| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
a Rush Alzheimer's Disease Center and Rush Institute for Healthy Aging and Departments of, Rush-Presbyterian-St. Luke's Medical Center, Chicago, IL
b Neurological Sciences, Rush-Presbyterian-St. Luke's Medical Center, Chicago, IL
c Psychology, Rush-Presbyterian-St. Luke's Medical Center, Chicago, IL
d Medicine, Rush-Presbyterian-St. Luke's Medical Center, Chicago, IL
e Health Care and Aging Studies Branch, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control, Atlanta, GA
Liesi E. Hebert, Rush Institute for Healthy Aging, 1645 West Jackson Blvd., Suite 675, Chicago, IL 60612 E-mail: LHebert1{at}rush.edu.
Decision Editor: Toni C. Antonucci, PhD
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
IN several cross-sectional studies of persons with Alzheimer's disease, women have shown greater impairment of language functions than men at apparently comparable levels of disease severity. Specifically, gender differences have been reported on tests of naming (Buckwalter, Rizzo, McCleary, Shankle, Dick, and Henderson 1996
; Henderson and Buckwalter 1994; Ripich, Petrill, Whitehouse, and Ziol 1995), verbal fluency (Henderson and Buckwalter 1994), and word recognition (Ripich et al. 1995). These findings are surprising because, among healthy adults, the language skills of women are generally believed to be equivalent or superior to those of men (Halpern 1997). The findings are important, if confirmed, because differential decline in language and communication skills might result in greater disability and need for care in affected women than men.
The basis for gender differences in language function in Alzheimer's disease is uncertain. It has been suggested that they might arise from gender differences in cerebral organization or in exposure to internal or external risk factors (Henderson and Buckwalter 1994). The possibility that hormonal influences are involved has been suggested, in part because there is some evidence that estrogen replacement therapy may selectively affect verbal functions (Sherwin 1998).
A longitudinal study design offers the most direct means of assessing how language functions change with disease progression. Unfortunately, only one previous study has examined the relation of gender to decline on tests of language functions (Ripich et al. 1995); no gender differences were observed, but only 23 persons participated in the longitudinal arm of the study. Researchers using measures of global cognitive function in longitudinal studies have generally not observed gender differences in rate of cognitive decline (Chui, Lyness, Sobel, and Schneider 1994; Mortimer, Ebbit, Jun, and Finch 1992; R. Stern et al. 1994; Y. Stern et al. 1994), but these results do not preclude the possibility of differential decline in selected linguistic functions.
In the present study, a large clinical sample of women and men with Alzheimer's disease had up to five annual evaluations that included multiple tests of language, memory, and perception. We used eight individual language tests to assess naming, repetition, verbal fluency, and comprehension. In analyses controlling for other demographic variables, we examined whether women experienced more rapid decline in language skills than men and if a similar or different pattern was observed for other cognitive functions. We also considered whether other conditions that impair cognition or mortality during follow-up affected the results.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
) we recruited participants from the Rush Alzheimer's Disease Center over a 12-month period using three eligibility criteria: (a) a clinical diagnosis of Alzheimer's disease (McKhann et al. 1984), (b) a score of 11 or more on the Mini-Mental State Examination (MMSE; Folstein, Folstein, and McHugh 1975), and (c) residence in the community. Of the 492 persons examined who met the criteria, 410 (83%) agreed to participate. Informed consent was obtained from participants and family members, according to procedures approved by the Institutional Review Board of Rush-Presbyterian-St. Luke's Medical Center. Baseline demographic characteristics on the 364 participants with follow-up data are presented in Table 1 .
|
) and conformed to the practice parameters recommended by the American Academy of Neurology (Quality Standards Subcommittee, 1994). The evaluation included a medical history, neurological examination, cognitive function testing, brief psychiatric evaluation, laboratory tests, and an interview with a knowledgeable informant. Magnetic resonance imaging scans were obtained for persons without a brain scan within the past 12 months or if otherwise clinically indicated. The evaluation was repeated at annual intervals for up to 4 years by examiners blind to previously collected data. The interval from first examination to each subsequent examination was measured in days, then divided by 365.25 to provide the analytic variable study time in units of years.
Diagnosis of Alzheimer's Disease
The diagnosis of Alzheimer's disease was based on the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association criteria (NINCDS/ADRDA; McKhann et al. 1984) and required a history of cognitive decline and evidence of impairment in memory and at least one other cognitive domain. In 30 persons, the neurologist identified an additional condition, such as stroke, that was believed to contribute to cognitive impairment. In NINCDS/ADRDA terminology, Alzheimer's disease was possible in these people and probable for the remaining 380 participants. As reported elsewhere (Wilson et al. 2000), of the 54 study participants who died and underwent brain autopsy, 52 (96%) met pathological criteria for Alzheimer's disease (Mirra, Hart, and Terry 1993).
Assessment of Language and Cognitive Function
Eighteen tests were administered as part of each evaluation. There were eight language tests, nine tests of memory or perception, and one measure of global cognitive function, the MMSE. We selected the language tests to assess functions typically impaired in Alzheimer's disease (Nebes 1989): naming, repetition, comprehension, and fluency. There were two measures of naming: a 15-item version of the Boston Naming Test (Morris et al. 1989) and Responsive Naming (Goodglass and Kaplan 1983). There were three measures of repetition: Word Repetition, High Probability Phrase Repetition, and Low Probability Phrase Repetition (Goodglass and Kaplan 1983). Verbal comprehension was assessed with Commands and Body Part Identification (Goodglass and Kaplan 1983). Fluency was evaluated with a semantic fluency task in which participants were asked to generate as many unique animal names as possible in 1 min (Morris et al. 1989).
There were six measures of episodic memory: immediate and delayed recall of the East Boston Memory Test (Albert et al. 1991); a three-alternative, forced-choice, delayed recognition memory test for the 15 Boston Naming Test items previously presented; and the Figural Memory Test, which involved three trials in which eight abstract line drawings (Kimura 1963) were presented for study, each followed by a two-alternative, forced-choice recognition memory test. Three visuoconstructional measures were administered: Constructional Praxis (Morris et al. 1989), which required participants to copy geometric designs, and Facial Recognition Test (Benton, Sivan, Hamsher, Varney, and Spreen 1994) and Figural Recognition, which assessed perception of faces and of stimuli from the Figural Memory Test, respectively, in a match-to-sample format.
Data Analysis
The longitudinal study design required an outcome measure covering a wide range of difficulty so that we could measure change over a wide range of function without encountering floor and ceiling effects. Because most individual tests do not have this characteristic, a composite measure of language function was used in primary analyses. We formed the measure by converting raw scores from each language test to z scores based on the distribution of each test at baseline; we averaged individual z scores to yield the composite language score. Secondary analyses considered each language test individually. Additional analyses examined composite measures of memory and visuoconstruction, which were formed like the composite language measure; a global measure combining all tests; and MMSE score. Composite scores were computed if at least half of the component tests had valid scores; otherwise the composite score was missing.
We used random effects regression models (Laird and Ware 1982) to test the association of gender with decline in language function over time. This approach has been used in several other studies of change in cognitive function (Jacqmin-Gadda, Fabrigoule, Commenges, and Dartigues 1997; Teri, Hughes, and Larson 1990). The approach permits simultaneous analysis of all observations for a person, allowing individuals to have different numbers of observations and different time intervals between observations. In addition, the models explicitly allow for differences among participants in initial score and in rate of change in score.
The classic random effects model is written as follows: Y = Xß = ZA + e, where Y is a vector for each person containing the test scores for each time. Similarly, X is a matrix of the values of all the predictors at each time. The predictors in our initial model were age at baseline, educational attainment, gender, race, and study time (time since baseline in years). Each coefficient (ß) was the expected difference in score for a one-unit increment of the predictor. For example, the coefficient for study time reflected the expected change in score for each year of follow-up. To test if the rate of change over time differed by demographic predictors, we added interaction terms: Age x Study Time, Gender x Study Time, and so forth.
The model also assumed that participants differed from one another in ways not explained by the predictors. These person-to-person differences were represented in the random effects vector A. In these analyses two person-specific factors were considered: differences in participants' initial score and in rate of change per year. The design matrix Z allowed an individual's score to start higher or lower than the average starting point described by the predictors and to change faster or slower than the average rate. The two random effects for each person were assumed to have a bivariate normal distribution. The e in the model represented a vector of error terms that described how the actual measurements for an individual differed from a smooth path; the terms for an individual were assumed to be independent over time and to have a common normal distribution.
The initial model included only the predictors listed previously. In additional models we added interactions of gender with age, race, and education to test whether the gender effects might differ across demographic groups. All relationships between scores and predictors were initially assumed to be linear. We checked this assumption by using both analytic techniques (including polynomial terms in models) and graphical techniques (plotting residuals against the predictor). We used univariate and bivariate residual plots and univariate summaries of residuals to check normality and homoscedasticity of residuals. We checked independence of errors within a person by using longitudinal correlation of the repeated measures.
As a confirmation of the results for the individual language tests, for which the normality assumption was not met, we repeated the analyses using general linear models fitted by a generalized estimating equation procedure (Liang and Zeger 1986). For seven of the tests, the logistic link function and binomial error structure distribution were used because the tests had a bounded range of possible scores. For Verbal Fluency, which had no upper bound on possible scores, the log link function and Poisson error structure were used.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
The language measure combined z scores of eight tests so that the initial distribution of the composite was close to normal, with a mean of zero. Because of correlation among the individual tests, the standard deviation was slightly smaller than 1 (0.7). Fig. 1 presents scores over the study interval for 20 randomly selected women and men. Table 2 presents the average annual decline by each year of follow-up for women and men. Most people experienced substantial decline during the follow-up interval. The standard deviations were larger than differences between either gender or follow-up year. This indicates large differences among individuals in rate of decline within each follow-up interval. Although the means suggest little difference between women and men, and a possible increase in rate of decline over time, this crude summary is not an adequate way to test these questions because it cannot show the individual patterns of decline or adjust for other factors contributing to these patterns.
|
|
). We found the overlapping concepts of age at observation, age at entry, and study time were best summarized in these data using the two variables age at entry and study time, based on previous analyses (Wilson et al. 2000). The results of the current analysis are listed in Table 3 . The coefficient for the effect of gender on initial level was .02. Multiplying this by the male gender indicator value of 1 for men and 0 for women indicated that a man had a baseline score 0.02 standard units (95% confidence interval [CI] = -0.17 to 0.20) higher than a woman of the same age, race, and education. Multiplying the coefficient for the effect of male gender on change in language score (Male Gender x Study Time) by the variable value for men (1) and women (0) indicated that men declined 0.03 standard units (95% CI = -0.18 to 0.12) more each year than comparable women. By summing the products of the coefficients and the variable values of male gender, study time, and Male Gender x Study Time, we obtained the estimated declines for a woman and man of average age and education. The estimated annual decline for men (0.74 units, 95% CI = 0.61 to 0.86) was 104% of the decline among women (0.71 units, 95% CI = 0.62 to 0.79). The sum of the products of coefficients multiplied by the values for all the variables provides the predicted language score.
|
|
To assess potential differences in the gender effect among the language tests, we repeated the analyses for each of the tests separately. Differences between women and men in the decline estimated on each of the tests were small, and none reached statistical significance (Fig. 3). Individual test scores were not normally distributed (Table 4 ), so the assumptions of random effects models were not fully met. Therefore, we repeated analyses of each test using generalized linear models that permit both nonlinear relationships and nonnormal probability distributions. There was no significant gender difference in the rate of decline on any test in these analyses. (Generalized linear model results are not shown.)
|
|
|
Because other dementing conditions may affect cognitive decline, and some, such as stroke, do not occur equally often in women and men, additional analyses excluded the 20 women and 10 men with possible rather than probable Alzheimer's disease. There was still no significant association between gender and decline on any of the measures of cognitive function.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
As noted previously, past longitudinal studies of Alzheimer's disease have generally not observed gender differences in decline on global cognitive measures (Chui et al. 1994; Mortimer et al. 1992; Ripich et al. 1995; R. Stern et al. 1994; Y. Stern et al. 1994; Teri et al. 1990). One study observed more rapid cognitive decline in women than in men (Lucca, Comelli, Tettamanti, Tiraboschi, and Spagnoli 1993) in a small sample (40 women, 16 men) with a brief observation period (1 year). A similar, although nonsignificant, finding was described in a preliminary report (Kramer-Ginsberg et al. 1988); however, later reports of this study, based on a larger sample size and longer period of observation, did not find gender differences (R. Stern et al. 1994). The present results are consistent with previous longitudinal findings and extend them, by showing that specific measures of language and of other cognitive functions appear to show similar rates of decline in men and women with Alzheimer's disease.
Evidence that Alzheimer's disease affects language differentially in women compared with men comes exclusively from cross-sectional studies (Buckwalter et al. 1996; Henderson and Buckwalter 1994; Ripich et al. 1995) that observed individuals at various stages of a disease that begins gradually and progresses at varying rates. Attempts to control for disease severity in cross-sectional studies are further complicated by the substantial correlation of most severity measures with the language measure being analyzed. Therefore, it is difficult for researchers to securely assess disease effects without directly measuring change, which requires repeated measurements over time.
Longitudinal studies have the potential of overcoming these limitations, but careful study design and analyses are needed to make the most of the data. One difficulty is the broad range in function seen as the disease progresses. Many tests do not cover a sufficient range individually, so in this study we combined a number of tests of varying difficulty to produce a composite measure that covered a broader range of function. Another challenge is how to measure change in function independent of initial level. To directly measure change, at least two observations are needed, and the more observations and the longer the period of follow-up, the better the ability to characterize individual change. There were up to five annual observations of individuals in this study. With multiple observations, analytic methods that take into account within-person correlations are needed. Repeated-measures analysis of variance software usually requires the same number of observations per person, which is difficult to maintain among Alzheimer's disease patients because of mortality. The composite measures used here had the distribution characteristics permitting use of random effects models. The random effects models controlled for within-person correlations and permitted a different number of observations and differing intervals between observations for individuals, so all data could be used. These models had the ability to distinguish between associations with initial level and with change in function and had the additional advantage of modeling initial level as an individual random effect instead of the usual fixed effect.
A significant limitation of these findings is that participants in this study, as in all other relevant studies of which we are aware, were selected from a clinical setting. This may be an important restriction because many persons with Alzheimer's disease do not come to medical attention. The factors that bring affected persons to medical settings are not well understood, but gender and other potentially relevant variables (e.g., age and severity of dementia) appear to be related to this decision (Ross et al. 1997). As a result, population-based longitudinal studies are needed to establish the generality of these findings. These analyses involve a greater number of observations than many previous studies, but they do not cover the entire course of disease. Although these data do not support any shape of decline more complex than linear, they do not preclude a curvilinear decline over the entire range of disease because relatively short increments of a curve may appear as linear. Other data that include more observations over a longer period might show a changing rate of decline, as discussed elsewhere (Beckett and Tancredi 1997; Wilson et al. 2000).
These analyses suggest that rate of decline in language function and other domains of cognitive function in Alzheimer's disease does not vary by gender. However, because women survive longer than men, women with Alzheimer's disease can be expected to reach lower levels of cognitive function before death than men.
|
Acknowledgments |
|---|
We thank Cheryl Bibbs and Vanessa Alston for coordinating study operations, Kenneth Tonnissen for programming, George Dombrowski for data management, and the staff of the Rush Alzheimer's Disease Center for accommodating the study.
Received for publication March 16, 1999. Accepted for publication March 24, 2000.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  J. J. Jalbert, L. A. Daiello, and K. L. Lapane Dementia of the Alzheimer Type Epidemiol. Rev., November 1, 2008; 30(1): 15 - 34. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z. Arvanitakis, R. S. Wilson, J. L. Bienias, D. A. Evans, and D. A. Bennett Diabetes Mellitus and Risk of Alzheimer Disease and Decline in Cognitive Function Arch Neurol, May 1, 2004; 61(5): 661 - 666. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
