| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Polisher Research Institute, Philadelphia Geriatric Center, Jenkintown, Pennsylvania
b Population Council, New York, New York
Vicki A. Freedman, Senior Research Scientist, Polisher Research Institute, Philadelphia Geriatric Center, The Pavilion, Suite 427, 261 Old York Road, P.O. Box 728, Jenkintown, PA 19046-7128 E-mail: vfreedman{at}pgc.org.
 |
Abstract |
|---|
 TOP Abstract Defining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
Methods. Using the 1993 Asset and Health Dynamics of the Oldest Old study (N = 7,443) and 1998 Health and Retirement Survey (N = 7,624), this study examined aggregate changes in the proportion of the noninstitutionalized population aged 70 and older with severe cognitive impairment. Impairment was measured for self-respondents using a modified version of the Telephone Interview Cognitive Screen; for proxy respondents, ratings of memory and judgment were used. Logistic regression was used to investigate potential explanations for aggregate changes.
Results. The percentage of older Americans with severe cognitive impairment declined from 6.1% in 1993 to 3.6% in 1998 (p < .001). The decline was statistically significant among self-respondents but not among those with proxy interviews. Improvements between 1993 and 1998 were not explained by shifts in demographic and socioeconomic factors or by changes in the prevalence of stroke, vision, or hearing impairments.
Discussion. As a group, older persons, especially those well into their 80s, appear to have better cognitive functioning today than they did in the early 1990s.
ALTHOUGH results are not completely uniform, and there is no assurance that the trend will continue, there is a growing consensus that during the 1980s and early 1990s physical functioning among older Americans improved (
Crimmins, Saito, and Reynolds 1997
;
Freedman and Martin 1998,
Freedman and Martin 2000;
Manton, Corder, and Stallard 1993,
Manton, Corder, and Stallard 1997;
Waidmann and Liu 2000). Thus far studies have focused on a spectrum of physical outcomes, including functional limitations, activities of daily living (ADLs), and instrumental activities of daily living (IADLs).
Comparisons within and across studies suggest that improvements in IADLs may have been larger than those for ADLs (
Crimmins et al. 1997;
Freedman and Soldo 1994;
Manton et al. 1997;
Waidmann and Liu 2000). Others have noted that IADLs, such as managing money, taking medications, and using the telephone, have a particularly strong relationship with cognition (
Barberger-Gateau et al. 1992;
Fitzgerald, Smith, Martin, Freedman, and Wolinsky 1993;
Kasper 1990), suggesting that cognitive functioning may also have improved among older Americans in recent decades. Consistent with this hypothesis is the work by
Manton, Stallard, and Corder 1995,
Manton, Stallard, and Corder 1998, which showed declines in the prevalence of dementia and in the debilitating effects of this condition among older Americans between 1982 and 1989.
A handful of studies in the cognitive psychology literature have shown improvements over historic time in a variety of intelligence measures (
Salthouse 1991).
Flynn 1987, for example, documented massive IQ gains for adults in 14 industrialized nations. Such studies have often focused on select samples (e.g., soldiers, school-aged children, young adults) that exclude older adults. A notable exception is
Schaie 1983, who found improvements among adults aged 25 to 67 from 1959 to 1991 on several cognitive dimensions in a longitudinal study of enrollees in a health maintenance organization in Seattle. In his age-specific presentation of Schaie's results,
Salthouse 1991 showed large improvements for 67-year-olds in two of five primary mental abilities, verbal and reasoning skills, over this 32-year period.
Primarily because of data limitations, to date no study has attempted to analyze national trends in cognitive functioning among older Americans. Such trends are of interest for several reasons. First, because cognitive deficits have a strong relationship with physical disability (
Carlson et al. 1999;
Kasper 1990;
Leveille et al. 1998), an investigation of cognitive trends may provide important insights into the existing literature on national disability trends. Second, severe cognitive impairment has been linked to differential patterns of medical and long-term care service use, with older impaired persons delaying care and substituting more expensive hospital care for physician visits (
Binder and Robins 1990;
Kasper 1995;
Larson, Kukull, and Katzman 1992); receiving more home health visits with, on average, longer hours per visit (
Coughlin and Liu 1989;
Kasper 1995); having greater numbers of long-term caregivers (
Kasper 1990); and entering nursing homes at higher rates (
Coughlin and Liu 1989). Thus, an investigation of recent shifts in cognitive functioning can provide important insights for predicting the future demand for medical and long-term care and costs associated with such care. Third, those with severe cognitive impairment living in the community are some of the most vulnerable members of society, at high risk for economic exploitation as well as physical neglect (
Larson et al. 1992), so tracking the size of this group and changes in who may be at risk for such abuses is of value in and of itself.
Our purpose in this study was to investigate national-level changes in the prevalence of severe cognitive impairment among older Americans. Drawing upon the first wave of the Asset and Health Dynamics of the Oldest Old study (AHEAD93) and the fourth wave of the Health and Retirement Survey (HRS98), we examined aggregate changes between 1993 and 1998 in the percentage of the noninstitutionalized population aged 70 and older with severe cognitive impairment. We also explored whether such aggregate changes can be attributed to recent shifts in the demographic and socioeconomic composition of the older population and to changes in selected health-related factors.
|
Defining and Measuring Cognitive Impairment |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
, p. 228). Studies of cognitive functioning across the life span often invoke a two-process categorization, contrasting fluid-like cognitive mechanics (e.g., processing speed, reasoning, memory) with crystallized pragmatics (e.g., knowledge, fluency;
Baltes 1993). Such distinctions are often blurred in the practice of measuring cognition, because many cognitive tasks employ both sets of skills (
Baltes and Baltes 1990). Recalling vocabulary, for example, requires pragmatic knowledge of a word as well as mechanical retrieval from memory.
An alternative way of conceptualizing cognitive impairment is found in the discussion of dementia in the most recent Diagnostic and Statistical Manual of Mental Disorders (DSM-IV;
American Psychiatric Association 1994). Although sometimes used interchangeably in the social-gerontological literature, cognitive impairment and dementia are not synonymous. Cognitive impairment is most often a manifestation of dementia or delirium, although a substantial proportion of cases with severe cognitive impairment do not present with a mental disorder (
Folstein et al. 1985). In contrast, dementia refers to a class of diagnoses based on clinical criteria including, but not limited to, cognitive deterioration.
Irrespective of these distinctions, deterioration of cognitive functioning, particularly of memory systems, is a hallmark of dementia and is therefore given considerable attention in the DSM-IV. The manual distinguishes five dimensions of cognitive impairment: memory impairment (both short and long term), aphasia (deterioration of language function), apraxia (impaired ability to execute motor activities), agnosia (inability to recognize objects), and disturbances in executive functioning (inability to think abstractly and process information). Poor spatial orientation and poor judgment are also identified as commonly accompanying declines in these areas because of dementia.
Researchers have developed numerous scales to capture these aspects of cognitive impairment in clinical settings (for a review, see
Teresi and Evans 1996). Perhaps the most widely used of these is Folstein's (
Folstein, Folstein, and McHugh 1975) Mini-Mental State Examination (MMSE), which taps each of the five dimensions of cognitive functioning in the DSM-IV and spatial orientation. Generally such measures, and the MMSE in particular, are more valid and reliable in identifying the severely cognitively impaired; others have reported considerable measurement error when such measures are used to identify the moderately impaired population (
Henderson 1992;
Kasper 1990). Estimates of severe cognitive impairment based on such scales among the older community-dwelling population have consistently ranged from 5% to 7% (
Kasper 1990).
Although many of these instruments require face-to-face contact with the patient, researchers have developed several instruments with the express purpose of screening for cognitive impairment on the telephone. The Telephone Interview Cognitive Screen (TICS), for example, which is based on Folstein's MMSE, measures short- and long-term memory, aphasia, apraxia, and disturbances in executive functioning, with the scale heavily weighted toward memory impairment. The original TICS consisted of 11 items with a maximum score of 41 (
Brandt, Spencer, and Folstein 1988). The scale has high reliability and excellent predictive validity in terms of identifying persons with clinically diagnosed dementia (
Brandt et al. 1988). Most often a cutoff of 30 or less is used as an indicator of cognitive impairment. The instrument has been modified and adapted in several population-based surveys, including the AHEAD93 and the HRS98; their version (described in the Methods section) emphasizes memory impairment and disturbance in executive function.
To avoid omitting some of the most cognitively impaired, community-based studies of cognitive functioning often allow informants (or proxies, in survey terminology) to report about the cognitive functioning of sampled persons (
Herzog and Rodgers 1999;
Kasper 1990). Researchers have developed a separate set of indicators to elicit the cognitive status of older individuals from such informants. Jorm (
Jorm 1994,
Jorm 1996;
Jorm and Jacomb 1989), for example, developed an informant-based scale for measuring cognitive decline in elderly persons, which measures changes in short- and long-term memory, aphasia, apraxia, agnosia, deterioration in executive functioning, and judgment. The scale correlates well with the MMSE in identifying the severely impaired population (
Jorm, Scott, Cullen, and MacKinnon 1991). Other studies, including the AHEAD93 and HRS98, have developed questions asking proxies to rate the cognitive performance (e.g., memory, judgment) of another person. These types of brief, informant-based measures have good predictive validity of clinical diagnoses of dementia and the MMSE (
Jorm 1996).
|
Influences on Cognitive Functioning |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
) and whether the analysis is of cross-sectional levels or longitudinal change (
Hultsch, Hertzog, Small, McDonald, and Dixon 1992). Studies showing that age effects are shared across a wide range of cognitive domains have been interpreted as evidence of broad age-related changes at the neurophysiological level (
Salthouse 1998;
Salthouse and Czaja 2000). Others (e.g.,
Starr, Deary, Inch, Cross, and MacLennan 1997) have argued that much of decline in old age actually reflects the association of age and disease.
Low educational attainment is another factor frequently associated with cognitive decline or poor performance on cognitive tests in old age, although the basis for the relationship is not clear (
Salthouse 1991;
Stern and Carstensen 2000). It could be that formal education results in cognitive practice that may affect later ability. Other elements of early and midlife experience, such as childhood socioeconomic status (which might influence opportunities beyond education), education of parents (which might reflect genetic endowment or childrearing practices), and adult occupation (which might reflect continued cognitive practice as well as environmental factors) may be related to both education and late-life cognitive functioning. However, it could be that these socioeconomic factors may simply serve as proxy measures of early health experiences and care. Moreover, in the case of educational attainment, the causation could run in the opposite direction, with early cognitive deficits relating to less schooling, which is then associated with poorer cognition later in life (
Gurland et al. 1997).
Another relationship frequently noted in the literature on cognition is with race and ethnicity. In the United States, older Blacks and Hispanics typically do worse on cognitive tests than do non-Hispanic Whites (
Albert et al. 1995;
Herzog and Wallace 1997;
Whitfield et al. 2000). A number of hypotheses have been put forth to explain such cultural differences (
Stern and Carstensen 2000). The association may reflect, for example, socioeconomic differences and differential susceptibility to diseases that affect cognitive functioning. In younger populations, the power of stereotypes has been shown to influence test scores (
Steele 1997;
Steele and Aronson 1995), and this influence may operate in older populations as well. It may be that older persons' performance is also influenced by cultural differences in attitudes toward aging and in culturally related differences in neurobiological processes developed early in life (
Park, Nisbett, and Hedden 1999). There is also evidence that racial and ethnic differences may be explained in part by bias in screening measures (
Mulgrew et al. 1999;
Teresi et al. 1995).
Sex differences in cognitive functioning in old age have also been identified, but the differential varies by aspect of cognition (
Hultsch et al. 1992). Little attention has been devoted to differentials by marital status, although it might be hypothesized that the social interaction associated with marriage might be beneficial to cognitive functioning (
Dixon and Gould 1998).
Health factors
Self-rated health and specific diseases have been associated with cognitive functioning, although even more than in the case of demographic and socioeconomic factors, it may be difficult to sort out causality. For example, depression is frequently associated with cognitive impairment, but as
Parmelee, Kleban, Lawton, and Katz 1991 have found, part of the link may be due to depression resulting from distress about cognitive decline. However, there are specific diseases such as stroke that have a clear physiological link to functioning. It is also plausible that abnormal glucose levels associated with diabetes might affect function, in particular memory (
Kumari, Brunner, and Fuhrer 2000), but population studies provide only mixed evidence depending in part on the aspect of functioning (
Strachan, Deary, Ewing, and Frier 1997), and it could be that some of the association is due to other diabetes-related disorders. Evidence regarding the association of hypertension, cardiovascular disease, or chronic obstructive pulmonary disease and cognitive performance is also mixed (
Zelinski, Crimmins, Reynolds, and Seeman 1998). Treatments, especially medications, for these and other diseases and conditions may also affect cognitive function (
Luisi, Owens, and Hume 1999).
Finally, cognitive functioning among older people has been shown to be highly correlated with sensory functioning, particularly vision and hearing (
Baltes and Lindenberger 1997;
Lindenberger and Baltes 1994;
Salthouse, Hancock, Meinz, and Hanbrick 1996;
Stein and Bienenfeld 1992). Once again, causality is not clear. It could be that cognitive decline affects sensory performance or that sensory impairment affects cognitive function, or at least performance, on tests. Alternatively, sensory and cognitive functions may be similarly affected by some unmeasured factors, or they may be part of a common neurobiological system (
Baltes and Lindenberger 1997;
Salthouse et al. 1996).
Testing factors.
Two other factors that may influence performance on cognitive functioning tests in population-based longitudinal surveys are mode of interview—in person or by telephone—and number of times the respondent has been tested before.
Herzog and Rodgers 1999 found no mode effects using experimental data from the second and third waves of AHEAD. With respect to training, there is evidence that several hours of training can improve the performance of older adults on tests of fluid, mechanical intelligence (
Baltes 1993). Evidence from the HRS for the population aged 51–61, however, indicates that among that age group learning effects as a result of multiple interviews are relatively small (
Herzog and Rodgers 1999).
|
Methods |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
; the HRS98 is described in detail at www.umich.edu/~hrswww/. Here we review the salient common features of the two surveys and emphasize important differences relevant to our analysis. The first wave of AHEAD (AHEAD93) included a nationally representative sample of initially noninstitutionalized persons born in 1923 or earlier (aged 70 and older in 1993). Sampled persons were re-interviewed in 1995 and again in 1998. The HRS began in 1992 with a cohort of noninstitutionalized persons born in 1931–1941 (aged 51–61 in 1992). Sampled persons and their spouses have been re-interviewed at 2-year intervals. In the fourth wave (the HRS98), the HRS and AHEAD surveys were merged and missing cohorts added to form a uniform survey of individuals aged 51 and older in 1998. Thus, the first wave of AHEAD and the fourth wave of the HRS provide nationally representative cross-sections of the noninstitutionalized population aged 70 and older in 1993 and 1998, respectively.
The AHEAD93 and HRS98 share several essential design features that make comparisons across the samples possible. Because the overarching purpose of both studies is to provide an understanding of the implications of health dynamics in late life for transitions in economic well-being and for reliance on public and private support systems, large sections of the questionnaires are comparable across surveys. In addition, in both surveys, households were selected on the basis of a multistage area probability sample plan, with oversampling of Blacks, Hispanics, and Florida residents. Weights provided for both years compensate for the differential probabilities of selection and are poststratified to the Current Population Survey (CPS) by age, sex, race, ethnicity, and marital status groups to account for differential nonresponse. Response rates were approximately 80% in the first wave of AHEAD. Early calculations suggest that response rates to the HRS98 are of a similar level.
Four potentially important differences between the surveys are worth noting. First, AHEAD93 is a survey of the noninstitutionalized population, but the HRS98 includes sampled persons from earlier waves who are in an institution at the time of the interview. To make the samples comparable, we excluded the institutionalized respondents in 1998 and adjusted the weights of remaining respondents so that the weighted noninstitutionalized population matches poststratification totals of the July 1998 CPS.
Second, important changes in the preferred mode of interview took place between the two surveys. In AHEAD93, the preferred mode for respondents 80 years and older was a face-to-face interview, whereas younger respondents were generally interviewed by telephone. In HRS98, the preferred mode for respondents 80 years and older and those interviewed for the first time (including most 70–74-year-olds) was a face-to-face interview, and other respondents were generally interviewed by telephone. Consequently, 61% of AHEAD93 respondents had a telephone interview whereas only 51% of HRS98 respondents did so. In both years, actual mode assignment was determined by a combination of factors including the health status of the respondent, with those interviewed in person generally in worse health than those interviewed by telephone. However, a randomized design to test mode effects (built into the second and third waves of AHEAD) has suggested that mode itself has no significant effect on cognitive functioning scores (
Herzog and Rodgers 1999). Given this finding and the possibility that mode in part reflects cognitive functioning (rather than vice versa), we omitted mode from our models. Sensitivity analyses (not shown) suggested our results are robust to the exclusion of this variable.
Third, the procedures according to which proxy respondents were recruited differed slightly between the two surveys. In the AHEAD93 interviewers were instructed to attempt to obtain interviews with the sampled person whenever possible. Individuals unable to complete the survey because of a physical or cognitive limitation (and less often those unavailable or with language difficulties) were allowed to have proxy informants report for them. In the HRS98 the same general approach was followed but an additional criterion was implemented: Those who scored less than a threshold score on a portion of the cognitive test triggered a signal to the interviewer that either a proxy should be sought or someone should be asked to assist the respondent for the remainder of the interview. In 27 cases in our sample, the proxy interview was completed, and proxy answers superseded responses provided by the sampled person (excluding their cognitive test scores). To avoid introducing bias into our comparisons, we used the cognitive test scores of these 27 respondents and classified them as self-respondents.
Finally, the surveys differ with respect to potential learning effects. Nearly three fourths of respondents to the HRS98 answered the cognitive tests in a prior interview, whereas no AHEAD93 respondents had the opportunity to do so (because AHEAD93 was the baseline wave of AHEAD). Although we were not able to control for both year and prior tests in our models, we did explore the effect of prior tests on the risk of severe cognitive impairment among self-respondents in 1998. We found that, controlling for other demographic, socioeconomic, and health-related factors, having one or more prior tests had only a small effect on the odds of being severely impaired (OR = 0.83) and this effect was not statistically different from 1.0 (95% CI = 0.43–1.6).
We used the public release version of the AHEAD93 and the preliminary release of the HRS98. The HRS98 data have not been cleaned and may contain errors that will be corrected in the final public release version of the data set. Sample sizes for the population aged 70 and older were 7,443 in the AHEAD93 and 7,624 in the HRS98.
Measures
Cognitive functioning.
The AHEAD93 and HRS98 surveys included identical tools to measure the cognitive functioning of sampled persons. In both surveys, researchers developed separate modules to gather information on the cognitive status of persons who answered for themselves (hereafter, self-respondents) and those who had responses provided about them by a proxy (hereafter, proxy respondents).
Questions to elicit the cognitive functioning of self-respondents in the AHEAD93 and HRS98 were from a modified version of the TICS instrument (
Brandt et al. 1988). The cognitive tests appeared in a slightly different order in the AHEAD93 and HRS98 but included the same items. In both years there were six tasks yielding a maximum of 35 points, with higher scores implying better functioning. Four memory and two executive functioning tasks were included: (a) an immediate recall test, which required recalling as many words as possible from a list of 10 commonly used words immediately after the interviewer read the list (10 points); (b) a delayed recall test of the same list, approximately 5 min after the list was read (10 points); (c) naming the day of the week and the date (i.e., month, day, and year; 4 points); (d) naming the object that people usually use to cut paper, the kind of prickly plant that grows in the desert, the president of the United States, and the vice president (4 points); (e) a serial 7s test that required subtracting 7 from 100 five times (5 points); and (f) counting backwards from 20 for 10 consecutive numbers (2 points). The major difference between the TICS and the AHEAD/HRS cognitive batteries is that the latter included an additional delayed word recall test and excluded tests of aphasia (asking respondents to repeat phrases), apraxia (asking respondents to tap on the phone), and a test of naming opposites. Thus, the AHEAD93 and HRS98 questions provided measures of both short- and long-term memory and executive functioning (serial 7s and counting backward), with the scale much more heavily weighted toward measuring memory impairments. In terms of the dichotomy between fluid and crystallized cognition, the scale probably taps both dimensions but is more heavily weighted toward measures related to fluid cognitive mechanics.
Missing information was handled in an identical fashion in both years. For those who refused an entire task, we followed
Herzog and Wallace 1997 suggestion of assigning those who refused the immediate recall task a score of 2 out of 10, those who refused the delayed recall a score of 0 out of 10, and those who refused the serial 7s a score of 1 out of 5. This imputation was based on Herzog and Wallace's finding that those who refused a task performed worse on other subscales than those who gave responses for all subscales. For those who refused or reported that they did not know a specific item within a subscale, we assigned the item a score of 0. We also explored alternative codings for nonresponse (e.g., assigning a score of 0 to subscales with missing items, deleting cases with missing items).
We used the total score from the six tasks to identify those with severe cognitive impairment. Because there is no accepted cutoff score for identifying those with severe cognitive impairment, we looked to the literature for guidance. The original TICS used 30 (out of 41) as a cutoff, but that scale did not include the more difficult delayed recall task. For the cognition scale in the AHEAD93 survey,
Herzog and Wallace 1997 suggested that a cutoff of 8 (out of 35) identifies a percentage of the population consistent with generally observed estimates of severe cognitive impairment. We adopted this latter suggestion in both years but also explored the sensitivity of our analysis to this definition.
We based our classification of those with proxy respondents on reports about the sampled person's memory and judgment. Proxies were asked "How would you rate [respondent's] memory at the present time? Would you say it is excellent, very good, good, fair, or poor?" and "How would you rate [respondent] in making judgments and decisions? Would you say he/she is excellent, very good, good, fair, or poor?" Among proxy respondents, we defined severe cognitive impairment as those who were reported to have poor memory and poor judgment. We also explored alternative coding schemes (e.g., weighting poor memory more heavily to be consistent with the self-respondent scale, using only poor memory to identify severe impairment).
Independent variables.
Our analyses were intended to address the question of whether aggregate changes between 1993 and 1998 in cognitive functioning among older persons are statistically significant, and if so, whether they remain significant after controlling for demographic and socioeconomic shifts and changes in selected health-related characteristics. The key variable of interest in our analyses was an indicator of survey year (i.e., coded 1 for respondents in the HRS98 and 0 for respondents in the AHEAD93).
In selecting control variables for this analysis, we aimed to include only questions that were identical in the AHEAD93 and HRS98. Nevertheless, we were able to control for many demographic, socioeconomic, and health-related factors.
Demographic variables included age (in 5-year age groups), sex, race (distinguishing non-White and White), Hispanic origin, and marital status (not married at the time of the survey vs currently married). Socioeconomic variables included a measure of whether the sample person or his or her spouse had any liquid assets (defined as a checking or savings account or money market funds) and completed level of education (less than high school, high school graduate, more than high school). We also explored the effects of finer gradations of education (e.g., including an indicator of completing less than 8 years of education as distinct from 8–11 years of completed education; distinguishing some college education from completing college; using a continuous indicator of years of completed education), but found our conclusions robust to such alternative parameterizations. In an attempt to control for intergenerational transmission of cognitive abilities, we also included an indicator of the sampled person's mother's educational level (<8 years, 
8 years, unknown). In doing so, we made no attempt to distinguish genetic correlations from environmentally based intergenerational transmission in early life, but we recognized that either could contribute to functioning.
We included three measures of health related to cognitive functioning: ever having a stroke, having poor eyesight, and having poor hearing. We also investigated numerous other chronic conditions (e.g., having ever had high blood pressure, diabetes, cancer, chronic lung disease, heart condition, arthritis) and lifestyle factors (e.g., current smoking status), but they generally did not have statistically significant effects on cognitive functioning nor did they change our conclusion about aggregate changes in cognitive functioning. We therefore excluded these variables for the sake of parsimony. We also explored the linkages among self-rated health, depression, and severe cognitive impairment but chose not to include these measures because of the uncertainty of the direction of their effects in cross-sectional samples (for discussion of this point with respect to depression, see
Parmelee et al. 1991).
Statistical Methods
Because the measures of severe cognitive impairment differed for self- and proxy respondents, we stratified all analyses by proxy status. This approach explicitly controls for the effects of proxy status while allowing for potentially different thresholds for identifying the severely impaired between the two groups.
We tested for changes over time in the percentage with severe impairment (among self-respondents and among proxy respondents) using chi-square tests for independence. Among self-respondents we also used t tests to assess changes over time in the mean score for each of the subscales and for the total cognitive score. We tested for changes in sample characteristics over time and changes in the percentage with cognitive impairment among various subgroups of self- and proxy respondents using chi-square tests for independence.
To explore whether changes in severe impairment were due to shifts in the composition of the population, we used logistic regression. We first estimated odds ratios and 95% confidence intervals from models including only a term for year (1998 vs 1993). We then estimated odds ratios and 95% confidence intervals from models controlling for demographic and socioeconomic characteristics, health-related factors, and testing factors. Statistical significance of differences between self- and proxy respondents were obtained from a single model in which proxy status was interacted with all covariates.
The AHEAD93 and HRS98 employed different sampling fractions for the youngest sample members (in the HRS98, persons aged 70–74 were sampled at 60% the rate of those in the AHEAD93). We therefore weighted all analyses.
Because the HRS98 is a preliminary release, information on sampling units has not yet been made publicly available. Consequently, it was not possible to adjust standard errors for the complex design of the surveys. All p values and confidence intervals presented were therefore based on the assumption that the AHEAD93 and HRS98 are simple random samples. We were able to explore the implications of this assumption for our analysis using design effects estimated for the AHEAD93 (for which sampling unit information is available). We found for self-respondents the design effects for the variables in our models ranged from 0.54 to 1.82; similarly, for proxy respondents the design effects ranged from 0.57 to 1.66. (We calculated these effects by taking the ratio of the variances produced by SUDAAN to the variances produced by SAS for the full logistic regression models.) Even if we assumed the design effects for changes over time were the size of the largest AHEAD93 design effect (i.e., 1.82), the overwhelming majority of tests significant at the .05 level in our tables remained so. Moreover, unless otherwise noted, our conclusions about aggregate changes over time in cognitive functioning remained robust to such design effects.
Finally, we explored an additional adjustment to compensate for the fact that many respondents (N = 4,812) appeared in both the AHEAD93 and HRS98 samples. We calculated these overlap effects, which increase the efficiency of estimates of changes over time, separately for the self- and proxy respondent samples using
Kish 1965(pp. 457–458) formula for variances of differences between two overlapping samples. In this application, the overlap effects were 0.81 and 0.90 for self- and proxy respondents, respectively. Applying these overlap effects to the design-effect adjusted variances for year in our logistic regression models did not change our conclusions about aggregate changes over time in cognitive functioning.
|
Results |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
|
Improvements among self-respondents were also unlikely to be accounted for by increases over time in the proportion of proxy respondents; on the contrary, the proportion of proxy respondents declined slightly from 10.5% in 1993 to 9.1% in 1998 (p < .01).
Improvements among self-respondents persisted irrespective of the coding scheme for refusals (see Table 2 ). Although the level of severe impairment changed depending on whether refusals were given low scores, assigned a zero, or deleted altogether, the percentage classified as severely impaired consistently declined over time, while the mean subscale and total scores increased.
|
|
|
|
|
A smaller set of variables was important in predicting severe cognitive impairment among proxy respondents. Moreover, many of those predictors had different effects for proxy respondents than for self-respondents, as indicated by the chi-square tests in the final column of Table 6 . For example, among the proxy respondent sample, those with less than a high school education were less likely than others to be described as having poor memory and judgment, whereas effects were just the opposite among self-respondents. Being non-White and having no liquid assets had effects only among self-respondents, but having poor vision had an effect only in the proxy sample.
|
Discussion |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
Our finding that improvements occurred among self-respondents but not proxy respondents is noteworthy. In both years, it appears that a core group of respondents could not answer for themselves because of severe cognitive impairment and thus participated in the survey through proxy respondents. Moreover, in both years, those for whom a proxy responded constitute a select group, less healthy according to measures we included, more likely to be from a minority population, and more likely to be of low socioeconomic status than those responding for themselves. It may be that the severity of impairment declined among the proxy respondent group, but our proxy-provided measures of severe impairment for this group may not be sensitive enough to identify subtle changes. Notwithstanding this potential weakness, our results suggest that the improvements we find among self-respondents are not due to a change in the proportion of proxy respondents over time.
How do our results compare to other studies of trends in cognitive functioning? Such comparisons are difficult to make, even for a single year, given that existing studies differ from ours with respect to years studied, age groups, and measures of cognitive impairment. Nevertheless, our findings agree with the general pattern of historical improvement summarized by
Salthouse 1991. For example, on the basis of
Schaie 1983, whose sample of adults aged 25–67 is biased toward upper socioeconomic groups,
Salthouse 1991 reported improvements from 1959 to 1991 across five primary mental abilities of, on average, 0.07 standard T-score units (a distribution with M = 50 and SD = 10) per year. Limiting this exercise to those in the upper age range for dimensions with significant improvement (verbal and reasoning skills), we calculate improvements on the order of 0.20 standard T-score units per year. Among self-respondents in our sample, who were aged 70 and older, we find much larger improvements: Mean scores increase from 19.6 in 1993 to 21.1 in 1998, estimates comparable to improvements of 0.55 standard T-score units per year.
The improvements we find are also somewhat larger in percentage terms than those observed during the 1980s and early 1990s for measures of physical disability, which have ranged roughly from 1% to 2% per year. Our estimates of decline are much larger: For the noninstitutionalized population aged 70 and older, relative annual declines are on the order of 8% per year over the 5-year period. In making such comparisons it is important to keep in mind that the base for calculating percentage change is much smaller for cognitive than for physical impairment, so even small percentage point improvements (2.5 percentage points in this case) appear to be quite large gains (more than 40% improvement over 5 years).
What could be causing such large improvements? Underlying the decline in severe cognitive impairment we observe are undoubtedly complex shifts in the incidence of and length of time lived in the community with severe cognitive impairment, the latter being a function of recovery rates, nursing home admission and discharge rates, and mortality rates. Although data are not yet available for us to calculate the relative contribution of these components to aggregate changes in cognitive impairment, our analysis permits us to speculate about several potential explanations for the shifts we observe. The first two possible explanations relate to survey design and selection issues; the remaining discussion focuses on potential environmental changes related to medicine, behavioral lifestyle factors, and cultural factors.
With respect to survey design matters, one possibility is that older persons with severe cognitive impairment are more likely in 1998 than in 1993 to be in institutional settings (and hence outside the scope of our analysis). However, there is evidence that the size of the nursing home population may have declined slightly as a percentage of the older population during the late 1980s and early 1990s (
Bishop 1999), so this explanation is unlikely. It may be that within nursing homes, severely cognitively impaired residents represent an increasing share of the residents, a hypothesis consistent with the increase in special care units devoted to Alzheimer's disease and related dementias during this period (
Rhoades and Krauss 1999). Our calculations suggest, however, that to fully explain the improvements we find the percentage of severely impaired residents in nursing homes would have had to increase from 40% to 100% over this 5-year period, an increase we believe is implausible, given estimates from the early 1990s of 60–70% of nursing home residents having dementia-related symptoms (
Rovner and Katz 1993).
Another possible explanation related to survey design is that those with cognitive limitations may have been more likely than others to have been lost to follow-up over time. At this writing, accurate data on nonresponse to the HRS98 are not yet available, so we explored this issue using the second wave of AHEAD. We found that among respondents to the first wave of AHEAD, 4.2% of those followed up in 1995 in the community were severely cognitively impaired in 1993, whereas 8.7% of those lost to follow-up (but not dead) in 1995 were severely impaired at baseline. We also found that controlling for variables used to poststratify the data (e.g., age, sex, marital status, race/ethnicity) did not fully explain these differentials. We interpreted these results for 1993 and 1995 as evidence that poststratification to the CPS in 1998 probably accounts for some but not all of the differential loss to follow-up by cognitive status. On the basis of these findings, we calculated what the rate of severe cognitive impairment might have been in 1998 had there been no loss to follow-up in the AHEAD between 1993 and 1998. In making these calculations we assumed that the ratio of deaths to missing cases was the same in 1998 as it was in 1995 (roughly 2:1), which yielded a nonresponse rate of approximately 11%. We also assumed the rate of severe impairment in 1998 among those who were lost to follow-up was equal to the rate in 1993 for those eventually lost to follow-up (8.7%). We found that, under these assumptions, the rate of severe impairment would have been 4.1% in 1998, rather than the 3.6% we found for that year. The revised aggregate change—from 6.1% in 1993 to 4.1% in 1998, or on average a decline of 7% per year—would likely still be statistically significant for a sample of this size.
Assuming survey design issues do not fully explain the improvements we are finding, are there environmental factors that may be responsible? Certainly changes in the treatment of dementia and related co-morbid conditions such as depression, stroke, and nutritional deficits (e.g., electrolyte imbalances) may be contributing to the better maintenance, and in some cases recovery, of cognitive functioning. In additional analyses limited to respondents appearing in both the AHEAD93 and HRS98, we observed roughly 60% of those classified as having severe impairment in 1993 were not severely impaired in 1998. Moreover, among respondents completing the cognitive test in both years, approximately half scored the same or better in 1998 than they did in 1993. Such individual-level improvements are consistent with the growing array of treatment options for older Americans with cognitive deficits and related co-morbidities, but with these data it is difficult to sort out specific treatments related to the improvements we observe. For example, the approval in the late 1990s of new drug treatments for Alzheimer's disease (e.g., Donepezil in 1997) might be expected to improve functioning (
McRae 1999); however, such treatments are generally limited to those in the early stages of Alzheimer's rather than those with severe cognitive impairment (the group we focused on in our analysis). Improvements in stroke treatment may be partly responsible; however, our findings suggest that changes in ever having a stroke explain little of the decline we observe in severe cognitive impairment. Improvements in cognition may also be the result of the expanded availability of treatments for depression; indeed, in exploratory analyses we found aggregate-level declines in the presence of depressive symptomology among self-respondents, and others have shown strong correlations between individual-level cognition and depression scores (
Parmelee et al. 1991). Analysis of panel data is required to untangle the complex relationship among treatment protocols, cognition, and co-morbidities, a topic that clearly needs further investigation.
Changes in other important behavioral and lifestyle factors may also be contributing to better maintenance of cognitive functioning in old age. We looked at some lifestyle factors (e.g., current smoking) but did not find any significant effect on the prevalence of severe cognitive impairment. Over this period, estrogen use, which may affect cognitive functioning—particularly memory—in postmenopausal women (
Shaywitz and Shaywitz 2000), also increased dramatically (
DuBois, Chawla, Neslusan, Smith, and Wade 2000). Our finding that both men and women experience improvements in cognitive functioning suggests estrogen use alone cannot fully explain observed improvements. Further research on the link between these and other potentially important behavioral and lifestyle factors (e.g., dietary habits, vitamin supplementation, exercise) and cognitive functioning in old age could prove to be illuminating.
A final possibility is that there may be broad social changes at work that are not captured by our models. Others have suggested that changes in the quality of education (e.g., in terms of curriculum, class size, teacher preparation) could account for some of the time-related improvements in cognitive ability, although there is little agreement on direction of such changes, and evidence based on school-aged children does not support such an argument (
Salthouse 1991). The fact that the improvements we observed were most salient among those with less than a high school education draws attention to the possibility that cultural forces outside the classroom may be at work. For example, changes in the culture of testing outside the classroom may make more recent cohorts of elderly persons—particularly those with less than a high school education—better equipped to answer cognitive tests. Others have noted the conflicting views on the role of multimedia and information technology on the minds of both young and old (Baltes's Foreword in
Schaie 1996), but our results suggest, on balance, no net negative effect for older Americans of the recent technological explosion in our society.
Irrespective of the underlying causes, this study provides a first step in understanding how cognitive functioning may have changed recently among older Americans. The preliminary evidence is quite positive: As a group, older persons, especially those well into their 80s, appear to have better cognitive functioning today than they did in the early 1990s. Although we have only two data points, making it impossible to draw conclusions about long-term trends, these findings suggest that the improvement in physical functioning observed in recent years among older Americans may have been accompanied by improvements in cognitive functioning. How the two trends interrelate is unclear. It may be that improved cognitive functioning accounts for much of the decline in IADL disability observed during the 1980s and early 1990s; alternatively, improvements in underlying health may be driving both physical and cognitive functioning. Untangling the complex relationship between improvements in the mind and in the body in old age is a fruitful area for further research.
A recent report by the National Research Council (
Stern and Carstensen 2000) on future directions for cognitive research on aging highlighted the plasticity of the human mind, even well into old age. Indeed, one of the most widely known models of cognitive functioning in old age, Baltes's model of selective optimization with compensation (
Baltes and Baltes 1990), describes a highly adaptive phenomena in which the cognitive processes of older persons can be enhanced under favorable environmental conditions. Our findings suggest that cognitive ability should also be considered a dynamic phenomenon at the aggregate level. This conclusion suggests in turn that policymakers interested in projecting medical and social service needs may well consider how future trends in cognitive functioning might affect demand for acute and long-term care services.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication July 10, 2000. Accepted for publication October 23, 2000.
|
References |
|---|
|
TOPAbstractDefining and Measuring Cognitive...Influences on Cognitive...MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  M. A. M. Rogers, B. L. Plassman, M. Kabeto, G. G. Fisher, J. J. McArdle, D. J. Llewellyn, G. G. Potter, and K. M. Langa Parental Education and Late-life Dementia in the United States J Geriatr Psychiatry Neurol, March 1, 2009; 22(1): 71 - 80. [Abstract] [PDF]
|
|
|
|
 |
|
 M. G. Parker and M. Thorslund Health Trends in the Elderly Population: Getting Better and Getting Worse Gerontologist, April 1, 2007; 47(2): 150 - 158. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Mejia, A. Miguel, L. M. Gutierrez, A. R. Villa, and F. Ostrosky-Solis Comparative Analysis of Cognitive Impairment Among Mexicans and Spanish-Speaking Immigrant's Elders J Aging Health, April 1, 2006; 18(2): 292 - 314. [Abstract] [PDF]
|
|
|
|
 |
|
 R. F. Schoeni, L. G. Martin, P. M. Andreski, and V. A. Freedman Persistent and Growing Socioeconomic Disparities in Disability Among the Elderly: 1982-2002 Am J Public Health, November 1, 2005; 95(11): 2065 - 2070. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. G. Parker, K. Ahacic, and M. Thorslund Health Changes Among Swedish Oldest Old: Prevalence Rates From 1992 and 2002 Show Increasing Health Problems J. Gerontol. A Biol. Sci. Med. Sci., October 1, 2005; 60(10): 1351 - 1355. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. M. Agree, V. A. Freedman, J. C. Cornman, D. A. Wolf, and J. E. Marcotte Reconsidering Substitution in Long-Term Care: When Does Assistive Technology Take the Place of Personal Care? J. Gerontol. B. Psychol. Sci. Soc. Sci., September 1, 2005; 60(5): S272 - S280. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
V. A. Freedman, H. Aykan, and L. G. Martin Another Look at Aggregate Changes in Severe Cognitive Impairment: Further Investigation Into the Cumulative Effects of Three Survey Design Issues J. Gerontol. B. Psychol. Sci. Soc. Sci., March 1, 2002; 57(2): S126 - 131. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
