Health, Life Expectancy, and Health Care Spending among the Elderly

Selection of Data

The Medicare Current Beneficiary Survey has been conducted continuously since 1991. The survey sample was drawn from Medicare enrollment files. Because Medicare covers over 96 percent of persons in the United States who are 65 years of age or older, the survey provides a very good representation of this population, especially because it includes institutionalized persons. The survey gathers information from about 12,500 Medicare beneficiaries on sociodemographic characteristics, use and costs of services covered by Medicare (services provided by inpatient and outpatient hospitals, skilled nursing facilities, hospice programs, physicians, and other practitioners, as well as some home health care) and services not covered (e.g., prescription drugs, nursing home care below the skilled nursing level as defined by Medicare, and dental care). Information on use and expenditures is gathered in three in-person interviews per year with a recall period of four months; memory aids (e.g., calendars and statements from Medicare and other insurers) are used to ensure completeness and accuracy. Expense data on Medicare-reimbursed services and mortality data are taken directly from Medicare records. Events reported by respondents are linked to claims, and imputation procedures are used to develop information on health care expenditures that is as accurate as possible.^15-17

The Medicare Current Beneficiary Survey follows a rotating panel design in which one third of the sample is replaced each year. Information on health status is gathered each fall. Persons included in the sample who neither drop out of the survey nor die have four fall interviews. If a respondent drops out after a fall interview and dies during the next year, the data for that person are included so that mortality rates are not underestimated. The response rate for the survey is about 70 percent.

We used survey data from 1992 through 1998. Our study was restricted to persons who were 70 years of age or older to avoid bias, because many persons newly enrolled in Medicare at 65 to 69 years of age may not be eligible to be interviewed in their first year of enrollment. Our study included 16,964 persons, with a total of 50,477 person-years.

We classified health status on the basis of responses to questions about five activities used as measures of physical functioning, developed by Nagi,¹⁸ six instrumental activities of daily living, and six activities of daily living. These measures are frequently used to characterize the health of the elderly.^2,4,19,20 The five Nagi activities are stooping, crouching, or kneeling; lifting or carrying objects weighing up to 6 kg (10 lb); extending the arms above the shoulder; grasping small objects; and walking two to three blocks. Respondents are asked how much difficulty, if any, they have with the activity, and the answers range from “no difficulty at all” to “not able to do it.” We counted persons who responded that they had any difficulty or that they were unable to perform the activity as having a limitation in physical functioning.

The six instrumental activities of daily living are using the telephone, doing light housework, doing heavy housework, preparing meals, shopping for personal items, and managing money. The six activities of daily living are bathing or showering, dressing, eating, getting into or out of a bed or a chair, walking, and using the toilet. For the purpose of our study, persons who reported having any difficulty or not being able to perform the activity for reasons of health were considered to have a limitation in the activity.

We defined states of health according to the following classification: no limitations, at least one Nagi limitation but no other limitations, a limitation in at least one instrumental activity of daily living but no limitations in activities of daily living, a limitation in at least one activity of daily living, institutionalization (e.g., in a nursing home), or death. Such hierarchical classifications are similar to those used in many disability models.^6,21,22 There were 15,278 changes in health state and 3462 deaths.

Active life was defined as life with no reported limitations or only Nagi limitations. Most institutionalized persons were in nursing homes, and most nursing home residents received assistance with one or more activities of daily living.²³ As a measure of health in separate analyses, we also used self-rated health status, for which responses ranged from excellent to poor.

Statistical Analysis

Table 1. Table 1. Probability of a Change in Functional Status after One Year among Medicare Beneficiaries 75 and 85 Years of Age, as Computed with the Use of Multivariate Hazard Models, for the Years 1992 through 1998 Combined.

We used multistate life-table methods to estimate total and active life expectancy. Multistate models allow for transitions among all states of health. Given our classification of functional status or self-rated health status into five states of health and death, there were 25 possible transitions from one state to another. Age-specific, first-order Markov transition probabilities were estimated with the use of a multivariate hazard model, with age and sex or age and race as the covariates — an approach that is similar to that used in previous studies.^13,24 The 25 equations, one for each possible transition, produced age-specific matrixes of annual transition probabilities. Examples of these probabilities at 75 and 85 years of age are shown in Table 1. As the table shows, the majority of persons at those ages will be in the same state of health after one year. The probability of institutionalization or death increases as the functional state worsens and is higher for those who are 85 years old at all initial functional states. Hazard estimates were weighted to reflect the sample design with the use of cross-sectional survey weights.

Health-expenditure matrixes were structured in a similar manner to the transition matrixes. Each cell of the matrix contains the average expenditures incurred when a person changed (or did not change) from one of the five initial health states to one of the six ending states. Expenditures were not modeled but were categorized according to age, sex, race, and type of transition. The expenditures associated with a change in health status from the fall of one year to the fall of the next year were the calendar-year expenditures for the later year. Expenditures were adjusted for inflation to 1998 dollars with the use of the rate of increase in Medicare per capita expenditures.⁹

We then used microsimulation to simulate a cohort (of 100,000 persons 70 years of age) whose changes in health status were governed by these estimated probabilities, and we recorded life expectancy and health care expenditures. The simulation approach is similar to an approach used previously,¹⁴ but we extended its application by associating annual health care expenditures with changes in health status.

Our estimates of life expectancy at the age of 70 years are somewhat lower (by 7 percent or less) than those of the National Vital Statistics System.²⁵ Other studies in which similar methods were used have also produced different estimates from those of the National Vital Statistics System.^24,26 This difference is likely to be due to our use of multistate life-table methods in our analysis. Health-state transition probabilities were estimated with 25 separate hazard-model equations and then used to produce a single estimate of life expectancy. Use of the same data with single-decrement life-table methods produces estimates of total life expectancy that are similar to those published by the National Vital Statistics System.

Because our estimates of life expectancy and cumulative expenditures are complex functions of the transition probabilities, we used the bootstrap method to estimate standard errors.²⁷ We sampled respondents from 67 primary sampling-unit groups. Within each group, we sampled Medicare beneficiaries with replacement with size equal to one less than the original group size. We then estimated the transition probabilities of this bootstrap sample with multivariate hazard models, as described above, and computed average life expectancy and expenditures on the basis of simulations of 25,000 persons at the age of 70. We performed this set of calculations 1000 times. Standard errors were computed from these 1000 estimates. Comparisons between groups were performed with the use of two-sample t-tests. All reported differences are significant at the level of P≤0.05 for a two-sided test. The relative standard errors for the functional state or self-reported state of health in the figures were less than 10 percent, except that in the figures showing life expectancy and expenditures in relation to functional state, the relative standard errors for years lived and expenditures incurred in noninstitutional states for persons institutionalized at age 70 were about 25 percent.

Table 2. Table 2. Years Spent in Different States of Health and Cumulative Health Care Expenditures from 70 Years of Age until Death, According to Sex and Race.

At 70 years of age, 28 percent of the study population had no functional limitations, 40 percent had only Nagi limitations, 12 percent had at least one limitation in an instrumental activity of daily living but no limitations in activities of daily living, 18 percent had a limitation in an activity of daily living, and 2 percent were institutionalized (data not shown). At age 70, total life expectancy was 13.2 years, of which 52 percent were active years (i.e., almost 7 years with either no limitations or only Nagi limitations) (Table 2). Total expenditures for medical care from age 70 to death were about $140,700. The average expenditures per year increased with worsening health status, from about $4,600 for persons reporting no limitations to about $45,400 for institutionalized persons. The expected expenditures for men were lower than those for women. Men actually had higher expenditures per year in every health state but had lower total expenditures because of a shorter life expectancy and also fewer years in the health states that incurred the greatest expenditures. Blacks had both a lower overall life expectancy and a lower active life expectancy than whites, but had similar levels of expenditures.

Expenditures incurred while a person had limitations in activities of daily living or was in an institution accounted for a large part of total costs from 70 years of age until death. For example, a person at age 70 could expect to live 34 percent of remaining life (4.5 years) with limitations in activities of daily living or in an institution but to incur 63 percent of medical expenditures (about $88,200) in these health states (Table 2).

Estimates of Life Expectancy and Health Care Expenditures According to Health Status

Figure 1. Figure 1. Life Expectancy at 70 Years of Age According to Functional State at the Age of 70.

The shading in the bars indicates the expected number of years lived in various functional states. For example, a person with no limitations at the age of 70 is estimated to live an additional 14.3 years, on average. Of those 14.3 years, 0.7 will be spent in an institution, 4.9 with a limitation in at least one instrumental activity of daily living (IADL) or activity of daily living (ADL), and 8.7 (61 percent of total life expectancy) with no limitation or only Nagi limitations. Instrumental activities of daily living, activities of daily living, and Nagi limitations are described in the Methods section.

Persons in better health at 70 years of age had a longer life expectancy than those in worse health (Figure 1). Persons with no limitations had the longest life expectancy, and institutionalized persons the shortest. Persons with better health were also expected to be active for a longer period. For example, the 28 percent of persons 70 years of age who had no limitations could expect to be active for 61 percent of their remaining years. In contrast, the 18 percent of persons 70 years of age who had a limitation in an activity of daily living could expect to be active for only 35 percent of their remaining 11.6 years.

Persons who were living in the community at age 70, regardless of their state of health, could expect to spend about 0.7 year in an institution. Persons in better health at age 70 might be expected to spend less time in an institution than persons with functional limitations, but persons in good health live longer, and longevity is associated with lack of social support (e.g., widowhood) and frailty, and thus with a high risk of institutionalization. However, in our study the annual risk of institutionalization was lower for those in better health at 70 years of age; they lived longer, but the expected time spent in an institution was the same as for persons in poorer health.

Figure 2. Figure 2. Life Expectancy at 70 Years of Age According to Self-Reported Health at the Age of 70.

The shading in the bars indicates the expected number of years lived in various states of health. For example, a person who reports excellent health at the age of 70 is estimated to live an additional 13.8 years, on average. Of those 13.8 years, 2.7 will be lived in fair or poor health, 3.7 in good health, and 7.3 (53 percent of total life expectancy) in very good or excellent health.

The same pattern of longer life for persons in better health was found when we used self-reported health status as a measure of health (Figure 2). Those who reported excellent health at 70 years of age had a life expectancy of 13.8 years, with most of that time spent in excellent or very good health. Those who reported poor health had a life expectancy of 9.3 years, with most of that time spent in fair or poor health.

Figure 3. Figure 3. Expected Expenditures for Health Care from 70 Years of Age until Death According to Functional State at the Age of 70.

Expenditures are in 1998 dollars. The shading in the bars indicates estimated health care expenditures for persons in various functional states. For example, a person with no limitation at the age of 70 is estimated to have cumulative health care expenditures of about $136,000 from the age of 70 until death. Of this amount, about $32,000 will be spent while the person is institutionalized, about $60,000 for care while the person has a limitation in at least one instrumental activity of daily living (IADL) or activity of daily living (ADL), and about $44,000 (32 percent of total expenditures) for care in the absence of limitations or with only Nagi limitations. Instrumental activities of daily living, activities of daily living, and Nagi limitations are described in the Methods section.

Figure 4. Figure 4. Expected Expenditures for Health Care from 70 Years of Age until Death According to Self-Reported Health at the Age of 70.

Expenditures are in 1998 dollars. The shading in the bars indicates health care expenditures for persons in various states of health. For example, a person reporting excellent health at the age of 70 is estimated to have cumulative health care expenditures of about $150,000 from the age of 70 until death. Of this amount, about $62,000 will be spent while the person is in fair or poor health, about $45,000 while the person is in good health, and about $43,000 (29 percent of total expenditures) while the person is in very good or excellent health.

Persons without functional limitations at 70 years of age who lived longer did not incur higher health care expenditures (Figure 3). Health care expenditures for persons 70 years of age or older who were living in the community at 70 years of age varied little according to initial health status. Persons without functional limitations incurred an estimated $136,000 in medical expenses from age 70 until death, as compared with an estimated $145,000 for persons with a limitation in at least one activity of daily living. Only those who were initially in an institution had much higher expenditures, which were the consequence of high nursing home costs. When we categorized persons only according to functional status, with no separate category for those institutionalized, and defined functional status as both having difficulty and receiving help with instrumental activities of daily living or activities of daily living, those in better functional states had greater longevity, but there was little variation in expected expenditures (data not shown). Similarly, health care expenditures from the age of 70 years and onward varied little according to the initial self-reported health state, despite differences in longevity (Figure 4).

By linking data on medical care expenditures to estimates of life expectancy for persons 70 years of age in various health states, we estimated the relations among health, longevity, and expected health care spending. Our analysis shows not only that persons in good health at 70 years of age can expect to live longer and to have more years of good health than those in poor health at age 70, but also that their total expected medical care expenses appear to be no greater than those for less healthy persons, even though healthier persons live longer. Lower annual expenditures from the age of 70 until death among healthier persons offset the greater time they have to accumulate health care costs — a finding hinted at in earlier research.^11,28

The possibility that better health among the elderly will moderate the expected increases in medical care spending for the elderly has been suggested by earlier studies.^6,7 Our results, however, raise questions about this possibility. For persons who reach the age of 70 in better health and who have more remaining years of life, the cumulative health care expenditures until death are similar to those for persons in poor health at the age of 70.

There are a number of limitations to our study. First, the age-specific probabilities of changes in health states that we used to produce our estimates were based on the period from 1992 to 1998. We did not take into account demographic and social changes or changes in medical care that might affect the relation between health status and expenditures. For example, changes are now occurring in long-term care, including a decrease in informal care, an increase in formal paid care, and an increase in the number of assisted-living facilities.^29,30 It is unclear how these changes may affect the costs of institutionalization, which make up a large part of health care costs for the elderly. In addition, future medical advances may increase costs while lowering rates of disability. Thus, caution is needed when projecting these patterns into the future.

Second, life-table methods assume a first-order Markov transition process. This assumption ignores the relation between the current health state and past states, except for the immediately preceding state. Third, our measures of health do not capture dimensions such as cognitive health, emotional health, or pain. Cognitive status is an important dimension of health in the elderly³¹; pain is also important — for instance, in the care of patients with cancer.³² Finally, our analysis captures only formal health care, not informal caregiving, which can be costly, both financially and emotionally, for family members of an elderly person.³³

Our study shows clearly that for the elderly, better health results in longer life but not in higher health care expenditures. Of course, there may be health care costs before the age of 70 years that enable people to reach old age in good health and in a good functional state. More research is needed to understand these factors.

It is not clear what the trends in the health of the elderly will be in the future. Favorable trends among the elderly in the areas of smoking cessation, education, and exercise compete with other trends toward increases in obesity and asthma among those under the age of 65. In any event, we believe that the patterns found in our study suggest that health-promotion efforts in the nonelderly population that have payoffs in better health and longer life for the elderly will not increase health care spending among the elderly.