Click for next page ( 95


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 94
4 Health Transitions and the Compression of Morbidity Public policy debates in the areas of health and medical care have emerged in recent years around the question of whether improve- ments in lifestyle and medical technology will delay the onset of chronic illness and disability and result in a compression of mor- bidity at older ages. If the incidence of chronic illness declines and morbidity is compressed to a very brief period at the end of the life span, there will be a growing number of healthy people at the older ages. One set of policy issues emerges from this scenario: What is the appropriate age for retirement? Will a relatively smaller working population be able to pay Social Security payroll taxes for the growing number and proportion of the retired elderly? Can productive employment be found for the growing number of healthy older persons in the face of relatively high and persistent unemployment? What is the viability of tax credits, reverse mortgages, inde- pendent retirement accounts (IRAs), and other saving plans in assisting the future elderly to pay for a greater share of their future health and long-term care costs? Will income maintenance and social welfare programs be ad- equate to meet the needs of the elderly with longer life ex- pectancy? Under another scenario relating to the compression of morbidity, improvements in survival may postpone chronic illness and disability to older ages, raising a different set of policy questions: 94

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OF MORBIDIlrY 95 . Will medical and social services be available to maintain the independence at home of the growing number of chronically ill and functionally disabled elderly people, avoiding ~nstitu- tionalization? Can we develop various types of supportive living arrange- ments for the growing number of women age 75 and older with no spouse who live alone? Will public and private income support and medical care pro- grams be adequate to meet the needs of the growing number of disabled persons? What type of rehabilitative programs should be designed and made available for physically impaired elderly populations? In assessing health transitions in old and very old populations, a number of conceptual, measurement, and analytic issues emerge that are not as evident in younger populations. These issues emerge for a number of reasons. First, and perhaps most important, is the different nature of disease, disease interactions, and physiology in older persons. These concerns focus on the nature of intraindividual changes in health at later ages. There are numerous age-related changes in the physiolog- ical state of the individual, for example, changes in metabolism, im- mune response, and organ function, a greater likelihood of accumulat- ing significant adverse environmental exposures, and the cumulative effects of multiple chronic conditions and diseases. Intraindividual changes in health at later ages involve these age-related changes and the effect of such changes on the manifestations of specific disease processes. Intrinsic to these concerns is the fact that the chronic diseases and conditions tend to manifest a different time scale than acute disease processes more typical at younger ages. In particular, the period of time over which chronic diseases become manifest in the elderly tends to become significant with respect to the total life span and to the amount of life expectancy at later ages. This greater period of time means that, even if the incidence rates of chronic disease were unchanged with age, there would be a greater probability that older individuals would accumulate multiple diseases, and the time over which the older person would have these multiple (and interacting) diseases simultaneously would be longer. Even if such chronic diseases become manifest at younger ages, the physiological reserves to preserve homeostasis are generally stronger at those ages, and the duration of disease expression may be shorter with disease

OCR for page 94
96 AGING POPULATION IN THE TWENTY-FIRST CENTURY control (if not cure) more completely and rapidly imposed by the physiology of younger persons. Alternatively, if a chronic disease is atypically manifest at younger ages, it may be due to particularly powerful genetic or other susceptibilities to these disease processes, in which case the natural course of the disease may be dramatically shortened by loss of homeostatic control and death. Thus, health transitions at later ages cannot be described sim- ply in terms of the incidence rate of independent disease events. Instead they must be viewed in terms of multiple, interacting disease processes that operate in a host with generally declining homeo- static reserve. It is clear that these concepts involve modeling health changes in the elderly person as changes in a complex multidimen- sional system evolving over a considerable period of time. To do so requires addressing formidable measurement and ana- lytic problems that are not well handled by our current data collection and analytic strategies. The practical need for such data and analytic strategies is already manifest by the underlying concepts in the de- velopment of clinical specialties in geriatric medicine (e.g., Minaker and Rowe, 1985; Besdine, 1984) and in concerns with the special problems in clinically managing and treating specific diseases (e.g., lung cancer, breast cancer, isolated systolic hypertension, diabetes) in older persons. It should also be noted that there are many examples of such complex system-based models in the physical sciences that could be the point of departure for developing appropriate health transition models for the elderly. The development of such models, however, requires an approach or philosophy of inference that differs from the approach typically employed in experimental or clinical trial designs; that is, we wish to develop an approximate mathematical description of a complex physical system rather than to test whether an observed effect would have reasonably occurred simply by chance. In addition to these differences in health states and their changes in elderly individuals, other issues emerge in modeling the health changes in the populations of the elderly. These arise because people who are alive at later ages represent only a portion of their cohorts, since significant mortality will have occurred in these cohorts" mortality that will systematically select out persons according to the health characteristics we wish to assess. Issues in measurement also occur because, though disease prevalence may increase, popu- lation prevalence (i.e., the number of persons available for study in

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OFMORBIDIlrY 97 any well-defined population) at very advanced ages is small. Fur- thermore, the description of the disease process becomes complex and the rate of changes in physiological state may accelerate when homeostatic mechanisms become weaker at advanced ages or due to the accumulation of multiple disease processes. This complexity of disease state may make even the development of crude indicators of health state at advanced ages more difficult (e.g., the need for multiple-cause mortality data rather than underlying-cause mortal- ity data to describe mortality patterns among the oldest-old). From the above discussion, it is clear that new concepts and models that are based on process and duration rather than on event and incidence need to be developed to assess health changes at later ages. Indeed, the many recent debates that have emerged in the biomedical and health policy literature about the direction of recent health changes at later ages may be due simply to the lack of common concepts and models. It is clear, however, that the complexity of the physical system being described makes it difficult to have a pars~rnonious and relatively simple conceptual model to help inform the debate and improve concepts and communication. In order to remedy this problem, a World Health Organization Scientific Advisory Group (World Health Organization, 1984) de- veloped a model of health transitions at later ages. This model is based on simple life-table concepts in order to develop a theoretical framework that would be readily understood at the policy level. The basic model is actually quite simple, as shown in Figure 4.1. Figure 4.1 contains a series of three survival curves labeled mor- tality, disability, and morbidity. The outermost curve (mortality) represents the effects of mortality on overall survival. We see that the proportion surviving from a cohort to each successive age de- clines (i.e. as one progresses to the right on the horizontal axis). In addition to the overall survival curve, there are survival curves for chronic morbidity and chronic disability. The lowest survival curve (morbidity) represents the probability of surviving to a given age free of chronic morbidity. The area labeled A under this curve represents the number of person-years that a person in this cohort could expect to survive free of disease (i.e., healthy life expectancy). The middle curve, labeled disability, describes the probability of surviving to a given age free of serious chronic disability. The formulation assumes that disability emerges as a result of the pro- gression of some underlying morbid process. This curve identifies two additional areas in the figure.

OCR for page 94
98 ._ .> AGING POPULATION IN THE TWENTY-FIRST CENTURY 80 60 50 20 10 0 10 DISABILITY MORTALITY __ - . As_ MORBIDITY ~ . ~ A" 1-- -' . 1 1 20 30 40 50 60 . ~ - \ . ~ \~\ ~'\~\\ %~-,\x `- ~ 80 90 100 110 `. ~ 'my_ Age 70 eO ~ ~ e60 50 eo and e60 are the number of years of autonomous life expected at birth and at age 60, respectively. Mso is the age to which 50 percent of females could expect to survive without loss of autonomy. FIGURE 4.1 The observed mortality and hypothetical morbidity and disabil- ity survival curvier for females in the United States of America in 1980. Source: World Health Organization (1984~. Reprinted by permission. The first, the area between the morbidity and disability survival curves labeled B. defines the number of person-years that can be expected to be lived by an individual from the cohort in a morbid, but disability-free, state. This portion of the figure has implications for acute health services, but relatively little direct impact on Tong- term care services. Many of the diseases that are fatal in middle age may produce a significant contribution to this portion of the figure- but relatively little to the disabled person-years to be experienced by the cohort. Examples of this type of disease process are cancer and myocardial infarctions. The second area defined by the disability curve lies between it and the mortality curve and is labeled C. This area represents the person-years that can be expected to be lived by a person from this cohort in a chronically morbid and disabled state. Recall that the

OCR for page 94
HEALTH IrRANSITIONS AND THE COMPRESSION OF MORBID 99 / total area under the disability curve (areas A and B) represents the number of person-years that can be expected to be lived in a disability-free state. This total area may be viewed therefore as a measure of active or autonomous life expectancy (e.g., Koizumi, 1982; Katz et al., 1983; Wilkins and Adams, 1983~. Certain dis- ease processes will have a magnum effect on active life expectancy; for example, onset of a serious stroke or Alzheimer's disease may immediately produce serious disability. Such diseases have serious consequences for long-term care as well as acute care services. With the three survival curves, it is possible to more precisely distinguish between the positions of different analysts about changes in health at advanced ages. For example, a major area of debate has centered on whether the health status of elderly persons has improved at each age as life expectancy at later ages increased. This debate was central to discussions about whether the entitlement age for social security could reasonably be increased from 65 to 67 to adjust for a two-year increase in life expectancy for males at age 65. Some analysts argued (e.g., Feldman, 1983) that life expectancy increased because chronically ill and morbid persons were surviving for longer periods of time. This could be represented in Figure 4.1 by the mortality survival curve moving to the right but the morbidity and disability curves remaining unchanged. This would increase the absolute numbers of years expected to be spent by a population in morbid and disabled states as well as the ratio of dependent morbid person-years of life to the total number of person-years expected to be lived. The majority opinion was that as the mortality curve moved to the right, both the morbidity and disability curves moved to the right in a reasonably "parallel" fashion. Thus, the probability of surviving to age 67 free of disease and disability was the same after the recent life expectancy increases at later ages as it had been at age 65 before the increase. This suggests that the ratio of disabled and morbid person-years to be lived to the total would decline, raising the questions of how this improvement could be turned to society's benefit. These discussions relate to a topic currently of considerable in- terest: the compression of morbidity. This concept suggests that different public health policies will have different effects on how rapidly each of the three curves moves to the right. For example, focusing exclusively on treating diseases after they emerge may cause life expectancy to increase but allow the age at onset of disease and

OCR for page 94
100 AGING POPULATION IN THE TWENTY-FIRST CENTURY disability to remain the same, with a net increase in the number of years lived in a morbid and disabled state. Alternatively, certain prevention and health promotion strategies may delay the age at onset of disease more than life expectancy Is increased. In this case, the period spent in a morbid state and the acute health resources required may be reduced. Interestingly, recent data suggest that the diseases responsible for most of the disability and dependence at advanced ages (e.g., arthritis, Alzheimer's disease) are not those typically targeted by prevention efforts (e.g., heart disease and can- cer). Thus, without careful attention to the effects of prevention on disability as well as disease, the morbid period may be decreased (decreasing acute health care costs) but the disabled period might be relatively untouched thus having little effect on aggregate needs for long-term care (Manton, 1986b). Alternatively, the rate of progress of the disease may be retarded so that the age at onset of disability may increase more than life expectancy. In this case, the disabled period is compressed or reduced. A specific example of the ejects of such compression was dis- cussed by Brody (1987~. He points out that the risk of hip fracture increases roughly exponentially with age. As a consequence, if the age at onset curve of hip fracture could be shifted five years to the right (i.e., the process retarded five years, perhaps by nutritional supplementation or hormonal therapy), the prevalence of hip frac- ture in the population would be cut in half, provided there were no corresponding increases in life expectancy. One extreme form of the compression of morbidity argument (Fries, 1980, 1983) suggests that the United States and other devel- oped countries currently have life expectancies near to their biological limits. As a consequence, the survival curve for mortality will have little potential to move much to the right with a corresponding rapid increase in the prevalence of death due to biological senescence, i.e., natural death (Fries, 1983~. Since natural death is argued to be due to senescent processes, it is also argued that it will be independent of the chronic disease processes that produce disability. Thus, while the mortality curve is fixed, the age at onset of disease and disability can be delayed by appropriate, independent action. This will cause the area between the survival curve for disease and disability and that for mortality to compress until, at the extreme, the age at onset of disease and disability is pushed beyond the biologically determined age at death, with the result that people die of senescence free of disease or disability.

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OF MORBIDIlrY 101 One problem with the formulation of the compression of mor- bidity mode! presented in the preceding paragraph is that it is hard to define "natural death." Indeed, Fries (1983) has suggested that the dividing line between natural and premature death may be an arbitrary one. Furthermore, the available evidence on the period of terminal decline leading to death suggests that it has not changed as life expectancy at later ages increased, nor is it highly age-variable (e.g., Riley et al., 1985; Roos et al., 1986~. Eries's optimistic argument differs considerably from those of many other investigators. Clearly it is different from the arguments of Feldman (1983) about what we have experienced. Such arguments do not seem to take into account that to achieve the goals set for the adoption of health promotion and disease prevention measures, major efforts are required in some population subgroups, in particular, lower socioeconomic status groups and in some minority groups such as blacks. Nonetheless, it may have utility in defining disability or morbidity prevention goals for selected population groups. Apart from practical concerns about the pure form of the com- pression of morbidity model, there are those who propose very dif- ferent mechanisms for health changes at later ages. For example, Kramer (1981) and Gruenberg (1977) suggest that we are currently increasing the number of years to be spent in morbid and disabled states. This is because they view the bulk of biomedical research as being directed toward reducing mortality and not toward decreasing morbidity and disability. Thus, they suggest that we have effective strategies for moving the mortality curve to the right but not for moving the morbidity and disability curves. StrehIer (1975), in contrast, suggests that we should intervene in the basic aging process. He suggests that within 25 years we should be able to increase the life span by 25 percent. He also suggests that, since we wiB intervene in basic aging processes, the number of years spent in morbid and disabled states will not increase. Thus, ah three curves will move to the right an equal amount so that only the number of person-years spent free of disease (area A) will increase, causing a massive decrease in the ratio of impaired to healthy years. Each of the above arguments is presented by its respective author as a conceptual model to illustrate a particular perception of the age-changing relation of morbidity, disability, and mortality. For the purposes of actually forecasting likely changes in health status and service needs, we need a more complete and feasible model to

OCR for page 94
102 AGING POPULA1TON IN THE TWENTY-FIRST CENTURY accommodate what is being learned from recent epidemiological and biomedical research on aging-related health changes. The development of the more general mode! could begin by rec- ognizing that the human organism is subject to multiple pathological processes at later ages. Furthermore, it should be recognized that different diseases have different associated periods of morbidity and disability. For example, cancer has a relatively long morbid period but a short disabled period. Myocardial infarction has short mor- bid and disabled periods (assuming effective interventions for these diseases exist). Alzheimer's disease has long morbid and disabled periods. Thus, the total morbid and disabled periods for the popula- tion can be compressed by targeting interventions to individuals at high risk of the disease processes with the longest disabled periods (assuming effective interventions for these diseases exist). This will change the mix of disease processes operating in the population and may reduce the disabled period more rapidly than life expectancy is increased. This is not to argue that efforts should not continue against the major fatal conditions but instead suggests that inad- equate attention has been paid to the potential prevention of the disease processes generating the most disability (e.g., Alzheimer's disease, arthritis, depression). Such a strategy of accelerating com- pression seems more plausible since it recognizes the multiplicity of chronic conditions affecting the elderly, does not require the concept of natural death, relates interventions to well-defined disease enti- ties, and allows interventions in the disease processes to have effects on life expectancy (i.e., does not require that disease and mortality processes operate independently). It also is arguably a much better representation of our best current scientific evidence. For example, we have little evidence that natural death exists. Autopsy studies of the extreme elderly do report increasing prevalence of multiple chronic diseases at later ages, although it is admittedly more diffi- cult to identify a single cause of death. Nonetheless, interventions in particular disease mechanisms may help to incrementally improve overall system performances. Verification of results from this type of mortality data also comes from longitudinal studies of aging populations. For example, I.akatta (1985) has studied cardiovascular functioning in the extremely elderly (i.e., persons in their eighties) and found that, in persons free of disease, cardiovascular function could be preserved to advanced ages. Similar evidence has emerged for other organ systems from studies in which individuals with manifest pathological changes have been

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OF MORBIDITY 103 removed from the analysis. For example, Katzman (1985) discusses the fact that cognitive functions appear to decline very slowly in longitudinally followed populations, especially in persons not close to death. He contrasts these results with those of cross-sectional studies, which show more rapid declines that appear to be artifactual. Such results are consistent with studies of cause-specific mortal- ity in the United States, which find that the same general mix of diseases is causing mortality at ages that are 2 to 4 years higher on average than they were 15 to 20 years ago. It is also interesting that studies of the variance of the age at death at advanced ages do not show the variance decline with life expectancy increase that would be expected if the upper tad! of the age at death distribution were truly being truncated by an absolute biological limit to life span at the current leveb of life expectancy. Although a general mode! is needed, the mode! presented in Fig- ure 4.1 is a simplified conceptualframework for studying age changes in the relation of morbidity, disability, and mortality in a population. It has merit because it helps us to understand the changing relation of those health events in a simple framework. Actual analyses will be more complex dealing with the multiplicity of chronic diseases at advanced ages, the possibility of reversals in chronic morbidity and disability, dependent competing risks, the dynamics of human aging processes and risk factors, systematic mortality selection, and the fact that there is a many-to-one mapping between chronic diseases and the type and level of disability. In addition to analytical issues, the mode] also raises questions about the nature of data that are required to evaluate the transition. In its simplest form, the mode} can be evaluated using population life tables, surveys of disability, and epidemiological studies of disease in- cidence. For example, Wilkins and Adams (1983) used the essence of the concept in assessing active life expectancy gains in the Canadian population. They found that of a Year gain in total life expectancy between 1951 and 1978, 4.7 years was in an activity-limited state. Recent life insurance studies suggest that the level of activity lirni- tation was quite low (Blanchet, 1986~. More detailed models would involve specific disease processes and their differential contributions to the years lived in each state. Thus, the general concept of the models can help guide us both in terms of needed data and analytic strategies. #

OCR for page 94
104 AGING POPULATION IN THE TWENTY-FIRST CENTURY DATA NEEDS AND RECOMMENDATIONS The statistical methodologies and analyses reviewed in this chap- ter also serve the purpose of identifying critical data needs. In order to describe health transitions at advanced ages, certain specific types of new data are needed and the development of new methodologies for analyzing longitudinal data is required. One need is for long-term follow-up of well-characterized co- horts. In order to describe health transitions among the elderly it is necessary to have repeated measurements of detailed physiolog- ical, social, and health characteristics at advanced ages (i.e., over 753. Such follow-up data would help us to understand the dynamics of physiological aging processes. For example, we might be able to determine whether certain risk factors such as socioeconomic sta- tus and overweight cease to be so strongly associated with disease risks at advanced ages, e.g., such follow-up data are necessary to help us disentangle the effects of systematic mortality selection from basic physiological aging processes. A cost-effective approach for generating such measurements at advanced ages would be to make such measurements of existing longitudinal study populations. It is recognized that a full application of the concept advanced may not be feasible, but a limited set of items on physiological and psy- chosocial functioning could be incorporated in local and national studies. Existing surveys, especially local and national pane} stud- ies, can be enhanced by repeated measurement of physiological and psychosocial functioning, capitalizing on the probability sample and baseline measurements available. For example, specific investigation of system functioning, such as cardiovascular function, can benefit by adding studies to existing panel studies with probability samples. Population estimates of physiological parameters at baseline and un- der conditions of stress (such as a treadmill test) can be derived from such add-on studies to existing cohorts. These data can also be user] in longitudinal predictive analyses to enhance usual demographic and self-reported data. Recommendation 4.1: The pane! recommends that detailed physiological assessments be added to existing longitudinal study populations when these cohorts can serve as an appro- priate sample to measure specific biological and psychosocial functioning (e.g., cardiovascular status and capacity for daily activities). Consideration needs to be given to the addition of physiological ~ ~. - r ~ 1

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OF MORBIDITY 105 assessments to the Longitudinal Study on Aging and to community- based special surveys. Examples of community-based epidemiological studies for which follow-up might be done are the Evans County (Cornoni-HuntIey et al., 1986) and Alameda County (Berkman and BresIow, 1983) studies. Nationally representative survey populations that might be considered are those in the Longitudinal Retirement History Survey and the National Long-Term Care Study. A second need is to determine morbidity and mortality risks among nationally representative samples of already morbid and dis- abled persons. This would help us resolve some of the questions about the effect on society of life expectancy gains at later ages. The prevalence of many chronic diseases is too low to permit the analysis of morbidity and mortality risks for morbid subpopulations using general health surveys. Adequate samples of morbid subpopulations can be obtained by use of a two-stage design in which a large popula- tion is screened in the first stage to identify individuals with certain chronic diseases a target group followed by a second phase of in- tensive interviewing of the target group. For example, in 1982 and 1984 a survey was conducted of chronically disabled elderly persons. In that survey persons drawn from the National Health Insurance Master file were screened by telephone to determine if they had a chronic (more than 90 days) ADL or lADL disability. If the answer was yes, then an intensive household survey was done. Screening approximately 36,000 persons identified 6,393 disabled persons. This design provided a cost-effective way to define a disabled subsample that could be followed for future morbidity changes and mortality. Sample members were followed for two years to examine their risk of death, their risk of institutionalization, and the trajectory of change in dependency level. Such data are important in forecasting future health service needs" forecasts that are essential to projecting the costs and resource requirements for health care of the elderly. These projections are required in considering funding and organization of health care. Recommendation d.2: The pane! recommends that spe- cially focused studies and surveys of specific chronically mor- bid and disabled elderly populations be conducted. A third type of useful data could be obtained by systematically extending ciata collection for clinical studies to measure comorbidi- ties (i.e., multiple morbid conditions). Not only is the issue of co- morbidity important for describing normal aging changes, but it is

OCR for page 94
106 AGING POPULATION IN THE TWENTY-FIRST CENTURY also important for determining what are the appropriate clinical re- sponses to disease at advanced ages. Currently most clinical data bases do not collect extensive data on comorbid conditions. Such data are important for assessing age-related differences in response to different types of medical and surgical treatments. Recommendation 4.3: The pane] recommends that stan- dard guidelines be developed for collecting clinical data on comorbidities at advanced ages and that the National Center for Health Statistics and various institutes of the National Institutes of Health promote the collection of such data in regularly funded clinical study programs. A fourth requirement is for data that systematically relate chron- ic disease diagnoses to level and type of disability. This would allow us to generate the mapping functions of diseases to disabilities and to utilize the extensive epidemiological data on disease risk factor associations in forecasting future demand for Tong-term care services. Recommendation 4.4: The pane! recommends that stan- dard health surveys and epidem~ological studies include ques- tions that help clarify the relationship between different dis- eases and different types and levels of disability at different advanced ages. Specifically, questions should be included that identify the diseases that are the major causes of different components of functional loss and disability. A fifth requirement is for increased geographic disaggregation of data. Geographic disaggregation would allow us to take better advan- tage of naturally occurring population differences in risk. Different subpopulations have very different exposures and disease risks. Such disaggregation would require more detailed tabulation of currently existing data. It is therefore a relatively low-cost way of increasing the usefulness of data that have been collected. Recommendation 4.5: The pane} recommends that more studies and analyses be conducted of the geographic varia- tion of morbidity and mortality risks. In order to be able to conduct such studies, it will be necessary to examine the question of identifying specific morbidity and mortality

OCR for page 94
HEALTH TRANSITIONS AND THE COMPRESSION OF MORBIDITY 107 events at a level of aggregation detailed enough that the area pop- ulations studied are relatively homogeneous. County-level mortality data need to be disaggregated to smaller areas. Probably no single data source is adequate to study detailed health transitions at advanced ages. The interactions of chronic dis- eases ax, 8 aging changes are such complex phenomena that they can best be studied by using information from multiple data sources. This suggests the need for biologically motivated models of the ba- sic physiological processes that can be used to integrate parameter estimates developed from different data sources. Such a strategy for integrating and evaluating multiple data sources must be pur- sued both as a way of systematizing our knowledge about health dynamics and as a strategy for identifying critical data needs. Recommendation 4.6: The panel recommends that the Na- tional Institute of Aging and the National Science Founda- tion conduct or sponsor studies to develop statistical pro- cedures that can utilize multiple data sources for studies of health transitions.