Leisure Activity and Cognitive Decline in Incident Alzheimer Disease FREE

High rates of leisure activity have been associated with reduced risk of Alzheimer disease (AD).^{1 - 5} Along with premorbid IQ, educational level, and occupational attainment, leisure activity may contribute to cognitive reserve (CR).^{6 - 7} Those with greater CR may tolerate a greater burden of Alzheimer neuropathologic conditions before clinical disease expression; however, relatively advanced neuropathologic disease by the time of diagnosis may result in faster disease progression.^{8 - 9} Proxies of higher CR, including higher educational level,^{8 ,10 - 11} higher occupational attainment,^{8 ,12} and higher premorbid reading activity,¹³ have been associated with faster cognitive decline in patients with dementia, although some studies^{14 - 15} found no association.

The Washington Heights-Inwood Columbia Aging Project (WHICAP) is a community-based study of cognitive aging that began in 1989. An earlier WHICAP investigation⁴ found that high rates of leisure participation were associated with lower AD incidence. The current investigation assesses the influence of leisure on the rate of cognitive decline among those who developed AD during study follow-up (ie, incident cases only). We hypothesized that higher prediagnosis leisure activity would be associated with faster cognitive decline, consistent with the hypothesis that leisure activity contributes to CR.

Participants in WHICAP were recruited in 2 similar cohorts. Recruitment began in 1992. The geographic study area was the 14 census tracts in Manhattan in New York, between (approximately) 155th and 181st streets. Lists of all Medicare or Medicaid beneficiaries in this area were obtained from the Health Care Financing Administration (now called Centers for Medicare & Medicaid Services). Potential participants were then drawn by systematic random sampling into 1 of 6 strata formed based on ethnicity (Hispanic, non-Hispanic black, and non-Hispanic white) and age (65-74 and ≥75 years). A total of 2125 participants were interviewed at baseline. A “refreshment” cohort of 2183 additional participants was formed in 1999 using generally similar methods. The main exceptions were as follows: new lists of beneficiaries were obtained but those drawn into the 1992 cohort were excluded, patients who reported being diagnosed as having dementia while arranging for the initial evaluation were excluded, and the study area was extended to the south and north to encompass all of Manhattan north of (approximately) 145th Street.

From the combined cohorts, 388 individuals developed AD during study follow-up. Of these, 318 had available follow-up assessments (15.0% of those without follow-up had not been seen for the next assessment wave). A total of 283 had prediagnosis leisure data and at least 2 cognitive assessments and were included in the analysis; of these, 133 had available postdiagnosis follow-up. Excluded patients with AD were older than the analysis sample (P = .04) but did not significantly differ by sex, race, APOE ε4 status, educational level, leisure participation, or baseline cognitive score. The study was approved by the Columbia University institutional review board, and written informed consent was obtained from all participants.

Diagnoses of AD were based on standardized, physician-administered physical and neurologic examinations along with a standardized neuropsychological battery. All assessments were administered at baseline and at subsequent follow-ups (roughly every 18 months). Assessments were conducted in English or Spanish based on participant preference. All available ancillary information (medical records, imaging studies, and evidence of social or occupational function deficits) was considered. Leisure information was not used in diagnosis.

Dementia diagnoses were made at consensus conferences attended by neurologists and neuropsychologists based on Diagnostic and Statistical Manual of Mental Disorders (Third Edition Revised) criteria. Probable or possible AD was diagnosed according to the criteria of the National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer Disease and Related Disorders Association.

The primary outcome was rate of change in a composite measure of cognition. A complete neuropsychological battery¹⁶ was administered, with tests that assessed 5 cognitive domains: (1) memory: total and delayed recall of the Selective Reminding Test¹⁷ and the recognition component of the multiple-choice version of the Benton Visual Retention Test¹⁸ ; (2) abstract reasoning: Wechsler Adult Intelligence Scale Revised similarities subtest¹⁹ and the identities and oddities subtest of the Dementia Rating Scale²⁰ ; (3) visual-spatial: 5 items from the Rosen Drawing Test²¹ and the matching component of the multiple-choice version of the Benton Visual Retention Test¹⁸ ; (4) language: 15-item Boston Naming Test,²² the 8 high-probability items from the repetition subtest of the Boston Diagnostic Aphasia Examination,²² and the first 6 items of the Boston Diagnostic Aphasia Examination comprehension subtest; and (5) executive speed: average scores for phonemic fluency assessed by the Controlled Oral Word Association Test and category fluency (animals, food, and clothing) mean scores.²³

The composite measure was derived as follows: the 12 raw scores were transformed into z scores based on means and standard deviations for each test calculated from baseline scores of 272 control subjects who were matched according to age, education level, and ethnicity to identify patients with incident AD. Individual test z scores for participants with dementia were averaged to create a z score for each cognitive domain. If fewer than half of the test z scores were missing for a given domain, the domain score was calculated with the available data. If more than half of the tests were missing, the domain score was considered missing and was excluded. The composite score was the average of the 5 domain scores, with missing data treated in the manner previously described.

The primary predictor variable was self-reported leisure activity. Although leisure was measured at multiple visits, the first available assessment was used for 2 reasons: (1) to increase response validity, since participants did not have dementia at this visit, and (2) to better approximate long-standing leisure habits (to better approximate CR). Participation in the previous month in 13 activities was considered in 4 categories according to previous work⁴ : (1) intellectual: reading magazines, newspapers, or books; going to classes; and playing cards, games, or bingo; (2) social: doing unpaid volunteer work; going to a club or center; going to movies, restaurants, or sporting events; attending church, synagogue, or temple; visiting friends or relatives; and being visited by friends or relatives; (3) physical: physical conditioning and walking for pleasure or excursion; and (4) other: knitting, music, or other hobby and watching television or listening to the radio. Participation in each activity was scored 1 point and summed to derive the total leisure activity score (range, 0-13). High vs low rates of leisure participation were based on a median split of 6 activities.⁴

We also considered factors that might influence cognitive decline or leisure participation. Baseline medical comorbidity was assessed with a modified²⁴ Charlson Comorbidity Index.²⁵ Myocardial infarction, congestive heart failure, peripheral vascular disease, hypertension, chronic obstructive pulmonary disease, arthritis, gastrointestinal disease, mild liver disease, and diabetes mellitus were weighted 1; chronic renal disease and systemic malignancy were weighted 2. History of stroke was considered separately. The APOE genotype was determined using established methods²⁶ and categorized based on the presence of at least 1 ε4 allele. We also considered ethnicity (non-Hispanic white, non-Hispanic black, and Hispanic), sex, education level, baseline cognitive performance, and study cohort.

Baseline characteristics by leisure category (low or high) were compared using χ² or t tests. Generalized estimating equations were used to examine influence of leisure on rates of postdiagnosis cognitive change. By treating each patient's repeated measures as a cluster, the generalized estimating equations account for the probable correlation of characteristics measured in the same individual over time. In the initial model, the dependent variable was the composite cognitive score. Predictor variables were total leisure score, time (years from diagnosis), and a leisure × time interaction. A significant leisure effect would suggest a difference in cognitive performance at diagnosis associated with each prediagnosis leisure activity. A significant time effect would suggest a change in cognitive scores over time (regardless of leisure score). A significant interaction term would suggest differential rates of cognitive change as a function of leisure.

Subsequently, multivariate-adjusted models were constructed. Since low leisure activity measured closer to diagnosis could reflect early dementia, we controlled for time between the leisure assessment and dementia diagnosis (leisure-to-diagnosis time) in all models. Models also included baseline cognitive score, ethnicity, educational level, sex, study cohort, comorbidity index, and stroke history. All covariates were simultaneously introduced into models. A separate model was built for each leisure category (total leisure, intellectual, social, physical, and other activities).

In supplementary analyses, we examined cognitive decline during the entire study follow-up (before and after diagnosis). This was done to more fully capture the natural course of the disease, since AD-associated neuropathologic conditions and subtle cognitive decline are present many years before diagnosis.^{27 - 28} Also, in this sample, rates of cognitive decline were similar before and after AD diagnosis.²⁸ In exploratory analyses, we examined leisure-associated decline within APOE ε4 strata.

At diagnosis, most participants (89%) had a clinical dementia rating of 1.0 (mild).²⁹ Leisure participation did not vary by education level, sex, ethnicity, or age (at baseline or diagnosis). Higher leisure activity was associated with higher baseline cognition, fewer medical comorbidities, and lower stroke prevalence (Table 1). Treating leisure as a continuous variable (number of activities), leisure activity was positively correlated with years of education (r = 0.004; P = .95), and intellectual leisure activity examined separately was positively correlated with years of education (r = 0.07; P = .26), although these findings did not reach statistical significance.

The mean ± SD follow-up time between baseline and last cognitive assessments was 5.3 ± 3.1 years (range, 1.0-13.9 years), with a mean ± SD of 3.6 ± 1.5 follow-up assessments (range, 2-5). The mean ± SD time between leisure assessment and AD diagnosis was 4.1 ± 2.8 years (range, 1.0-13.1 years). Those with postdiagnosis follow-up (n = 133) had a mean ± SD postdiagnosis follow-up of 3.3 ± 2.1 years (range, 1.1-9.8 years) and a mean ± SD of 2.4 ± 0.8 postdiagnosis assessments (range, 1-5).

As expected in this sample with dementia, the composite cognitive score declined during study follow-up. In a generalized estimating equations model, the β associated with the composite cognitive score was −.09 (P <.001), indicating a 9% decline per year. Rates of decline were similar before (β = −.08; P <.001) and after (β = −.09; P <.001) AD onset. Quadratic terms for time were added to these models to test whether cognitive change was nonlinear. These terms were nonsignificant and thus were not included in the final models.

The primary models examined cognitive change after diagnosis and were limited to 133 participants. Higher total leisure scores were associated with faster cognitive decline, although this effect did not reach statistical significance (β = −.005; P = .17). Leisure categories (physical, social, intellectual, and other) were subsequently examined separately. After multivariate adjustment, no statistically significant effects by leisure category were seen (Table 2).

Models that examined cognitive decline during the entire study follow-up (before and after diagnosis) included 283 participants. In this much larger sample, each leisure activity was associated with an additional yearly decline of 0.005 of a z-score unit (P = .03) after multivariate adjustment. Figure 1 illustrates predicted decline in composite cognitive score based on generalized estimating equations by tertile of baseline leisure activity. In category-specific models, only intellectual activity was associated with faster decline, with an additional decline of 0.03 of a z-score unit per activity per year (P <.001) (Table 2).

Since APOE ε4 has been associated with differential rates of cognitive decline, we repeated models after stratifying by ε4 status. Because of incomplete APOE data, we were underpowered to test this association using only postdiagnosis cognitive change; thus, these models measured change during the entire study. Higher total leisure activity was associated with faster cognitive decline among non-ε4 carriers only (β = −.007, P = .04; among ε4 carriers: β = .003, P = .46; fully adjusted models).

We performed the analysis again with the total leisure score as the primary predictor, examining change in specific cognitive domains. In models of postdiagnosis change, differential rates of change were seen for the executive speed domain (β = −.10; P = .03) but not in other domains (memory: β = −.004, P = .47; visuospatial: β = −.003, P = .58; abstract reasoning: β = −.001, P = .82; language: β = −.005, P = .30). In models of cognition from baseline on, differential rates of change were seen in the memory (β = −.009; P = .05) and executive speed (β = −.009; P = .02) domains but not the others (visuospatial: β = −.004, P = .42; abstract reasoning: β = −.001, P = .73; language: β = −.007; P = .27).

To our knowledge, this study is the first to examine the influence of premorbid leisure activity on cognitive decline among incident AD cases. In models of postdiagnosis cognitive change, leisure categories were associated with faster cognitive decline, although the effect did not reach statistical significance. Supplementary analyses that measured cognitive decline during the entire study follow-up, potentially capturing earlier disease stages and more than doubling the sample size, revealed that higher total leisure and intellectual activity (but not social or physical activity) was associated with significantly faster cognitive decline, even after multivariate adjustment.

We hypothesized that leisure activity serves as a proxy of CR. People with high CR may have more efficient neural networks or may use alternate networks more effectively after neurologic insult. These people theoretically compensate better for AD-associated neuropathologic conditions, allowing for a longer period of normal function before diagnosis. Thus, those with higher CR may have lower or delayed AD incidence. However, since neuropathologic conditions are relatively advanced by diagnosis,^{24 ,30 - 32} they may have a faster subsequent disease course.⁶

Support for this theory comes from studies that demonstrated faster cognitive decline among patients with AD with higher education levels^{8 ,10 ,24} and higher premorbid reading activity.¹³ Furthermore, imaging studies suggest that CR may modulate the relationship between neuropathologic disease and clinical manifestations. Adjusting for clinical severity, more highly educated patients with AD tolerated greater cerebral blood flow deficits (an indirect measure of AD),³⁰ and patients with AD with higher occupational attainment and higher premorbid intelligence demonstrated lower cerebral metabolism.^{31 - 32} In our earlier positron emission tomographic study of patients with AD, those with a high leisure activity rate tolerated greater cerebral blood flow deficits (suggesting greater neuropathologic disease), even adjusted for disease severity, educational level, and IQ.²⁴ Our current findings are consistent with imaging results, suggesting that higher premorbid leisure activity is linked to more advanced neuropathologic disease, resulting in faster subsequent progression. Our findings also mirror the recent results regarding educational level from this cohort,²⁴ despite the lack of correlation between leisure activity and educational attainment, suggesting that leisure activity may represent an aspect of CR distinct from education level. A theoretical model of leisure as a proxy of CR is presented in Figure 2.

We found no association between physical activity and cognitive decline in patients with AD. The mental stimulation associated with intellectual and social activities may influence CR differently than physical activity. Alternatively, our negative findings with physical activity may be due to our admittedly limited physical activity assessment.

Exploratory models stratified by APOE ε4 status revealed that leisure-associated cognitive decline during the entire study follow-up was more pronounced among non-ε4 carriers, suggesting that the effect of APOE ε4 may override any contributing effect of leisure on cognitive decline among carriers.

A major strength of this study is that the sample was limited to cases of incident AD, allowing us to capture more of the disease course, including the earlier stages. Other strengths include the high quality of our diagnostic procedure and our use of multiple cognitive assessments to provide more accurate slope calculations. Finally, our ethnically diverse, community-based sample increases generalizability of our findings.

We acknowledge several limitations. Although we used the earliest available leisure measurement, this was on average only 4 years before diagnosis. If early dementia reduced leisure activity, this might work against our findings. We attempted to account for this by controlling for time between the leisure assessment and dementia incidence for each participant. In future investigations, it would be better to assess leisure earlier in life. Unlike others,⁵ we did not measure frequency and intensity of activities, possibly reducing the sensitivity of our leisure assessment. Finally, we cannot exclude the possibility that leisure acts as a surrogate for some other factor not included in our models that is truly associated with faster cognitive decline. Our results add to the growing body of evidence that disease course in AD may vary as a function of CR.

Correspondence: Yaakov Stern, PhD, Gertrude H. Sergievsky Center, 622 W 168th St, PH 19th Floor, New York, NY 10032 (ys11@columbia.edu).

Accepted for Publication: February 1, 2007.

Author Contributions:Study concept and design: Helzner, Scarmeas, and Stern. Acquisition of data: Scarmeas and Stern. Analysis and interpretation of data: Helzner, Scarmeas, Cosentino, Portet, and Stern. Drafting of the manuscript: Helzner. Critical revision of the manuscript for important intellectual content: Scarmeas, Cosentino, Portet, and Stern. Statistical analysis: Helzner, Scarmeas, and Cosentino. Study supervision: Scarmeas and Stern.

Financial Disclosure: None reported.

Funding/Support: This work was supported in part by grants P01-AG07232 and RR00645 from the National Institute on Aging and grant 5T32NS007153-22 from the National Institute of Neurological Disorders and Stroke.

Role of the Sponsor: The funding bodies had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.

Additional Contribution: Nicole Schupf, PhD, provided a thoughtful review of the manuscript.