American Journal of Epidemiology
Copyright © 2002 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved
Vol. 156, No. 5
Printed in U.S.A.
DOI: 10.1093/aje/kwf060
Physical Activity and Menstrual Cycle Characteristics in Two Prospective Cohorts
Barbara Sternfeld1, Marlena K. Jacobs1, Charles P. Quesenberry Jr.1, Ellen B. Gold2, and
MaryFran Sowers3
1
Division of Research, Kaiser Permanente, Oakland, CA.
Department of Epidemiology and Preventive Medicine, School of Medicine, University of California at Davis, Davis, CA.
3 Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, MI.
2
Received for publication November 2, 2001; accepted for publication May 7, 2002.
Relations between physical activity and prospectively collected menstrual cycle characteristics were examined
in two large cohorts. One cohort consisted of women employed in the semiconductor industry in 1989 who
participated in a prospective study of reproductive outcomes (n = 367). The other consisted of women living in
Tecumseh, Michigan, who completed both the 1992–1993 and 1993–1994 examinations for the Michigan Bone
Health Study (n = 328). Mean cycle length, variability of cycle length, and mean bleed length were calculated from
daily diaries (Semiconductor cohort) or monthly menstrual calendars (Michigan cohort) for a median of five and
11 cycles, respectively. Physical activity was assessed by self-report at baseline and expressed as metabolic
equivalent-minutes per week. In the Semiconductor study, women also reported daily minutes of vigorous
exercise in their diaries. In the Michigan cohort, total physical activity, total recreational physical activity, and
vigorous recreational activity were positively associated with cycle length. The magnitude of these associations
declined as body mass index increased. In the Semiconductor cohort, the minutes of daily vigorous exercise were
positively associated with cycle length only in a repeated-measures analysis. These findings lend modest support
to the hypothesis that moderate levels of physical activity can lengthen the menstrual cycle. Am J Epidemiol
2002;156:402–9.
exercise; menstrual cycle; menstruation; prospective studies
Abbreviations: BMI, body mass index; MET, metabolic equivalent; SE, standard error.
Numerous epidemiologic studies have observed an inverse
association between physical activity and the incidence of
breast cancer (1–4). Because breast cancer is a hormonally
mediated disease, one potential biologic mechanism
accounting for these observations is a reduced lifetime exposure to cyclic estrogen and progesterone as a result of regular
physical activity (5, 6). The evidence for such hormonal
effects of physical activity comes primarily from studies of
athletes, who have been found to have a higher prevalence of
amenorrhea or oligomenorrhea (7–12) and a greater likelihood of anovulatory cycles (13, 14) compared with sedentary women. The athletes may also experience a shortened
luteal phase, even when no other changes in the menstrual
cycle are observed (15–19). The hormonal changes that
underlie these alterations may include lower levels of
follicular-phase estradiol, lower luteal-phase progesterone,
and the absence of the midcycle luteal hormone surge (16,
18, 20).
The extent to which such menstrual cycle alterations occur
with the lower levels of more moderate intensity physical
activity typically seen in the population of women at large is
not well understood. A few epidemiologic studies suggest
that women participating in moderate recreational activity
have longer and more variable cycles than do sedentary
women (21, 22), and amenorrhea has been induced in previously sedentary women with the initiation of a vigorous
training program (23). From a public health perspective,
understanding the effects on the menstrual cycle of moderate
levels of physical activity is most relevant because few
women engage in high-level, vigorous exercise, but as many
as 20 percent regularly participate in moderate-intensity
activity (24). Therefore, the purpose of this investigation was
Reprint requests to Dr. Barbara Sternfeld, Division of Research, Kaiser Permanente, 3505 Broadway, Oakland, CA 94611 (e-mail:
[email protected]).
402
Physical Activity and the Menstrual Cycle 403
to determine the influence of physical activity on menstrual
cycle characteristics in two epidemiologic cohorts on whom
menstrual cycle data were prospectively collected. Specifically, this study examined the relations between physical
activity and mean cycle length, variability of cycle length,
and mean bleed length in an ethnically diverse cohort of
women employed in the semiconductor industry and in a
cohort of White women participating in a longitudinal
community health study. The extent to which these relations
varied by body fat was also addressed.
MATERIALS AND METHODS
Study sample
Participants came from two distinct epidemiologic study
cohorts. The first cohort consisted of women who were
employed in the semiconductor industry in 1989 and who
were eligible for and participated in a prospective study of
spontaneous abortion and other menstrual and reproductive
health outcomes. Details of the recruitment process and
eligibility criteria have been presented previously (25). In
brief, the sampling frame consisted of all women between
ages 18 and 44 years working in the fabrication rooms at five
semiconductor companies in three metropolitan areas of two
states (California and Utah) and an equal number of “nonfab” women workers frequency matched at each company by
ethnicity and 5-year age categories. To be eligible, women
could not be pregnant, lactating, using oral contraceptives or
other steroid hormones, or sterilized. They must have had a
period in the previous 3 months, had intercourse in the
previous 2 months, and had a working freezer (for storage of
daily urine samples). Of the 739 women who were eligible,
481 completed the baseline interview, and 402 completed
collection of daily diary data and urine samples for at least
one nonconceptive menstrual cycle. Women were excluded
from the current analysis if they did not have both valid cycle
length and bleed length data for at least one menstrual cycle
(n = 35). This left a total sample size of 367 women, referred
to here as the Semiconductor cohort.
The second cohort consisted of women aged 24–48 years
in 1992, who were living in Tecumseh, Michigan, and who
completed both the 1992–1993 and 1993–1994 examinations for the Michigan Bone Health Study. Recruitment for
the Michigan Bone Health Study has been described previously (26). Originally begun in 1988, the Study recruited 542
nonpregnant women between ages 20 and 40 years who were
identified either from family records of participants in the
original Tecumseh Community Health Study in 1959–1960,
or from various community outreach efforts. In 1992, an
additional 122 community women between ages 35 and 45
years were identified and recruited into the study, resulting
in a potential cohort of 664 women. Of these, 64 members of
the 1988 cohort refused follow-up in 1992 or had moved,
leaving 600 examinees in 1992–1993. For our analysis, 272
were excluded for the following reasons: 1) oral contraceptive, fertility drug, or other hormone use (n = 113); 2) pregnant or lactating (n = 29); 3) hysterectomy, bilateral
oophorectomy, or other gynecologic surgery (n = 100); 4)
naturally postmenopausal (n = 4); 5) missing menstrual cycle
Am J Epidemiol
Vol. 156, No. 5, 2002
data (n = 8); or 6) follow-up refused in 1993–1994 or moved
away (n = 18). After exclusions, 328 women remained,
referred to here as the Michigan cohort.
Assessment of physical activity
Physical activity was assessed in the Semiconductor
cohort during the baseline interview. Women were asked
about their participation during the previous month in 62
different recreational physical activities, such as aerobics,
running/jogging, bicycling, hiking, and walking. For each
activity in which a respondent participated, she was asked
how many times a month and for how many minutes, on
average, she did that activity. Using the Compendium of
Physical Activities (27), we assigned each activity a metabolic equivalent (MET) value that reflected the energy
expenditure or intensity of that activity. Intensity was then
multiplied by frequency and duration and summed over all
activities, resulting in a summary measure of recreational
physical activity expressed in MET-minutes per week. A
similar measure was constructed for vigorous physical
activity by limiting the sum to only those activities with a
MET value of six or greater.
In addition, women in the Semiconductor cohort were
asked to keep a daily diary of menstrual bleeding in which
they noted how many minutes of vigorous exercise they had
performed each day. Two variables were constructed from
these data: a per-woman average number of minutes of daily
vigorous exercise over the length of study participation and
a per-cycle average number of minutes of daily vigorous
exercise for each cycle.
In the Michigan cohort, the Minnesota Leisure Time Physical Activity survey, a quantitative activity history with
established reliability and validity that has been widely used
in epidemiologic studies (28, 29), was adapted by adding
occupational and household activities and deleting many
activities uncommon in women (e.g., hunting). In contrast to
the previous-month time frame used in the Semiconductor
interview, the time frame for the Michigan survey was July
and January prior to the baseline interview, allowing for the
appreciable seasonal variation in physical activities that may
occur in an environment such as that in Michigan. For each
month, participants were asked about the number of times
per week they performed each activity and the average duration of each session. Activities were assigned MET values,
and intensity, duration, and frequency were used to compute
an overall measure of total activity in MET-minutes/week by
averaging the two monthly measures of total activity. To
create measures directly comparable with those of the Semiconductor study, the MET-minutes per week in recreational
physical activity were computed by considering only recreational activities, and the MET-minutes per week in
vigorous recreational physical activity were computed from
only the recreational activities with a MET value of six or
higher.
Assessment of menstrual cycle characteristics
Starting immediately after the baseline interview, the
Semiconductor cohort used daily diaries to record the pres-
404 Sternfeld et al.
ence or absence of menstrual bleeding on each day. The
number of menstrual cycles for which daily diaries were
completed ranged from one to eight, with a median of five
completed cycles and with 62.9 percent of the cohort
providing data for four or more cycles. The total number of
cycles used in our analysis was 1,527. Three variables were
created based on the daily diary: 1) mean cycle length,
defined as the per-woman mean cycle length over all cycles;
2) variability of cycle length, defined as the per-woman standard deviation of cycle length, computed for women with at
least two valid cycles; and 3) mean bleed length, defined as
the per-woman mean bleed length over all cycles.
In the Michigan cohort, monthly menstrual calendars on
which women marked each day of menstrual bleeding
provided data for construction of the same three variables. A
total of 3,497 cycles of menstrual calendar data were available for computing cycle length and variability, and 3,095
cycles were available for computing bleed length. The perwoman mean was 10.7 cycles of valid cycle data (median =
11) and 9.4 cycles of valid bleed data (median = 10), with a
per woman range of 1–15 and 0–15, respectively.
activity in the Michigan cohort) and menstrual cycle characteristics. Multivariable linear regression was used to model
separately each menstrual cycle characteristic (cycle length,
cycle variability, and bleed length) as a function of each
physical activity variable, adjusting for potential confounders. Covariates included in the final models were those that
showed a bivariate association with both physical activity
and cycle characteristics (p < 0.10). To explore whether the
relation between a physical activity variable and a menstrual
cycle characteristic varied by BMI, age, or ethnicity (Semiconductor cohort only), models were constructed that
included the appropriate cross-product terms or that stratified on the variable of interest. To examine relations between
cycle-specific menstrual cycle or bleed length and the
amount of vigorous exercise either in the same cycle or in the
immediately preceding cycle, multivariable linear regression
with repeated measures, with correlation structure specified
as unstructured, was performed using the Semiconductor
data.
All analyses were stratified by cohort.
RESULTS
Covariates
In both cohorts, age, ethnicity, education, marital status,
parity, smoking status, and alcohol consumption were selfreported at baseline in response to either interviewer-administered or self-administered questions. Total caloric intake
was assessed with the Block Food Frequency Questionnaire
(30) for 285 (86.9 percent) women in the Michigan cohort
and 288 (78.5 percent) women in the Semiconductor cohort.
Body mass index (BMI) was computed as weight (kg)/height
(m2) from self-reported height and weight in the Semiconductor cohort and from measured height and weight in the
Michigan cohort.
Data analyses
The distributions of all variables were examined for
outliers and violations of assumptions of normality. Because
the distribution of the per-woman standard deviation of cycle
length, a measure of the within-woman variability of cycles,
was highly skewed, it was log-transformed for analysis by
using a natural log and adding 0.25 to all standard deviations
before transformation to account for women with a cycle
variability of zero. Summary variables of physical activity
were also highly skewed, but transformation did not substantially change the distribution. Since these were considered as
independent variables, no transformation was used. Differences in the characteristics of the two cohorts were
compared by using the chi-square test for differences in
proportions for categorical variables, the t test for differences in means in continuous variables, and the Wilcoxon
rank sum test for differences in medians for the physical
activity variables.
Spearman correlation coefficients were used to evaluate
the bivariate relations between physical activity measures
(recreational physical activity and vigorous recreational
activity in both cohorts, per-woman average daily vigorous
exercise in the Semiconductor cohort, and total physical
As shown in table 1, the two cohorts differed from each
other in many respects. As noted above, the Michigan cohort
consisted of White women, while almost 20 percent of the
Semiconductor cohort were Filipinas, and more than 50
percent were of a race/ethnicity other than White. The Michigan cohort was older and had a higher BMI, reflecting, to a
large extent, the different racial/ethnic compositions of the
two cohorts; the Michigan women also appeared to be more
physically active. Mean cycle length was almost 2 days
longer in the Semiconductor cohort, but variability of cycle
length and mean bleed length were similar.
In both cohorts, all correlation coefficients between cycle
length and total activity (Michigan cohort only), total recreational activity, vigorous recreational activity, or per-woman
minutes of vigorous exercise (Semiconductor cohort only)
were less than 0.10 and were not statistically significant
(data not shown). In the Semiconductor cohort, adjustment
for age, BMI, race/ethnicity, and smoking also failed to
reveal any significant associations between mean cycle
length and physical activity. In these models, an increase of
30 MET-minutes/week of total recreational activity was
insignificantly associated with a 0.001-day increase in mean
cycle length (standard error (SE) = 0.007, p = 0.86). This
would be equivalent to, for instance, 10 minutes per week of
a three-MET activity, such as brisk walking on level ground.
Similarly, in a model with vigorous recreational activity as
the outcome, 30 MET-minutes/week was associated with a
0.009-day decrease in mean cycle length (SE = 0.009, p =
0.29). This is equivalent to, for instance, 5 minutes per week
of a six-MET activity, such as aerobic dancing.
However, the repeated-measures analysis revealed a more
complex relation. Table 2, which presents the adjusted association between cycle-specific cycle length and per-cycle
mean minutes of daily vigorous exercise in the concurrent
cycle, shows a statistically significant positive association.
The model suggests that an increase of 10 minutes per day of
vigorous exercise during a given menstrual cycle would be
Am J Epidemiol
Vol. 156, No. 5, 2002
Physical Activity and the Menstrual Cycle 405
TABLE 1. Characteristics of the Semiconductor cohort in 1989 and Michigan cohort, 1992
Semiconductor cohort
No.
Age (years) (mean (SD†))**
367
33.6 (5.1)
Michigan cohort
328
37.4 (4.4)
Race/ethnicity (no. (%))
White, non-Hispanic
166 (45.2)
Filipino
67 (18.3)
Other Asian
46 (12.5)
Hispanic
52 (14.2)
Black/other
36 (9.8)
328 (100.0)
Education, (years) (no. (%))**
≤12
113 (30.8)
157 (48.0)
13–15
144 (39.2)
97 (29.7)
≥16
110 (30.0)
73 (22.3)
Marital status (no. (%))
Not married
55 (15.0)
61 (18.6)
311 (85.0)
267 (81.4)
0
122 (33.2)
43 (13.2)
1
93 (25.3)
41 (12.6)
2
100 (27.2)
126 (38.7)
≥3
52 (14.2)
116 (35.6)
239 (65.5)
195 (59.8)
68 (18.6)
75 (23.0)
<1
168 (46.4)
255 (78.0)
1–<7
170 (47.0)
45 (13.8)
24 (6.6)
27 (8.3)
Married
Parity (no. (%))**
Smoking (no. (%))
Nonsmoker
Former
Current
Alcohol (no. of drinks/week) (%))**
≥7
Energy intake (kcal/day) (mean (SD))*
Body mass index (mean (SD))**
1,470.9 (595)
24.5 (5.5)
1,563 (514)
27.0 (6.0)
Physical activity (MET†-minutes/week) (median (IQ† range))
Total
Total recreational**
7,104 (4,138–11,276)
504 (126–1,344)
Vigorous recreational**
59.5 (0–672)
Per woman mean minutes of vigorous exercise (minutes/week)
26.7 (1.63–84.6)
1,643 (936–2,955)
566 (180–1,328)
Menstrual cycle characteristics (mean (SD))
Mean cycle length (days)**
30.04 (6.2)
28.8 (3.5)
Variability (SD) of cycle length (days), median (IQ range)
2.4 (1.6–4.1)
2.8 (1.8–5.5)
Mean bleed length (days)
5.0 (1.2)
5.0 (1.2)
* p < 0.05, significant difference between Semiconductor and Michigan cohorts based on chi-square test for
difference in proportions, t test for difference in means or the Wilcoxon rank sum test for difference in medians.
** p < 0.001, significant difference between Semiconductor and Michigan cohorts based on chi-square test
for difference in proportions, t test for difference in means, or the Wilcoxon rank sum test for difference in
medians.
† SD, standard deviation; MET, metabolic equivalent; IQ, intelligence quotient.
Am J Epidemiol
Vol. 156, No. 5, 2002
406 Sternfeld et al.
TABLE 2. Multivariable association between cycle-specific menstrual cycle
length and per-cycle mean minutes of daily vigorous exercise in the
concurrent cycle* in the Semiconductor cohort, 1989
Variables in model
Per-cycle vigorous exercise in
concurrent cycle (per minutes/day)
β
(SE)†
p
0.0245
(0.0106)
0.02
–0.1494
(0.0559)
<0.01
0.1076
(0.0737)
0.14
Filipino
2.0536
(0.9573)
0.04
Other Asian
2.4680
(0.9174)
<0.01
Hispanic
0.6011
(0.7991)
0.45
–0.7204
(0.7169)
0.31
Age (years)
BMI† (weight (kg)/height (m2))
Race/ethnicity
White (reference)
Black/other
Smoking status
Never (reference)
Former
0.4275
(0.7600)
0.57
Current
–0.2501
(0.9424)
0.79
* Recorded in daily diary and averaged over cycle concurrent with cyclespecific cycle length, using repeated measures analysis.
† SE, standard error; BMI, body mass index.
associated with an increase in the length of that cycle of
approximately two tenths of a day. When this model was
stratified by race/ethnicity, the magnitude of the association
between vigorous exercise and the length of that cycle was
greater in Filipinas (β = 1.2752, SE = 0.384, p = 0.02) and
other Asians (β = 2.472, SE = 0.952, p = 0.01) than in
Whites (β = 0.1355, SE = 0.036), and the association was not
statistically significant in Hispanics (β = 0.0097, SE = 0.023,
p = 0.26). The mean minutes of daily vigorous exercise for
the previous cycle were also positively and significantly
related to cycle length after adjustment for covariates (β =
–0.035, SE = 0.015); the magnitude of this association did
not appear to differ by race/ethnicity.
In the Michigan cohort, statistically significant positive
associations between physical activity and cycle length were
observed after adjusting for covariates and allowing for
effect modification by BMI. As shown in table 3, vigorous
recreational activity was directly related to cycle length, but
the magnitude of the association decreased as BMI
increased. At a BMI equal to 24, the mean change in cycle
length associated with, for example, 50 minutes per week of
aerobic dancing (equal to 300 MET-minutes/week) was an
increase of three hundredths of a day, independent of other
variables. In contrast, at a BMI of 30, the mean change in
cycle length associated with the same amount of activity was
a decrease of almost two tenths of a day. Similar relations
were observed in the Michigan cohort when total physical
activity and total recreational activity were considered,
although the relation between total recreational activity and
cycle length and the interaction term with BMI were not
statistically significant (data not shown).
Neither bivariate correlations nor multivariable analysis
revealed any significant associations in either cohort
between cycle variability and any of the physical activity
measures. In the Semiconductor cohort, associations were
also not observed between any of the physical activity
measures and bleed length. In contrast, in the Michigan
cohort, modest, but statistically significant, positive correlations with bleed length were observed (r = 0.14, 0.20, and
0.12 for total activity, total recreational activity, and
vigorous recreational activity, respectively). As shown in
table 3, the positive association of bleed length with vigorous
physical activity persisted independently of age, BMI, and
smoking status. Similar results were obtained when either
total activity (β = 0.001, SE = 0.0003, p = 0.023) or total
recreational activity (β = 0.004, SE = 0.001, p = 0.003) was
substituted for vigorous recreational activity.
DISCUSSION
This study of two relatively large cohorts of women of
reproductive age showed a modest, direct, although not
entirely consistent, relation between physical activity and
menstrual cycle length. In the Semiconductor cohort, the
Am J Epidemiol
Vol. 156, No. 5, 2002
Physical Activity and the Menstrual Cycle 407
TABLE 3. Multivariable associations of vigorous recreational activity with menstrual cycle length or bleed length
in the Michigan cohort, 1992*
Menstrual cycle length (days)
Variables in model
Baseline vigorous recreational activity† (per 30
MET-minutes/week)
β
(SE)*
p
Bleed length (days)
β
(SE)
p
0.075
(0.028)
0.008
0.004
(0.002)
0.031
–0.090
(0.045)
0.04
–0.014
(0.016)
0.38
0.206
(0.040)
<0.001
–0.005
(0.012)
0.70
–0.003
(0.001)
0.008
Former
0.291
(0.514)
0.57
0.127
(0.187)
0.50
Current
0.413
(0.469)
0.38
0.483
(0.169)
0.004
Age (years)
BMI* (weight (kg)/height (m2))
Activity × BMI
Smoking status
Never (reference)
* SE, standard error; BMI, body mass index.
† Self-reported for previous January and July during baseline interview, limited to recreational physical activities of six
metabolic equivalents (METs) or more.
association appeared to be confined to physical activity
either in the immediate cycle under consideration or in the
preceding cycle. Neither baseline activity level, reported for
the month preceding entry into the study, nor daily exercise,
reported over the entire study period, was associated with
mean cycle length. In the Michigan cohort, all measures of
activity, including total activity (consisting of household,
occupational, and recreational activities), recreational physical activity of all intensities, and vigorous recreational
activity, were positively associated with cycle length.
However, the magnitude and direction of that association
depended on BMI. This study found no relation between
physical activity and cycle variability and a direct relation
with bleed length in only the Michigan cohort.
The finding of longer cycles related to greater physical
activity is generally in agreement with previous work.
Numerous studies have reported greater prevalence of amenorrhea and oligomenorrhea among female athletes compared
with sedentary women (7, 10, 11), and anovulation and
subsequent absence of menstrual bleeding have been
induced with the initiation of vigorous exercise training in
some studies (23), although not in others (31, 32). In a study
of the relation of more moderate amounts of activity and
menstrual cycle characteristics in college women, greater
activity was associated with slightly increased probability of
cycles longer than 43 days and was more strongly associated
with long cycles in women who experienced both long and
regular-length cycles (21). In a study of women aged 29–31
years, daily vigorous activity was associated with increased
cycle variability and a nonstatistically significant increase in
risk of at least one long cycle but was not associated with
mean cycle length as a continuous variable (22).
These findings suggest that moderate activity, like
vigorous activity, may have hormonal effects that may
lengthen the menstrual cycle, resulting, over a lifetime, in
Am J Epidemiol
Vol. 156, No. 5, 2002
lower levels or less cyclic fluctuations of estrogen and
progesterone.
The finding in the Semiconductor cohort of a relation only
between cycle-specific exercise and cycle length may
suggest that an increase in menstrual cycle length is more of
an acute response to physical activity than an adaptive
chronic response. This interpretation could be consistent
with evidence of athletic amenorrhea, since an acute
response would occur repeatedly in women engaged in
regular exercise. The finding may also indicate that the
intraindividual variability in physical activity in this cohort
was great enough that using activity in 1 month (the baseline
measure) as indicative of activity level over the course of the
study resulted in a nondifferential misclassification and
attenuation to the null. Misclassification and attenuation
toward the null could also have occurred by averaging daily
reports of vigorous exercise over all cycles if activity varied
greatly within woman from cycle to cycle. Finally, the
finding may have been observed only in the repeatedmeasures analyses because of the greater statistical power it
provided, as well as the greater precision of exposure
measurement.
The fact that no association between activity and cycle
length was observed in the Michigan cohort before adjustment for BMI may illustrate a type of confounding in which
the bivariate relation is attenuated rather than inflated; this
could occur because BMI and activity are inversely associated, but BMI and cycle length are directly related. On the
other hand, the effect modification by BMI may have a
biologic basis in the contribution of energy conservation to
exercise-associated menstrual cycle alterations. Although it
is now well established that these alterations are not caused
by low body fat (33), evidence supports the theory that when
energy output far exceeds energy intake, a stable body
weight may be maintained by shutting down specific
408 Sternfeld et al.
biologic processes, such as reproductive capability, to
conserve energy (34, 35). Such a mechanism may be more
operative in women with lower stores of body fat.
The direct association between all measures of physical
activity and bleed length observed in the Michigan cohort is
unexpected and is not in accord with two previous studies
that reported inverse associations between activity and days
of bleeding (22, 36). The explanation for this discrepancy is
not readily apparent; the finding of our study may be due to
chance or may point to a true association. Although days of
menstrual bleeding were well documented, heaviness of
flow was not, so it is unknown whether or not longer
bleeding was accompanied by heavier flow. In either case,
more research in this area is warranted.
This investigation suffered from one major limitation that
deserves attention: The limitation of assessment of physical
activity is based on self-report. Although the quantitative
physical activity history approach used in both cohorts, in
which type of activity, frequency, and duration were
reported for a defined time frame, is probably one of the
most reliable and valid methods of physical activity
measurement for most epidemiologic studies, it still has
many sources of measurement error (37). Most important,
this approach depends on the ability of the participants to
recall and report their activity accurately. It also depends on
the comprehensiveness of the questionnaire (38); inclusion
or exclusion of specific activities and domains of activity can
result in either systematic under- or overreporting (39). The
apparent difference in activity level between the Semiconductor and Michigan cohorts, which precluded classifying
activity levels in the two cohorts into meaningful and
comparable categories, may be due in part to these measurement problems. For instance, overreporting was likely in
both cohorts. National surveys estimate that fewer than 15
percent of women of childbearing age participate in regular,
vigorous recreational physical activity (24), while more than
50 percent of the women in the Michigan cohort and more
than 30 percent in the Semiconductor cohort reported the
equivalent of this level of activity. However, within each
cohort, it is likely that these sources of error were nondifferential and did not affect the relative ranking of persons. As a
result, the measurement errors are unlikely to have systematically biased the results.
This study also had two notable strengths. In both cohorts,
menstrual cycle characteristics were collected prospectively,
with daily diaries in the Semiconductor cohort and monthly
menstrual calendars in the Michigan cohort. This allowed for
more accurate ascertainment of the outcomes than could be
achieved with retrospective reporting. In addition, the
cohorts provided a larger sample size and, in the case of the
Semiconductor cohort, a more racially/ethnically diverse
sample than most previous studies of this question. In addition, both cohorts were broadly representative of the populations from which they were drawn, with the Michigan cohort
constituting a community-based sample that drew from all
socioeconomic and demographic strata. As a result, it may
be reasonable to generalize these findings to other populations of women in the United States.
The finding that more physical activity was related to
longer menstrual cycles is suggestive of a mechanism by
which physical activity could reduce risk of breast cancer.
From a public health perspective, the finding is supportive of
the benefits of promoting regular participation in physical
activity in the population as a whole. However, these data do
not allow any firm conclusion about whether the observed
associations imply a meaningful, clinically relevant difference in reproductive hormone levels. Future studies should
be directed toward assessment of the influence of physical
activity on the underlying cyclical hormone feedback loops
that govern ovulation and the more overt characteristics of
the menstrual cycle.
ACKNOWLEDGMENTS
Funded in part by National Institute of Child Health and
Development (1R01 HD36250), the National Institute of
Environmental Health Sciences (5-K04-ES0000202), the
National Institute of Arthritis and Musculoskeletal Diseases
(R01 AR20557), and the Semiconductor Industry Association.
The authors acknowledge the following persons for their
contributions to study design, data collection, and data analysis of the original cohort studies: Dr. Marc Schenker, Dr.
Bill L. Lasley, Dr. Steven Samuels, Dr. Brenda Eskenazi, Dr.
Farla Kaufman, Marianne O’Neill Rasor, and Mary
Janausch.
REFERENCES
1. Frisch RE, Wyshak G, Albright NL, et al. Lower lifetime occurrence of breast cancer and cancers of the reproductive system
among former college athletes. Am J Clin Nutr 1987;45:328–
35.
2. Bernstein L, Henderson BE, Hanisch R, et al. Physical exercise
and reduced risk of breast cancer in young women. J Natl Cancer Inst 1994;86:1403–8.
3. Thune I, Brenn T, Lund E, et al. Physical activity and the risk of
breast cancer. N Engl J Med 1997;336:1269–75.
4. Friedenreich CM, Bryant HE, Courneya KS. Case-control
study of lifetime physical activity and breast cancer risk. Am J
Epidemiol 2001;154:336–47.
5. Henderson BE, Ross RK, Bernstein L. Estrogens as a cause of
human cancer: the Richard and Hinda Rosenthal Foundation
award lecture. Cancer Res 1988;48:246–53.
6. Henderson BE, Ross RK, Pike MC. Toward the primary prevention of cancer. Science 1991;254:1131–8.
7. Loucks AB, Horvath SM. Athletic amenorrhea: a review. Med
Sci Sports Exerc 1985;17:56–72.
8. Frisch RE, Wyshak G, Vincent L. Delayed menarche and
amenorrhea in ballet dancers. N Engl J Med 1980;303:17–19.
9. Dale E, Gerlach DH, Wilhite AL. Menstrual dysfunction in distance runners. Obstet Gynecol 1979;54:47–53.
10. Glass AR, Yahiro JA, Deuster PA, et al. Amenorrhea in Olympic marathon runners. Fertil Steril 1987;48:740–5.
11. Feicht CB, Johnston TS, Martin BJ, et al. Secondary amenorrhea in athletes. Lancet 1978;2:1145–6.
12. Shangold MM, Levine HS. The effect of marathon training
upon menstrual function. Am J Obstet Gynecol 1982;143:862–
9.
13. Bernstein L, Ross RK, Lobo RA, et al. The effects of moderate
Am J Epidemiol
Vol. 156, No. 5, 2002
Physical Activity and the Menstrual Cycle 409
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
physical activity on menstrual cycle patterns in adolescence:
implications for breast cancer prevention. Br J Cancer 1987;55:
681–5.
Russell JB, Mitchell D, Musey PL, et al. The relationship of
exercise to anovulatory cycles in female athletes: hormonal and
physical characteristics. Obstet Gynecol 1984;63:452–6.
Kaiserauer S, Snyder AC, Sleeper M, et al. Nutritional, physiological, and menstrual status of distance runners. Med Sci
Sports Exerc 1989;21:120–5.
Loucks AB, Mortola LF, Girtoon L, et al. Alterations in the
hypothalamic-pituitary-ovarian and the hypothalamic-pituitaryadrenal axes in athletic women. J Clin Endocrinol Metab 1989;
68:402–11.
Ellison PT, Lager C. Moderate recreational running is associated with lowered salivary progesterone profiles in women. Am
J Obstet Gynecol 1986;154:1000–3.
Beitins IZ, McArthur JW, Turnbull BA, et al. Exercise induces
two types of human luteal dysfunction: confirmation by urinary
free progesterone. J Clin Endocrinol Metab 1991;72:1350–8.
Pirke KM, Schweiger U, Broocks A, et al. Luteinizing hormone
and follicle stimulating hormone secretion patterns in female
athletes with and without menstrual disturbances. Clin Endocrinol 1990;33:345–53.
Broocks A, Pirke KM, Schweiger U, et al. Cyclic ovarian function in recreational athletes. J Appl Physiol 1990;68:2083–6.
Harlow SD, Matanoski GM. The association between weight,
physical activity, and stress and variation in the length of the
menstrual cycle. Am J Epidemiol 1991;133:38–49.
Cooper GS, Sandler DP, Whelan EA, et al. Association of physical and behavioral characteristics with menstrual cycle patterns
in women age 29–31 years. Epidemiology 1996;7:624–8.
Bullen BA, Skrinar GS, Beitins IZ, et al. Induction of menstrual
disorders by strenuous exercise in untrained women. N Engl J
Med 1985;312:1349–53.
US Department of Health and Human Services. Physical activity and health: a report of the Surgeon General. Atlanta, GA:
US Department of Health and Human Services, Centers for
Disease Control and Prevention, National Center for Chronic
Disease Control and Prevention, 1996.
Gold EB, Eskenazi B, Lasley BL, et al. Epidemiologic methods
for prospective assessment of menstrual cycle and reproductive
characteristics in female semiconductor workers. Am J Ind
Med 1995;28:783–97.
Am J Epidemiol
Vol. 156, No. 5, 2002
26. Sowers MF, Crutchfield M, Jannausch ML, et al. Longitudinal
changes in body composition in women approaching the
midlife. Ann Hum Biol 1996;23:253–65.
27. Ainsworth BE, Haskell WL, Leon AS, et al. Compendium of
physical activities: classification of energy costs of human
physical activities. Med Sci Sports Exerc 1993;25:71–80.
28. Taylor HL, Jacobs DR Jr, Schucker B, et al. A questionnaire for
the assessment of leisure time physical activities. J Chronic Dis
1978;31:741–55.
29. LaPorte RE, Montoye HJ, Caspersen CJ. Assessment of physical activity in epidemiologic research: problems and prospects.
Public Health Rep 1985;100:131–46.
30. Block G, Hartman AM, Dresser CM, et al. A data-based
approach to diet questionnaire design and testing. Am J Epidemiol 1986;124:453–69.
31. Bonen A. Recreational exercise does not impair menstrual
cycles: a prospective study. Int J Sports Med 1992;13:110–20.
32. Rogol AD, Weltman A, Weltman JY, et al. Durability of the
reproductive axis in eumenorrheic women during 1 yr of endurance training. J Appl Physiol 1992;72:1571–80.
33. Sanborn CF, Albrecht BH, Wagner WW Jr. Athletic amenorrhea: lack of association with body fat. Med Sci Sports Exerc
1987;19:207–12.
34. Loucks AB. The reproductive system. In: Bar-Or O, Lamb DR,
Clarkson PM, eds. Perspectives in exercise science and sports
medicine. Vol 9. Carmel, IN: Cooper Publishing Group, 1996:
41–66.
35. Williams NI, Caston-Balderrama AL, Helmreich DL, et al.
Longitudinal changes in reproductive hormones and menstrual
cyclicity in cynomolgus monkeys during strenuous exercise
training: abrupt transition to exercise-induced amenorrhea.
Endocrinology 2001;142:2381–9.
36. Harlow SD, Campbell BC. Host factors that influence the duration of menstrual bleeding. Epidemiology 1994;5:352–5.
37. LaPorte RE, Black-Sandler RB, Cauley JA, et al. The assessment of physical activity in older women: analysis of the interrelationship and reliability of activity monitoring, activity
surveys, and caloric intake. J Gerontol 1983;38:394–7.
38. Ainsworth BE. Issues in the assessment of physical activity in
women. Res Q Exerc Sport 2000;71:S43–6.
39. Irwin M, Ainsworth B, Conway J. Estimation of energy expenditure from physical activity measures: determinants of accuracy. Obes Res 2002;9:517–23.