American Journal of Epidemiology
Copyright O 1997 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 146, No. 7
Printed In U.S.A
Ethnic Differences in the Length of the Menstrual Cycle during the
Postmenarcheal Period
Sioban D. Harlow,1 Ben Campbell,2 Xihong Lin, 3 and Jonathan Raz3
In 1989,125 African-American and 123 European-American girls aged 12-14 years were enrolled in a 2-year
study in which they maintained a menstrual calendar, recording the date and amount of menstrual bleeding.
Weight, exercise, and stress during the previous week were recorded at the start of each menstrual cycle.
Although only minor ethnic differences were observed in expected cycle length (29.3 vs. 28.8 days for
European-American and African-American girls, respectively), more prominent differences were observed in
the between-subjects standard deviation of cycle length (2.9 vs. 2.2 days, respectively) and in the odds of
having a cycle longer than 45 days (odds ratio=1.86, 95% confidence interval 1.17-2.97) for EuropeanAmerican compared with African-American girls. Low weight for height and high levels of exercise increased
the probability of having a cycle longer than 45 days and decreased expected cycle length of 13- to 45-day
cycles. Additional investigation of potential ethnic differences in menstrual cycle characteristics is warranted.
Aw J Epidemiol 1997; 146:572-80.
body weight; ethnic groups; exertion; menarche; menstrual cycle; menstruation; reproduction; stress,
psychological
Data on patterns of hysterectomy in the United
States suggest that significant ethnic differences may
exist in the incidence of menstrual dysfunction. Although the total hysterectomy rate is only slightly
higher for African-American (61.7/10,000) compared
with European-American (56.5/10,000) women (1),
the mean age at hysterectomy is significantly younger
(42.0 vs. 46.1 years, respectively) (2). Moreover,
African-American women are twice as likely to have a
diagnosis of fibroids as the indication for hysterectomy (1-3). Recently, we (S. D. H. and B. C.) reported
ethnic differences in the duration of menstrual bleeding during the postmenarcheal period (4). Coupled
with data on ethnic differences related to age at menarche (5, 6), the evidence suggests that differences in
menstrual function may begin at menarche and be
sustained throughout reproductive life. In addition to
problems of menstrual morbidity, menstrual cycle
characteristics may also influence women's long-term
risk of chronic disease (7-9). Yet, despite this evidence of potentially important ethnic differences, large
gaps exist in our knowledge about population variability in menstrual cycle length.
Information on the length and variability of the
menstrua] cycle comes predominantly from four seminal menstrual diary studies, three of which were conducted in populations of predominantly European descent (10-12) and one of which was conducted among
Japanese women (13). More recently, the World
Health Organization (14-16) has published a series of
short-term studies of menstrual bleeding patterns from
industrialized and developing countries including
studies of adolescents (17, 18). A scattering of other
smaller studies has provided data on menstrual function in, for example, adolescent Swiss (19) and Nigerian (20) girls and Indian (21) women. Although
marked geographic differences have been reported in
the timing of puberty and menopause (22) and in the
duration of menstrual bleeding (14, 17, 18), there is
less evidence of a marked regional difference in menstrual cycle length (14). Several studies in the United
States have examined the effect of weight and exercise
on menstrual cycle length and the probability of amenorrhea in teenage girls (23, 24); however, none of
these studies has examined potential ethnic differences
in menstrual cycle length.
Received for publication December 6, 1996, and accepted for
publication June 10, 1997.
Abbreviations: BMI, body mass index; Cl, confidence interval;
GEE, generalized estimating equation; MET, metabolic equivalent;
OR, odds ratio.
1
Department of Epidemiology, School of Public Hearth, University of Michigan, Ann Arbor, Ml.
2
Department of Anthropology, Northwestern University, Evanston, IL
3
Department of Biostatistlcs, School of Public Health, University
of Michigan, Ann Arbor, Ml.
Reprint requests to Dr. Sioban D. Harlow, Department of Epidemiology, University of Michigan, 109 Observatory Street, Ann Arbor,
Ml 48109-2029.
572
Ethnic Differences in Menstrual Cycle Length
This paper examines the relation between menstrual
cycle length and weight, exercise, and stress in a
population-based sample of postmenarcheal AfricanAmerican and European-American girls. In addition to
examining mean cycle length and the probability of
long cycles (25), we also use recently developed statistical techniques to model the within-subject variance in menstrual cycle length (26).
MATERIALS AND METHODS
hi 1989, 125 African-American and 123 EuropeanAmerican girls living in North Carolina were enrolled in
a 2-year study of adolescent behavior and development
This paper is based on menstrual diary data collected
as one component of that study, the details of which
have been reported previously (4). In the parent study,
respondents completed an extensive self-administered
questionnaire at each of five semiannual interviews. Anthropometric measurements were taken and an 18-ml
blood sample was also collected. In addition to keeping a
menstrual diary, a behavioral checklist and saliva sample
were also obtained monthly.
A local county school district provided enrollment
lists of all girls in grades 7 and 8, which included information on ethnicity. A random sample of AfricanAmerican and European-American girls was drawn after
stratifying the list by ethnicity. Girls were contacted
sequentially, evaluated for eligibility, and, if eligible,
asked to participate. To be eligible, girls had to have
reached menarche. Informed consent was obtained first
from the parent and then from the girl herself according
to guidelines of the Institutional Review Board of the
University of North Carolina at Chapel Hill. As expected,
given differences in age at menarche, fewer AfricanAmerican girls (6.3 percent) than European-American
girls (12.4 percent) were premenarcheal and thus ineligible to participate. For the eligible girls, 22 percent of
parents refused permission to contact their daughters,
with no difference in willingness to give permission
observed by ethnicity. For the girls we had permission
to contact, 88 percent of the African-Americans and
74.9 percent of the European-Americans agreed to
participate. Thus the final recruitment rate for eligible
African-Americans was 69.8 percent and for EuropeanAmericans was 58.2 percent The mean age of participants and eligible nonparticipants did not differ.
Girls completed a two-part self-administered questionnaire at the time of enrollment, which provided
demographic data, including self-reported ethnicity, as
well as information about age at menarche, contraceptive use, pregnancy status, and self-reported height and
weight. Height and weight measurements were also
taken. At the semiannual follow-up visits, information
on contraceptive use and pregnancy was reascertained.
Am J Epidemiol
Vot. 146, No. 7,1997
573
Interviewers received standardized training for all
components of the study, including the menstrual
diary.
The girls maintained a weekly menstrual diary. Each
day they experienced bleeding was marked with an X.
On the second day of each menstrual period, women
completed a nine-item questionnaire that included information on their current weight, whether they had
lost weight, gained weight, or dieted in the past week
(yes/no), and whether they had participated in any of
22 different athletic activities such as running, tennis,
or aerobics and, if so, for how many hours. The
four-item Perceived Stress Scale of Cohen et al. (27)
was also included. The time frame of reference for
each variable was within the past week. A case worker
visited each girl's home once a month and collected
her menstrual diary from the previous month.
Menstrual data
Definitions recommended by the World Health Organization were used to summarize the bleeding data
(28). Menstrual cycle length was calculated by counting the number of days from the first day of one
bleeding episode up to and including the day before
the next bleeding episode. A bleeding-free interval
was defined as no bleeding for at least 3 days. A
bleeding episode could not consist of a single day of
spotting. To be eligible for this analysis, a girl had
to contribute at least three menstrual cycles. Seven
girls never returned a month of menstrual diaries,
one girl was pregnant from the time of enrollment until
she dropped out, and 10 girls contributed fewer than
three eligible cycles. Fourteen (11.2 percent) AfricanAmerican girls did not contribute any eligible data as
compared with four (3.3 percent) European-American
girls (p = 0.016).
For the remaining 230 girls included in this analysis,
the menstrual record was censored when a girl
dropped out of the study, when she became pregnant
and continued the pregnancy, when she started birth
control pills and continued using contraception
throughout the study period, or at the end of the 2-year
study period. Seventy-five women (32.6 percent) were
censored before the end of the study, 45 (19.6 percent)
because they withdrew, 25 (10.9 percent) because they
started using hormonal contraception, and five (2.2
percent) because of pregnancy. Twenty-three girls had
missing data at some point during the study period.
African-American girls were more likely to be censored before the end of the study and more likely to be
censored because they started using birth control pills
{p < 0.001). Menstrual characteristics including cycle
length and variability and bleed length did not differ
between girls who dropped out and girls who com-
574
Harlow et al.
pleted the study. Girls who were censored because
they began using birth control pills tended to have a
shorter cycle length (1.7 days mean difference in median cycle length, p = 0.06) during their eligible
cycles than girls who completed the study.
Covariates
Ethnicity was determined by self-reported response
to the question "Are you Black, White/Caucasian,
Asian, Hispanic, American Indian/Alaskan Native, Pacific Islander, Other?". All women defined their ethnicity as either Black or White/Caucasian. Gynecologic age was calculated by subtracting age at
menarche from current chronological age in months.
We measured perceived stress in the past week using
the four-item Perceived Stress Scale of Cohen et al.
(27). This scale was designed to be a global measure
of the extent to which situations in one's life are
appraised as stressful. The items tap the degree to
which respondents find their lives unpredictable, uncontrollable, and overloaded. The four-item version
has been found to be reliable (alpha = 0.72) and is
correlated with life event scores, depressive and physical symptomatology, and social anxiety. Each item is
scored from 0 to 4 and then the four-item scores are
summed. Dieting to lose weight is considered to be a
marker of cognitive dietary restraint, which previously
has been found to influence menstrual function (29,
30); and it was measured by asking the single question, "Did you diet to lose weight in the past week?
(yes/no)". In addition to summing the number of reported physical activities, total moderate to very hard
leisure energy expenditure in the past week was calculated by summing the products of the metabolic
equivalent (MET) rating of each activity and the hours
spent doing diat activity (31). (One MET is one unit
of measurement of heat production by the body.)
Since the distribution of METs is highly skewed,
log-transformed values were used in all regressions.
Weight (kg) divided by height (m) squared was used
as an index of body mass (BMI), with measured height
at baseline and monthly self-reported weight used to
calculate BMI each month. The correlation between
measured weight and self-reported weight at the baseline examination was 0.9. All changes of weight that
exceeded 10 pounds were reviewed for potential errors, and three girls who had four or more unreliable
weight observations are excluded from risk factor
analyses. A self-reported change in weight of more
than 5 pounds during the previous menstrual cycle,
calculated by taking the difference of the previous and
current weight measurements, was used to measure
reported change in weight.
Analysis
Our approach to analyzing menstrual diary data is
more fully described elsewhere (25, 26). In brief,
when the distribution of menstrual cycle lengths is
examined, the bulk of the cycles fall between 20 and
40 days and form an approximate Gaussian distribution, and the remainder create an extremely long nonGaussian right tail. We separately examine the influence of ethnicity and other covariates on each of these
two parts of the distribution, which we refer to as
standard and extreme cycles, respectively. Using the
median and interquartile range to estimate the Gaussian distribution, less than 1 percent of cycles would be
expected to fall beyond 45 days and less than 1 percent
would be fewer than 13 days (25). Hence for these
data, we have chosen 45 days as the cutoff and have
defined cycles of longer than 45 days as extreme.
Since we have multiple observations of bleed characteristics for each girl, standard error estimates for the
regression coefficients obtained from standard regression procedures are inaccurate. Therefore we use linear and logistic models appropriate for repeated measures data. These models adjust the standard error
estimates for the correlation in the data and do not
require an equal number of observations per subject
(32). The regression coefficients are interpreted as one
would in a standard regression analysis.
First, we evaluated covariate effects on the probability of having an extreme cycle, i.e., a cycle length
greater than 45 days. A generalized estimating equation (GEE) approach (32) was used with the SAS IML
macro GEE, assuming exchangeable correlation (i.e.,
that the dependence between two observations within
the individual are the same no matter how far apart in
time they occur). An attractive feature of the GEE
approach is that correct standard errors of the regression coefficient estimates can be obtained using sandwich estimates even when the correlation structure is
misspecified.
We then examined how covariates affect the mean
length of standard cycles using the data of cycle
lengths between 13 and 45 days. The analysis was
carried out by fitting linear mixed models (33) using
SAS procedure PROC MIXED. The impact of categorical covariates on between-women variance in cycle length was evaluated by allowing the random effects for different groups to have different variances.
An additional log linear regression model was constructed to study the association between withinwoman variance and covariates of interest (26). Note
that the between-women variance represents variation
in cycle length among women in a population,
Am J Epidemiol
Vol. 146, No. 7, 1997
Ethnic Differences in Menstrual Cycle Length
575
American girls was 6 months greater at the baseline
examination than it was for European-American girls.
BMI ranged from 14.8 to 44.6 (median = 21.6)
during the study period, with African-Americans
having significantly higher BMIs than EuropeanAmerican girls {p=0.002 at baseline). The median
change in weight from cycle to cycle was 0 pounds,
with a change greater than 5 pounds reported in 10.7
percent of observed cycles. Dieting to lose weight was
reported during only 12.0 percent of menstrual cycles.
Perceived stress scores ranged from 0 to 16, with a
median observed score of 5 and a score of 9 at the 90th
percentile. No difference in perceived stress scores
was observed by ethnicity. The number of activities
reported ranged from 0 to 15 (median = 2), with seven
or fewer activities reported in 99 percent of menstrual
cycles. No exercise was reported during 23.0 percent
of episodes, and the median number of hours of exercise reported was 4. The median total leisure MET
score was 20, which represents 2 hours of very hard
activity or 5 hours of moderate activity. The 95th
percentile for MET scores was 120 or 12 hours of very
hard activity. European-Americans reported doing
more activities with a higher median energy expenditure than did African-Americans (2.5 vs. 1.5 activities,
p <0.001, and 35.8 vs. 16.1 METs, p < 0.001, respectively) (4).
Girls contributed a median of 21 cycles (range three
to 32 cycles) for a total of 4,476 cycles that ranged in
length from 7 to 256 days (median = 28 days). Eighty-
whereas the within-woman variance is variation in
cycle length over time for a woman in a specific
population.
In the unadjusted analyses, the effect of each covariate on expected cycle length and on the probability
of an extreme cycle was determined by introducing
each covariate separately into linear and logistic regression models, respectively. We considered a shift in
expected cycle length of one-half day or more to
represent a meaningful difference. Adjusted regression
models were then built by introducing all covariates
with a statistically significant effect in crude analyses
into the regression models. All possible interactions of
interest were evaluated. This analysis evaluates acute
effects, addressing how differences in risk factor values at the physiologic onset of the menstrual cycle
affect length of that cycle.
RESULTS
At the time of enrollment, girls ranged in age from
12 to 14 years and by the end of the study were 14-16
years old. In table 1, the distribution of covariate
values is summarized by ethnicity. The mean age at
menarche was 11.7 years (range 8-14 years); and
during the course of the study period, gynecologic age
ranged from 0 to 6 years. The mean age at menarche
was 11.5 years for African-American girls compared
with 11.9 years for European-American girls (p <
0.001)), and the mean gynecologic age for African-
TABLE 1. Menstrual characteristics, weight, activity, and stress levels of 111 African-American and 119 European-American girls
aged 12-16 years at baseline and during the 2-year study period, Durham, North Carolina, 1989-1991
African-American girts
European-American girls
Population
Mean
Population
Range
P
value
Mean
Range
Age at baseline (years)
111
13.4
12-14
119
13.3
12-14
0.13
Ago at menarche (years)
111
11.5
8-14
119
11.9
9-14
<0.001
Gynecologic age at baseline (months)
111
23.7
0-57
119
17.1
0-52
<0.001
Median menstrual cycle length* (days)
111
29.8
21-85
119
30.4
22-53
0.49
111
109
22.7
23.6
16.0-36.7
15.8--M).7
CD
%
20.8
21.4
14.6-32.6
15.4-41.0
Weight
Body mass index at baseline
Median*
Ever reported weight change >5
pounds (2.27 kg)
Dieting to lose weight
111
111
Activity
Median no. of activities*
Median total leisure METs*,t
111
111
1.5
16.1
0-7
0-112
119
119
2.5
35.8
0-7
0-136
<0.001
< 0.001
Stress
Median perceived stress score*
111
4.9
0-10
119
4.8
0-11
0.86
O
CD CD
72.1
46.0
* The value for each girl is the median of all her observed values across the study period,
t MET, unit of measurement of heat production by the body.
Am J Epidemiol
Vol. 146, No. 7, 1997
0.002
63.0
55.5
0.14
0.15
576
Harlow et al.
five percent of the study participants reported 12 or
more cycles. For these analyses, 68 cycles from four
women who had unreliable weight data and 30 cycles
less than 13 days are excluded, leaving a total of 4,378
eligible cycles from 226 women. A total of 123
women (54 percent) reported at least one cycle longer
than 45 days (n=294), including 49 African-American
and 73 European-American girls.
Probability of having a cycle longer than 45 days
When the effect of each covariate on the probability of
an extreme cycle was examined, European-American
girls were observed to be 1.76 times more likely to
have a cycle longer than 45 days than were AfricanAmericans (95 percent confidence interval (CI) 1.152.70) (table 2). In this population, an impact of weight
was observed only among the thinnest women. Being
at the lowest 10th percentile of weight for height
(BMI < 17.8 kg/m2) and having a change of more
than 5 pounds in self-reported weight in the current
cycle each increased the odds of having a long cycle
(odds ratio (OR) = 1.75, 95 percent CI 1.17-2.61 and
OR = 1.66, 95 percent CI 1.26-2.18, respectively).
Unlike the previous study of Harlow and Matanoski
(24), dieting to lose weight decreased rather than increased the odds of a long cycle almost twofold.
Exercise was linearly related to the probability of
having a long cycle, such that women at the 75th
percentile of energy expenditure (e.g., 5 hours of very
hard exercise per week) had an odds ratio of 1.2
compared with women at the mean level of energy
expenditure (e.g., 1.1 hours of very hard exercise per
week). As would be expected, girls who were less than
a year postmenarche were much more likely to have a
long cycle (OR = 1.77, 95 percent CI 1.24-2.52).
Girls who had menarche at age 13.5 years or older
(OR = 1.36) were also more likely to have a long
cycle, although the confidence interval for the latter
was wide and did not exclude 1.0. Neither length of
the previous cycle nor level of perceived stress influenced the odds of having a long cycle in these data
(OR = 1.0 for each covariate, data not shown).
The final adjusted model for the probability of having a long cycle is also presented in table 1. Ethnicity,
low weight for height, a large change in weight, high
levels of exercise, reported dieting to lose weight, and
time since menarche remained in the model, and each
appears to be independently associated with the probability of having a cycle longer than 45 days. Only two
odds ratios changed substantially after adjustment.
The odds ratio for being at the 10th percentile of
weight for height decreased from 1.75 to 1.48, and that
for being within 1 year of menarche decreased from
1.77 to 1.44. There was some suggestion of a possible
interaction between ethnicity and gynecologic age,
with the effect of being in the first year after menarche
possibly limited to European-American girls. However, since only 17 girls were observed immediately
after menarche, it was not possible to evaluate this
potential interaction further.
TABLE 2. Logistic regression of ethnicity and other risk factors on the probability of having a cyde longer than 45 days with
robust variance estimation using GEE* for 107 African-American and 119 European-American giris aged 12-16 years, Durham,
North Carolina, 1989-1991
Unadjusted model
OR*
95% CI*
0.57
1.00
1.76
1.15-2.70.
3,740
0.56
1.75
Change in weight > 5 pounds (2.27 kg)
3,838
0.50
Diet to lose weight
4,051
Log leisure METs*4
4,118
Age at menarche > 13.5 years
Time since menarche < 12 months
OR
95% CI
0.62
1.00
1.86
1.17-2.97
1.17-2.61
0.39
1.48
1.00-2.21
1.66
1.26-2.18
0.63
1.88
1.38-2.56
-0.57
0.56
0.37-0.87
-0.64
0.53
0.32-0.85
0.12
1.13
1.05-1.22
0.13
1.14
1.05-1.24
4,378
0.31
1.36
0.82-2.25
4,378
0.57
1.77
1.24-2.52
0.36
1.44
1.00-2.07
4,378
Ethnicity
African-American
European-American
10th percentile of wetght-for-height
(BMI* < 17.8 kg/m*)
Regression
coefficient
Adjusted modeit
Regression
coefficient
No. of
episodes
2
* GEE, generalized estimating equation; OR, odds ratio; CI, confidence interval; BMI, body mass index; MET, unit of measurement of heat
production by the body.
t Includes aD variables shown in table except age at menarche for 3,725 cycles of 105 African-American and 116 European-American
girls.
$ Centered on mean value (11.3 METs).
Am J Epidemiol
Vol. 146, No. 7, 1997
Ethnic Differences in Menstrual Cycle Length
Length of 13- to 45-day cycles
When linear mixed models were used to model the
length of standard (13- to 45-day) cycles (table 3),
ethnicity was observed to have only a slight influence
on expected cycle length, with European-American
girls having cycles 6/10 of a day longer on average
than African-American girls, although the confidence
interval included 0. As was true in Harlow and
Matanoski's previous study (24), fewer risk factors
were found to influence expected length of standard
cycles than influenced the probability of a long cycle.
Girls who reached menarche after age 13.5 years had
cycles that were on average 1.15 days longer than girls
with an earlier age at menarche. Mean cycle length
increased with increasing BMI, such that girls at the
75th and 90th percentiles of BMI (24.6 and 28.5
kg/m2, respectively) had cycles 0.24 and 0.55 days
longer than girls at the median BMI (21.5 kg/m2) in
this population. Similarly, girls at the 25th and 10th
percentiles of BMI (19.2 and 17.6 kg/m2) had cycles
0.17 and 0.31 days shorter, respectively. Mean cycle
length decreased with increasing levels of physical
activity, such that girls at the 75th and 90th percentiles
of activity had cycles on average 0.16 and 0.21 days
shorter than girls at the mean level of exercise,
whereas girls who reported no exercise tended to have
cycles 0.28 days longer. Dieting to lose weight, large
fluctuations in weight, level of perceived stress, and
gynecologic age did not influence mean cycle length
(data not shown).
In the final adjusted model (table 3), AfricanAmerican girls with menarche before 13.5 years, who
had a BMI of 21.5 kg/m2 and an energy expenditure of
11.3 METs, had an expected cycle length of 28.75
days, compared with an expected cycle length of 29.32
days for similar European-American girls. Weight for
577
height and exercise had similar magnitudes of effects
in the adjusted and unadjusted models. Late age at
menarche was somewhat more strongly associated
with having a longer mean cycle length: Girls with
menarche at age 13.5 or older had an expected cycle
length 1.7 days longer than other girls. However, the
confidence interval for this parameter was wide. The
sample size was not sufficient to evaluate potential
interactions between later age at menarche and both
perceived stress and ethnicity.
Variance of 13- to 45-day cycles
The between-women standard deviation in cycle
length was found to differ between these two ethnic
groups (table 3), with African-American girls exhibiting less variability across the population compared
with European-American girls (2.22 vs. 2.95 days,
respectively). The likelihood ratio test for including a
separate between-women variance for each ethnic
group, as opposed to a common variance, was significant (x 2 = 5.43, 1 df). (Note that whether or not
the confidence intervals for African-American and
European-American between-women standard deviations overlap is not a test of the standard deviations
being significantly different) We also examined
potential differences in within-woman standard
deviations. The within-woman standard deviation for
African-American women and European-American
women did not differ significantly (standard deviations = 4.20 and 4.57 days, respectively;/* = 0.13). In
contrast, the within-woman variability did appear to be
greater for women who experienced menarche after
age 13.5 years compared with women who experienced menarche earlier. The within-woman standard
deviations are 5.54 days for women with later men-
TABLE 3. Linear regression of ethnicity and other risk factors on cycle length using a mixed effects model: unadjusted and
adjusted models for 13- to 45-day cycles for 107 African-American and 119 European-American girls aged 12-16 years, Durham,
North Carolina, 1989-1991
Unadjusted model
No. of
ep bodes
Regression
coefficient
(dayajt
Adjusted model*
95% CI*
Regression
coefficient
(days)t
95% Cl
Betweenwomen SDt
(days)
95% Cl
28.97
29.57
28.45 to 29.49
28.99 to 30.15
28.75
29.32
28.24 to 29.27
28.69 to 29.94
2.22
^95
1.78 to 2.59
2.45 to 3.38
0.09
0.02 to 0.16
-0.11
-0.21 to 0.00
1.70
0.40 to 3.00
Ethnicity
African-American
European-American
4,084
Weight-for-height^
3,497
0.08
0.00 to 0.15
Log leisure METs±
3,853
-0.11
-0.21 to -0.02
Age at menarche > 13.5 years
4,084
1.15
-O.08 to 2.39
* Includes all variables shown in table for 3,497 13- to 45-day cycles of 105 African-American and 116 European-American girts,
t Interpreted as expected cycle length (ethnicity) and deviation from expected cycle length (covariates).
i Cl, confidence interval; SD, standard deviation; MET, unit of measurement of heat production by the body, centered on mean value (11.3
METs).
§ Centered on median value (21.5 kg/m2).
Am J Epidemiol
Vol. 146, No. 7, 1997
578
Harlow et al.
arche and 4.25 days for women with earlier menarche
(/?=0.001).
DISCUSSION
This study revealed large ethnic differences in the
probability of having a cycle longer than 45 days and
in the between-women variance of 13- to 45-day cycles during the early postmenarcheal period. More
specifically, among postmenarcheal girls aged 12-16
years, European-Americans have an increased probability of having an extreme cycle, a slightly longer
mean cycle length, and a larger between-women cycle
variance than do African-American girls. Higher levels of exercise and low weight for height each increased the probability of having an extremely long
cycle and of being associated simultaneously with
shorter than average length of standard cycles. Later
age at menarche was associated with a longer mean
cycle length and a larger within-woman variance of
cycle length.
In an earlier report from this same population (4),
differences in bleed duration and the probability of a
heavy bleed were also observed between AfricanAmerican and European-American girls. Age at menarche has also previously been reported to differ between these two ethnic groups within the United States
(5, 6). Thus, ethnicity appears to influence every parameter of the menstrual cycle definable by the phenomenon of menstrual bleeding. The most obvious
ethnic difference, timing of pubertal maturation (5, 6)
as represented by age at menarche, does not appear to
explain the observed differences in menstrual cycle
length. Although age at menarche was associated with
expected cycle length, and time since menarche influenced the probability of having an extremely long
cycle, ethnic differences in cycle length parameters
persisted after adjustment for these parameters.
Ethnic differences also persisted after adjustment
for other risk factors known to influence cycle length,
including weight for height and level of exercise. It is
possible that differences in the distribution of unmeasured risk factors may account for the observed ethnic
differences in cycle length. More detailed measures of
body composition or fat distribution, both of which are
known to influence menstrual function (7, 34-36),
should be included in future studies. Similarly, we had
only one measure of stress in this dataset, which may
not have been sufficient to account for differences in
the experience of stressful experiences across the two
populations (7, 24). African-Americans and other minorities have been found to report higher levels of
perceived stress in a national probability sample of US
adults (37); however, no such difference was observed
in this adolescent sample. Moreover, in contrast to
findings from previous studies (7, 24), perceived stress
did not appear to influence the probability of having a
cycle greater than 45 days. However, since in an
earlier report from these data (4) perceived stress was
found to influence both the duration and the amount of
menstrual bleeding, it is unclear whether the current
result reflects something about the impact of stress on
cycling among postmenarcheal girls or the validity of
this stress measure in adolescents. In either case, additional investigation that includes a more comprehensive stress battery to test the validity of this measure in
adolescent populations may be warranted. Several
other factors such as composition of the diet or history
of exposure to metals or estrogen-disrupting chemicals
may also be found to account for the observed ethnic
differences, aldiough fewer data are available concerning the role of these factors on menstrual function (7).
The possibility of systematic reporting differences
cannot be completely ruled out. However, instructions
for the use of menstrual calendars were standardized,
and differential measurement error is unlikely to completely account for the findings reported here.
Weight for height, exercise, and time since menarche are known to influence menstrual cycle length;
and our findings are consistent with previous studies
(7). The relative importance of low versus high body
mass index varies depending on the weight distribution in the population (7, 24, 34, 38) and probably also
depends on relative distributions of fat versus lean in
the population as well as on fat distribution as measured by the waist/hip ratio (7, 34, 36, 38). Future
studies should measure both of these parameters. The
fact that dieting to lose weight was associated with a
decrease in the probability of having a long cycle and
with no effect on expected cycle length is contrary to
previous findings (24, 30). Again, whedier this finding
reflects something about the origin of long cycles
among adolescents (39-41) or a failure to adequately
capture dieting behaviors in this age group deserves
additional attention.
The importance of examining extremely long cycles
separately from standard cycles (25) was illustrated
here, as both low weight for height and high levels of
exercise were found to increase the probability of
having a very long cycle while also decreasing expected length of standard cycles. This tendency for a
risk factor to both increase and decrease menstrual
cycle length is consistent with proposed models for
ovarian ecology, which suggest a continuum of effects
from altered luteal function, to anovulation, to amenorrhea (42). As we learn more about the distribution of
menstrual cycle lengths and develop models for separately analyzing mean and variance, it should become
Am J Epidemiol
Vol. 146, No. 7, 1997
Ethnic Differences in Menstrual Cycle Length
possible to examine this oppositional effect without
dividing the data.
This is one of the first studies to explicitly model
population differences in the variance in cycle length.
It is of interest to note that different covariates appear
to influence between-women and within-woman variability. The finding that late age at menarche is associated with greater within-woman variance is consistent with previous reports that history of long cycles is
associated with greater within-woman variance (26),
with a greater frequency of oligomenorrhea (39, 40),
and with a greater frequency of endocrine and menstrual abnormalities into adulthood (41).
Generalizability of these findings is clearly limited
by the narrow age range of the sample. Research on
older women should be undertaken because although
these data are consistent with data from 17- to 19-yearold women (24), both studies focused on adolescents.
In addition, no hormonal data were available to evaluate potential differences in the probability of ovulation. Although participation rates by the girls themselves were quite high and little dropout occurred, the
fact that only 70 percent of eligible African-American
and 58 percent of eligible European-American girls
were recruited raises the possibility of selection bias.
For selection bias to have occurred, participation
would also have to have been associated with menstrual cycle length. The one study that examined factors associated with teenagers' participation in menstrual diary studies found no association between
menstrual characteristics and willingness to participate
(24, 43). However, age at menarche and time since
menarche, the menstrual parameters that appear most
likely to affect a girl's decision to participate in a study
of adolescent development, were associated with menstrual cycle length. More information about the potential impact of selection in menstrual diary studies is
needed.
The fact that African-American women are twice as
likely as European-American women to have a diagnosis of fibroids as the reason for hysterectomy (2) and
that African-American women also appear to have an
earlier age at menopause (44) suggest that differences
in uterine and ovarian function may also exist at the
end of reproductive life. Thus the finding of an apparent ethnic difference in menstrual function soon after
menarche suggests the possibility that risk factors for
later menstrual dysfunction may be identifiable early
in reproductive life and warrants further investigation.
In conclusion, the strength of the ethnic difference
in menstrual cycle length and variability observed in
this study is noteworthy. This study has also demonstrated the importance of examining differences in
variance as well as measures of central tendency when
Am J Epidemiol
Vol. 146, No. 7, 1997
579
studying the menstrual cycle. Additional investigation
of potential differences in menstrual characteristics
among different ethnic groups may help elucidate reasons for differences in risk of menstrual and uterine
abnormalities in older women (2).
ACKNOWLEDGMENTS
This work was supported by grant HD30373 from the
National Institute of Child Health and Development. The
data for this analysis were collected as part of a project
funded by grant HD12806 from the Center for Population
Research, National Institute of Child Health and Human
Development, to J. Richard Udry, Principal Investigator,
University of North Carolina at Chapel Hill.
The authors thank Dr. Carolyn Halpern, Project Manager,
University of North Carolina at Chapel Hill, for management of the data collection and Jean Humrich for manuscript preparation.
REFERENCES
1. Wilcox LS, Koonin LM, Pokras R, et al. Hysterectomy in the
United States, 1988-1990. Obstet Gynecol 1994;83:549-55.
2. Kjerulff KH, Guzinski GM, Langenberg PW, et al. Hysterectomy and race. Obstet Gynecol 1993;82:757-64.
3. Bernstein SJ, McGlynn EA, Kamberg CJ, et al. Hysterectomy:
a literature review and ratings of appropriateness. Santa
Monica, CA: The RAND Corporation, 1992.
4. Harlow SD, Campbell B. Ethnic differences in the duration
and amount of menstrual bleeding during the postmenarcheal
period. Am J Epidemiol 1996;144:980-8.
5. McMahon B. Age at menarche. Vital Health Stat Series 11,
1973:133.
6. Harlan WA, Harlan EA, Grillo GP. Secondary sexual characteristics of girls 12-17 years of age: the US Health Examination Study. J Pediatr 1980;96:1074-8.
7. Harlow SD, Ephross SA. Epidemiology of menstruation and
its relevance to women's health. Epidemiol Rev 1995;17:
265-86.
8. Velebil P, Wingo PA, Zhisen X et al. Rate of hospitalization
for gynecologic disorders among reproductive-age women in
the United States. Obstet Gynecol 1995;86:764-9.
9. Whelan EA, Sandier DP, Root JL, et al. Menstrual cycle
patterns and risk of breast cancer. Am J Epidemiol 1994;140:
1081-90.
10. Chiazze L, Brayer FT, Macisco JJ, et al. The length and
variability of the human menstrual cycle. JAMA 1968;203:
377-80.
11. Treloar AE, Boynton RE, Behn BG, et al. Variation of the
human menstrual cycle through reproductive life. Int J Fertil
1967; 12:77-126.
12. Vollman RF. The menstrual cycle. Philadelphia, PA: WB
Saunders and Company, 1977.
13. Matsumoto S, Nogami Y, Ohkuri S. Statistical studies on
menstruation: a criticism on the definition of normal menstruation. Gunma J Med Sci 1962; 11:294-318.
14. Belsey EM, Peregoudov S, and Task Force on Long-Acting
Systemic Agents for Fertility Regulation. Determinants of
menstrual bleeding patterns among women using natural and
hormonal methods of contraception. I. Regional variation.
Contraception 1988;38:227-42.
15. Belsey EM, d'Arcangues C, Carlson N, et al. Determinants of
menstrual bleeding patterns among women using natural and
580
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
Harlow et al.
hormonal methods of contraception, n. The influence of individual characteristics. Contraception 1988;38:243-57.
World Health Organization Task Force on Psychosocial Research in Family Planning, Special Programme of Research,
Development and Research Training in Human Reproduction.
Women's bleeding patterns: ability to recall and predict menstrual events. Stud Fam Plann 1981 ;12:17-27.
World Health Organization Task Force on Adolescent Reproductive Health. World Health Organization Multicenter Study
on Menstrual and Ovulatory Patterns in Adolescent Girls. II.
Longitudinal study of menstrual patterns in the early postmenarcheal period, duration of bleeding episodes and menstrual
cycles. J Adolesc Health Care 1986;7:236-44.
World Health Organization Multicenter Study on Menstrual
and Ovulatory Patterns in Adolescent Girls. World Health
Organization Task Force on Adolescent Reproductive Health.
I. A multicenter cross-sectional study of menarche. J Adolesc
Health Care 1986;7:229-35.
Flug D, Largo RH, Prader A. Menstrual patterns in adolescent
Swiss girls: a longitudinal study. Ann Hum Biol 1984;11:
495-508.
Thomas KD, Okonofua FE, Chiboka O. A study of the menstrual patterns of adolescents in Ile-Ife, Nigeria. Int J Gynecol
Obstet 1990;33:l-4.
Jeyaseelan L, Rao PSS. Correlates of menstrual cycle length in
South Indian women: a prospective study. Hum Biol 1993;65;
627-34.
Gray RH, Doyle PE. The epidemiology of conception and
fertility. In: Barren SL, Thomson AM, eds. Obstetrical epidemiology. London: Academic Press, 1983:25-60.
Bernstein L, Ross RK, Lobo RA, et al. The effects of moderate
physical activity on menstrual cycle patterns in adolescence:
implications for breast cancer prevention. Br J Cancer 1987;
55:681-5.
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.
Harlow SD, Zeger SL. An application of longitudinal methods
to the analysis of menstrual diary data. J Clin Epidemiol
1991 ;44:1015-25.
Lin X, Raz J, Harlow SD. Linear mixed models with heterogeneous within-cluster variances. Biometrics 1997;53:
169-81.
Cohen S, Kamarck R, Mermelstein R. A global measure of
perceived stress. J Health Soc Behav 1983; 24:385-96.
Belsey EM, Farley TMM. The analysis of menstrual bleeding
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
patterns: a review. Appl Stochastic Models Data Anal 1987;
3:125-50.
Harlow SD, Campbell B. Host factors that influence the duration of menstrual bleeding. Epidemiology 1994;5:352-5.
Schweiger U, Tuschi RJ, Platte P, et al. Everyday eating
behavior and menstrual function in young women. Fertil Steril
1992;57:771-5.
Sallis JF, Haskell WL, Wood PD, et al. Physical activity
assessment methodology in the Five-City Project Am J Epidemiol 1985;121:91-106.
Zeger S, Liang KY. Longitudinal data analysis for discrete and
continuous outcomes. Biometrics 1986;42:121-30.
Laird NM, Ware JH. Random-effects models for longitudinal
data. Biometrics 1986;38:963-74.
Symons JP, Sowers MR, Harlow SD. The relationship of body
composition measures and menstrual cycle length. Ann Hum
Biol 1997 ;24:107-16.
Carlberg KA, Buckman MT, Peake GT, et al. Body composition of oligo/amenorrheic athletes. Med Sci Sports Exerc
1983;15:215—17.
Hartz AJ, Rupley DC, Rimm AA. The association of girth
measurements with disease in 32,856 women. Am J Epidemiol
1984;119:71-80.
Cohen S, Williamson GM. Perceived stress in a probability
sample of the United States. In: Spacapan S, Oskamp S, eds.
The social psychology of health. Newbury Park, CA: Sage
Publications, Inc., 1988:31-67.
Cumming DC, Wheeler GD, Harber VJ. Physical activity,
nutrition and reproduction. Ann NY Acad Sci 1994;709:
55-76.
Gardner J. Adolescent menstrual characteristics as predictors
of gynaecological health. Ann Hum Biol 1983;10:31-4O.
Gardner J, Valadian I. Changes over thirty years in an index of
gynaecological health. Ann Hum Biol 1983;10:41-55.
Venturoli S, Porcu E, Fabbri R, et al. Postmenarcheal evolution of endocrine pattern and ovarian aspects in adolescence
with menstrual irregularities. Fertil Steril 1987;48:78-85.
Ellison PT, Cabot TD. Human ovarian function and reproductive ecology: new hypotheses. Am Anthropol 1990;92:
933-52.
Harlow SD, Park M. A longitudinal study of the occurrence,
duration and severity of menstrual cramps in a cohort of
college women. Br J Obstet Gynecol 1996;103:1134-42.
Bromberger JT, Matthews KA, Kuller LH, et al. Prospective
study of the determinants of age of menopause. Am J Epidemiol 1997;145:124-33.
Am J Epidemiol
Vol. 146, No. 7, 1997