Challenges in Estimating Trends in Late-Life Disability from the American Community Survey

To the Editor:

We read with interest the article by Fuller-Thomson and colleagues (1) and their conclusion that estimates from the American Community Survey (ACS) showed an increase in difficulty with activities of daily living (ADLs) and a plateau in physical activity limitation between 2000 and 2005 for the U.S. population aged 65 years and older. Although the authors provided caveats in their interpretation of their findings, we believe that additional caution is warranted.

We recently considered using the 2000–2007 ACS to assess late-life disability trends. However, we decided that because of changes in survey administration over time, we could not reliably estimate trends from the ACS. Our greatest concern was the change in interview mode. The proportion of ACS interviews that were completed via mail declined from 58% in 2000 to 52% in 2005 (and to 48% in 2007), and the proportion completed through computer-assisted interviews concomitantly increased (2–4). For people aged 16–64 years, estimates of limitation are higher using computer-assisted interviews versus mail. For example, in 2003, the differences for ADL limitation (difficulty dressing, bathing, or getting around inside) were 1.9% versus 1.8%, respectively, and for physical functional limitation (substantial limitation in walking, climbing stairs, reaching, lifting, or carrying) were 7.4% versus 6.5%, respectively (5). We were not able to find mode-specific estimates for the older population. However, if the older group experienced similar patterns, the increase in the proportion of computer-assisted interviews would likely account for some of the estimated increase in limitations. Unfortunately, the public ACS micro data sets do not provide an indicator of interview mode to use in modeling trends. Accordingly, we chose not to use the ACS.

Another factor that appears to have contributed to an upward bias in the trends found by Fuller-Thompson and colleagues is aging of the population. They mentioned in their discussion the changes in the age distribution within the population aged 65 years and older from 2000 to 2005, especially the rapid increase of those aged 80 years and older. But they were not able to adjust for age in their analyses because they simply fitted regression lines across the published aggregate prevalences of limitations for each of the six survey years. We were not able to access from the internet the prevalence figures that Fuller-Thompson and colleagues used, but based on the micro data sets for each year, we calculated our own estimates of the aggregate prevalences for those aged 65 years and older, which track closely the data points graphed in figure 1 of their article. Indeed, when we fitted trend regression models using our prevalences, we found that ADL limitation increased at a rate of 1.56% per year in comparison to the 1.81% per year found by Fuller-Thomson and colleagues (calculated by dividing their regression coefficient on [year−2000] of .16 by the intercept of 8.84). However, when we calculated age-specific prevalences by 5-year age groups from the micro data, standardized the overall 65+ annual prevalence rates to the 5-year age distribution from various populations (2000, 2005, and an average of 2000–2005), and then estimated models based on these standardized prevalences, we found for ADL limitation that the p value on the trend coefficient was reduced from .005 to a range from .07 to .09, and the estimated annual change was reduced to 0.7%–0.8%. For functional limitations, the results of Fuller-Thomson and colleagues implied an annual increase of 0.37%, and we found annual change of 0.24% in our model of unadjusted prevalences and −0.14% to −0.16% for our models of age-standardized prevalences [none of the trend coefficients for functional limitations significantly different from 0, as in (1)].

As an alternative test of the importance of adjusting for age, we ran logit regressions of individual-level ADL and functional limitations on trend using the micro data set. For ADLs, our results implied an annual change of 1.67% in the unadjusted model and 0.85% in the model with 5-year age group controls. For functional limitations, the estimates were 0.34% and −0.23%, respectively. (All logit trend coefficients were statistically different from 0.) Accordingly, it appears that controlling for age results in a halving of the rate of increase for ADLs and in a trend downward, as opposed to upward, for functional limitations.

We applaud Fuller-Thomson and colleagues for their attempt to shed light on the most recent disability trends and especially for their care in accounting for the institutionalized population. We also agree that tracking trends going forward is critical. However, estimated trends may be contaminated by changes over time in a variety of factors, some of which they note, such as survey design and administration, use of proxies, population coverage, question wording, and age distribution (6). Moreover, differences across surveys may also affect comparisons of trends (7).

In sum, we do not think that the analysis of the ACS by Fuller-Thomson and colleagues provides sufficient evidence to reach even tentative conclusions about recent late-life disability trends.