caring for a nondemented person with one of these conditions.
Analytical Methods
For much of our analysis, we compared persons across the three ADRD groups (severe, moderate, and none), either in terms of disability or cost, then estimated multivariate models to more fully adjust for multiple predictors of total cost. For the descriptive statistical analyses, we compared the median total cost of care stratified by the three ADRD categories and the number of limitations in ADL, categorized as follows: no limitations, one or two limitations, three or four limitations, and five or six limitations. We compared the median values of the total cost variable because of the right-skewness of the total cost of care. To determine whether there were any significant differences between the median annual cost of care among the severe, moderate, and no ADRD groups, we conducted a nonparametric significance test of the null hypothesis that the sample medians in all three groups were equal (severe = moderate = none), with the probability of equality calculated using Fisher's exact test and operationalized in Stata version 7.0 (StataCorp 2001).
For the multivariate models, we used median regression with bootstrapped standard errors (Chernick 1999; Efron and Tibshirani 1993; Gould 1992; StataCorp 2001). As previously mentioned, the total cost of care variable displayed a right-skewed distribution, and for that reason we chose to focus our analyses on the median. In comparison with ordinary least squares (OLS), median regression analyzes variation in the median of the dependent variable over the independent variables (rather than variation in the mean) and median regression fits a regression surface that minimizes the sum of the absolute residuals (rather than minimizing the sums of squares of the residuals). We calculated standard errors via the bootstrap method (with 3,000 replications) to minimize the effects of heteroskedasticity (Gould 1992; Rogers 1992). Although we present only the results of the median regression analyses, we also estimated a standard OLS model with bootstrapped standard errors and robust regression that downweights the importance of outliers. The substantive results of the median regression did not differ from the results of these alternative models.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Persons with severe ADRD had median annual total costs that were nearly 10 times as high as those for persons who did not have ADRD ($10,234 for severe, $4,318 for moderate, and $2,268 for none, p < .001, Table 2 ). Mean costs showed a similar pattern whereby mean total costs were higher among the severe ADRD group compared with the none group, with costs for the moderate ADRD group in between. Item-specific costs were greater for persons with severe ADRD compared with those without ADRD for all cost items except prescription drugs. Thus, the level of costs among groups differed more than did the relative contribution of cost items to total costs.
|
Disability, Comorbidities, and Cost
There was a hierarchical ordering on total cost by the number of limitations in ADLs, classified into four groups and stratified by ADRD status. The number of limitations in ADLs had a stronger effect on total costs than did ADRD status. Within a given ADRD stratum, the median total costs were more than 10 times greater for persons with five or six limitations in ADLs relative to persons with no ADL limitations (Table 3 ). However, total costs always varied by a factor of 5 or less when we compared severe versus no ADRD within a given ADL limitation strata. For example, median costs were $552 for the nondemented group versus $2,634 for those with severe ADRD within the strata of persons who had no ADL limitations. The same comparison of no ADRD (median costs $16,486) versus severe ADRD ($28,302) among persons with five or six limitations in ADLs showed a similar relationship.
|
|
Severe ADRD not only had a direct effect on total costs, but also served as an effect modifier for the impact of disability, but not for comorbid health conditions, on cost (Table 5 ). The interaction of severe ADRD and the number of ADL limitations showed an increase in total cost of $2,180 (p < .001) for each limitation in an ADL experienced by a person who was severely demented. None of the interactions between comorbidities and severe ADRD were significant. Thus, for these conditions severe ADRD did not serve as an effect modifier of cost.
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
; Ernst and Hay 1994; Hu et al. 1986; Huang et al. 1988; Max 1993; Ostbye and Crosse 1994; Rice et al. 1993; Stommel et al. 1994; Weinberger et al. 1994). Because the sample we used oversampled disabled elderly persons, our primary contribution is not calculation of per capita costs by ADRD status, but identification of the relative impact on cost of ADRD, disability, and comorbidity. Further, we show that the presence of ADRD serves as an effect modifier for limitations in ADLs, a finding worthy of more study. Disability is a more powerful predictor of total costs than is ADRD severity in both descriptive and multivariate analyses. It is likely that ADRD is related to level of ADL limitations, but our study cannot shed light on any causative link between severe ADRD and disability because of our cross-sectional measurement of these two variables. Not surprisingly, both recently occurring comorbid conditions (heart attack and stroke occurring within the past year) and COPD prevalent on the interview date increase total annual costs among persons who have no or moderate ADRD. We estimated the effect of these conditions on cost because they are common causes of morbidity and mortality, and as the population continues to age, their effect on costs both individually as well as in conjunction with ADRD will be important determinants of the overall cost of caring for community-dwelling elderly persons.
We tested whether severe ADRD serves as an effect modifier of the relationship between four common health conditions and cost. Some past work has shown that persons with ADRD receive less Medicare-financed health care compared with others with similar conditions, raising the possibility that demented individuals receive less intensive formal health services as a result of their mental status (Taylor and Sloan 2000). We do not find that this is the case, using a broader definition of costs.
Our study replicates in a national sample what has previously been found in local samples regarding composition of costs for persons with ADRD: Medicare-financed health care services do not compose the majority of total costs of caring for community-dwelling elderly persons: Caregiving services do. Although the magnitude of such costs differs by ADRD status, the relative composition of cost items is roughly the same. We find this to be true in spite of two factors that both work in the opposite direction. First, we used fairly conservative hourly price estimates for caregiving hours taken from other studies because the self-reported information on whether care was paid for or not, and if yes, how much, is too incomplete to be reliable. The magnitude of the average cost estimate per person by ADRD status is sensitive to the hourly price used, but the multivariate work identifying the relative importance of individual characteristics on costs is not. Second, we have fairly precise measurement of payment by Medicare for formal health services because we used Medicare claims data. Other studies, such as that of Hu and colleagues 1986, for example, did not use such data. The use of Medicare claims plausibly results in a more precise estimate of cost of acute services compared with estimates of such costs based on self-report, but any undercounting of such costs by self-reports in past studies was not sufficient to make such care a larger proportion of total costs relative to informal care in our study.
Several limitations of our study should be noted. First, we studied the costs of ADRD only among community-dwelling persons aged 65 and older. Approximately 10–15% of persons with ADRD are aged less than 65. Such persons were not included in our sample. It is likely that cost differentials between persons with and without ADRD would be even larger for persons aged less than 65 because the age impact on morbidity strengthens with age. Second, the screening procedure used in the NLTCS resulted in a more disabled sample compared to the entire population aged 65 and older making our per capita cost estimates in particular likely overestimates. However, our models included more than 1,700 community respondents who did not report a limitation in an ADL in the 1994 NLTCS. Thus, we had a substantial amount of variation in terms of disability in the community sample. Having more healthy elderly persons in our sample would likely serve to increase cost differences between persons with ADRD and the no ADRD group and would likely increase the relative impact of ADRD, disability, and comorbidities on cost. Third, our cost estimates did not account for the burden on caregivers in terms of physical and emotional strain, delay of care seeking, and reduced employment, among other hardships involved in providing care to demented individuals (Gwyther 1998; Jones and Peters 1992). Such costs are real, but they are very subjective from the standpoint of monetizing them. Fourth, our cost accounting approach ensured that each person had costs measured for 1 year exactly, which is good but introduces a survivor bias to the sample because costs just before death are generally elevated, likely causing overestimates of the point estimates of median annual cost of care. The annualization procedure we used should not bias the multivariate findings, because multiplying a weekly cost by 52 simply introduces a constant to the dependent variable for each case in the study. Finally, we were not able to assess costs in institutions such as nursing homes in a similar manner to allow us to identify the effect of residential location on costs, or to see what factors may predict cost variation within nursing home.
Further work is needed that compares the adjusted cost of caring for persons in community settings, traditional nursing homes, and other residential long-term care settings that are specifically designed to care for persons with ADRD. Work that analyzes the cost of caring for a person and also considers any treatment benefits and/or benefits to caregivers would be of the most use for families who must make treatment decisions as well as for clinicians and policymakers who are trying to hold down costs while maintaining an appropriate level of care for persons with ADRD.
|
Acknowledgments |
|---|
We thank Mike Ezell for excellent research assistance and three anonymous reviewers for their comments on earlier drafts.
Address correspondence to Don Taylor, 122 Old Chemistry Bldg., Center for Health Policy, Law and Management, Duke University, Durham, NC 27708. E-mail:
Received for publication February 4, 2000. Accepted for publication March 14, 2001.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  C. T. Cigolle, K. M. Langa, M. U. Kabeto, Z. Tian, and C. S. Blaum Geriatric Conditions and Disability: The Health and Retirement Study Ann Intern Med, August 7, 2007; 147(3): 156 - 164. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. D. Wolinsky, D. K. Miller, E. M. Andresen, T. K. Malmstrom, J. P. Miller, and T. R. Miller Effect of Subclinical Status in Functional Limitation and Disability on Adverse Health Outcomes 3 Years Later J. Gerontol. A Biol. Sci. Med. Sci., January 1, 2007; 62(1): 101 - 106. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Stuart, A. L. Gruber-Baldini, C. Fahlman, C. C. Quinn, L. Burton, I. H. Zuckerman, J. R. Hebel, S. Zimmerman, P. K. Singhal, and J. Magaziner Medicare Cost Differences Between Nursing Home Patients Admitted With and Without Dementia Gerontologist, August 1, 2005; 45(4): 505 - 515. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. S. Bloom, S. Chhatre, and R. Jayadevappa Cost effects of a specialized care center for people with Alzheimer's disease American Journal of Alzheimer's Disease and Other Dementias, July 1, 2004; 19(4): 226 - 232. [Abstract] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
