NBER WORKING PAPER SERIES EVIDENCE FOR SIGNIFICANT COMPRESSION OF MORBIDITY IN THE ELDERLY U.S. POPULATION David M. Cutler Kaushik Ghosh Mary Beth Landrum Working Paper 19268 http://www.nber.org/papers/w19268 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 July 2013 We are grateful to the National Institute of Aging for research support (P01-AG005842) and to Dan McFadden for helpful comments. The views expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau of Economic Research. At least one co-author has disclosed a financial relationship of potential relevance for this research. Further information is available online at http://www.nber.org/papers/w19268.ack NBER working papers are circulated for discussion and comment purposes. They have not been peerreviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications. (C) 2013 by David M. Cutler, Kaushik Ghosh, and Mary Beth Landrum. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including (C) notice, is given to the source. Evidence for Significant Compression of Morbidity In the Elderly U.S. Population David M. Cutler, Kaushik Ghosh, and Mary Beth Landrum NBER Working Paper No. 19268 July 2013 JEL No. I1,J11 ABSTRACT The question of whether morbidity is being compressed into the period just before death has been at the center of health debates in the United States for some time. Compression of morbidity would lead to longer life but less rapid medical spending increases than if life extension were accompanied by expanding morbidity. Using nearly 20 years of data from the Medicare Current Beneficiary Survey, we examine how health is changing by time period until death. We show that functional measures of health are improving, and more so the farther away from death the person is surveyed. Disease rates are relatively constant at all times until death. On net, there is strong evidence for compression of morbidity based on measured disability, but less clear evidence based on disease-free survival. David M. Cutler Department of Economics Harvard University 1875 Cambridge Street Cambridge, MA 02138 and NBER dcutler@harvard.edu Kaushik Ghosh NBER 1050 Massachusetts Ave. Cambridge, MA 02138 ghoshk@nber.org Mary Beth Landrum Harvard Medical School Department of Health Care Policy 180 Longwood Avenue Boston, MA 02115-5899 landrum@hcp.med.harvard.edu Older Americans are living longer. Life expectancy at age 65 has increased about 2 years in the past two decades. But are we living healthier? This issue is vital for health policy and economic reasons. Longer life is valuable to people, but it is even more valuable if the additional years lived are in good health. For the public sector as well, the consequences of longer lives depend on their quality. Medical spending for healthy seniors is modest; spending for the severely disabled is much greater. Thus, if morbidity is being compressed into the period just before death, the impacts of population aging are not as severe as if additional life involves many years of expensive care. This question of whether morbidity is being compressed into the period just before death has been at the center of health debates in the United States for some time. Fries (1980) first put forward the argument that the United States was undergoing a compression of morbidity. His work was provocative, and others took different views. Gruenberg (1977) argued that reduced disease mortality would extend unhealthy life, while Manton (1982) posited a dynamic equilibrium where both morbidity and mortality are falling, leading to indeterminate impacts on disability-free and disabled life expectancy. Empirical evidence on trends in morbidity is also unclear. Some authors argue that morbidity is being compressed into the period just before death (Cai and Lubitz, 2007; Manton et al., 2008), while others believe that the period of disabled life is expanding (Crimmins and Beltr?n-S?nchez, 2010) or that the evidence is more mixed (Crimmins et al., 2009). There are three reasons for this disagreement. First, there is not a single definition of morbidity. Some studies look at whether people report specific chronic conditions, which have 1 increased over time, while other studies look at functioning. As a result, studies differ in the morbidity trends they incorporate. Second, it is often difficult to link health to the stage of life of the individual. If people are reporting more chronic disease, is that in the period just before the end of life, in which case the additional disease does not encompass many years? Or is the disease occurring in periods of time far from the end of life, in which case it represents many years of poor health? To answer this question, one needs data on quality of life matched to time until death. Most cross-section data sources do not have such a link, however, and thus they need to make assumptions about the disease process to generate lifetime disease-prevalence estimates. These assumptions can have large impacts on the results. Third, the data samples that tend to be used often focus on a particular subset of the population, for example the non-institutionalized. Since there are changes in the residential location of the elderly population over time, focusing on population subsets can give biased results. In this paper, we examine the issue of compression of morbidity, addressing these three concerns. Our primary data source is the Medicare Current Beneficiary Survey, or MCBS. We have MCBS data for a representative sample of the entire elderly population between 1991 and 2009. The sample sizes are large, over 10,000 individuals annually. Further, the MCBS data have been linked to death records through 2008, and hence all deaths can be matched. Importantly, this includes deaths that occur after the person has left the survey. Thus, we can form morbidity measures by time until death for a large, representative share of the elderly population. 2 We use these data in two ways. First, we examine trends in various measures of morbidity by time until death. We consider a number of different metrics: the presence of disease; whether the person reports ADL or IADL disability; and various summary measure of functioning that draw together 19 different dimensions of health (Cutler and Landrum, 2011). We show trends overall and by time until death. As is well known, the MCBS data from the 1990s and 2000s show a reduction in the share of elderly people who report ADL or IADL limitations (Freedman et al., 2004, 2013). Our first result is that this reduction in disability is most marked among those with many years until death. Health status in the year or two just prior to death has been relatively constant over time; in contrast, health measured three or more years before death has improved measurably. We then translate these changes into disability-free life expectancy and disabled life expectancy. We show that disability-free life expectancy is increasing over time, while disabled life expectancy is falling. For a typical person aged 65, life expectancy increased by 0.7 years between 1992 and 2005. Disability-free life expectancy increased by 1.6 years; disabled life expectancy fell by 0.9 years. The reduction in disabled life expectancy and increase in disability-free life expectancy is true for both genders and for non-whites as well as whites. Hence, morbidity is being compressed into the period just before death. The paper is structured as follows. We begin in the next section by defining the compression of morbidity and showing how disability and mortality changes jointly affect disability-free and disabled life expectancy. The second section describes the data we use. The third section presents simple trends in health status by time until death. The fourth section calculates disabled and disability-free life expectancy. The last section concludes. 3 I. The Compression of Morbidity The question we wish to examine is whether morbidity has been compressed into the period just before death, or whether it is accounting for a greater part of the life of elderly individuals. While this goal is clear, the empirical implementation needs a more precise definition. We consider two definitions of a compression of morbidity. One definition, dating back to Fries (1980), is whether the life table is 'rectangularizing' - that is, whether disabled life expectancy is falling over time. A second definition is more modest: the share of remaining life that is non-disabled is increasing over time. Note that in this latter formulation, disabled life expectancy may be increasing as well, just not as rapidly as non-disabled life expectancy. In situations where only morbidity or mortality is changing, these two measures will always move together. In situations where both mortality and morbidity are changing, however, trends in the two measures of compression of morbidity may be different. To see this, consider a simple example presented in Table 1. The first column depicts a person who lives for five years, the first three of which are without disability, and the fourth and fifth are with a disability. To be concrete, suppose that the person has heart disease in the fourth year and develops chronic obstructive pulmonary disease in the fifth, which results in death six months later. The specific diseases do not matter, but as is typical in the data, we reflect disability as occurring progressively over life and generally do not consider recovery. In forming life tables, people who die during a year are assumed to die halfway through the year. Thus, the baseline life expectancy1 is 4.5 years, of which the first 3.0 years is disability-free and the latter 1.5 years is disabled. 1 We refer to life expectancy even though this is a life table for a single person. It is easier to show the point this way than to consider a population distribution. 4 Now imagine that morbidity declines (column 2). To be specific, suppose that because of improved medical treatment of cardiac risk factors, the person does not suffer a coronary event in the fourth year and thus is not disabled in that year. In year 5, however, the person still suffers lung disease and dies. As the last rows show, overall life expectancy is unchanged, but disability-free life expectancy has increased to 4.0 years and disabled life expectancy has fallen to 0.5 years. By either definition above, disability has been compressed into the period before the end of life. The third column shows the impact of a reduction in mortality. We imagine that the medical system gets better at treating the combination of heart disease and lung disease, and thus the person survives an additional year with both conditions, albeit they are still disabled. Total life expectancy has increased by one year in this example, all of which is associated with disability. Further, the share of life that is disabled has increased. Thus, there is an expansion of disability by either measure. Note that in this example, the person is still better off; it is just that the disabled part of life has increased. The final column shows a combination of disability reductions (the person does not suffer the coronary event) and mortality reductions (the person survives an additional year with lung disease). Life expectancy has increased by 1 year, relative to the baseline. The increase is entirely in disability-free life; disabled life starts one year later but ends one year later. In this scenario, whether morbidity has been compressed depends on the definition employed: disabled life expectancy has not declined, but a greater share of life is spent in the non-disabled state. In general, the impact of combined morbidity and mortality changes on disability-free and disabled life expectancy depends on how rapid each change is and when in the course of life it occurs. All of this we need to evaluate empirically. 5 II. Medicare Current Beneficiary Data Our primary data source is the Medicare Current Beneficiary Survey (MCBS). The MCBS, sponsored by the Centers for Medicare and Medicaid Services (CMS), is a nationally representative survey of aged, disabled, and institutionalized Medicare beneficiaries that oversamples the very old (aged 85 or older) and disabled Medicare beneficiaries. Since we are interested in health among the elderly, we restrict our sample to the population aged 65 and older. A number of surveys have measures of disability in the elderly population (Freedman et al., 2004), including the National Health Interview Study and the Health and Retirement Study. Still, the MCBS has a number of advantages relative to these other surveys. First, the sample size is large, about 10,000 to 18,000 people annually. In addition, the MCBS samples people regardless of whether they live in a household or a long-term care facility, or switch between the two during the course of the survey period. Third, the set of health questions is very broad, encompassing health in many domains. Fourth, and most importantly, individuals in the MCBS have been matched to death records. As a result, we can measure death for over 200,000 people, even after they have left the survey window. Death data are available through 2008. The MCBS started as a longitudinal survey in 1991. In 1992 and 1993, the only supplemental individuals added were to replace people lost to attrition and to account for newly enrolled beneficiaries. Beginning in 1994, the MCBS began a transition to a rotating panel design, with a four year sample inclusion. About one-third of the sample was rotated out in 1994, and new members were included in the sample. The remainder of the original sample was rotated out in subsequent years. We use all interviews that are available for each person from the start of the survey in 1991 through 2009. We ignore the panel structure of the MCBS interviews 6 and treat each survey year as a repeated cross-section that has been linked to mortality information. The MCBS has two samples: a set of people who were enrolled for the entire year (the Access to Care sample) and a set of ever-enrolled beneficiaries (the Cost and Use sample). The latter differs from the former in including people who die during the year and new additions to the Medicare population. The primary data that we use are from the health status questionnaire administered in the fall survey, which defines the Access to Care sample. We thus use the Access to Care data. We compute time until death from the exact date at which the Access to Care survey was administered to the person. The MCBS population becomes older and less white over time, as the elderly population changes demographically. We do not want to show trends that are influenced by these demographic changes. We thus adjust survey weights so that the MCBS population in each year matches the population in the year 2000 by age, gender, and race. All of our tabulations are weighted by these adjusted weights. Recall that our death dates are available through 2008. For each individual interviewed in 1991-2007, therefore, we can determine if they died in the next 12 months or survived that period. Similarly, we can categorize individuals through 2006 as dying between 12-24 months or not, and individuals through 2005 as dying between 24 and 36 months or not. Death at 36 months or beyond is also known for the population through 2005. Trends in the distribution of time until death are shown in Figure 1. The share of the population that is within one year of death is about 5 percent on average. Reflecting the overall reduction in mortality, this share is declining over time (this will be true of the population 1-2 years from death and 2-3 years from death as well). Between 1991 and 2007, the decline is 1 7 percentage point, or 18 percent. Correspondingly, the share of the population that is 3 or more years from death increased by about 3 percentage points, also shown in figure 1. The MCBS asks extensive health questions. The first set of health questions are about medical events the person has experienced. These include cardiovascular conditions (heart disease, stroke), diseases of the central nervous system (Alzheimer's disease, Parkinson's disease), musculoskeletal problems (arthritis, broken hip), pulmonary disease, and cancer. For purposes of disability assessment, we divide these diseases into four groups, based on their likely association with death and disability (Lunney et al., 2003). The first disease is cancer. Once past the acute phase of cancer treatment, people with cancer tend to have reasonably high quality of life until the last few months of life, when health deteriorates markedly. The second group is permanently disabling conditions that get progressively worse. Alzheimer's disease, Parkinson's disease, and pulmonary disease fall into this category.2 The third group is acute conditions for which recovery is possible but not assured. This includes heart disease, strokes, and hip fractures. Finally, we group diabetes and arthritis as common disabling but generally non-fatal conditions. Table 2 shows the prevalence of these conditions across all years of the survey, the annual percentage point change in the prevalence over time, and the disability rate conditional on having the disease (defined as whether the person reports an ADL or IADL limitation; see below). Non-fatal conditions are the most common. Over half of the elderly population reports a prior diagnosis of arthritis, the prevalence of which is increasing by 0.3 percentage points annually. Nearly one in five elderly people has diabetes. Acute conditions for which recovery is possible are next most common, ranging in prevalence from 4 percent of the population (hip fracture) to 26 percent (ischemic heart disease). Perhaps owing to better prevention, the 2 Congestive heart failure is natural to add to this list but is only asked about from 2003 on. 8 prevalence of both heart disease and heart attacks is declining over time. About 18 percent of the elderly population has a history of cancer, which is increasing over time. Degenerative diseases are relatively less common, though pulmonary disease affects about one-seventh of the elderly population. People with these conditions are extremely likely to report having an ADL or IADL impairment. The MCBS also asks a number of questions about the impact of morbidity on a respondent's ability to function and perform basic tasks, shown in Table 3. The first category of questions is about physical functioning, such as difficulty walking a reasonable distance (1/4 mile or 2-3 blocks) or carrying moderate-weight objects. Difficulty in these areas ranges from one-quarter to three-quarters of the elderly population. The second and third categories are impairments in Activities of Daily Living (ADLs, such as bathing or dressing) and Instrumental Activities of Daily Living (IADLs, such as doing light housework or managing money). Six questions are asked about each of ADL and IADL limitations. Because limitations in these areas reflects more severe impairment, the share of the elderly population reporting difficulty in these areas is lower than the share reporting difficulty with functional limitations. The final category is sensory impairments, including trouble seeing and hearing. In the case of vision, the difficulty refers to even with correction such as glasses or contact lenses, and for hearing it is with hearing aid. The possible responses to the vision and hearing questions changed in 2002. Prior to 2002, the responses for each question were: no trouble, a little trouble, and a lot of trouble. Starting in 2002, a more severe category was added to each: no usable vision and Deaf. After this change, more people reported less severe vision and hearing impairments - most likely, they judged themselves less severely disabled relative to the more 9 severe categories now being offered as a response. The share of people reporting difficulty with vision and hearing each fell by 4 percentage points in 2002, far larger than in any other year. To adjust for this, we create a counterfactual time series for difficulty with vision and hearing assuming that the trend in each variable in the year the survey changed was the same as the trend in the prior three years. We then extend this aggregate estimate back to 1991. At the individual level, we randomly choose individuals who reported that they had a little trouble seeing or hearing and recategorize their responses to having no trouble, to match the adjusted aggregate totals. With these adjustments, about one-third of the elderly population reports vision and hearing impairments on average. The health status questions are generally the same for the community population and the institutional population, with the exception that the institutionalized are not asked about three IADLs limitations - light housework, preparing meals, and heavy lifting. On average, 5 percent of people are in a nursing home. In order to utilize these questions, we assume that everyone in a nursing home has difficulty with these activities.3 Summary Health Status Measures The most common single measure of disability in the literature is any difficulty with ADL or IADLs. We follow this in our analysis and define "disability" as an ADL or IADL impairment. While simple to implement, this measure lacks a rigorous theoretical foundation. Moreover, a binary measure does not capture heterogeneity in the population. For many purposes, we care about finer gradations in the distribution of health. There is a literature (e.g. 3 With regard to the other IADLs, 61 percent of people living in institutions report difficulty using the telephone and 85 percent report difficulty shopping for personal items and managing money. Over 90 percent report difficulty with basic activities such as stooping, crouching or kneeling, or carrying a 10 lb. object (Cutler and Landrum, 2011). 10 Verbrugge and Jette, 1994) arguing for a distinction between functional status (measures of specific physical functioning) and disability (the ability to engage in the activities typically expected of a person). Within this latter spirit, we examine the different dimensions of health among the elderly. In particular, we estimate a factor analytic model of the different domains of functioning and choose the number of domains that best summarize the data. Formally, denote yij as the response to question j for individual i. Suppose there are J questions total (J=19 in our setting). We imagine that these health states are a linear function of K different unobserved factors, denoted Fik. We fit a factor analytic model of the form (e.g., Bartholomew, 1987; and Knol and Berger, 1991): yij = ?0j + ?1jFi1 + ?2jFi2 + ?3jFi3 + ... + ?KjFiK, (1) where yij is a 0 or 1 outcome variable, ?0j is a threshold parameter that accounts for varying prevalence of limitations in the population (for example, limitations climbing stairs are more common that limitations in bathing) and the ?kj's are factor loadings that describe the relationship between unobserved factor k and question j. Unobserved factors are assumed to follow a multivariate normal distribution. The latent variable model described by (1) is similar to the factor analyses and Grade of Membership models that have been previously used to describe dimensions of disability (Lamb, 1996; Manton et al., 1994, 1998; Woodbury et al., 1978). We can fit this model provided K= High School 0.018 0.080*** 0.166*** 0.313*** 0.093*** 0.116*** 0.176*** 0.278*** 0.381*** 0.0421*** -0.055*** -0.077*** (0.010) (0.011) (0.013) (0.015) (0.010) (0.010) (0.011) (0.012) (0.012) (0.008) (0.006) (0.005) 0.002 0.048*** 0.157*** 0.283*** 0.073*** 0.094*** 0.150*** 0.247*** 0.375*** 0.016* -0.038*** -0.091*** (0.009) (0.010) (0.012) (0.014) (0.009) (0.009) (0.009) (0.010) (0.011) (0.007) (0.005) (0.005) Conditions Alzheimer's Parkinson's Broken Hip Stroke Pulmonary IHD Diabetes Arthritis Cancer 0.246*** 0.212*** 0.149*** 0.187*** 0.160*** 0.139*** 0.134*** 0.152*** 0.063*** (0.012) (0.019) (0.011) (0.008) (0.007) (0.006) (0.007) (0.005) (0.007) 0.323*** 0.253*** 0.188*** 0.156*** 0.163*** 0.068*** 0.130*** 0.139*** 0.044*** (0.010) (0.018) (0.012) (0.007) (0.006) (0.006) (0.006) (0.005) (0.006) Constant 0.205*** (0.010) 0.187*** (0.009) N 31,374 38,880 2 R 0.250 0.218 Note: The table shows regressions for reporting an ADL or IADL impairment in either 1991-93 (the first columns) or 2004-06 (the second columns). Table 6: Impact of Demographics and Medical Conditions on Health Health change Measure of Health (percentage points) Disability F1 F2 F3 Total change -7.4% -.138 -.091 -.201 Effect of changes in X's Demographics Condition prevalence -1.4% 0.5% -.025 .008 -.034 .014 -.026 .010 Effect of changes in ?'s Conditions -2.9% -.185 -.083 -.063 Demographics -2.1% -.084 -.074 -.039 Constant -1.8% .148 .090 -.093 Note: The table is a decomposition of changes in the measure of health indicated in the columns. For each health measure, we estimate equations of the form: Hit = Xit?t + ?it, for two time periods: 1991-93 and 2004-06. The first row, total change, shows the percentage point change in Hit over time. The remaining rows show the predicted percentage point change in Hit resulting from changes in the X variables, decomposed into demographics and condition prevalence, and changes in the ?'s, decomposed into those for conditions, those for demographics, and the constant term. Table 7: Decomposition of Disability Over Time, By Time Until Death Disability Decomposition of total Measure change in group change in disability Total change ---6.3% Effect of survival --- -0.7% Change within time periods <=12 months -0.4% -0.0% 13-24 months -3.7% -0.2% 25-48 months -10.0% -0.8% 49-72 months -9.4% -0.6% 73-96 months -12.3% -0.7% >96 months -15.9% -3.3% Note: The first column shows the percent change in disability rate for people in each category of time until death. The change is taken from 1991-93 to the latest 3 years available. The second column decomposes the total change in disability. The first row, total change, shows the percentage point change in disability over time from 1991-93 to 1998-00. The second row shows the change in disability resulting from changes in the share of people with different periods of time until death. The remaining shows show the change in disability resulting from changes in the disability rate in each time-until-death category. Group All Table 8: Changes in Disabled and Disability-Free Life Expectancy at age 65 1991-93 2003-05 Change Disability Disability Disability Total Free Disabled Total Free Disabled Total Free 17.5 8.8 8.7 18.2 10.4 7.8 0.7 1.6 Men Women 15.5 19.2 9.2 8.4 6.2 10.8 16.7 19.4 10.9 10.0 5.8 9.4 1.3 0.2 1.7 1.6 Disabled -0.9 -0.4 -1.4 White 17.6 9.0 8.6 18.3 10.6 7.7 0.7 1.6 -0.9 Non-white 15.8 7.0 8.9 16.7 8.8 7.9 0.9 1.8 -1.0 Note: The table shows total life expectancy, disability-free life expectancy, and disabled life expectancy, in years. Disability is an indicator for the presence of an ADL or IADL limitation. Figure 1: Population Distribution by Time Until Death Share within one year of death >36 months 8% 85% 7% 84% 6% 83% 5% <12 months 82% Share more than three years from death 86% 9% 81% 4% 1991 1993 1995 1997 1999 2001 2003 2005 2007 Note: Data are from the Medicare Current Beneficiary Survey, 1991-2009 and are weighted to the population distribution in 2000 by age, sex, and race, as are all subsequent figures. Figure 2: Factor Loadings hearing_trouble ia_tele 1.00 ia_light 0.80 vision_trouble ia_heavy 0.60 fl_walk ia_meal 0.40 0.20 0.00 fl_write ia_shop -0.20 -0.40 fl_reach ia_money fl_lift a_bath fl_stoop a_dress a_toilet a_eat a_walk Factor1 a_transfer Factor2 Factor3 Note: The figure shows the factor loadings for the first three factors of the health status questions. Data are from the Medicare Current Beneficiary Survey, 1991-2009. Figure 3: Trends in Disease Prevalence 70% 60% 50% 40% 30% 20% 10% 1991 1993 1995 1997 1999 Any major disease Cancer Recoverable acute events 2001 2003 2005 2007 2009 Chronic disabling Non-fatal Note: major diseases include cancer, chronic disabling conditions, and recoverable acute events. Specific conditions in the chronic disabling, recoverable acute event, and non-fatal condition categories are in table 2. Figure 4: Any Major Disease Prevalence by Time Until Death 100% 90% 80% 70% 60% 50% 40% 1991 1993 Total 1995 1997 <12 Months 1999 2001 12-24 Months 2003 2005 24-36 Months 2007 2009 >36 Months Note: Major diseases include cancer, chronic disabling conditions, and recoverable acute conditions. Figure 5: Any Minor Disease Prevalence by Time Until Death 80% 70% 60% 50% 1991 Total 1993 1995 <12 Months 1997 1999 12-24 Months Note: Minor diseases include arthritis and diabetes. 2001 2003 2005 24-36 Months 2007 2009 >36 Months Figure 6: Trend in Functioning 70% 90% Any functional limitation 60% 80% Any ADL or IADL 50% 70% Any IADL 40% 60% Any ADL 30% 50% 20% 40% 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Note: Specific questions used in functional limitations and ADL/IADL limitations are shown in Table 2. Figure 7: ADL/IADL Disability by Time Until Death 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 1991 1993 1995 1997 1999 2001 2003 all <12 months 13-24 months 49-72 months 73-96 months 2005 2007 2009 25-48 months > 96 months Note: Specific ADL and IADL questions are defined in Table 2. Figure 8: Percent Change in Disability by Time Until Death <12 months 13-24 months 25-48 months 49-72 months 73-96 months > 96 months 0% -5% -10% -15% -20% -25% Note: The data are based on Figure 7 and represent changes from 1991 through 2000. Figure 9: Percent Change in Disability by Time Until Death (a) Men (b) Women <12 13-24 25-48 49-72 73-96 > 96 months months months months months months <12 13-24 25-48 49-72 73-96 > 96 months months months months months months 0% 0% -5% -5% -10% -10% -15% -15% -20% -20% -25% -25% -30% -30% (c) White (d) Non-White <12 13-24 25-48 49-72 73-96 > 96 months months months months months months <12 13-24 25-48 49-72 73-96 > 96 months months months months months months 0% -5% -5% -10% -10% -15% -15% -20% -20% -25% -25% -30% 0% -30% Figure 9 (continued) (e) High School Degree or Less (f) Some College or More <12 13-24 25-48 49-72 73-96 > 96 months months months months months months <12 13-24 25-48 49-72 73-96 > 96 months months months months months months 0% 0% -5% -5% -10% -10% -15% -15% -20% -20% -25% -25% -30% -30% Note: The data are for 1991-2000 and are based on data like those in figure 7. Figure 10: Functional Limitations by Time Until Death 100% 90% 80% 70% 1991 1993 Total 1995 1997 <12 Months 1999 2001 12-24 Months 2003 2005 24-36 Months 2007 2009 >36 Months Note: Functional limitations are defined in Table 2. Figure 11: Trend in Factor Scores 0.2 0.1 0.0 -0.1 -0.2 1991 1993 1995 1997 F1 (ADL/IADL) 1999 2001 F2 (Fnl Limit) 2003 2005 F3 (Sensory) Note: F1, F2, and F3 are based on the factor analysis displayed in Table 3. 2007 2009 Figure 12: Trend in Factor Scores by Time until Death (a) F1 (ADL/IADL impairments) 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 1991 1993 Total 1995 1997 <12 Months 1999 2001 12-24 Months 2003 2005 24-36 Months 2007 2009 >36 Months (b) F2 (Functional Limitations) 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 1991 1993 Total 1995 <12 Months 1997 1999 2001 12-24 Months 2003 2005 24-36 Months 2007 >36 Months 2009 (c) F3 (Sensory) 0.6 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 1991 1993 Total 1995 <12 Months 1997 1999 2001 12-24 Months 2003 2005 24-36 Months Note: F1, F2, and F3 are based on the factor analysis displayed in Table 3. 2007 >36 Months 2009 Figure 13: Life Expectancy at Age 65 21 0.4 years 20 19 18 1.5 years 17 16 15 1991 1993 1995 1997 1999 Women 2001 2003 Men Note: Data are from the National Center for Health Statistics. 2005 2007 2009 Figure 14: Trend in Disabled and Disability-Free Life Expectancy 20 18 16 14 7.8 8.7 12 10 8 6 4 8.8 10.4 2 0 1991-93 Disability-free Lifeyears 2003-05 Disabled Lifeyears Note: The figure combines life expectancy data from the NCHS with imputed disability rates by age and time until death. Figure 15: Trend in Disabled and Disability-Free Life Expectancy at 65, by Gender and Race 20 18 16 14 12 5.8 10.8 9.4 7.7 8.6 7.9 6.2 8.9 10 8 6 4 9.2 10.9 8.4 10.0 10.6 9.0 7.0 8.8 2 0 1991-93 2003-05 1991-93 2003-05 Men Women Disability-free Lifeyears 1991-93 2003-05 1991-93 2003-05 White Non-white Disabled Lifeyears Note: The figure combines life expectancy data from the NCHS with imputed disability rates by age and time until death. Figure 16: Trend in Disease-Free Life Expectancy and Life With Disease 20 18 16 14 12 9.7 9.5 10 8 6 8.0 8.6 1991-93 4 2003-05 2 0 Disease-free Lifeyears Years with Disease Note: The figure combines life expectancy data from the NCHS with imputed disease rates by age and time until death.