Introduction

The purpose of this study is to examine changes in the prevalence of functional limitation and disability among older adults in five settings in East and Southeast Asia: Indonesia, the Philippines, Singapore, Taiwan, and the Beijing Municipality. Asia in general, and the selected societies in particular, provide an excellent setting for this study. First, due to unprecedented reductions in fertility and substantial improvements in old-age mortality in recent decades, population aging throughout much of Asia is extremely rapid (Hermalin, 2000). In 2005 there were an estimated 188 million persons age 60 or over in East Asia, and 44 million in Southeast Asia. This represents an approximate doubling of the number of older persons since 1980, and these numbers are expected to double again before 2025 (United Nations, 2005). This makes these regions among the fastest aging in the world. The percent of persons age 60 or over has increased at an annual rate of about 3% between 1980 and 2005 and is expected to continue at a slightly higher rate through 2030. This rate of growth far exceeds that experienced in the United States or Europe (Hermalin, 2000). Future growth will be most accelerated for the oldest-old (age 80 and older), for whom the prevalence of health problems is highest. In both East and Southeast Asia, the oldest old are expected to more than double between 2000 and 2050 as a percent of both the total population and the elderly population (United Nations, 2005).

These trends have policy makers in Asia concerned because of their potential implications for future disease burden and associated informal and formal care demands (Hermalin, 2000; Interministerial Committee on the Ageing Population, 1999; Ogawa & Retherford, 1997). The precise implications of population aging for future levels of health and health care utilization depend on whether the increases in life expectancy experienced in the region are accompanied by an increase or decrease in health problems in later life. Arguments for both scenarios as well as a more intermediate view have been advanced (Fries, 1980; Gruenberg, 1977; Kramer, 1980; Manton, 1982). Even under the best of circumstances with regard to declines in disease and disability rates, however, the sheer growth in the older population that is occurring in Asia and other parts of the world will lead to increases in the absolute number of disabled persons (Manton, 1997; Mayhew, 1999). The health infrastructure in the study settings has been oriented toward problems of infectious diseases and maternal and child health and is less well-equipped to handle the health care needs of the older population (Hermalin, 2002). Gaining a better understanding of the trends and determinants of health in later life is thus critical for future planning. In addition, by investigating the universality of findings across different environments, this study can add significantly to our understanding of health processes in later life.

Background

According to projections reported recently by Mayhew (1999), the number of disabled individuals will grow substantially concurrently with worldwide population aging. Consequently, Mayhew argues that aging is poised to overtake population growth as the main factor for expanding health expenditures on a worldwide basis. He further projects that costs associated with health problems will be highest in societies where the proportion of those in older age is greatest, and will increase most rapidly in societies undergoing rapid population aging, such as those included in the proposed study. Whether these projections are accurate for Asian societies will depend to a great extent not only on the size of the increase in the older population, but on the changes that occur in the health status of the older populations as these societies continue to develop. These can be viewed in two ways—as changes that occur to populations measured by prevalence rates, and changes that occur in individual states of health measured by transition probabilities.

Fries (1980) suggested that improvements in health for older populations are a natural outcome of a decrease in the incidence of disease. As a consequence, periods of morbidity for older adults are ‘compressed’ into the very end of life and older adults experience longer lives in better health. An opposing view, advanced by Kramer (1980) and Gruenberg (1977) holds that, along with increases in life expectancy, come more years in states of disability, as lethal sequelae of diseases are eliminated without concomitant changes in the effects of disease on quality of life. Manton (1982) has proposed an alternate scenario under which contradictory cycles of improvement and decline in health occur, leading to a state of ‘dynamic equilibrium.’

The question of whether older adults are experiencing longer life with improving or worsening health has been addressed in a number of studies that examine trends and transitions in the health of older adults in the United States. Evidence from the 1970s and early 1980s indicated that the gains in life expectancy that were realized during this period were accompanied by an increasing prevalence of disability (Colvez & Blanchet, 1981), suggesting a trade-off between longer life and worsening health (Verbrugge, 1984). Waidmann, Bound and Schoenbaum (1995) have questioned the validity of disability reports amidst a changing health environment, suggesting that the increase in disability rates were a function of better diagnostic techniques and earlier detection of chronic disease. There now appears to be fairly consistent evidence that since the late 1980s or early 1990s there has been substantial improvement in older adult health, particularly with respect to disability. For instance, evidence from several recent studies suggests that rates of functional disability among older Americans declined during the 1980s and 1990s (Crimmins, Saito, & Reynolds, 1997; Freedman & Martin, 1998; Freedman, Martin, & Schoeni, 2002; Manton, Corder, & Stallard, 1993; Ofstedal, Madans, & Feldman, 1994; Schoeni, Freedman, & Wallace, 2001; Waidmann & Liu 2000). In a comprehensive review of the literature, Freedman et al. (2004) note that ADL disability declined on the order of 1–2.5% per year during the mid and late 1990s. Other research indicates that declines have occurred for IADLs and basic physical tasks of the Nagi variety (e.g., climbing stairs, walking 1/4 mile, lifting) (Freedman et al., 2002).

These above cited studies conducted in the United States use large panel or repeated cross-sectional surveys, such as the Medicare Current Beneficiaries Survey, the National Health Interview Surveys, the National Long-Term Care Survey, the Longitudinal Study of Aging, and the Health and Retirement Study. Studies using these data sources can provide the basis for the development of hypotheses about health elsewhere, but we cannot assume the relationships found in the United States are universal without comparisons across different political and cultural regimes, and areas characterized by different levels of socioeconomic development. For example, we may expect patterns in Asia to differ from those of the United States due to differing levels of baseline health (Mathers, Murray, Lopez, Salomon, & Sadana, 2003; Zimmer, Natividad, Ofstedal, & Lin, 2002), which has implications for the potential magnitude and rate of change. In addition, the aging of populations in Asia has been accompanied by extremely rapid changes in socioeconomic status on both an individual and a societal level (Knodel, Ofstedal, & Hermalin, 2002). It is possible that health trends will vary across Asian countries themselves for these same reasons.

Much less research on trends in health has taken place in Asia and so it is difficult to draw conclusions about what changes are taking place there. Some research on individual level transitions has examined populations in China, Japan and Taiwan (Brown et al., 2002; Gu & Zeng, 2004; Liang, Liu, & Gu, 2001; Tang, Jiang, & Futatsuka, 2002; Zimmer, Liu, Hermalin, & Chuang, 1998); however, examinations of population-wide functional status trends for Asia are in their relative infancy. Schoeni et al. (2006) recently demonstrated declines in prevalence rates for several functional limitation and disability indicators for Japan similar to what has been seen in the United States, while and Zimmer, Martin, and Chang (2002) show less favorable trends occurring for Taiwan. Cross-sectional examinations of health status are somewhat more common. Recent examples include a comparison of a variety of physical and mental health outcomes across the Philippines, Thailand, Taiwan and Singapore (Zimmer et al., 2002), an examination of the health of the oldest-old in China (Zeng,Vaupel, Zhenyu, Chunyuan, & Yuzhi, 2002) and several writings about the health of older adults in India (Alam, 2006; Rajan, Mishra, & Sarma, 2001). While cross-sectional studies are informative of levels and subgroup differences in health, they do not give us a good sense of how functional health has been changing over time. This paper makes use of longitudinal data in several Asian settings to examine whether the recent improvements in disability that have been observed in the United States are also taking place in a less developed part of the world.

Subjects and Methods

Data sources

The data to be used in this paper come from representative panel surveys of older persons in Taiwan, Indonesia, the Philippines, Singapore, and the Beijing Municipality in China. The names of the surveys and key design features (including the full age range covered in each survey and total sample sizes) are provided in Appendix Table VII and brief descriptions of the surveys are provided below.

The Beijing data are from the Beijing Multidimensional Longitudinal Study of Aging conducted by the Capital University of Medical Science in Beijing. A total of 3,257 adults aged 55 and older were interviewed in the baseline wave in 1992, yielding a response rate of 90%. This sample consisted of individuals living in one of three administrative areas within the Beijing municipality. The first, Xuan Wu, is a district located in metropolitan Beijing. The other two areas, Da Xing and Huai Ruo, are rural agricultural areas located up to 100 km from the city. According to documentation regarding the initial sampling, these three areas were chosen based on their ability to represent the total municipal region with respect to socioeconomic and demographic characteristics (Department of Social Medicine, 1995). The Beijing rural area is made up of two distinct geographic environments, one being mountainous and one being plains. The two rural areas represent one of each environment. Follow-up interviews were conducted in 1994 and 1997.

The Indonesian Family Life Survey (IFLS) is a large-scale, nationally representative panel survey of about 7,200 households in Indonesia. The household sample was drawn from 13 provinces spanning multiple islands in the Indonesian archipelago; together these provinces account for approximately 83% of the Indonesian population. The survey was conducted by the RAND Corporation, in collaboration with Lembaga Demografi at the University of Indonesia. A primary focus of the survey, and the one that is of most relevance to this study, is on the health, economic and social functioning of the older population. As a result, the survey design included interviews with just under 5,000 randomly selected individuals age 50 or over (plus their spouses) from each household that contained requisite members. The baseline response rate was 91%, and follow-up interviews with the full sample were conducted in 1997 and 2000.

The Philippine Elderly Survey (PES) is a nationally representative survey of middle-aged and older adults that was launched in 1996. The sample for the 1996 wave is comprised of 2,285 men and women 50 years of age or older, representing a response rate of 85%. A follow-up interview was conducted in 2000–2001 with a subsample of respondents in two sizable regions in the Philippines (the greater Manila area and Leyete, a large rural province in the north). The total number of respondents in the baseline sample for the longitudinal component is 932. A total of 167 of these respondents were determined to have died prior to follow-up, and 644 were successfully interviewed. The gender distribution of the Philippines sample is somewhat skewed in favor of women (63% in the sample vs. 55% in the population based on estimates published by the United Nations, 2005). The gender distribution in the sample is similar in both survey waves, so the fact that it is skewed should have minimal influence on our analysis, although we must still be somewhat cautious in our interpretation. The major objective of the survey was to assess how rapid demographic change has affected older Filipinos, particularly with respect to physical and psychological health, economic well being, and familial relations and support.

The National Survey of Senior Citizens in Singapore was conducted in 1995. The survey is representative of men and women age 55 years or over in Singapore. A total of 4,750 persons were interviewed. The response rate at baseline was 60%. In 1999 a follow-up survey, which repeated and expanded key health, health care utilization, socioeconomic, and informal support measures, was conducted as part of the project, “Transitions in Health, Wealth, and Welfare of Elderly Singaporeans: 1995–1999.” Interviews were completed with 1,977 individuals and 557 respondents were determined to be deceased based on informant reports during the data collection period.

The Taiwan data come from the Survey of Health and Living Status of the Middle Aged and Elderly in Taiwan, a national survey of men and women 50 years of age and older in 1996, conducted by the Bureau of Health Promotion in Taiwan (formerly the Taiwan Provincial Institute of Family Planning). This survey builds on a panel that began in 1989, following a cohort of persons age 60 years and over through seven waves of data collection, including five in-depth personal interviews conducted in 1989, 1993, 1996, 1999, and 2003, and two abbreviated telephone interviews in 1991 and 1995. In 1996, the panel sample (comprised of persons who were then age 67 or older) was supplemented with a new representative sample of persons 50 to 66 years of age. Both panels were sampled from the national household register, which includes institutions as special households. The baseline response rate was 92% for the original sample of individuals age 60 or over in 1989, and 81% for the younger cohort of individuals age 50–66 who were added in 1996.

In the current paper, we examine trends for individuals aged 60 and older across two waves of each survey. For each wave, we restrict the analysis to individuals who are age 60 or older at the time the interview for that wave conducted. For example, when evaluating the trend in prevalence of functional limitation in Indonesia, we compare the prevalence rate for persons who are age 60 and over in 1993 to the prevalence rate for persons who are age 60 and older in 1997. The waves were chosen so as to cover roughly the same interval length and the same period of time across surveys: 1993 and 1997 for Indonesia, 1994 and 1997 for Beijing, 1995 and 1999 for Singapore, 1996 and 1999 for Taiwan, and 1996 and 2000–2001 for the Philippines.

Variables

Dependent variables

Most studies that have examined trends in functional health have used dichotomous variables indicating whether or not an individual has any limitation within a specific domain (Freedman et al., 2004; Manton et al., 1993; Schoeni et al., 2006). We follow this in the current study by measuring limitation using dichotomous variables indicating whether or not the respondent reported having difficulty performing at least one Activity of Daily Living (ADL), at least one Instrumental Activity of Daily Living (IADL), and at least one Nagi physical functioning task (0 = no, 1 = yes), respectively. The number and types of activities queried differed across the surveys and are listed in Appendix Table VIII. The precise question wording also varied somewhat across the surveys (as shown in Appendix Table IX). For these reasons we do not focus on cross-setting differences in prevalence in this paper, but rather on trends in prevalence within settings over time and on how the trends compare across settings in terms of their direction and significance.

Question wording was consistent across waves within settings, with two exceptions. The first exception occurred in the Taiwan survey, for which the 1996 questionnaire included a skip and stem question structure for the ADL measures. Specifically, in 1996, respondents skipped the individual ADL questions if they had no Nagi difficulties or if they reported “no” to a global/stem question concerning ADL difficulties (see Appendix Table IX). In contrast, every respondent was asked the individual ADL questions in 1999. The second exception occurred in Singapore, for which the administration of the ADL and IADL questions differed across waves. The questions used in the 1995 survey deviated from the conventional approach. For ADLs, no explicit question wording was provided for interviewers and they were instructed to record the respondent’s status with respect to mobility, feeding, toileting, and personal grooming and hygiene. Respondents were asked about their IADL status in the 1995 wave, but the question wording differed from most other surveys (see Appendix Table IX). In 1999, the question wording for both ADLs and IADLs followed a more conventional approach, whereby respondents were asked whether they have difficulty with the activities because of a mental or physical health problem and, if so, how much difficulty they have. In addition, in 1999, a skip pattern was introduced such that respondents who did not report any out of a list of 10 health conditions (stroke, high blood pressure, diabetes, cancer, chronic lung disease, heart disease, arthritis, permanent loss of memory or mental ability, kidney problems, cataract or glaucoma) were assumed to have no difficulty with functioning and were skipped out of the ADL and IADL questions. Hence, results pertaining to changes in the prevalence of ADL limitation in Taiwan and, particularly in Singapore, must be interpreted with some caution.

Independent variables

The independent variable of key interest is a dummy variable representing the survey wave (coded 0 = baseline, 1 = follow-up). In addition, to assess the extent to which any observed change in the prevalence of limitation is attributable to changes in the composition of the older population, we include controls for age (coded as continuous variable ranging from 60 to 99), sex (0 = male, 1 = female), marital status (0 = not currently married, 1 = currently married), and education (primary, secondary or higher versus no formal education).

Analysis methods

To determine whether or not there were significant changes in the prevalence of limitations over time within a setting, we pool cross-sectional samples for the baseline and follow-up waves. We then fit two logistic regression models that estimate the log-odds that individuals report each type of limitation. The first model includes only a dummy variable for survey wave, using the baseline as the comparison category. The second model adds the compositional variables described above in order to assess the extent to which they account for any observed changes in functional limitation. These regressions take the form of:

  1. Model 1:

    \( \ln {\left( {P \mathord{\left/ {\vphantom {P {1 - P}}} \right. \kern-\nulldelimiterspace} {1 - P}} \right)} = \alpha _{0} + \alpha _{1} T2 \)

  2. Model 2:

    \( \ln {\left( {P \mathord{\left/ {\vphantom {P {1 - P}}} \right. \kern-\nulldelimiterspace} {1 - P}} \right)} = \alpha _{0} + \alpha _{1} T2 + \alpha _{x} x_{2} \ldots \alpha _{k} x_{k} , \)

where α1 represents the log odds for the difference in prevalence rates between time periods and α2...αk represent the effects of compositional variables.

All analyses are weighted to account for differential sampling probabilities and non-random attrition due to non-response at follow-up. In addition, since repeated observations are being used in this analysis to estimate population rates at two different time periods, we need to be sensitive to the likelihood of correlated error terms. We use STATA 7.0 to conduct the regression analyses, which is capable of handling such data with its robust command by employing the Huber/White/sandwich estimator of variance (Lin & Wei, 1989).

Results

Table I presents distributions on key sample characteristics for persons age 60 years or over in the specified year for each of the five study settings. The average age of the samples is slightly lower in Indonesia and Beijing (67.3 and 68.2 respectively for the first wave) than in the other three settings (ranging from 69.2 in Singapore to 69.7 in the Philippines). Females predominate in the Philippines and Singapore (62.8 and 53.6% female for the first wave, respectively), whereas males predominate in Taiwan (54.8% male). (The male predominance in Taiwan is due to the large migration of male soldiers from Mainland China to Taiwan following the Chinese civil war, which altered the sex distribution for this age cohort.) The proportion married at baseline also varies across settings, ranging from about one-half in the Philippines and Singapore to over two-thirds in Beijing. The level of education is quite low in all settings except the Philippines. One-half or more of older adults in Singapore, Indonesia and Beijing and two-fifths of those in Taiwan have no formal education. The large majority of older Filipinos received a primary level education, and one-fifth to one-quarter completed secondary or higher education.

Table I Sample Characteristics for Each Country by Survey Wave

Table II presents the percentage of persons age 60 or over in each setting in the designated year who reported having difficulty performing one or more ADL, IADL, and Nagi activity. Recall that the items used to measure limitation differed across surveys (see Appendix Table VIII), so we are not able to compare prevalence levels across settings. Our interest here is to compare the prevalence of limitation over time within each setting.

Table II Prevalence of ADL, IADL and Nagi Limitation by Survey Wave

With the exception of Singapore, each study setting experienced a significant increase in the prevalence of limitation in at least one of the three domains. The most marked increases occurred with respect to Nagi limitation. In all four of the settings that included Nagi measures, significant increases in the percent reporting Nagi limitation were observed. The increases ranged from 17% in Taiwan to 48% in Beijing. In addition, Indonesia and Taiwan experienced significant increases in the prevalence of ADL limitation, and Beijing and the Philippines experienced significant increases in the prevalence of IADL limitation. The Philippines also showed an increase in ADL limitation, however due to small sample size, the difference was not statistically significant. Singapore is the only country that did not exhibit a significant change in either of the functional domains measured-ADLs or IADLs.

Tables III through V present results from multivariate logistic regression models predicting ADL, IADL, and Nagi limitation, respectively, for each setting. Results are presented as odds-ratios. As noted previously, data from both survey waves within a setting were pooled, and a dummy variable representing the survey wave (0 = baseline, 1 = follow-up) was included in the model. In addition, the samples were weighted to be representative of the target populations, and standard errors were adjusted to account for repeated observations on the same individuals. The top panel of the tables presents results from bivariate models that include only survey wave as a predictor; the bottom panel presents results from multivariate models that adjust for compositional effects of key sample characteristics. We also present, in Table VI, a qualitative summary of the findings, which compares the direction and general significance of trends across functional domains and countries.

Table III Odds-Ratios for the Effects of Survey Wave and Key Compositional Variables on ADL Limitation

Focusing first on ADL limitation (Table III), the unadjusted odds-ratios in the top panel simply reproduce the bivariate results in Table II and show that the prevalence of ADL limitation for persons age 60 years or older increased significantly in Indonesia and Taiwan, increased slightly (but not significantly so) in the Philippines, and did not change in Singapore or Beijing. Controlling for key compositional factors leads to slight declines in the odds-ratios. Although the declines are not large, they are sufficient enough to reduce the cross-wave differences in Indonesia and Taiwan to insignificance. With regard to the effects of covariates on ADL limitation, age shows a strong positive association with ADL limitation in all settings. The effects of other variables are less consistent, with females showing higher ADL limitation than men in Singapore, married persons showing higher levels of ADL limitation than unmarried persons in both Beijing and Singapore, and those with secondary or higher education in Taiwan showing lower levels than those with no education.

Parallel results for IADL limitation are shown in Table IV. Both Beijing and the Philippines experienced significant increases in IADL limitation, whereas no change was observed in Taiwan or Singapore (top panel). In this case, the odds-ratios for survey year did not change when compositional characteristics were added to the models (bottom panel). Hence, the observed increases in prevalence in Beijing and the Philippines cannot be attributed to changes in the composition of the population between the two survey waves, at least in terms of characteristics included in the model. Again, age shows a strong positive association with IADL limitation in all settings, and females are more likely than males to report IADL limitation in Beijing, Singapore and Taiwan. Education is also an important predictor of IADL impairment for both Beijing and Taiwan, such that those with any education (primary or secondary up) are less likely than those with no education to be limited in IADLs.

Table IV Odds-ratios for the Effects of Survey Wave and Key Compositional Variables on IADL Limitation

Lastly we turn to results for Nagi limitation (Table V). The unadjusted results show substantial increases in the prevalence of Nagi limitation in all four settings for which data are available. As was the case for IADL limitation, the odds-ratios for survey year were not affected by the inclusion of compositional variables in the model. Here again, age shows a strong positive association with Nagi limitation in all settings. Sex is also an important predictor of Nagi limitation in all settings, with women being substantially more likely then men to be limited. And, consistent with results for IADL limitation, education is an important predictor of Nagi limitation for Beijing and Taiwan, with more educated individuals reporting lower levels of limitation.

Table V Odds-ratios for the Effects of Survey Wave and Key Compositional Variables on Nagi Limitation

We conducted additional analyses (not shown here) to assess the robustness of results across individual items within the ADL, IADL and Nagi domains, and to examine whether the observed trends differ by age, gender, marital status, and education. Analysis of the individual items failed to reveal any specific items that were driving (or suppressing) the overall trend in the corresponding domain. Although there were slight differences in the precise magnitude of the change across items, the direction of the change was remarkably consistent. To evaluate subgroup differences, we estimated a set of models for each functional outcome in each country that included interactions between the subgroup variable of interest (age, gender, marital status, education in turn) and survey wave. Of all of the interactions tested across settings, only one was significant (a gender interaction for IADL limitation in Singapore), suggesting that the trends are similar (or at least not sufficiently different) across subgroups.

Discussion

Taken together, the findings suggest that the Asian societies represented in this study do not appear to be experiencing the improvements in physical functioning that have been observed during recent decades in the United States. On the contrary, every setting except Singapore experienced significant increases in functional limitation in at least one domain, and none experienced significant decreases in any domain (see Table VI). The most consistent and dramatic changes were observed for Nagi limitation, for which increases occurred in Beijing, Indonesia, the Philippines, and Taiwan. Two of the four settings that included IADL measures experienced increases in IADL limitation (Beijing and the Philippines), and two of the five settings experienced increases in ADL limitation (Indonesia and Taiwan). Changes in the composition of the older population did not account for increases in the prevalence of either IADL or Nagi limitation.

Table VI Comparison of the Trends in ADL, IADL and Nagi Limitation Across Settings

It is noteworthy that there are inconsistencies in trends across measures, that is, strong increases in the prevalence of limitation based on Nagi measures, but very little change based on ADLs, with IADLs somewhere in between. Is it reasonable to find differences in trends across different domains of function? Other studies have found this to be the case. For instance, Waidmann and Liu (2000) found for the United States improvements in ADL and IADL limitation concurrently with deteriorations in more basic physical measures. Other studies in the United States (Freedman et al., 2002) and in Japan (Schoeni et al., 2006) have found trends in some, but not other domains. Indeed, the three domains we examined differ in a number of ways, including the extent to which they are influenced by disease, physical impairment, environment, social support, and other factors that cause functional limitation and disability. Further analyses we conducted (not shown) showed moderate correlations between ADL, IADL and Nagi limitations (ranging from about 0.3 to about 0.6 depending on the setting and the specific pair of measures), indicating that although the three are highly related, each provides some unique information. Therefore, it is not surprising that studies have found different patterns of change over time in these domains.

How do we reconcile the seemingly contradictory findings between Asia and the United States? Or are the findings actually contradictory? As noted earlier, findings from the United States in the 1970s and early 1980s suggest that disability among older persons actually increased for a period of time, before it started to decline in the 1980s and 1990s. It is plausible that Asia (or at least the settings examined here) is going through a similar stage. Social and economic development and advances in the health infrastructure occurred much later in these settings than in the United States. Thus, in comparison to older Americans, older Asians may have experienced substantially more difficult living and working environments and poorer health earlier in their lives, including greater exposure to infectious diseases. These experiences are likely to have health consequences later in life. In addition, recent improvements in survival that have been experienced in all of the settings examined here may have disproportionately benefited those in poorer health and with limitations. Evidence suggests that this is the case during the 1990s in Taiwan (Zimmer et al., 2002; Zimmer, Martin, & Lin, 2005).

Disproportionate improvement in survival is unlikely to explain all of the increases we observe in the prevalence of functional limitation and disability. Another reason one might find an increase in prevalence is due to changes in the ways in which people perceive their functional health. For instance, as the general health of a population improves and health services become more widespread and increasingly sophisticated, expectations about what it means to be in good health may be transformed and people may become more aware and less tolerant of their limitations. This would result in increases in disability and functional limitation that are not necessarily tied to changes in underlying health and objective impairments. Moreover, it is not reasonable to expect that sharp increases in prevalence would continue at the same level into the future. Future cohorts of elderly in Asia will be very different in terms of their early life experiences, education, and economic status, and we may begin to see improvements in functioning at the societal level for older adults over the next decade or two.

At the same time, the relatively consistent results we have observed in this study occur across settings that themselves differ quite substantially. Singapore and Taiwan, for instance, are much closer to the United States in terms of socioeconomic development than China, Indonesia or the Philippines. Although Singapore displayed very little change in prevalence of ADLs and IADLs over time, Taiwan showed an increase in the prevalence of Nagi limitation that is consistent with the less developed societies that we examined. While it is true that Singapore and Taiwan are more similar to the United States in terms of their levels of socioeconomic development, this development occurred much later and more rapidly in Asia than in the United States. The timing and pace of economic development, in combination with transitions in the orientation and coverage of the health care system as well as other societal changes, may influence societal trends in health, net of the level of development.

The study has some important limitations. A major limitation is the short time period over which we are trying to discern trends, and the use of only two waves of data. It will be important to replicate this study as new waves of data become available, to assess whether the increases in prevalence that we observed over a very short period of time persist over a longer period. A separate study that focused only on Taiwan, using three waves of data (1993, 1996, 1999) and more limited set of measures found evidence of longer-term trends (Zimmer et al., 2002). One of the planned extensions of this work is to conduct longer-term investigations in other settings where data are available. In the meantime, we must be very cautious about interpreting these findings as true long-term population trends.

Ideally, for this study, we would have used data from repeated cross-sectional surveys that draw fresh samples of the population at each time point. However, such surveys are rare, particularly those containing sufficient measures of health and disability. Panel data can and are often used to study trends (Manton et al., 1993; Schoeni et al., 2006; Waidmann & Liu, 2000), but this can be problematic for two reasons. The first has to do with non-random non-response in follow-up waves. Over time, a panel sample is likely to become less representative of the target population. We would generally expect higher levels of attrition among the more frail and thus, if anything, panel attrition would tend to bias towards improvements in functioning. In the current study, panel attrition is addressed by developing sample weights that account for non-random attrition due to non-response in the follow-up wave. Another potential problem is that participation in a panel study may influence responses in subsequent waves. This can be particularly problematic with regard to test-related material, such as cognitive performance tests, for which studies have documented a learning effect in repeated administrations (Jacqmin-Gadda, Fabrigoule, Commenges, & Dartigues, 1997; Unger, van Belle, & Heyman, 1999; Zelinski & Burnight, 1997), but is generally considered to be less problematic for most other types of survey questions, such as reports of health problems and symptoms.

The current study used one of several potential ways of operationalizing functional limitation, which involved a simple dichotomization of any versus no limitation for each of the three domains (ADL, IADL and Nagi). We followed this approach in order to maintain comparability with other studies, as well as to maximize consistency across surveys that used different question wording, items, and response categories to measure limitation. For each individual survey, it would be possible to construct other variables based on number of difficulties and severity of difficulties. Examining trends in this way could potentially lead to alternate conclusions. For instance, it is possible that the probability of having one or more limitations is increasing but severity of limitations is decreasing in a population. Under this scenario, the interpretation of our findings could be quite different, suggesting some improvement in overall functional health. We suggest that further research examine severity within settings in more detail.

As pointed out previously, there are differences in the measurement of functional limitation and disability across surveys. These differences make comparisons of levels of functional limitation and disability and of the precise magnitude of change in these levels problematic, and for this reason we specifically avoid making such comparisons here. Rather, we focus our discussion on a comparison of the direction and general significance of change across settings. Although it would be ideal to have identical measures across settings, to our knowledge there are no surveys that currently exist across Asia that provide such data. Furthermore, even identically worded questions can be subject to differences in interpretation and influenced by differences in environmental circumstances across countries, both of which may confound comparisons. In short, there is virtually no perfect way to compare survey items of health across disparate settings, and yet such comparisons are important for gaining insights into the relevance of theoretical frameworks in different settings and the generalizability of findings across countries.

The specific way in which ADLs were measured changed slightly in Taiwan between waves as did the way in which ADLs and IADLs were measured in Singapore. For Taiwan, individuals were asked about ADLs if they reported any Nagi limitation for the first wave, whereas in the second wave all respondents were asked about ADLs. The former approach assumes that respondents with no Nagi difficulties would not report any ADL difficulty. At least in Taiwan, this assumption appears warranted; only two individuals (0.1%) who had no Nagi limitations in 1999 reported an ADL limitation. Thus, this difference in measurement should have little impact on the results. In Singapore, the changes in question wording were more substantial, and we cannot rule out the possibility that our null finding of no change in prevalence across waves may be due to the changes in question wording.

Future analyses should include several other extensions of this work. First, it would be possible to extend the analyses for Indonesia and Taiwan to incorporate data from the most recent survey waves (2000 for Indonesia, 2003 for Taiwan). Also, for purposes of comparability we restricted our analyses to those aged 60 and older and to waves that were conducted in similar years, but several surveys contain data for individuals age 50+ or 55+ and it would be valuable to examine trends for that age group. Finally, it would be possible to examine trends in other health indicators and mortality patterns to better understand the dynamics that underlie the observed increases in functional limitation and disability in these settings.

In conclusion, if indeed Asian populations are experiencing increases in the prevalence of functional limitation and disability, this would have profound implications for a number of important policy concerns including quality of life, demand for informal and formal health care and retirement. In addition, increasing health problems could exacerbate health care and other cost increases that are expected within societies undergoing very rapid population aging. Still, increasing prevalence rates for functional limitation are unlikely to continue indefinitely and, like the United States, we might expect a period of increasing health problems to be followed by a period of decline. Thus, it will be important to monitor trends in functional health and other health indicators in Asia and other regions as the data to do so become increasingly available.