This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

F U T U R E  E L D E R L Y
D I S A B I L I T Y  &  S P E N D I N G  

26 September 2005  
  

Disability And Health Care Spending 
Among Medicare Beneficiaries

Improved disability status among the elderly is unlikely
to eliminate cost pressures as the number of beneficiaries
continues to rise.

By Michael E. Chernew, Dana P. Goldman, Feng Pan, and
Baoping Shang

ABSTRACT:

This paper forecasts the impact of changing disability rates on spending by Medicare beneficiaries. We adjust for differential changes in spending by the disabled because the composition of the disabled population and the intensity of their treatment are changing. Among community-dwelling elderly, spending growth among the least disabled grew more quickly than among the most disabled, which offsets some of the cost savings associated with declining disability rates. Using estimates of spending trends by disability category, we project that the cost savings associated with improved disability rates will not dramatically slow Medicare spending in the long run.

With the coming influx of the aging U.S. population into Medicare, the fiscal challenges posed by a larger elderly population will depend crucially on their health status. Studies have found that seniors’ medical spending is largely determined by their disability status.¹ Compared with sociodemographic, psychosocial, and diagnosis variables, functional status demonstrated the strongest association with the use of medical care.² Per capita spending for people with five or more limitations in activities of daily living (ADLs) is nearly five times the amount incurred by those with limitations in only instrumental activities of daily living (IADLs).³

Recent studies have shown that disability among the elderly has been falling over time.⁴ According to Kenneth Manton and XiLiang Gu, the decline in chronic disability prevalence accelerated from 1994 to 1999 compared with 1989 to 1994, and institutionalization among the elderly also dropped substantially during that period.⁵ Because less disabled people have lower mortality rates, the decline in disability may be associated with a decline in mortality, which would contribute to the increase in the elderly population. Timothy Waidmann and Korbin Liu suggest that if disability rates continue their current decline, the number of disabled elderly people will not grow either in absolute terms or relative to the size of the working-age population, even in the face of the dramatic growth in the elderly population.⁶

A reduction in disability rates will tend to reduce health care spending because less disabled people spend less than those with greater disability. A recent study by James Lubitz and colleagues suggests that the savings of improved health might offset health care costs as the result of longer life.⁷ Specifically, if disability trends continue, the expected cumulative health expenditures per elderly person might not increase in the future, despite greater longevity. However, whether or not disability trends among the elderly will continue to improve is uncertain. Evidence suggests that disability rates among the young population rose during the past fifteen years.⁸ As this population ages, it will likely generate a greater rate of disability among the elderly.

What is missing from all of this discussion is whether relative spending on the disabled is changing over time. The extent of the cost savings associated with lower rates of disability will depend on how much the less disabled spend. Our goal in this paper is to address this missing piece by investigating changes over time in the relationship between disability and spending.

The underlying technological and medical advances that have led to rising health care spending and improved health status among the elderly might not affect all subpopulations equally and therefore might change the relationship between spending and disability. In fact, if spending reduces rates of transition to more disabled health states, the decline in rates of disability might reflect the effects of greater spending on the least disabled.

David Cutler and Ellen Meara provide empirical support for the hypothesis that growth in medical spending may vary across subpopulations.⁹ They find that medical spending for people age 85 or older rose more rapidly than for those ages 65–69, as a result of increased use of postacute care services among the oldest old. Moreover, they report that spending among the elderly increased rapidly between 1984 and 1995, despite the decline of disability rates during that period. Therefore, understanding the spending trends conditional on disability is crucial for the projection of future Medicare spending.

Study Methods And Data

We used data from the 1990s to estimate how the relationship between disability and spending changed during that period. Our analysis of disability is based on the two building blocks used to measure disability in most empirical work: limitations in ADLs and IADLs.¹⁰ Our analysis is based on six ADLs (eating, bathing, dressing, transferring from bed to chair, walking, and using the toilet) and six IADLs (using the phone, doing light housework, doing heavy housework, making meals, shopping, and managing money).¹¹ The level of ADL limitation reflects not only one’s health status but also one’s environment and social situation.

Methods. The impact of improving disability rates on Medicare spending will reflect two offsetting factors: a case-mix effect and a mortality effect. The case-mix effect arises because people with less disability spend less, which suggests that improved disability rates will reduce spending. The mortality effect captures the fact that people with less disability tend to have lower mortality rates. As a result, program spending will rise as improved disability leads to the existence of more beneficiaries. The magnitude of these effects depends on the differential costs across disability groups and the differential mortality rates. Both the cost differential and the mortality differential might change over time and influence the extent to which disability trends influence spending.

Our analysis of the impact of improved disability rates on spending by Medicare beneficiaries relies on a microsimulation model developed to forecast future disease, functional status, and spending of the elderly.¹² It is based on age-specific transitions across health and disability states.¹³ Spending is tied to health and demographic factors. The model allows us to alter the trajectory of disability and spending conditional on disability, to forecast future spending. It also allows us to alter the hazard rates associated with transitions into disability states. In the model, those hazard rates are a function of demographics and disease. Thus, we can manipulate the transition probabilities in aggregate (for example, reduce the probability of becoming disabled by 10 percent) or alter the impact of any given illness on the probability of becoming disabled (for example, assume that the relationship between heart disease and the likelihood of disability weakens).

The model has several key components. First, it begins tracking people when they are age sixty-five. For this reason, assumptions about entering cohorts are important. We assumed that the disability status of existing cohorts reflects the current disability status of that age group and that entering cohorts enter with the same disability profile of current sixty-five-year-olds. Second, following the assumption of the Social Security Administration, we assumed that mortality rates decline at a rate of 0.68 percent per year.¹⁴ Although mortality rates differ by disability category, we assumed that the relative decline in mortality is the same across disability states.

To estimate the trends in spending by disability group among the elderly, we modeled total spending by Medicare beneficiaries (program spending plus beneficiaries’ out-of-pocket spending and spending by other payers) as a function of disability, other covariates (including disease burden), and a set of parameters. Changes in Medicare spending over time will reflect either changes in the distribution of covariates (including disability) or changes in the coefficients that relate those covariates to spending.

We estimated a single-equation generalized linear model (GLM).¹⁵ Our initial specification included interactions of all explanatory variables with a linear time trend, thereby allowing the relationship between spending and all covariates (including disability and disease) to change over time. After running the fully interacted model, we dropped the time interactions for domains and disease states in which the estimates suggested that the coefficients were stable over time. These dropped interactions include those of time with age, education, sex, race, region of residence, marital status, and smoking history. The dropping of the interaction terms signifies stability of the effects of these variables over time. The corresponding variables not interacted with time were retained in the model and were often important predictors of spending. The exclusion of these interactions did not substantively change the estimated cost trends by disability category.

We tested this model against other models of spending that made different functional form and distributional assumptions for spending.¹⁶ Our tests were based on both a split sample approach and a set of models that were estimated on data from 1992 through a given year (t) and then used results to predict spending in periods after t through 1999. We estimated several such models using different years to define t. Our model-selection criterion was based on mean average prediction errors.¹⁷ We also tested a model using a nonlinear time trend, ln (t), instead of a linear term. The nonlinear time trend model had a slightly greater mean average prediction error, so we used the linear time trend model for our analysis.

Because disability is a marker of disease, a portion of the association between spending and disability might not be causal. Higher spending on people with disabilities could reflect efforts to treat the underlying disease that caused the disability. If people with disabilities suffer disproportionately from diseases (or disease severity) that we cannot control for in our analysis, then higher spending could reflect efforts to treat the underlying medical conditions not otherwise controlled for.

To estimate the effects of disability trends on spending by Medicare beneficiaries, we estimated trends in program spending under three scenarios. First, we estimated spending under our baseline model, which assumes that the transition probabilities between disability and disease states match those observed in the 1990s. This model allows cost trends to differ by disability categories and other enrollee traits.

Second, we simulated spending assuming that the hazard rates adjust, so that the prevalence among community-dwelling elderly of having one or more IADL limitations, one or two ADL limitations, three or four ADL limitations, and five or more ADL limitations declines by about 20 percent in steady state. Third, we simulated spending assuming that disability does not get worse (the hazard associated with a worsening level of disability is set to 0).

Data. We used data from the Medicare Current Beneficiary Survey (MCBS), a nationally representative study designed to ascertain the Medicare population’s use of and spending for health care services. The survey, a rotating panel design of twelve interviews over three years, has been ongoing since 1991. We used data from 1992 through 2000. The MCBS sample frame consists of aged and disabled beneficiaries enrolled in Medicare Part A or Part B or both, although we used only the aged for this analysis, and the oldest old (age eighty-five or older) are oversampled. The MCBS contains demographic data (age, sex, race, and educational attainment) and detailed self-reported information on health, including the prevalence of various conditions, and measures of physical limitations in ADLs and IADLs.

Measuring costs. Our primary measure of health care spending represents total spending on health care by Medicare beneficiaries (including program expenditures, out-of-pocket costs, and spending by other payers such as Medicaid and supplemental plans). We included spending for all health care services (inpatient, outpatient, physician care, drugs, and nursing home and home care). The cost data are based on Medicare claims data, linked to the MCBS, combined with respondents’ self-reports.¹⁸ For services covered by Medicare, the data capture spending by Medicare, other payers including Medicaid, and the beneficiary. Spending for services not covered by Medicare is based on self-reports and could be underreported. All spending was adjusted to 2000 dollars using the medical care component of the Consumer Price Index (CPI), which results in a conservative estimate of cost growth.¹⁹

Measuring disability. For this study, IADL disability was defined as requiring any supervision or assistance with any of the six IADLs. ADL disability was defined as requiring any supervision or assistance with any of the ADLs. Given the number of ADLs and IADLs, a global measure of disability must aggregate the individual measures. Research has shown that a simple count of ADL or IADL limitations provides a reasonably good proxy for the hierarchical nature of the ADL items and of the relative severity of a person’s disabilities.²⁰ Our measure of the severity of disability for the noninstitutionalized population was based on five commonly used categories: nondisabled, those with IADL limitations only, those with one or two ADL limitations, those with three or four ADL limitations, and those with five or more ADL limitations.

The MCBS does not consistently collect ADL counts for people residing in nursing homes. Thus, we treated nursing home residence as an additional category of disability above those living in the community, but we recognized that some community-dwelling elderly people might suffer from greater disability than some nursing home residents.

Measuring disease. The MCBS contains a wide array of self-reported diagnoses. We included binary variables measuring the presence of several major diseases often linked to expenditures: diabetes, cancer, heart disease, stroke, Alzheimer’s disease, hypertension, osteoarthritis, and lung disease.

Other covariates. We chose additional binary variables that have been found to be associated with health care spending. These include age (65–69, 70–74, 75–79, 80–84, 85+), sex, marital status, race (white, black, Hispanic), education level (fewer than 11 years, 12–15 years, 16 years or more), geographical region (Midwest, West, Northeast, South, Puerto Rico, or unknown), whether the person ever smoked, body mass index (BMI) category (obese, overweight, underweight), and supplemental health insurance coverage (Part A only, Part B only, Medicaid, employer supplemental, private supplemental, or health maintenance organization.

Study Results

Trends in disability among the Medicare population have fallen over time (Exhibit 1). During the study period, the average number of ADL limitations per community-dwelling beneficiary fell from 0.68 to 0.61, and the average number of IADL limitations also dropped about 10 percent. Moreover, the percentage of community-dwelling beneficiaries with at least one ADL limitation fell from 30.4 percent to 27.8 percent (data not shown).

In aggregate, spending per Medicare beneficiary rose 11 percent during this period. The increase in spending reflects an increase in spending within less disabled groups (Exhibit 2). Among community-dwelling beneficiaries, costs per beneficiary in the IADL-only group and the no-disability group rose 92 percent and 82 percent, respectively, compared with 58 percent for those with one or two ADLs, 45 percent for those with three or four ADLs, and 44 percent for those with five or more ADLs. As a result, the ratios of spending in the ADL disabled categories, relative to spending among the nondisabled, fell during the study period. After a rapid rise in spending among nursing home residents early in the study period, there was a decline starting in 1997. This might reflect changes in reimbursement rules. Overall, there was an 8.0 percent increase in spending among institutionalized beneficiaries.

The evidence that spending rose most rapidly among the least disabled may partially reflect changing disease mix within the disability categories. There is some evidence that the less disabled had a greater increase in the prevalence of certain diseases than did those in categories of greater disability. For example, between 1992 and 1999, a period prior to changes in the survey question regarding heart disease, there was a 22 percent increase in self-reported heart disease among the nondisabled, compared with only 12 percent among the most disabled.

Between 1992 and 2000, self-reported diabetes within the nondisabled category increased 31 percent, compared with only 5 percent among people with three or four ADLs. The evidence for cancer is mixed. The percentage with cancer rose 12 percent in the group with three or four ADLs, dropped 4 percent in the group with one or two ADLs, rose 6 percent in the nondisabled group, and dropped 6 percent in the group with five or more ADLs.

The multivariate analysis, which formed the basis for our inference and simulations, allowed us to examine differences in cost levels by disability group—as well as cost trends—after adjusting for disease and other factors, including health behavior, insurance status, and sociodemographic status. Consistent with the literature, we found that costs among the more disabled were higher than among the less disabled. Among all of the health and demographic characteristics, the variables for three or four ADLs, five or more ADLs, and nursing home status, whose coefficients are significantly greater than 1, demonstrate the strongest associations with total medical spending.²¹ The results indicate that total medical care spending of elderly with one or two ADLs was about twice that of the nondisabled. Elderly people with five or more ADLs incurred four to five times the medical care spending of the nondisabled.

We interacted disability status with the time trend to determine whether there were statistically significant differences in cost growth from 1992 to 2000. The results indicate that spending grew most rapidly among the least disabled groups. Adjusted spending by the nondisabled and IADL disabled groups rose 23 percent and 28 percent, respectively, compared with a 10 percent increase for those with one or two ADLs, 0.6 percent for those with three or four ADLs, and a 10 percent decrease for the most disabled. As a result, the ratio of spending among the ADL disabled groups, relative to the nondisabled, declined over the study period.

These findings hold in the GLM model, regardless of whether we also allowed differential trends in spending by age or region. Thus, changes in spending are not explicitly tied to changes in the way we treat the oldest old or diseases such as heart disease; this lends credence to a causal interpretation.

These results have implications for the potential cost savings from the overall decline in elderly disability. Exhibit 3 illustrates the beneficial impact of reduced disability on costs per beneficiary, assuming that the trends estimated in our model persist. The baseline curve reflects the status quo hazard rates. Scenario A adjusts the hazard rates so that the prevalence among community-dwelling elderly people of having one or more IADLs, one or two ADLs, three or four ADLs, and five or more ADLs declines by about 20 percent in steady state. Scenario B eliminates the hazard rate associated with a worsening level of disability and has the greatest impact on spending. Notice in both cases that the savings associated with improved disability diminish over time. This is because the cost growth is greatest for the least disabled groups.

In aggregate, our projections suggest that improved disability trends will not slow total spending for Medicare beneficiaries (Exhibit 4). Improvements in disability may reduce current costs, but because of the associated greater longevity and rapid cost growth among the less disabled, total expected lifetime costs might not drop. Essentially, efforts to reduce disability are valuable, but if current trends in spending by disability category continue, success at reducing disability will not result in substantial cost savings per beneficiary.

Discussion

Study limitations. Our approach has several important limitations. First, our projections of spending trends were based on Medicare’s experience from 1992 to 2000, which might not be typical. For example, our model did not adjust directly for Medicare reforms that undoubtedly affected the relationship between disability and spending over time. For example, we did not directly adjust for changes in physician payment methodologies, nor did we adjust for the Balanced Budget Act (BBA) of 1997, which influenced payments, particularly for long-term care. Payment systems such as those implemented in the BBA can have a major effect on the convergence in spending. As a result, the trends imposed might not generalize to future experience, which will reflect future changes to the Medicare payment systems (including the effects of the Medicare Prescription Drug, Improvement, and Modernization Act, or MMA, of 2003, which we did not capture).

In fact, projections of past trends suggest that during our projection period, spending might no longer rise monotonically with disability category. This might not be realistic, but we would need more detailed knowledge of the forces driving spending growth in the less disabled groups to better assess the plausibility of the convergence in spending by disability category.

Second, we did not differentiate mortality trends by disability category. Differential mortality trends could also affect the impact of changing disability rates on cost trends. Third, we did not distinguish between spending by the Medicare program and spending by the beneficiary. In part this is because future trends might differ in the share paid by the beneficiary because of provisions of MMA.

Finally, our measure of disability was very crude (based only on a rough aggregation of IADL and ADL limitations). From this analysis, we could neither provide greater insight regarding the clinical factors driving spending trends by disability group nor distinguish clearly between causally related effects of disability trends on costs and effects arising because of important unobservable factors that relate to our disability measures. Although we controlled for major disease categories, unmeasured case-mix changes within disability categories could have influenced results.

Policy implications. With society aging, policymakers must be concerned with the fiscal responsibilities associated with an older population. Existing trends in disability suggest that the costs associated with aging will be lower than simple projections because the elderly of the future will likely be less disabled than the current cohort of seniors. However, the optimism contained in spending forecasts that assume substantial savings associated with reduced disability rates might be overstated. Our analysis suggests that although the less disabled spend less than the more disabled, the differential is narrowing over time. Significant cost growth has occurred in the least disabled population; if this is predictive of future trends, it will offset some of the savings associated with improved disability status.

A fundamental question in predicting future spending is whether the convergence of spending across disability groups among the elderly will continue. In the future, convergence will reflect the changes in Medicare reimbursement policy and benefit design. It might also be the case that technical innovations affect the medical spending among the elderly unevenly across disability groups. As the number of Medicare beneficiaries continues to grow, policymakers will be challenged to design systems that promote efficiency in the delivery of their care. Our analysis indicates that it is unlikely that improved disability status among the elderly will eliminate cost pressures, which suggests that tough choices will be necessary and political pressures will be great.

This research was funded by the National Institute on Aging (NIA) and the Lasker Foundation through a grant to the National Bureau of Economic Research (NBER), with additional funding from NIA through its support of the RAND Roybal Center for Health Policy Simulation (P30AG024968) and the RAND Center for the Study of Aging (P30AG12815). The authors are grateful to participants in the NBER disability workshop meetings in Charleston, South Carolina, and Jackson Hole, Wyoming, for helpful comments.

NOTES

1. K. Liu, S. Wall, and D. Wissoker, “Disability and Medicare Costs of Elderly Persons,” Milbank Quarterly 75, no. 4 (1997): 461–493; T.R. Fried et al., “Functional Disability and Health Care Expenditures for Older Persons,” Archives of Internal Medicine 161, no. 21 (2001): 2602–2607; and D.M. Cutler and E. Meara, “The Concentration of Medical Spending: An Update,” NBER Working Paper no. w7279 (Cambridge, Mass.: National Bureau of Economic Research, 1999).
2. Fried et al., “Functional Disability and Health Care Expenditures.”
3. Cutler and Meara, “The Concentration of Medical Spending.”
4. Measuring disability and trends in disability is complex for a range of data and conceptual reasons. For an overview of some of the literature in this area, see V.A. Freedman, L.G. Martin, and R.F. Schoeni, “Recent Trends in Disability and Functioning among Older Adults in the United States: A Systematic Review,” Journal of the American Medical Association 288, no. 24 (2002): 3137–3146; V.A. Freedman et al., “Resolving Inconsistencies in Trends in Old-Age Disability: Report from a Technical Working Group,” Demography 41, no. 3 (2004): 417–441; and B.C. Spillman, “Changes in Elderly Disability Rates and the Implications for Health Care Utilization and Cost,” Milbank Quarterly 82, no. 1 (2004): 157–194
5. K.G. Manton and X. Gu, “Changes in the Prevalence of Chronic Disability in the United States Black and Nonblack Population above Age Sixty-five from 1982 to 1999,” Proceedings of the National Academy of Sciences (U.S.) 98, no. 11 (2001): 6354–6359.
6. T.A. Waidmann and K. Liu, “Disability Trends among Elderly Persons and Implication for the Future,” Journals of Gerontology, Series B: Psychological Sciences and Social Sciences 55, no. 5 (2000): S298–S307.
7. J. Lubitz et al., “Health, Life Expectancy, and Health Care Spending among the Elderly,” New England Journal of Medicine 349, no. 11 (2003): 1048–1055.
8. D.N. Lakdawalla, J. Bhattacharya, and D.P. Goldman, “Are the Young Becoming More Disabled?” Health Affairs 23, no. 1 (2004): 168–176.
9. Cutler and Meara, “The Concentration of Medical Spending.”
10. C.J. Evashwick, ed., The Continuum of Long-Term Care, 2d ed. (Albany, N.Y.: Delmar Thomson Learning, 2001).
11. Three of the IADL limitations are not asked about in interviews with respondents who did not reside in the community.
12. D.P. Goldman et al., Health Status and Medical Treatment of the Future Elderly: Final Report, Pub. no. TR-169-CMS (Santa Monica, Calif.: RAND, 2004).
13. The simulation is a first-order Markov Model, which imposes the condition that transitions between health states at any given time are a function of only the health status at the start of the period and do not reflect the entire history of health status.
14. Social Security Administration, The 2004 Annual Report of the Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds, 2004, www.ssa.gov/OACT/TR/TR04/tr04.pdf (25 July 2005), 73–74.
15. Our model assumes that spending, conditional on the covariates, is an exponential function of a linear combination of the covariates plus an additive stochastic term that follows a normal (Gaussian) distribution.
16. We tested our model against an ordinary-least-squares (OLS) model and a model that assumed a Gamma distribution for spending instead of a Gaussian distribution.
17. The nonlinear model with the Gaussian distribution had the lowest mean average prediction errors.
18. F. Eppig and G.S. Chulis, “Matching MCBS and Medicare Data: The Best of Both Worlds,” Health Care Financing Review 18, no. 3 (1997): 211–229.
19. The medical care CPI is published by the Bureau of Labor Statistics within the U.S. Department of Labor; it measures consumer price changes for a market basket of medical goods and services. Our adjustment is conservative because the medical care CPI rises more rapidly than the total CPI.
20. Liu et al., “Disability and Medicare Costs of Elderly Persons.”
21. Results are available in an online appendix, content.healthaffairs.org/cgi/content/full/hlthaff.w5.r42/DC2.

Michael Chernew (mchernew{at}umich.edu) is a professor in the Department of Health Management and Policy, University of Michigan School of Public Health, in Ann Arbor. Dana Goldman is corporate chair and director of health economics at RAND in Santa Monica, California. Feng Pan is a doctoral candidate in the Department of Health Management and Policy. Baoping Shang is a fellow at the Pardee RAND Graduate School.

Access the table of contents for this package

DOI: 10.1377/hlthaff.W5.R42
©2005 Project HOPE–The People-to-People Health Foundation, Inc.