This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation

Trends in Rates of Onset of and Recovery From Disability at Older Ages: 1982–1994

¹ Center for Policy Research, Syracuse University, Syracuse, New York.
² Rush Institute for Healthy Aging, Rush-Presbyterian–St. Luke's Medical Center, Chicago, Illinois.
³ Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland.

Address correspondence to Douglas A. Wolf, Center for Policy Research, 426 Eggers Hall, Syracuse University, Syracuse, NY 13244. E-Mail: dawolf{at}maxwell.syr.edu

	Abstract

TOP Abstract Methods Results Discussion References

 
Objectives. Although there is substantial evidence of declining prevalence of disability among the older population during the late 1980s and 1990s, evidence on trends in the underlying dynamics of disability is lacking. For this study, we estimated models of transitions between discrete disability and vital states that incorporated simple linear time trends.

Methods. We analyzed data from the 1982–1994 interviews of the New Haven Established Populations for Epidemiologic Studies of the Elderly study and used three alternative measures of disability status. We estimated separate models of disability prevalence and disability transitions by gender.

Results. Eleven of 12 estimated trends in transition rates were statistically significant. For men and women, and for three alternative disability indicators, we found downward trends in rates of both onset of and recovery from disability among people aged 75 and older. We did not find any consistent pattern of trends in disability among those aging into the 75 and older group during this period.

Discussion. Our findings are consistent with declining population-level disability prevalence only if any downward trend in onset outweighs the downward trend in recovery. These findings are also consistent with a trend toward more severe disability problems among the disabled population.

A SUBSTANTIAL body of evidence indicates that the prevalence of disability among the older population of the United States fell during the late 1980s and 1990s. Manton and Gu's (2001) analysis of data from the 1982, 1984, 1989, 1994, and 1999 National Long Term Care Survey (NLTCS), for example, showed declining prevalence throughout the period, with average annual rates of decline greater in the 1990s than in the 1980s. Spillman's (2004) reanalysis of NLTCS data also showed declining disability prevalence but pointed out that the declines occurred primarily for instrumental activities such as financial management and shopping. Analyses of other national survey data sources, such as the Medicare Current Beneficiary Survey (Waidmann & Liu, 2000) and the National Health Interview Survey (Schoeni, Freedman, & Wallace, 2001), have also demonstrated declining levels of disability at older ages, although for somewhat different periods, at various annual rates, and for different population subgroups. Prompted by the apparent inconsistencies across population-level data sources, a working group set out to apply common definitions and analytic approaches to five such data sources (Freedman et al., 2004). The latter study dealt exclusively with the more severe types of disabilities with activities of daily living (ADLs), such as bathing, dressing, feeding, toileting, and transfer. The group's findings included mixed evidence on prevalence trends during the 1980s but consistent evidence of a steady decline during the 1990s.

The data sources used in these studies provide a series of static measures of old-age disability. However, in order for those static measures to indicate changes over time, there must also be changes in the dynamic process of disablement that generates those static measures. Indeed, if other factors were held constant, then a decline in the prevalence of disability among the older population could result from any one of three underlying trends, operating in isolation: (a) falling rates of onset of disability, (b) rising rates of recovery from disability, or (c) falling disability prevalence among the new elders that are added to the older population each year. Each of these distinct factors would, by itself, imply falling disability prevalence rates, although even large changes in the third factor—the prevalence of disability among new elders—would most likely produce very small changes in the prevalence of disability among the general older population. Trends in these components of disability dynamics may, in turn, reflect compositional change in the older population along measured or unmeasured dimensions such as behavior and predisposition, as well as trends in access to services or practice patterns (Cutler, 2001; Wolf, 2001).

Other theoretically possible trends could also produce a downward change in disability prevalence—for example, upward trends in death rates among disabled persons, or, somewhat more plausibly, downward trends in death rates among nondisabled persons in combination with stable death rates among disabled individuals. There could, of course, be many possible combinations of trends in two or more of the identified components of disability dynamics. And more complicated patterns could underlie the observed trends in prevalence: For example, rates of disability onset could rise rather than fall—a change that, on its own, would cause the stock of disabled persons to grow—but recovery rates could also rise even more rapidly, offsetting the unfavorable trends in onset rates. Finally, any of several possible trends in rates of onset or recovery could combine so as to produce an apparent absence of trends in disability prevalence.

Despite the necessary link between trends in disability prevalence and the various incidence rates that constitute disability dynamics, there has been very little research on temporal patterns of change in the dynamics of disability. Moreover, the two major studies published so far have reached contradictory conclusions. Manton, Corder, and Stallard (1993) used data on disability transitions observed between the 1982 and 1984 and the 1984 and 1989 interviews of the NLTCS. They also defined seven categories of disability severity. After they adjusted the data to produce 2-year incidence rate estimates from the two observed transitions, their results indicated that nearly all of the onset rates (i.e., rates of movement to more disabled states) declined, and all of the recovery rates (i.e., rates of movement to less disabled states) also declined between the 1982–1984 and 1984–1989 periods. Crimmins, Saito, and Reynolds (1997) used Longitudinal Survey of Aging data, which permitted comparisons of 1984–1986, 1986–1988, and 1988–1990 disability transitions. Like Manton and colleagues, Crimmins and associates (1997) found falling rates of disability onset (1988–1990 compared to 1984–1986), but in contrast to the results of the former study, the latter study's results showed rising rates of recovery from disability over time.

The contradictory nature of past research on trends in disability dynamics suggests that further inquiry would be useful. This article investigates several trends in components of change in disability prevalence rates during the 1980s and early 1990s. We examined trends in rates of disability onset and recovery using an extension of the modeling approach introduced in Laditka and Wolf (1998), introducing a simple dependence of disability transition rates on calendar time. We also investigated trends in the disability status of persons joining the elderly population each year. Our data spanned a longer time period (12 years, compared to the 6–7 year periods of Crimmins et al., 1997, and Manton et al., 1993, respectively) and contained more frequent assessments (i.e., annual rather than at 2- or 5-year intervals), and therefore more numerous observed transitions, than the data used in past research. However, our data covered a period of time during which, according to the best available evidence, ADL disability was not declining nationally. Moreover, we used local-area rather than national data in our analysis. Thus, although we found consistent and strong evidence on trends in the transition rates, our ability to explain national-level trends in disability prevalence is limited.

	METHODS

TOP Abstract Methods Results Discussion References

 
Sample
We used data from the New Haven site of the Established Populations for Epidemiologic Studies of the Elderly (EPESE) project. The New Haven EPESE enrolled a stratified sample of 2,812 participants, representing a participation rate of 82%, drawn from the population of noninstitutionalized persons aged 65 and older in the city of New Haven (Mendes de Leon et al., 1999). Researchers conducted a baseline interview in 1982 and followed this with annual interviews through 1989 and with two additional interviews in 1991 and 1994. Participation exceeded 90% for all follow-up interviews. Attrition due to loss among survivors to follow-up was under 3%.

The EPESE design is for a cohort study (i.e., persons initially 65 and older in 1982 were 66 and older in 1983, 67 and older in 1984, and so on). However, for our analysis, in which we wished to estimate trends over the period 1982–1994 for a specified age group, we selected only that portion of each respondent's observed history that began at age 75. Thus, with the exception of 1994, our data came from a population aged 75 and older in each year, with the age-75 group replenished each year. In 1994, our study population was aged 77 and older, a lower limit imposed by the initial restriction to persons aged 65 and older in 1982.

Our analysis of transition rates used sequential pairs of disability indicators. Each respondent provided information on as many as nine transitions (i.e., 1982–1983, 1983–1984, ... , 1991–1994). If a respondent missed one or more interviews, we used sequential pairs of nonmissing values, however long the time interval between useable measures (indeed, in a few cases we used pairs of values separated by a full 12 years). A few observations—roughly 3% of the potential sample—were lost because a multiyear interval (i.e., one containing one or more missing values between two successive nonmissing values) began at an age less than 75 and ended at an age greater than 75. In such instances, the status at age 75 was unknown, requiring us to discard that potential observation. The number of transitions included in each of three sets of analyses, using three alternative disability definitions, varied due to the pattern of missing values found for each disability indicator.

The EPESE baseline sample excluded the institutional population, although follow-up interviews did identify survivors who had moved into an institutional setting. This distorts the trend in the prevalence of institutionalization in the sample, compared to what it would have been in a population sample. Therefore, we modeled disability transitions among the noninstitutionalized population. Interviewers asked respondents, at each follow-up interview, whether they had spent any time in a nursing home since the previous contact. In our analysis we coded people as having entered the institutional population if they reported in a follow-up interview having been in a nursing home for 3 months or more. We treated "institutionalized" as an absorbing state.

We also modeled transitions to the absorbing state "dead." EPESE researchers ascertained respondents' deaths through a daily monitoring of newspaper obituaries and hospital admission records and through annual contact with the participants or their next of kin, and confirmed all deaths through linkage with the National Death Index. We could not confirm vital status for 27 participants, who we assumed to be alive at the end of follow-up.

Measures
We adopted three widely used self-report measures of basic daily functions as indicators of disability. The first measure (Katz) focuses on ADLs and assesses the ability to perform six basic self-care tasks (e.g., bathing, dressing, eating) without help (Branch, Katz, Kniepmann, & Papsidero, 1984). ADLs represent basic activities related to personal care, and limitations in ADLs reflect a greater severity of disability than limitations in instrumental ADLs, which refer to higher level functional abilities. Previous work on temporal trends in disability has mostly focused on ADL and instrumental ADL disability, but the New Haven EPESE data set contained no data on instrumental ADL disability. Instead, we focused on two other measures of functional disability. Both measures assess limitations in basic physical functions that researchers think are more proximate to the underlying disease processes that give rise to age-related disability (Pope & Tarlov, 1991; Verbrugge & Jette, 1994). They may therefore be more direct markers of the impact of age-related diseases on functional abilities. The Rosow measure assesses the ability to perform tasks requiring mobility and strength and includes three items (walking up and down stairs, walking half a mile, and doing heavy work around the house; Rosow, Breslau, & Guttman, 1966). The Nagi measure assesses the ability to perform four basic functions (pulling or pushing large objects; stooping, crouching, or kneeling; reaching or extending arms above shoulder level; and writing or handling small objects; Nagi, 1976). The Katz and Rosow measures define disability as a reported inability to perform one or more of the indicated tasks without assistance; the Nagi measure indicates disability as a reported difficulty performing one or more of the indicated tasks. Together, these measures cover a broad spectrum of functional tasks and abilities that experts consider to be aspects of the entire disability profile in older age.

Another key variable in our analysis was the number of months between successive recorded disability measures. Past studies of disability transitions based on these data have treated the period 1982–1989 as a series of exactly 12-month intervals (e.g., Mendes de Leon et al., 1999). However, as in any longitudinal field study there is inevitably some variability in the timing of interviews. Furthermore, a few respondents missed one or more follow-up interviews but subsequently were re-interviewed, creating unusually long (if unintended) between-interview intervals. And, the final two intervals (1989–1991 and 1991–1994) spanned longer periods than the first seven, all of which spanned 12 months. Consequently, the effective distribution of time elapsed between interviews exhibited considerable variability. Although 12 months was, indeed, the modal elapsed time between interviews in our sample, fewer than half of the intervals are that length. Moreover, a test of association between the binary variable change (versus no change) in disability status and the interval-length variable (measured in months) produced a chi-square statistic of 315.6 (df = 80, p <.0001), indicating the inadvisability of ignoring between-interval variation. The modeling approach we used allowed us to use the full range of between-interview intervals while controlling for variability in interval length.

We estimated separate models of disability prevalence and disability transitions for men and women. All models included variables representing age and the passage of time (coded 1982 = 0, 1983 = 1, and so on). We also controlled for interview mode. The EPESE study employed in-person interviews in 1982, 1985, 1988, and 1994, and telephone interviews in the remaining years. Given the tendency, found in some studies, for telephone surveys to underreport disability in comparison to in-person interviews (Herzog & Rodgers, 1988), researchers have remained uncomfortable with the incorporation of such mode changes in panel studies (Freedman, Martin, & Schoeni, 2002). The mode variable used in our models of disability prevalence indicated a telephone interview. We expected a change from in-person to telephone mode to overstate recovery while understating onset of disability, and vice versa. In view of the hypothesized symmetry of the mode-change effect, our disability-transition models included a mode (change of mode) variable coded as 1 (indicating a change from in-person to telephone interview), –1 (indicating a change from telephone to in-person interview), and 0 (indicating no change).

Analysis
Although we were principally concerned with disability transitions rather than prevalence, and we used local-area rather than national data, we began by comparing trends in disability prevalence found in the New Haven EPESE sample to those found in national-level data from comparable time periods.

Model of disability prevalence
We used logistic regression to estimate simple time trends in disability prevalence for each of our disability indicators. For example,

For each combination of gender and outcome variables we considered three different samples: respondents aged 75 and older (for the 9-year period 1982–1991), respondents aged 77 and older (for the 12-year period 1982–1994), and respondents aged 75 (i.e., those aging into the 75-and-older population each year, for the period 1982–1991). We included the first and third of these samples because of our emphasis on the 75-and-older population, chosen so as to maximize our use of available EPESE data. We included the 77-and-older sample in order to expand the set of potential comparisons of our local-area sample to published studies based on national data. We limited all three prevalence samples to community-dwelling individuals.

Model of disability dynamics
We modeled disability transitions for individuals aged 75 and older, focusing on the effects of age and time. With the addition of the absorbing states described previously, our data reflected a four-state system: nondisabled (State 1), disabled (State 2), institutionalized (State 3), and dead (State 4). The possible transitions were 1 2, 1 3, 1 4, 2 1, 2 3, and 2 4 (as well as the nontransitions 1 1 and 2 2). Building on the approach described in Laditka and Wolf (1998), we modeled 1-month transition probabilities using multinomial logistic regressions, a typical element of which is

We normalized the parameters of the non-transition probabilities (e.g., p₁₁) to zero. We arranged the transition probabilities into a Markov transition matrix P, as follows:

The elements of the last two rows of P reflect the fact that institutionalized and dead were absorbing states. We coded both age (at last birthday) and time (in calendar years) as integers, producing a two-dimensional lattice of age- and time-specific transition rates and transition matrices, within which we assumed monthly transition probabilities to be constant.

Following Laditka and Wolf (1998), we estimated the parameters of the embedded Markov chain using maximum likelihood techniques. In the simplest possible model of this type, a transition is characterized by an initial status (i), a final status (j), and the number of months spanned by the interval (m). By the Markov assumption, the probability of observing such a transition is given by the i, j^th element of the matrix power P^m. The exponent in this matrix power serves to control for the time between measures (for additional discussion of this type of model, see Craig & Sendi, 2002, or Lièvre, Brouard, & Heathcote, 2003).

Our introduction of both age and temporal variation, however, substantially complicated the specification and estimation of the model. For example, suppose a respondent of age a₁ is observed to occupy state i in month m₁ of year t, and to occupy state j in month m₂ of the following year. If the month of birth (mob) precedes m₁, then the contribution to the likelihood is the i, j cell of the matrix product

where P_a,t expresses the dependence of the transition probability matrix (i.e., of its elements) on age and time, and the superscripts denote matrix powers. If the birth month follows the interview month, the corresponding expression is somewhat different:

If, as frequently happens in panel data, there are d (> 0) calendar years separating two successive interviews, then we must insert additional terms into these matrix products. For example,

The i, 4 cell (4 = dead) in a multiperiod transition matrix P^m is the probability of having died by month m, given that one was initially in state i, not the probability of dying in month m. Therefore, the contribution to the likelihood for someone known to have died in month m (m > 1 months after the start of an interval) is the i, 4 element of P^m minus the i, 4 element of P^m-1.

We estimated all models using sampling weights. We computed standard errors using a version of the "sandwich" (robust) covariance matrix that takes sampling weights into account (Manski & Lerman, 1977). This covariance matrix also corrected for the clustering of transitions within respondents (Cole, Bonetti, Zaslavsky, & Gelber, 2005).

	RESULTS

TOP Abstract Methods Results Discussion References

 
Models of Disability Prevalence
Table 1 shows the estimated trends—the regression coefficients corresponding to the time variable—from several alternative disability-prevalence regressions. With respect to the prevalence of ADL disabilities (our Katz measure), we found no evidence of trends among the community-dwelling older population for either the 1982–1991 period (for those aged 75 and older) or the slightly longer 1982–1994 period (for those aged 77 and older). These results are consistent with recent findings for the U.S. population as a whole. Freedman and colleagues' (2004) analysis used five different data sources and a number of different criteria for measuring ADL disability in the 70-and-older community-dwelling population. Their results indicated that ADL disabilities among this group did not decline during the 1980s but did decline beginning in the late 1990s. Our results also agree with Spillman's (2004) reanalysis of the NLTCS: Spillman found no statistically significant decreases in the prevalence of ADL disability during 1984–1994 for either the 75–84 or the 85–and-older population.

Models of Disability Transitions
Table 2 shows parameter estimates and standard errors for all transition models. Preliminary estimation indicated that the maximum-likelihood solution was one in which p₁₃—the probability of making a direct transition from being noninstitutionalized and nondisabled to being institutionalized—approached 0. Although it is indeed possible to be a nondisabled community-dweller one month and a nursing home resident the next, such sequences are likely to be quite rare in the population. Our data were unlikely to include any such direct transitions, given the EPESE design with its 1-year (or more) follow-up intervals. Examining our data, we found that the smallest interval between recorded states i = nondisabled and j = institutionalized was 6 months. Therefore we re-estimated all models with the restriction p₁₃ = 0.

The key findings for purposes of this article were the presence of significant time trends in all estimated models. Over the period 1982–1994, we found statistically significant downward trends in 5 of 6 onset risks (the sixth such estimate, for men in the Katz model, also had a negative coefficient but a p value of.062). Such trends, other things being equal, would imply a falling prevalence of disability in the population. However, we also found significant downward trend in recovery risks for all six combinations of gender and disability indicators. By itself, and other things being equal, such trends would imply a rising prevalence of disability among elderly adults rather that the falling prevalence found in past research. These estimated trend effects also controlled for interview-mode effects, a majority of which were themselves statistically significant.

	DISCUSSION

TOP Abstract Methods Results Discussion References

 
We have noted that trends in the prevalence of disability among the older population—including flat trends or nontrends in prevalence—can be decomposed into trends in disability onset, in recovery from disability, and in the presence of disability among new entrants into the older population. Our analysis performed this decomposition using three different definitions of disability for the 75-and-older population included in the New Haven EPESE data.

We did not find a consistent pattern of trends in disability prevalence in the EPESE data. For ADL disability (our Katz measure), there were no statistically significant trends. For the Rosow measure of functional status, we found a significant downward trend for men, but we found no trend for women. However, for the Nagi measure we found statistically significant upward trends in disability prevalence for both men and women. In general, the trends for our New Haven sample agreed with published findings for national-level trends in comparable indicators, study populations, and time period. We also failed to find consistent trends in disability among those entering the 75-and-older population each year.

Despite these mixed patterns of trends in prevalence, we found remarkably strong and consistent evidence for downward trends in disability onset, accompanied by rather surprising, and in all cases statistically significant, downward trends in the probability of recovering from disability. The former, by itself, would suggest improved population health, but the latter, in isolation, would imply the opposite. The fact that three different patterns of prevalence trends all decomposed into the same pattern of declining incidence rates calls attention to the fact that the relative magnitudes of trends in incidence rates is extremely important. An apparently unfavorable trend in recovery rates—rising, rather than falling, rates—can be masked by even larger downward rates of onset, producing a net decline in overall prevalence.

It is possible that compositional change, namely a trend away from institutional residence, biased our estimated trends in recovery rates among the noninstitutionalized population. Increasing use of assisted living facilities and of home care might encourage community residence among persons whose disabilities would, in earlier times, have led to institutionalization. A growing presence of such people in the community would, in turn, make it appear that recovery rates have fallen. Lacking data on institutional residence trends for the EPESE study area, we turned to Bishop's (1999) analysis of 1985 and 1995 National Nursing Home Survey data. Over this 10-year period, which corresponds reasonably well with the period spanned by our study, the prevalence of nursing home occupancy fell from 0.0575 to 0.0459 among 75–84 year olds and from 0.2194 to 0.1988 among persons aged 85 and older. Weighting these by the baseline age distribution in our sample (77.7% in the 75–84 group and 22.3% in the 85-and-older group) changes the estimated prevalence of nursing home occupancy for our sample by a tiny fraction (–0.017, from 0.094 to 0.077). This, in turn, implies an average monthly reduction of 0.0018 in the log-odds of nursing home occupancy over a 10-year period. Yet we estimated monthly reductions in the log-odds of a recovery transition ranging from 0.06 to nearly 0.12. Therefore, compositional change is unlikely to explain much of our estimated trends in recovery rates. Moreover, by analogy to the prevalence of disability itself, the prevalence of nursing home occupancy can fall over time principally because either (a) the rate of nursing home entry by disabled people falls, or (b) the rate of discharge to the community among disabled nursing home occupants rises. Regarding the first of these possibilities, our model allowed for trends in the rate of nursing home entry but found no clear pattern of such trends (only one of the six such coefficients, that for men in the Katz model, was negative and significantly different from 0). By treating institutionalization as an absorbing state, we ruled out the second such possibility. Thus, we are confident that our findings concerning falling recovery rates are likely to be a reflection of true change in disability dynamics (in our sample) and are relatively unaffected by compositional change.

Our results are consistent with those for the United States as a whole as reported by Manton, Corder, and Stallard (1993), who compared disability transitions observed in the 1982–1984 NLTCS interviews to those observed in the 1984–1989 NLTCS. Research has shown that the 1982–1989 period was a time during which ADL disability prevalence did not fall (Freedman et al., 2004). However, our findings indicate that the same sort of trends in incidence rates—falling rates of onset in combination with falling rates of recovery—could lie behind the more recently observed drop in ADL disability prevalence as well.

Past research has shown that the greatest reductions in disability prevalence have occurred for the least severe disability categories (Spillman, 2004). Our estimates of trends in disability onset and recovery rates for the New Haven EPESE sample are consistent with such a trend. If disability onset rates fell in the United States during the 1980s and 1990s, but those who became disabled were worse off in terms of health and functioning than were disabled persons of previous years, then the chances for recovery within the disabled older population would likely fall over time as well. If the trends in onset were sufficiently large to outweigh the trends in recovery, then on balance prevalence rates—the stocks of persons disabled at any given time—would fall during this time period as well. This is, in other words, an interpretation that relies heavily on the idea of selective risks of transitions that vary according to unmeasured attributes (i.e., unmeasured heterogeneity). As usual, it is impossible to judge whether such an interpretation is warranted without further research. A deeper question, but one that lies outside the scope of this article, is why trends in both onset and recovery rates might simultaneously be falling. It seems likely that very different factors are implicated in the onset of and the recovery from disability. A broad range of factors, among which are lifestyle and behavioral patterns, activity levels, social interactions, random events (e.g., accidents), and preventive strategies, contribute to the onset of disability. Many of those same factors promote recovery as well, but recovery also seems to be closely associated with interventions such as medical treatments and the provision of various services.

Researchers often use estimated rates of change in disability status to compute active life expectancy, an estimate of the average number of years someone from a specified population group will live in a disability-free state. Virtually all existing estimates of active life expectancy assume that the underlying disability–dynamics process is stationary (i.e., unchanging over time; see, for example, Crimmins, Hayward, and Saito, 1994, or Land, Guralnik, and Blazer, 1994). Yet, just as a continuation of downward trends in death rates during the future implies that the average remaining life years of today's population exceeds life expectancy computed using current-period death rates, our results suggest that active life expectancy for today's nondisabled older population exceeds that implied by recent period-based estimates of disability-transition rates.

As noted earlier, our study is limited in several ways: (a) we used local, rather than national, data; (b) our use of what was originally inception-cohort data to represent a continuously replenished fixed-age group sample required that we discard some sample information; and (c) we had to treat nursing home occupancy as an absorbing state despite well-established knowledge that the disability characteristics of the institutionalized population are quite different from those of the community-dwelling population. Bearing these limitations in mind, we found strong evidence of parallel downward trends in rates of both onset of and recovery from disability among community-dwelling persons aged 75 and older. The net effect of the two offsetting trends implies a level or slightly downward trend in the prevalence of disability at older ages. These findings are consistent with population-level findings of increasing levels of severity among the disabled population during the 1980s and 1990s. If research were to show that the population-level trends in disability prevalence were accompanied by the same sorts of incidence trends that we found for the EPESE sample, a prudent conclusion would seem to be that the recent downward trends in population-level disability prevalence will become progressively harder to sustain.

	Acknowledgments

 
Earlier versions of this article were presented at the National Center on Health Statistics workshop "Progress in Estimating Active Life Expectancy," Hyattsville, Maryland, October 9, 2002; at the International Conference on Health Policy Research, Chicago, Illinois, October 17–19, 2003; and at the TRENDS Network's "Workshop on Trends in Disability and Health," Ann Arbor, Michigan, February 17, 2006. We have received helpful comments from workshop participants, three anonymous reviewers, and the editor of this journal.

	Footnotes

 
Decision Editor: Kenneth F. Ferraro, PhD

Received for publication June 23, 2005. Accepted for publication June 16, 2006.

	References

TOP Abstract Methods Results Discussion References

 

• Bishop, C. E. (1999). Where are the missing elders? The decline in nursing home use, 1985 and 1995. Health Affairs, 18,146-155.[Abstract]
• Branch, L., Katz, S., Kniepmann, K., Papsidero, J. A. (1984). A prospective study of functional status among community elders. American Journal of Public Health, 74,266-268.[Abstract/Free Full Text]
• Cole, B. F., Bonetti, M., Zaslavsky, A. M., Gelber, R. D. (2005). A multistate Markov chain model for longitudinal, categorical quality-of-life data subject to non-ignorable missingness. Statistics in Medicine, 24,2317-2334.[Medline]
• Craig, B. A., Sendi, P. P. (2002). Estimation of the transition matrix of a discrete-time Markov chain. Health Economics, 11,33-42.[Medline]
• Crimmins, E. M., Hayward, M. D., Saito, Y. (1994). Changing mortality and morbidity rates and the health status and life expectancy of the older population. Demography, 31,159-175.[Medline]
• Crimmins, E. M., Saito, Y. (2000). Change in the prevalence of diseases among older Americans: 1984–1994. Demographic Research, 3..
• Crimmins, E. M., Saito, Y., Reynolds, S. L. (1997). Further evidence on recent trends in the prevalence and incidence of disability among older Americans from two sources: The LSOA and the NHIS. Journal of Gerontology: Social Sciences, 52B,S59-S71.[Abstract]
• Cutler, D. M. (2001). Declining disability among the elderly. Health Affairs, 20,11-27.[Abstract/Free Full Text]
• Freedman, V. A., Crimmins, E., Schoeni, R. F., Spillman, B. C., Aykan, H., Kramarow, E., et al. (2004). Resolving inconsistencies in trends in old-age disability: Report from a technical working group. Demography, 41,417-441.[Medline]
• Freedman, V. A., Martin, L. G. (1998). Understanding trends in functional limitations among older Americans. American Journal of Public Health, 88,1457-1462.[Abstract/Free Full Text]
• Freedman, V. A., Martin, L. G., Schoeni, R. F. (2002). Recent trends in disability and functioning among older adults in the United States. Journal of the American Medical Association, 288,3137-3146.[Abstract/Free Full Text]
• Herzog, A. R., Rodgers, W. L. (1988). Interviewing older adults: Mode comparisons using data from a face-to-face survey and a telephone survey. Public Opinion Quarterly, 52,84-99.[Abstract]
• Laditka, S., Wolf, D. A. (1998). New methods for modeling and calculating active life expectancy. Journal of Aging and Health, 10,214-241.[Abstract/Free Full Text]
• Land, K. C., Guralnik, J. M., Blazer, D. G. (1994). Estimating increment–decrement life tables with multiple covariates from panel data: The case of active life expectancy. Demography, 21,297-319.
• Lièvre, A., Brouard, N., Heathcote, C. (2003). The estimation of health expectancies from cross-longitudinal surveys. Mathematical Population Studies, 10,211-248.
• Manski, C. F., Lerman, S. R. (1977). The estimation of choice probabilities from choice based samples. Econometrica, 45,1977-1988.
• Manton, K. G., Corder, L. S., Stallard, E. (1993). Estimates of change in chronic disability and institutional incidence and prevalence rates in the U.S. elderly population from the 1982, 1984, and 1989 National Long Term Care Survey. Journal of Gerontology: Social Sciences, 48,S153-S166.
• Manton, K. G., Gu, X. (2001). Changes in the prevalence of chronic disability in the United States black and nonblack population above age 65 from 1982 to 1999. Proceedings of the National Academy of Sciences, 98, USA, 6354–6359.
• Mendes de Leon, C. F., Glass, T. A., Beckett, L. A., Seeman, T. E., Evans, D. A., Berkman, L. F. (1999). Social networks and disability transitions across eight intervals of yearly data in the New Haven EPESE. Journal of Gerontology: Social Sciences, 54B,S162-S172.[Abstract]
• Nagi, S. (1976). An epidemiology of disability among adults in the United States. Milbank Memorial Fund Quarterly/Health & Society, 54,439-467.
• Pope, A. M., Tarlov, A. R., (Eds.). (1991). Disability in America: Towards a National Agenda for Prevention. Washington, D.C.: National Academy Press.
• Rosow, I., Breslau, N., Guttman, A. (1966). A health scale for the aged. Journal of Gerontology, 21,556-559.
• Schoeni, R. F., Freedman, V. A., Wallace, R. B. (2001). Persistent, consistent, widespread, and robust? Another look at recent trends in old-age disability. Journal of Gerontology: Social Sciences, 56B,S206-S218.[Abstract/Free Full Text]
• Spillman, B. C. (2004). Changes in elderly disability rates and the implications for health care utilization and cost. Milbank Quarterly, 82,157-194.[Medline]
• Verbrugge, L. M., Jette, A. M. (1994). The disablement process. Social Science and Medicine, 38,1-14.
• Waidmann, T. A., Liu, K. (2000). Disability trends among elderly persons and implications for the future. Journal of Gerontology: Social Sciences, 55B,S298-S307.[Abstract/Free Full Text]
• Wolf, D. A. (2001). Population change: Friend or foe of the chronic care system? Health Affairs, 20,28-42.[Abstract/Free Full Text]

This Article Abstract Full Text (PDF) Services Download to citation manager PubMed PubMed Citation