The new england journal of medicine

original article

Leisure Activities and the Risk of Dementia

in the Elderly
Joe Verghese, M.D., Richard B. Lipton, M.D., Mindy J. Katz, M.P.H.,
Charles B. Hall, Ph.D., Carol A. Derby, Ph.D., Gail Kuslansky, Ph.D.,
Anne F. Ambrose, M.D., Martin Sliwinski, Ph.D., and Herman Buschke, M.D.

abstract

background
From the Einstein Aging Study (J.V., R.B.L., Participation in leisure activities has been associated with a lower risk of dementia. It is
M.J.K., C.B.H., C.A.D., G.K., M.S., H.B.) unclear whether increased participation in leisure activities lowers the risk of dementia
and the Departments of Neurology (J.V.,
R.B.L., C.B.H., C.A.D., G.K., H.B.), Epidemi- or participation in leisure activities declines during the preclinical phase of dementia.
ology and Social Medicine (R.B.L., C.B.H.),
and Physical Medicine and Rehabilitation methods
(A.F.A.), Albert Einstein College of Medi-
cine, Bronx, N.Y.; and the Department of We examined the relation between leisure activities and the risk of dementia in a pro-
Psychology and the Center for Health and spective cohort of 469 subjects older than 75 years of age who resided in the communi-
Behavior, Syracuse University, Syracuse, ty and did not have dementia at base line. We examined the frequency of participation
N.Y. (M.S.). Address reprint requests to
Dr. Verghese at the Einstein Aging Study, in leisure activities at enrollment and derived cognitive-activity and physical-activity
Albert Einstein College of Medicine, 1165 scales in which the units of measure were activity-days per week. Cox proportional-haz-
Morris Park Ave., Bronx, NY 10461, or at ards analysis was used to evaluate the risk of dementia according to the base-line level
[email protected].
of participation in leisure activities, with adjustment for age, sex, educational level, pres-
N Engl J Med 2003;348:2508-16. ence or absence of chronic medical illnesses, and base-line cognitive status.
Copyright 2003 Massachusetts Medical Society.

results
Over a median follow-up period of 5.1 years, dementia developed in 124 subjects (Alz-
heimers disease in 61 subjects, vascular dementia in 30, mixed dementia in 25, and oth-
er types of dementia in 8). Among leisure activities, reading, playing board games, play-
ing musical instruments, and dancing were associated with a reduced risk of dementia.
A one-point increment in the cognitive-activity score was significantly associated with
a reduced risk of dementia (hazard ratio, 0.93 [95 percent confidence interval, 0.90 to
0.97]), but a one-point increment in the physical-activity score was not (hazard ratio,
1.00). The association with the cognitive-activity score persisted after the exclusion of
the subjects with possible preclinical dementia at base line. Results were similar for Alz-
heimers disease and vascular dementia. In linear mixed models, increased participation
in cognitive activities at base line was associated with reduced rates of decline in memory.

conclusions
Participation in leisure activities is associated with a reduced risk of dementia, even af-
ter adjustment for base-line cognitive status and after the exclusion of subjects with pos-
sible preclinical dementia. Controlled trials are needed to assess the protective effect of
cognitive leisure activities on the risk of dementia.

2508 n engl j med 348;25 www.nejm.org june 19, 2003

 leisure activities and dementia

he incidence of dementia in-

t creases with increasing age.1,2 Although

the prevention of dementia has emerged
as a major public health priority, there is a paucity
of potential preventive strategies.3-5 Identifying pro-
study population
methods

The study design and recruitment methods of the

Bronx Aging Study have been described previous-
tective factors is essential to the formulation of ef- ly.19,20 Briefly, the study enrolled English-speaking
fective interventions for dementia. Cross-section- subjects between 75 and 85 years of age who resid-
al studies report associations between dementia and ed in the community. Criteria for exclusion includ-
reduced participation in leisure activities in midlife, ed severe visual or hearing impairment and a previ-
as well as between cognitive status and participa- ous diagnosis of idiopathic Parkinsons disease,
tion in leisure activities in old age.6,7 Katzman pro- liver disease, alcoholism, or known terminal illness.
posed that persons with higher educational levels Subjects were screened to rule out the presence of
are more resistant to the effects of dementia as a dementia at base line and were included if they
result of having greater cognitive reserve and in- made eight or fewer errors on the Blessed Informa-
creased complexity of neuronal synapses.8 Like ed- tionMemoryConcentration test.19-21 This test has
ucation, participation in leisure activities may lower a high testretest reliability (0.86), and its results
the risk of dementia by improving cognitive re- correlate well with the stages of Alzheimers dis-
serve.9-15 ease.22,23 At the inception of the study, the cohort
In observational studies, elderly persons who was middle-class, most subjects were white (91 per-
had participated to a greater extent in leisure ac- cent), and the majority were female (64 percent).
tivities had a lower risk of dementia than those who Written informed consent was obtained at enroll-
had participated to a lesser extent.10-15 Although ment. The local institutional review board approved
these results suggest that leisure activities have a the study protocol.
protective role, an alternative explanation is possi- The study enrolled 488 subjects between 1980
ble. In most types of dementia, there is a long pe- and 1983. Subjects underwent detailed clinical and
riod of cognitive decline preceding diagnosis.16-18 neuropsychological evaluations at enrollment and
Reduced participation in activities during this pre- at follow-up visits every 12 to 18 months. The po-
clinical phase of dementia may be the consequence tential study period consisted of the 21-year period
and not the cause of cognitive decline. Resolution from 1980 to 2001. We excluded 2 subjects without
of this issue requires a long period of observation documented leisure activities and 17 subjects who
before diagnosis to enable researchers to disen- moved or declined to return for follow-up. After
tangle the potential effects of preclinical dementia. these subjects had been excluded, 469 subjects
Base-line cognitive status, educational level, and (96.1 percent) were eligible. In 1992, 73 surviving
level of depression may confound the relation be- subjects were still having study visits in our current
tween leisure activities and dementia.10-15 More- project, the Einstein Aging Study.
over, most studies have not assessed the associa-
tions between leisure activities and particular types clinical evaluation
of dementia.10-14 During the study, subjects were interviewed with
The Bronx Aging Study provided us with the op- the use of a structured medical-history question-
portunity to study the influence of leisure activi- naire and were examined by study clinicians.19,20
ties on the risk of dementia over a long period19,20 Functional limitations on 10 basic and instrumen-
while accounting for previously identified con- tal activities of daily living were rated on a 3-point
founders. This community-based study has fol- scale for each activity (range of total scores, 10 to
lowed a cohort of persons who did not have demen- 30 points), with 1 point indicating no limitation,
tia at base line, with the use of detailed clinical and 2 points indicating does activity with difficulty,
neuropsychological evaluations performed at in- and 3 points indicating unable.19,20 A spouse or
tervals for up to 21 years.19,20 We examined the in- family member accompanied most subjects or was
fluence of individual and composite measures of contacted for confirmation of the history.
cognitive and physical leisure activities on the risk
of the development of dementia. neuropsychological evaluation
An extensive battery of neuropsychological tests
was administered at study visits.18-20 We examined

n engl j med 348;25 www.nejm.org june 19, 2003 2509

 The new england journal of medicine

performance on the Blessed InformationMemo- ical-activity scale correlated with functional status
ryConcentration test (range of scores, 0 to 33),21 (Spearman r=0.293, P=0.001) but not with scores
the verbal and performance IQ according to the on the Blessed test (Spearman r=0.021, P=0.65).21
Wechsler Adult Intelligence Scale,24 the Fuld Ob-
ject-Memory Evaluation (range of scores, 0 to 10),25 diagnosis of dementia
and the Zung depression scale (range of scores, 0 to At study visits, subjects in whom dementia was
100).26 These tests were used to inform the diagno- suspected on the basis of the observations of mem-
sis of dementia at case conferences. bers of the study staff, results of neuropsychologi-
cal tests, or a worsening of the scores on the Blessed
leisure activities test21 by four points or a total of more than seven
At base line, subjects were interviewed regarding errors underwent a workup including computed to-
participation in 6 cognitive activities (reading books mographic scanning and blood tests.19,20 A diag-
or newspapers, writing for pleasure, doing cross- nosis of dementia was assigned at case conferences
word puzzles, playing board games or cards, partic- attended by study neurologists, a neuropsycholo-
ipating in organized group discussions, and play- gist, and a geriatric nurse clinician, according to the
ing musical instruments) and 11 physical activities criteria of the Diagnostic and Statistical Manual of Men-
(playing tennis or golf, swimming, bicycling, danc- tal Disorders, third edition (DSM-III) or, after 1986,
ing, participating in group exercises, playing team the revised third edition (DSM-III-R).27-29 Updated
games such as bowling, walking for exercise, climb- criteria for the diagnosis of dementia and particular
ing more than two flights of stairs, doing house- types of dementia were introduced after the study
work, and babysitting). Subjects reported the fre- had begun.
quency of participation as daily, several days per To ensure uniformity of diagnosis, all cases were
week, once weekly, monthly, occasionally, or discussed again at new diagnostic conferences held
never. We recoded these responses to generate a in 2001 and involving a neurologist and a neuro-
scale with one point corresponding to participa- psychologist who had not participated in diagnostic
tion in one activity for one day per week. The units conferences between 1980 and 1998.29 Dementia
of the scales are thus activity-days per week; the was diagnosed according to the DSM-III-R crite-
scales were designed to be intuitively meaningful ria.28 Reduced participation in leisure activities was
to clinicians and elderly persons and to be useful in used to assess functional decline, but the leisure-
the design of intervention studies or public health activity scales were not available to the raters assess-
recommendations. For each activity, subjects re- ing such decline. Disagreements between raters
ceived seven points for daily participation; four were resolved by consensus after the case was pre-
points for participating several days per week; one sented to a second neurologist, with blinding main-
point for participating once weekly; and zero points tained. Cases of dementia were classified according
for participating monthly, occasionally, or never. We to the criteria for probable or possible Alzheimers
summed the activity-days for each activity to gener- disease published by the National Institutes of Neu-
ate a cognitive-activity score, ranging from 0 to 42, rological Disorders and Stroke and the Alzheimers
and a physical-activity score, ranging from 0 to 77. Disease and Related Disorders Association30 and
The estimates of the overall level of participa- the criteria for probable, possible, or mixed vascu-
tion were consistent with good testretest reliability lar dementia published by the Alzheimers Disease
for scores obtained on entry and at the next visit a Research Centers of California.31
year later on the cognitive-activity scale (Spearman
r=0.518, P=0.001) and the physical-activity scales statistical analysis
(Spearman r=0.410, P=0.001). There was no direct Continuous variables were compared with use of
measurement of the time spent in activities, al- either an independent-samples t-test or the Mann
though participation was verified by family mem- Whitney U test, and categorical variables were com-
bers or friends. The scores were not correlated with pared with use of the Pearson chi-square test.32 In
age. Scores on the cognitive-activity scale correlat- primary analyses, we studied the association be-
ed with scores on the Blessed test21 (Spearman tween cognitive and physical activities and the risk
r=0.286, P=0.001), but not functional status of dementia and specific types of dementia using
(Spearman r=0.042, P=0.77). Scores on the phys- Cox proportional-hazards regression analysis to es-

2510 n engl j med 348;25 www.nejm.org june 19 , 2003

 leisure activities and dementia

timate hazard ratios, with 95 percent confidence in-

results
tervals.33 The time to an event was defined as the
time from enrollment to the date of a diagnosis of demographic characteristics
dementia or to the final contact or visit for sub- During 2702 person-years of follow-up (median fol-
jects without dementia. All multivariate models re- low-up, 5.1 years), dementia developed in 124 sub-
ported include the following covariates unless oth- jects (Alzheimers disease in 61, vascular dementia
erwise specified: age at enrollment, sex, educational in 30, mixed dementia in 25, and other types of de-
level (high school or less vs. college-level educa- mentia in 8). By the end of the study period, 361 sub-
tion), presence or absence of chronic medical ill- jects had died, 88 subjects had dropped out (mean
nesses, and base-line scores on the Blessed test. [SD] follow-up, 6.64.9 years), and 20 subjects
Presence of the following self-reported chronic were still active. When the cohort was divided into
medical illnesses was individually entered in the thirds according to the cognitive-activity score or
models: cardiac disease (angina, previous myocar- the physical-activity score, no significant differences
dial infarction, or cardiac failure), hypertension, di- in the length of follow-up were found among these
abetes mellitus, stroke, depression, and hypothy- subgroups.
roidism. We also divided the study cohort into thirds On average, subjects in whom dementia devel-
on the basis of their scores on the two activity scales oped were older, had lower levels of education, and
and determined the risk of dementia according to had significantly lower scores on the cognitive-
these groups. We examined the role of individual activity scale, but not on the physical-activity scale,
leisure activities by comparing subjects who partic- than subjects in whom dementia did not develop
ipated in an activity several days or more per week (Table 1). Although their scores on neuropsycho-
(frequent participation) with subjects who partici- logical tests were in the normal range (Table 1), sub-
pated weekly or less frequently (rare participation) jects in whom dementia later developed had poorer
and, in the full models, adjusted for participation in cognition on the Blessed test (P=0.001)21 and a low-
other leisure activities. er base-line performance IQ (P=0.07).24 The fre-
In secondary analyses, we examined the influ- quency of chronic medical illnesses did not differ
ence of base-line cognitive status and possible pre- significantly between subjects in whom dementia
clinical dementia. First, we sequentially excluded developed and those in whom it did not. Nineteen
from the full models subjects in whom dementia de- subjects were receiving antidepressants at enroll-
veloped during the first two, four, seven, and nine ment. Cognitive-activity scores were inversely cor-
years of follow-up in order to avoid confounding by related with the Zung depression scale26 (Spearman
a possible influence of preclinical dementia on par- r = 0.215, P<0.001), as were physical-activity scores
ticipation in leisure activities. Second, we used linear (Spearman r=0.254, P<0.001), indicating that low-
mixed models controlled for age, sex, and educa- er levels of participation were associated with in-
tional level to assess the relation between cognitive creasing levels of depression.
activities and base-line cognitive status and the an- A smaller proportion of the 361 subjects with a
nual rate of change in cognitive status.34 We ana- high-school education or less than of the 108 sub-
lyzed verbal IQ as well as specific cognitive domains, jects who had attended college participated in read-
including episodic memory (with the Buschke Se- ing (90 percent vs. 96 percent, P=0.05), writing (67
lective Reminding test [range of scores, 0 to 72, with percent vs. 80 percent, P=0.01), doing crossword
lower scores indicating worse memory]35 and the puzzles (21 percent vs. 36 percent, P=0.001), and
Fuld Object-Memory Evaluation25) and executive playing musical instruments (7 percent vs. 15 per-
function (with the DigitSymbol Substitution sub- cent, P=0.02). There was no difference according
test of the Wechsler Adult Intelligence Scale [range to educational level in the proportion of subjects
of scores, 0 to 90, with lower scores indicating worse who played board games or participated in group
cognition]).24 Each model included terms for the discussions.
cognitive-activity score, time, and the interaction
between the two. The assumptions of the models leisure activities
were examined analytically and graphically and were Among cognitive activities, reading, playing board
adequately met. games, and playing musical instruments were as-

n engl j med 348;25 www.nejm.org june 19, 2003 2511

 The new england journal of medicine

Table 1. Base-Line Characteristics of Subjects in Whom Dementia Developed and Subjects in Whom It Did Not.*

Subjects in Subjects in
Whom Dementia Whom Dementia
Did Not Develop Developed
Variable (N=345) (N=124) P Value
Age (yr) 78.93.1 79.73.1 0.01
Female sex (%) 63 67 0.51
White race (%) 92 91 0.60
Duration of follow-up (yr) 5.64.1 5.94.1 0.52
High-school education or less (%) 74 84 0.02
Functional rating 10.91.9 11.52.1 0.01
Physical-activity score 13.67.6 12.88.2 0.31
Cognitive-activity score 10.65.8 7.55.5 <0.001
Neuropsychological tests
Blessed InformationMemoryConcentration test 2.11.9 3.52.4 0.001
Performance IQ 105.812.4 97.513.8 0.07
Verbal IQ 111.015.5 103.915.6 0.35
Fuld Object-Memory Evaluation 7.51.2 6.71.5 0.001
Zung depression scale 46.410.4 48.410.9 0.32
Medical illnesses (%)
Hypertension 52 45 0.12
Cardiac disease 29 23 0.72
Stroke 7 3 0.99
Diabetes 11 12 0.87
Thyroid illness 14 9 0.12
Depression 17 19 0.10

* Plusminus values are means SD. P values for scales and tests were calculated by the MannWhitney U test. The func-
tional rating ranges from 10 to 30, with higher scores indicating better function; scores on the physical-activity scale range
from 0 to 77, with higher scores indicating greater participation; scores on the cognitive-activity scale range from 0 to 42,
with higher scores indicating greater participation; the range of scores on the Blessed InformationMemoryConcentra-
tion test is 0 to 33, with higher scores indicating worse general cognitive status; the normal ranges of performance IQ
and verbal IQ are 85 to 115; the range of scores on the Fuld Object-Memory Evaluation is 0 to 10, with higher scores in-
dicating better memory; and the range of scores on the Zung depression scale is 0 to 100, with higher scores indicating
a greater level of depression.

sociated with a lower risk of dementia (Table 2). Alzheimers disease (hazard ratio, 0.93 [95 percent
Dancing was the only physical activity associated confidence interval, 0.88 to 0.98]), vascular demen-
with a lower risk of dementia. Fewer than 10 sub- tia (hazard ratio, 0.92 [95 percent confidence in-
jects played golf or tennis, so the relation between terval, 0.86 to 0.99]), and mixed dementia (haz-
these activities and dementia was not assessed. ard ratio, 0.87 [95 percent confidence interval, 0.78
to 0.93]). The frequency of participation in cog-
cognitive activities nitive activities was related to the risk of demen-
When the cognitive-activity score was modeled as tia. According to the model in which we adjusted
a continuous variable (Table 3), the hazard ratio for the base-line score on the Blessed test, the haz-
for dementia for a one-point increment in this score ard ratio for subjects with scores in the highest
was 0.93 (95 percent confidence interval, 0.89 to third on the cognitive-activity scale, as compared
0.96). Adjustment for the base-line score on the with those with scores in the lowest third, was
Blessed test in a second model (Table 3) did not at- 0.37 (95 percent confidence interval, 0.23 to 0.61)
tenuate the association. Participation in cognitive (Table 3).
activities was associated with a reduced risk of In additional analyses, adjustment for intellec-

2512 n engl j med 348;25 www.nejm.org june 19 , 2003

 leisure activities and dementia

tual status with the use of the verbal IQ24 did not
Table 2. Risk of Development of Dementia According to the Frequency
alter the association between participation in cog- of Participation in Individual Leisure Activities at Base Line.*
nitive activities and the risk of dementia (hazard
Subjects Hazard Ratio
ratio, 0.92 [95 percent confidence interval, 0.87 with All for Dementia
to 0.97]). Participation in cognitive activities was Leisure Activity and Frequency Dementia Subjects (95% CI)
also associated with a reduced risk of dementia no.
among the 361 subjects with a high-school edu- Cognitive activities
cation or less (hazard ratio, 0.94 [95 percent con- Playing board games
fidence interval, 0.91 to 0.98]). The association of Rare 108 366 1.00
cognitive activities with dementia was not affected Frequent 16 103 0.26 (0.170.57)
by adjustment for functional status, the restriction Reading
Rare 40 87 1.00
of the analyses to subjects with scores of less than Frequent 84 382 0.65 (0.430.97)
5 on the Blessed test,21 or the exclusion of subjects Playing a musical instrument
who died during the first year after enrollment. Rare 120 452 1.00
Frequent 4 17 0.31 (0.110.90)
Doing crossword puzzles
physical activities Rare 117 407 1.00
The physical-activity score was not significantly as- Frequent 7 62 0.59 (0.341.01)
sociated with dementia, either when analyzed as a Writing
continuous variable or when the study cohort was Rare 104 382 1.00
Frequent 20 87 1.00 (0.611.67)
divided into thirds according to this score (Table 3).
Participating in group discussions
Rare 117 437 1.00
influence of preclinical dementia Frequent 7 32 1.06 (0.482.33)
The presence of preclinical dementia might reduce Physical activities
participation in leisure activities,6,7 leading to the Dancing
overestimation of its protective influence. The asso- Rare 99 339 1.00
Frequent 25 130 0.24 (0.060.99)
ciation between the base-line cognitive-activity score Doing housework
and dementia was significant even after the exclu- Rare 39 106 1.00
sion of 94 subjects in whom dementia was diag- Frequent 85 363 0.88 (0.601.20)
nosed during the first seven years after enrollment Walking
Rare 19 65 1.00
(hazard ratio, 0.94 [95 percent confidence interval, Frequent 105 404 0.67 (0.451.05)
0.88 to 0.99]) (Table 4). The association was no Climbing stairs
longer significant after the exclusion of the 105 sub- Rare 44 153 1.00
jects in whom dementia was diagnosed during the Frequent 80 316 1.55 (0.962.38)
first nine years after enrollment. However, only 19 Bicycling
Rare 116 443 1.00
subjects were given a diagnosis of dementia after Frequent 8 26 2.09 (0.974.49)
this point. Swimming
We used linear mixed models to examine the in- Rare 108 386 1.00
fluence of participation in cognitive activities on the Frequent 16 83 0.71 (0.222.29)
annual rate of change in cognitive function.34 In Playing team games
Rare 120 450 1.00
these models (Table 5), the term for the cognitive- Frequent 4 19 1.00 (0.147.79)
activity score represents the cross-sectional associ- Participating in group exercise
ation between the cognitive activities and the scores Rare 88 330 1.00
Frequent 36 139 1.18 (0.721.94)
on the selected tests administered at enrollment.
Babysitting
These results indicate that subjects with increased Rare 114 429 1.00
participation in cognitive activities at entry had bet- Frequent 10 40 0.81 (0.116.01)
ter overall cognitive status. Analysis with use of the
term for time indicates that cognitive performance * The frequency of participation in leisure activities was categorized as frequent
if the subject participated at least several times per week and as rare if the sub-
declines linearly as a function of follow-up time. ject participated once per week or less frequently. Hazard ratios were adjusted
The term for the interaction between the cognitive- for age, sex, educational level, presence or absence of medical illnesses, score
activity score and time represents the longitudinal on the Blessed InformationMemoryConcentration test, and participation or
nonparticipation in other leisure activities. For each activity, rare participation
effect of the base-line measure of participation in was used as the reference category. CI denotes confidence interval.
cognitive activities on the annual rate of decline in

n engl j med 348;25 www.nejm.org june 19, 2003 2513

 The new england journal of medicine

nual rate of decline on the Fuld Object-Memory

Table 3. Risk of Dementia According to the Base-Line Scores
Evaluation is reduced by 0.006 point (P=0.04).
on the Cognitive-Activity Scale and the Physical-Activity Scale.*

No. of Hazard Ratio for Dementia discussion

Leisure Activity Subjects (95% CI)

Model 1 Model 2 This prospective, 21-year study demonstrates a sig-

Cognitive-activity score nificant association between a higher level of par-
ticipation in leisure activities at base line and a de-
1-Point increment 0.93 (0.890.96) 0.93 (0.900.97)
creased risk of dementia both for Alzheimers
<8 Points 182 1.00 1.00
disease and for vascular dementia. A one-point in-
811 Points 137 0.50 (0.310.75) 0.48 (0.290.74) crement in the cognitive-activity score, which cor-
>11 Points 150 0.33 (0.210.51) 0.37 (0.230.61) responds to participation in an activity for one day
Physical-activity score per week, was associated with a reduction of 7 per-
1-Point increment 0.99 (0.911.01) 1.00 (0.981.03) cent in the risk of dementia. The association be-
tween cognitive activities and the risk of dementia
<9 Points 162 1.00 1.00
remained robust even after adjustment for poten-
916 Points 157 1.06 (0.671.65) 1.44 (0.912.28)
tial confounding variables such as age, sex, educa-
>16 Points 150 0.92 (0.581.45) 1.27 (0.782.06) tional level, presence or absence of chronic medical
illnesses, and base-line cognitive status. Increased
* Model 1 was adjusted for age, sex, educational level, and the presence or absence
of chronic medical illnesses; model 2 includes the variables in model 1 and the
participation in leisure activities was associated with
base-line score on the Blessed InformationMemoryConcentration test. For a lower risk of dementia. Subjects with scores in the
each scale, scores in the lowest third were used as the reference category. highest third on the cognitive-activity scale (more
CI denotes confidence interval.
than 11 activity-days) had a risk of dementia that
was 63 percent lower than that among subjects with
scores in the lowest third.
We identified three possible explanations for the
Table 4. Risk of Dementia per 1-Point Increment in the Base-Line
Cognitive-Activity Score, with the Sequential Exclusion of Subjects
association between greater participation in leisure
in Whom Dementia Developed during the First Nine Years of Follow-up.* activities and a decreased risk of dementia. First, the
presence of preclinical dementia may decrease par-
Excluded Subjects ticipation in leisure activities. Second, unmeasured
Subjects in Whom
with Dementia Hazard Ratio confounding may influence the results. Third, there
Analysis Dementia Developed (95% CI) may be a true causal effect of cognitive activities. We
used several strategies to test the hypothesis that re-
no.
duced participation in leisure activities appears to be
Overall 0 124 0.93 (0.900.97) a risk factor for, but is in fact a consequence of, pre-
With exclusion of subjects with clinical dementia. Adjustment for base-line scores
a diagnosis of dementia on cognitive tests, which predict dementia, did not
Diagnosis during first 2 yr 36 88 0.94 (0.900.97) alter the association between participation in cog-
Diagnosis during first 4 yr 63 61 0.94 (0.890.98) nitive activities and dementia. We have reported that
an accelerated decline in memory begins seven years
Diagnosis during first 7 yr 94 30 0.94 (0.880.99)
before dementia is diagnosed.18 The exclusion of
Diagnosis during first 9 yr 105 19 0.96 (0.891.04) subjects in whom dementia was diagnosed during
the first seven years after enrollment should elim-
* Hazard ratios are adjusted for age, sex, educational level, presence or absence inate most subjects who had preclinical dementia
of chronic medical illnesses, and score on the Blessed InformationMemory
Concentration test. CI denotes confidence interval. at enrollment. However, participation in cognitive
activities predicted dementia even among those in
whom it developed more than seven years after en-
performance on the selected tests; this effect was rollment. Results from the linear mixed models that
significant only for the tests of episodic memory. analyzed cognitive function over time corroborate
The estimates show that for a one-point increment the findings of previous studies13,36 and show that
in the cognitive-activity score, the annual rate of de- increased participation in cognitive activities is as-
cline in scores on the Buschke Selective Reminding sociated with slower rates of cognitive decline, espe-
test is reduced by 0.043 point (P=0.02), and the an- cially in terms of episodic memory.

2514 n engl j med 348;25 www.nejm.org june 19 , 2003

 leisure activities and dementia

Because of the observational nature of our study,

Table 5. Association of Participation in Cognitive Leisure Activities
there is a possibility of residual or unmeasured con- with Base-Line Cognitive Function and Rate of Change in Cognitive Function.*
founding. The observed association appears to be
independent of educational level and intellectual Estimated
Change in
level, which may influence the choice of leisure ac- Cognitive Test Test Score (SE) P Value
tivities. Perhaps reduced participation in leisure ac-
Buschke Selective Reminding test
tivities is an early marker of dementia that precedes Cognitive-activity score (per 1-point increment) 0.3830.092 <0.001
the declines on cognitive tests.13 Alternatively, par- Time (per 1-year increment) 1.5780.211 <0.001
ticipation in leisure activities may be a marker of Interaction 0.0430.018 0.02
behavior that promotes health. But the specificity Fuld Object-Memory Evaluation
Cognitive-activity score (per 1-point increment) 0.0280.011 0.007
of our findings for cognitive activities and not phys- Time (per 1-year increment) 0.9610.466 0.04
ical activities argues against this hypothesis. We Interaction 0.0060.003 0.04
did not study the effect of apolipoprotein E geno- DigitSymbol Substitution test
type, which may influence the rates of cognitive de- Cognitive-activity score (per 1-point increment) 0.5690.097 <0.001
Time (per 1-year increment) 0.9980.172 <0.001
cline.15,36 Hence, despite the magnitude and con- Interaction 0.0010.014 >0.05
sistency of the associations, our findings do not Verbal IQ
establish a causal relation between participation in Cognitive-activity score (per 1-point increment) 0.7890.131 <0.001
leisure activities and dementia, and controlled trials Time (per 1-year increment) 4.4721.980 0.02
Interaction 0.0200.012 >0.05
are therefore needed.
If there is a causal role, participation in leisure * Associations were assessed by linear mixed models, controlled for age,
activities may increase cognitive reserve, delaying sex, and educational level. The term for the interaction between the cognitive-
the clinical or pathological onset of dementia.8,37,38 activity score and time represents the longitudinal effect of the base-line
measure of participation in cognitive activities on the annual rate of decline in
Alternatively, participation in cognitive activities performance on the given test.
might slow the pathological processes of disease
during the preclinical phase of dementia. Our find-
ings do not imply that subjects who were less active tially limiting the generalizability of our results.
cognitively increased their risk of dementia. Although standard criteria and well-established
The role of individual leisure activities is not well procedures were used to make diagnoses, some
known, since most studies have used composite misclassification is inevitable. Time spent in each
measures. In a French cohort, knitting, doing odd activity was not directly measured, although the his-
jobs, gardening, and traveling reduced the risk of de- tory was verified by family members or other infor-
mentia.10 In the Nun Study, low density of ideas and mants. Duration and cognitive demand are both im-
low levels of grammatical complexity in autobiog- portant in the assessment of an activity. It is difficult
raphies written in early life were associated with to assign weights to the cognitive demands of lei-
low cognitive test scores in later life.39 Reading, sure activities, since such demands vary among ac-
playing board games, playing musical instruments, tivities and among the persons who engage in each
and dancing were associated with a lower risk of de- activity. Leisure activities were arbitrarily classified
mentia in our cohort. There was no association be- as cognitive or physical. For instance, doing house-
tween physical activity and the risk of dementia. Ex- work requires not only a certain functional status
ercise is said to have beneficial effects on the brain by but also the ability to plan, prepare, and adapt to
promoting plasticity, increasing the levels of neuro- changes in circumstances and the environment.
trophic factors in the brain, and enhancing resist- The leisure activities we studied reflect the interests
ance to insults.40 Cognitive and physical activities of our cohort, and it is quite likely that activities oth-
overlap, and therefore it is not surprising that pre- er than the ones we studied are also protective.10-15
vious studies have disagreed on the role of physical Participation in leisure activities is associated
activities.10-15 Although physical activities are clearly with a reduced risk of development of dementia,
important in promoting overall health,41 their pro- both Alzheimers disease and vascular dementia.
tective effect against dementia remains uncertain. The reduction in risk is related to the frequency of
Our study has several limitations. Ours was a co- participation. According to our models, for exam-
hort of volunteers who resided in the community; ple, elderly persons who did crossword puzzles four
whites and subjects older than 75 years of age were days a week (four activity-days) had a risk of demen-
overrepresented, as compared with the general pop- tia that was 47 percent lower than that among sub-
ulation of those over 65 years of age, thus poten- jects who did puzzles once a week (one activity-day).

n engl j med 348;25 www.nejm.org june 19, 2003 2515

 leisure activities and dementia

Clinical trials are needed to define the causal role of er the risk of dementia that parallel current recom-
participation in leisure activities. A recent study re- mendations for participation in physical activities
ported reduced cognitive declines after cognitive to reduce the risk of cardiovascular diseases.42,43
training in elderly persons without dementia.36 If Supported by a grant (AGO3949-15) from the National Institute
confirmed, our results may support recommenda- onPresented Aging.
in part at the 127th annual meeting of the American
tions for participation in cognitive activities to low- Neurological Association, New York, October 1216, 2002.

refer enc es
1. Fratiglioni L, De Ronchi D, Aguero-Torres lation-based sample of older persons. Neu- 30. McKhann G, Drachman D, Folstein M,
H. Worldwide prevalence and incidence of rology 2002;59:1910-4. Katzman R, Price D, Stadlan EM. Clinical
dementia. Drugs Aging 1999;15:365-75. 16. Small BJ, Fratiglioni L, Viitanen M, Win- diagnosis of Alzheimers disease: report of
2. Jorm AF, Jolley D. The incidence of blad B, Backman L. The course of cognitive the NINCDS-ADRDA Work Group under
dementia: a meta-analysis. Neurology 1998; impairment in preclinical Alzheimer disease: the auspices of Department of Health and
51:728-33. three- and 6-year follow-up of a population- Human Services Task Force on Alzheimers
3. Seshadri S, Beiser A, Selhub J, et al. based sample. Arch Neurol 2000;57:839-44. Disease. Neurology 1984;34:939-44.
Plasma homocysteine as a risk factor for 17. Elias MF, Beiser A, Wolf PA, Au R, White 31. Chui HC, Victoroff JI, Margolin D, Jag-
dementia and Alzheimers disease. N Engl RF, DAgostino RB. The preclinical phase of ust W, Shankle R, Katzman R. Criteria for
J Med 2002;346:476-83. Alzheimer disease: a 22-year prospective the diagnosis of ischemic vascular dementia
4. Engelhart MJ, Geerlings MI, Ruitenberg study of the Framingham Cohort. Arch Neu- proposed by the State of California Alzhei-
A, et al. Dietary intake of antioxidants and rol 2000;57:808-13. mers Disease Diagnostic and Treatment
risk of Alzheimer disease. JAMA 2002;287: 18. Hall CB, Ying J, Kuo L, et al. Estimation Centers. Neurology 1992;42:473-80.
3223-9. of bivariate measurements having different 32. Fleiss JL. Statistical methods for rates
5. Morris MC, Evans DA, Bienias JL, et al. change points, with application to cognitive and proportions. 2nd ed. New York: John
Dietary intake of antioxidant nutrients and ageing. Stat Med 2001;20:3695-714. Wiley, 1981:38-46.
the risk of incident Alzheimer disease in a 19. Katzman R, Aronson M, Fuld P, et al. 33. Cox DR. Regression models and life-
biracial community study. JAMA 2002;287: Development of dementing illnesses in an tables. J R Stat Soc [B] 1972;34:187-220.
3230-7. 80-year-old volunteer cohort. Ann Neurol 34. Laird NM, Ware JH. Random-effects
6. Friedland RP, Fritsch T, Smyth KA, et 1989;25:317-24. models for longitudinal data. Biometrics
al. Patients with Alzheimers disease have 20. Aronson MK, Ooi WL, Morgenstern H, 1982;38:963-74.
reduced activities in midlife compared with et al. Women, myocardial infarction, and 35. Buschke H. Selective reminding for anal-
healthy control-group members. Proc Natl dementia in the very old. Neurology 1990; ysis of memory and learning. J Verbal Learn
Acad Sci U S A 2001;98:3440-5. 40:1102-6. Verb Behav 1973;12:543-50.
7. Kondo K, Niino M, Shido K. A case-con- 21. Blessed G, Tomlinson BE, Roth M. The 36. Ball K, Berch DB, Helmers KF, et al.
trol study of Alzheimers disease in Japan association between quantitative measures Effects of cognitive training interventions
significance of life-styles. Dementia 1994;5: of dementia and of senile change in the cer- with older adults: a randomized controlled
314-26. ebral grey matter of elderly subjects. Br J Psy- trial. JAMA 2002;288:2271-81.
8. Katzman R. Education and the preva- chiatry 1968;114:797-811. 37. Wilson RS, Bienias JL, Berry-Kravis E,
lence of dementia and Alzheimers disease. 22. Fuld PA. Psychological testing in the dif- Evans DA, Bennett DA. The apolipoprotein
Neurology 1993;43:13-20. ferential diagnosis of the dementias. In: E varepsilon 2 allele and decline in episodic
9. Rogers RL, Meyer JS, Mortel KF. After Katzman R, Terry RD, Bick KL, eds. Alzhei- memory. J Neurol Neurosurg Psychiatry 2002;
reaching retirement age physical activity sus- mers disease: senile dementia and related 73:672-7.
tains cerebral perfusion and cognition. J Am disorders. Vol. 7 of Aging. New York: Raven 38. Stern Y, Albert S, Tang M-X, Tsai W-Y.
Geriatr Soc 1990;38:123-8. Press, 1978:185-93. Rate of memory decline in AD is related to
10. Fabrigoule C, Letenneur L, Dartigues 23. Grober E, Dickson D, Sliwinski MJ, et al. education and occupation: cognitive reserve?
JF, Zarrouk M, Commenges D, Barberger- Memory and mental status correlates of mod- Neurology 1999;53:1942-7.
Gateau P. Social and leisure activities and ified Braak staging. Neurobiol Aging 1999; 39. Snowdon DA, Kemper SJ, Mortimer JA,
risk of dementia: a prospective longitudinal 20:573-9. Greiner LH, Wekstein DR, Markesbery WR.
study. J Am Geriatr Soc 1995;43:485-90. 24. Wechsler D. Manual for the Wechsler Linguistic ability in early life and cognitive
11. Laurin D, Verreault R, Lindsay J, MacPher- Adult Intelligence Scale. Washington, D.C.: function and Alzheimers disease in late life:
son K, Rockwood K. Physical activity and risk Psychological Corporation, 1955. findings from the Nun Study. JAMA 1996;
of cognitive impairment and dementia in eld- 25. Fuld PA, Masur DM, Blau AD, Crystal H, 275:528-32.
erly persons. Arch Neurol 2001;58:498-504. Aronson MK. Object-memory evaluation for 40. Cotman CW, Berchtold NC. Exercise:
12. Scarmeas N, Levy G, Tang MX, Manly J, prospective detection of dementia in normal a behavioral intervention to enhance brain
Stern Y. Influence of leisure activity on the functioning elderly: predictive and normative health and plasticity. Trends Neurosci 2002;
incidence of Alzheimers disease. Neurology data. J Clin Exp Neuropsychol 1990;12:520-8. 25:295-301.
2001;57:2236-42. 26. Zung WWK. Depression in the normal 41. Paffenbarger RS Jr, Hyde RT, Wing AL,
13. Wilson RS, Mendes De Leon CF, Barnes aged. Psychosomatics 1967;8:287-92. Lee I-M, Jung DL, Kampert JB. The associa-
LL, et al. Participation in cognitively stimu- 27. Diagnostic and statistical manual of men- tion of changes in physical-activity level and
lating activities and risk of incident Alzhei- tal disorders, 3rd ed.: DSM-III. Washington, other lifestyle characteristics with mortality
mer disease. JAMA 2002;287:742-8. D.C.: American Psychiatric Association, 1980. among men. N Engl J Med 1993;328:538-45.
14. Wang H-X, Karp A, Winblad B, Fratigli- 28. Diagnostic and statistical manual of 42. American College of Sports Medicine
oni L. Late-life engagement in social and lei- mental disorders, 3rd ed. rev.: DSM-III-R. Position Stand: exercise and physical activity
sure activities is associated with a decreased Washington, D.C.: American Psychiatric for older adults. Med Sci Sports Exerc 1998;
risk of dementia: a longitudinal study from Association, 1987. 30:992-1008.
the Kungsholmen project. Am J Epidemiol 29. Verghese J, Lipton RB, Hall CB, Kuslan- 43. Christmas C, Andersen RA. Exercise and
2002;155:1081-7. sky G, Katz MJ, Buschke H. Abnormality of older patients: guidelines for the clinician.
15. Wilson RS, Bennett DA, Bienias JL, et al. gait as a predictor of non-Alzheimers demen- J Am Geriatr Soc 2000;48:318-24.
Cognitive activity and incident AD in a popu- tia. N Engl J Med 2002;347:1761-8. Copyright 2003 Massachusetts Medical Society.

2516 n engl j med 348;25 www.nejm.org june 19 , 2003

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.