Participants

Participants were enrolled in a longitudinal study of aging and dementia at the University of Southern California Alzheimer Disease Research Center. Enrolling entailed commitment to semiannual retesting. On enrollment, each participant received a neurological assessment, medical evaluation, and neuropsychological testing. Informed consent was obtained from all participants or their caregivers using procedures approved by the Institutional Review Board at the University of Southern California. AD was diagnosed according to the Diagnostic and Statistical Manual for Mental Disorders, 3rd edition–revised (DSM-IIIR) criteria for dementia, and National Institute of Neurologic and Communicative Disorders and Stroke/AD and Related Disorders Association (NINCDS/ADRDA) criteria for AD (McKhann et al., 1984). Dementia severity was graded using the Clinical Dementia Rating Scale (CDR; Burke et al., 1988), a 5-point scale used to discriminate individuals without dementia (0), and with questionable (0.5), mild (1), moderate (2), and severe (3) dementia. Criteria for inclusion in the analyses were CDR of 1 or greater at baseline or at retesting, clinical diagnosis of AD, at least one follow-up testing after AD diagnosis, baseline MMSE scores greater than 5 to avoid floor effects on retesting, and information about education and occupation obtained either from a knowledgeable informant or from the participant before AD diagnosis.

In all, 171 AD patients met these criteria; 144 were diagnosed with AD at their baseline assessment, while 27 were diagnosed with AD at retesting. For these 27, this retest occasion served as the baseline for the present analyses. These 171 patients were identified from a total of 974 individuals who were tested as part of the longitudinal study. Of these 974, 509 were diagnosed with AD at initial or follow-up testing. Among those with AD, 10 had only self-reported information collected when or after AD diagnosis was established, 160 had baseline scores of 5 or lower on the MMSE, and 14 died before the first follow-up testing could be scheduled. Another 88 had missing education and/or occupational data. Of the remaining 237 AD patients, 171 (72%) had at least one follow-up testing.

During the first testing with an AD diagnosis, average MMSE score was 17.7 (SD = 5.5), 123 patients received a CDR score of 1, 43 received a score of 2, and 5 received a score of 3. The design called for retesting subjects every 6 months if demented. In this study, the average number of follow-ups after AD diagnosis was 3.7 (SD = 3.5) over an average of 2.5 years (SD = 1.9 years). Therefore, the actual average follow-up interval was 8.1 months. Delays in follow-ups were mostly due to scheduling issues and illness. Follow-up for cognitive testing was ended due to death in 15 participants. In 146 participants, cognitive test follow-up was ended due to another medical reason or inability to complete a cognitive test battery. Ten participants were still active in the longitudinal cognitive testing.

The sample was 61% Caucasian, 27% Hispanic, 11% African American, and 1% Asian/Pacific Islander. Information about ethnicity was missing for one participant. Women composed 57% of the sample; 78% were English-speaking and 22% were Spanish-speaking. Previously, the Spanish versions of the neuropsychological battery (Taussig et al., 1992) and the MMSE (Taussig et al., 1996) showed no measurable cultural bias and reliably discriminated AD cases from normal controls.

Measures

Rate of cognitive decline was measured with the Mini-Mental State Examination (MMSE; Folstein et al., 1975), a 30-point test of global cognition that includes orientation to time and place, language, memory, and visuomotor performance.

The main independent variables were education and complexity of occupation. Information about education, lifetime occupational history, and demographics was obtained from participants when they were cognitively intact (9%), or otherwise from a spouse (38%), son/daughter (22%), brother/sister (4%), or another, unspecified person (27%). For participants who gave self-reports, average interval between time of report and AD diagnosis was 6.0 years.

Occupations were coded by using categories from the 1970 U. S. Census. To measure complexity, we used scores for substantive complexity of work, and for complexity of work with data, people, and things calculated for the 1970 U. S. Census occupational categories by Roos and Treiman (1980) based on ratings in the Dictionary of Occupational Titles (see Miller et al., 1980). Once occupation was coded with its 1970 U. S. Census category, occupational complexity ratings are available in a directory. For the substantive complexity factor, scores range from 0 to 10. Interpretation of scales for complexity of work with data, people, and things is presented in the Appendix. These measures are “intended to reflect the levels of complexity at which a worker in a particular occupation functions in relation to data, people, and things” (Miller et al., 1980, p. 19).

The scores used in our analyses reflect the average complexity of occupations held by each participant throughout adulthood. We multiplied the complexity score for each reported occupation by years spent in that occupation, summed the values across all occupations, and divided the total sum by the total number of years spent in all occupations. Years spent as a homemaker were not included in these calculations because complexity ratings for this category were not available.

Substantive complexity was correlated with complexity of work with data (r = .94, p < .01) and people (r = .71, p < .01), and complexity of work with data was correlated with complexity of work with people (r = .66, p < .01). Complexity of work with people was negatively correlated with complexity of work with things (r = −.30, p < .01). Substantive complexity, complexity of work with data, and complexity of work with people were correlated with education (r = .61, .57, .57, respectively, p < .01).

Analysis

Education and measures of occupational complexity were dichotomized into high (1) and low (0) categories. Education was classified as 12 years or less versus more than 12 years. Substantive complexity was dichotomized at the median, as scores greater than or equal to 4.9 versus less than 4.9. Complexity of work with data, people, and things were dichotomized along complexity levels (see the Appendix) as follows: Complexity of work with data was classified as Analyzing or more complex work versus Compiling or less complex work, complexity of work with people as Speaking/Signaling or more complex work versus Serving or Taking Instructions, and complexity of work with things as Manipulating or more complex work versus Tending or less complex work.

We used mixed-effects models in the SAS procedure MIXED (SAS Institute, 2003) to test whether average rate of decline on MMSE was predicted by levels of education or occupational complexity. Education and each complexity measure were entered into separate models. A mixed-effects model yields estimates of overall fixed effects, or average effects for the group, as well as random effects that reflect individual deviation from the fixed effect. Random effects were calculated using an unstructured covariance matrix and included baseline MMSE (intercept) and rate of decline (slope). The procedure accounts for intraindividual correlation among MMSE scores over time. All models controlled for age (centered at the mean), gender, native language (English/Spanish), AD diagnosis at initial testing vs. at retesting, and dementia severity at baseline (CDR score). Models with occupational complexity also controlled for years of education.