Skip to Main Content

Sign In
Register

Issues
More Content
Submit
Purchase
Alerts
About

search filter
search input
Advanced Search

Volume 160
Issue 2

15 July 2004
Article Contents

Abstract
MATERIALS AND METHODS
RESULTS
DISCUSSION
ACKNOWLEDGMENTS
References

< Previous
Next >
Factors Affecting Menstrual Cycle
Characteristics
Yan Liu Ellen B. Gold Bill L. Lasley Wesley O. Johnson
American Journal of Epidemiology, Volume 160, Issue 2, 15 July 2004, Pages 131140,
https://doi.org/10.1093/aje/kwh188
Published:
15 July 2004

Views

PDF
Cite
Permissions

Share

Abstract
This 19891991 study in California and Utah used daily urinary metabolites of estrogen and
progesterone and computer algorithms to assess ovulatory status and day of ovulation. The
authors examined the associations of risk factors with menstrual cycle characteristics for 309
working women aged 2044 years who collected a median of five cycles each of daily urine
samples. Linear mixed models were used to assess continuous menstrual outcomes.
Compared with women less than age 35 years, women aged 35 years or older had a
significantly decreased (0.94 days, 95% confidence interval: 1.83, 0.05) adjusted mean
cycle length. Age modified the effects of smoking, physical activity, ethnicity, and alcohol
consumption on mean follicular phase length. Asian women had a significantly longer (1.65
days, 95% confidence interval: 0.54, 2.76) adjusted mean cycle length compared with
Caucasian women. Compared with women who did not consume alcoholic drinks, women
who did had a significantly shorter (1.26 days, 95% confidence interval: 2.21, 0.31)
adjusted mean cycle length. Mean cycle and phase lengths were significantly associated with
length of the prior luteal phase. These results indicate that potentially modifiable risk factors,
as well as immutable host factors, are associated with menstrual cycle characteristics that
may in turn be related to subsequent disease risk.

follicular phase; hormones; life style; linear models, statistical; luteal phase; menstrual cycle;
urine
Abbreviations: AOR, adjusted odds ratio; BMI, body mass index; CI, confidence interval;
FSH, follicle-stimulating hormone.
Topic:

physical activity
smoking
estrogen
alcohol drinking
alcoholic beverages
ethnic group
menstrual cycle, proliferative phase
luteal phase
menstrual cycle
menstruation
ovulation
urinary tract
whites
progesterone
urine
ovulatory cycle
metabolites
asian

Issue Section:
ORIGINAL CONTRIBUTIONS

Received for publication March 27, 2003; accepted for publication February 10, 2004.

The length and regularity of menstrual cycles reflect changes in ovarian steroid production (1,
2). Premenopausal steroid hormone levels may have an important influence on subsequent
development of chronic diseases in women (15). Therefore, studies to assess factors
affecting menstrual cycle characteristics may identify high-risk women and risk factors that
may be modified to reduce menstrual cycle disturbances related to subsequent disease risk.

However, self-reported bleeding patterns cannot distinguish ovulatory and anovulatory cycles
or timing of ovulation in ovulatory cycles. Ovulatory and anovulatory menstrual intervals (3)
and different phases of ovulatory cycles (6) may relate to risk factors and long-term disease
risk differently. Thus, assays of prospectively collected daily urine samples for metabolites of
estrogen and progesterone have been used in epidemiologic studies to assess ovulatory status
and timing of ovulation. In addition, if an undetected pregnancy and loss occurs, menstrual
cycle length may be misclassified if self-reported information is used alone (7). Therefore,
metabolites of daily urinary human chorionic gonadotropin must also be assayed to detect
early pregnancy loss (8, 9).

Previous epidemiologic studies have examined the effects of psychological stress in the
workplace (10), caffeine consumption (11), smoking (12), and occupation (13) on menstrual
function by using daily urinary hormone metabolites. In addition, various statistical models
have been used to explore the relations among characteristics of menstrual cycles and phases
(1416). However, these models have been fitted by using data lacking endocrinologic
measures of ovarian function. Murphy et al. (17) examined the effects of time-varying
covariates on menstrual function, incorporating characteristics of the previous cycle, by using
hormone measures assayed in daily saliva samples collected for a fixed time period for all
participants.

The present study examined the effects and potential interactions of lifestyle and
demographic factors on menstrual cycle characteristics. We assessed prospectively collected
daily diary information and assays of daily urine samples for metabolites of reproductive
hormones in a non-clinic-based, free-living sample of working women expected to participate
in the study for a fixed number of menstrual cycles rather than for a fixed time period.

MATERIALS AND METHODS

Study sample

Study participants were part of a larger study (the Semiconductor Health Study in California
and Utah) to examine the effect of work-related activities and exposures on risk of
spontaneous abortion and menstrual cycle disruption among women aged 2044 years who
worked in the semiconductor industry. Recruitment methods and eligibility criteria (18) as
well as laboratory assay methods (8) have been described in detail elsewhere. Briefly, women
were eligible to participate if they had had sexual intercourse in the past 2 months, were not
sterilized, were not using hormonal contraceptives or contraceptive devices, and had male
partners who were not sterilized. The 403 women recruited collected daily diary information
and daily urine samples for a total of 2,015 cycles. Of these women, 338, providing
information on 1,139 cycles, completed daily diaries and daily urine collections for at least
one complete menstrual cycle, had fewer than 3 days of missing data in any 5-day rolling
window, and had complete covariate information. Cycles in which conceptions and losses
occurred were not included in these samples.

Definition of menstrual outcomes

Each urine sample was assayed for metabolites of estrogen (estrone conjugates) and
progesterone (pregnanediol-3-glucuronide) by enzyme immunoassays (19) and was adjusted
for urinary dilution by dividing by creatinine concentration. Menstrual segments were
identified by a recording in the daily diary of 2 or more consecutive days of bleeding or
spotting (with >7 days between bleeding episodes). Determinations of ovulatory status and
day of ovulation were based on algorithms by Waller et al. (20). Menstrual cycle or segment
length was calculated from the first day of menses through the day before onset of the next
menses. Follicular phase length was computed as the first day of menses through the day of
ovulation. Luteal phase length was computed by subtracting follicular phase length from the
corresponding cycle length.

Daily steroid levels were also examined by our study endocrinologist (B. L.) for computer-
determined ovulatory cycles with long (>16 days) or short (<5 days) luteal phases or long
(>28 days) follicular phases or cycles without a determined day of ovulation. As a result of
this review, of those cycles with short or long luteal or follicular phases, 17 cycles were
reclassified as anovulatory cycles, and 25 cycles appeared to be two consecutive cycles for
which information was missing on menstrual bleeding for the beginning of the second cycle.
Twelve cycles without a determined day of ovulation were also reclassified as anovulatory.
These 54 cycles, combined with 142 computer-determined anovulatory cycles, were excluded
from the present analyses of ovulatory cycles. Thus, 29 women were excluded because they
did not contribute information on at least one ovulatory menstrual cycle. For the remaining
309 women (943 ovulatory cycles), day of ovulation for 36 cycles with short or long luteal or
follicular phases was manually redetermined based on the hormone plots (by B. L.) to better
fit the steroid patterns. The endocrinologist also assigned a day of ovulation to each of the 15
ovulatory cycles for which the computer algorithm could not determine that information
because the estrogen/progesterone ratio peaks did not meet the Waller et al. (20) criteria. All
such individual determinations were made without knowledge of the womans risk factor
characteristics.

In addition to examining cycle characteristics as continuous variables, using definitions in the

literature (1012), which identified the fifth and 95th percentiles, the following endpoints
were examined as dichotomous variables: short cycle (<25 days), long cycle (>35 days), short
luteal phase (<11 days), and long follicular phase (>23 days). We also defined long luteal
phase (>15 days) and short follicular phase (<13 days) based on the 10th and 90th percentiles
of their distributions in our data.

Statistical analyses

The linear mixed model (21) for continuous measures and the marginal logistic regression
model for dichotomous outcomes, used in conjunction with generalized estimating equations
(22, 23) to account for the intrawoman correlation of outcomes, are appropriate statistical
tools to analyze menstrual cycle data with multiple cycles for each participant (7, 1012).
Using the linear mixed model (21), we examined the effect of risk factors on cycle and phase
lengths. We modeled the between-woman and within-woman variances and the intrawoman
correlation for continuous outcomes by using the same model. The linear mixed models were
implemented by using the SAS MIXED procedure (SAS Institute, Inc., Cary, North
Carolina). Using the marginal logistic regression model, we computed adjusted odds ratios
and 95 percent confidence intervals to assess the effects of risk factors on binary endpoints.
Analyses were performed by using the SAS GENMOD procedure (SAS Institute, Inc.).

The linear mixed model contains fixed effects, random effects, and errors. Models are
constructed as a composite of terms selected from each component (figure 1). The linear
mixed model was fitted separately for each outcome. To examine the influence of the prior
cycle on the subsequent cycle, final models were also fitted first by using all 943 cycles and
then by using the subset of 634 cycles for women who contributed more than one cycle.
Standard steps for the process of model selection and fitting were as follows: 1) selection of
maximum likelihood or restricted maximum likelihood methods, 2) selection of the
appropriate covariance structure for the data using the method selected, and 3) selection of
covariates based on biologic and statistical considerations using the covariance structure
selected.

We fitted models with all possibilities illustrated in figure 1, as further described below, and
compared results across models. We first fitted the models by using the restricted maximum
likelihood method. We then reran the analyses by using the maximum likelihood method to
compare the results with those obtained from the restricted maximum likelihood method. The
only random effects included were intercept terms, bi. A first-order autoregressive, within-
woman covariance structure (24) was used because menstrual cycle and phase lengths were
less likely to be correlated for nonconsecutive cycles than for consecutive cycles. Because
anovulatory cycles were excluded in the present analyses, ovulatory cycles were not
necessarily consecutive; that is, outcome measures could be missing between any two
observed cycles for a woman. The first-order autoregressive structure takes into account such
intermittent missing data (25). We also implemented other within-woman covariance
structures, for example, the variance components structure (24), which assumes that measures
are independent within women, or the compound symmetry structure (24), which assumes
that measures are equally correlated within women. These covariance structures were
selected by using the Akaike Information Criterion (26).

Age, ethnicity, body mass index (BMI), education, smoking, alcohol and caffeine
consumption, and physical activity were considered potential determinants of outcomes and
were analyzed as categorical variables. Model selection for fixed covariate effects was based
on the p values in the type III F tests and factors that the literature indicated were potentially
important. We implemented a step-forward model selection procedure to choose the variables
important in our data, entering them in order of their p values, the smallest first. Significance
levels of 0.25 for entering and 0.3 for staying were the criteria for including covariates;
significance levels of 0.10 for entering and 0.15 for staying were the criteria for including an
interaction. All two-way interactions were examined.

For cycle length, a model including age, ethnicity, alcohol consumption, and smoking was
obtained by using the stepwise procedure. For follicular phase length, the model included
age, ethnicity, alcohol consumption, smoking, and interactions of smoking and physical
activity with age. For luteal phase length, the model included ethnicity and smoking. In the
literature, BMI and physical activity were consistently shown to be associated with menstrual
function (2, 6, 27). Education and caffeine consumption were neither important in the present
data nor consistently demonstrated to be important in the literature. Thus, age, ethnicity,
BMI, smoking, alcohol consumption, and physical activity were included in the final models
to assess cycle and luteal phase length. The final model assessing follicular phase length
included ethnicity, BMI, smoking, alcohol consumption, and physical activity and was fitted
separately for women aged 34 years or younger and for women aged 35 years or older.

The continuous measures of menstrual outcomes were assessed without transformation since
the residual terms from the models based on nontransformed measures were reasonably
normally distributed and not improved by logarithmic or other transformation.

In a separate analysis, we examined the effect of length of the prior luteal phase on
subsequent cycle length by adding length of previous luteal phase as an additional
covariate to the above-described final models. Dependence between repeated observations on
luteal phase length is built into the model by repeated conditioning of the current response on
the previous response; thus, only the variance component within-woman covariance structure
was used when luteal phase length was also assessed as the response variable. Moreover, to
assess this effect, we used data on a subset of women who provided more than one cycle (239
women, 634 cycles). The linear mixed models were also fitted by using this subset of data
without adding length of previous luteal phase. The potential impact of excluding
participants with only one cycle from the estimation of the effects of risk factors was
investigated by comparing results.

Outcomes were also divided into categories based on the corresponding cutoffs for the
endpoints; for example, cycle length was classified as short (<25 days), normal (2535 days),
or long (>35 days). We used generalized estimating equations to analyze two binary logistic
regression models instead of running a trinomial logistic model because the multinomial
version of the generalized estimating equations method for categorical responses is not
supported by SAS software, version 8.2 (SAS Institute, Inc.). Each adjusted odds ratio was
computed in comparison with a referent: 2535 days for cycle length, 1323 days for
follicular phase, and 1115 days for luteal phase. For comparison with results obtained from
the linear mixed model, the same risk factors (i.e., age, ethnicity, BMI, smoking, alcohol
consumption, and physical activity) were included in the final marginal logistic regression
models.

Both linear mixed models and marginal logistic regression models handle unbalanced data, as
in the case of the present data for which women contributed unequal numbers of ovulatory
cycles for our analyses. In general, results based on linear mixed model and marginal logistic
regression model analyses are valid if outcomes are missing at random or missing completely
at random, respectively (28, 29). The strict criterion (3 days of missing data for any 5-day
rolling window) used for excluding cycles is likely to make the majority of missingness of
data completely at random, that is, independent of menstrual characteristics. Since missing
completely at random is a special case of missing at random, estimates based on both the
linear mixed model and the marginal logistic regression models were valid in the present
analyses.

The differences among all linear mixed models we fitted were relatively minor, and the form
of all linear mixed models was equally complex. Two simple principles were used in our
model-building process: 1) model fit, evaluated by using the likelihood ratio test or
information-based criteria, for example, the Akaike Information Criterion; and 2) biologic
plausibility. Consideration of robustness of results to correlation presumed in the linear
mixed model motivated our use of the marginal logistic regression model. The generalized
estimating equations approach makes no assumption about the form of potential correlation
among responses, although it is designed to accommodate correlation regardless of its form.

RESULTS
Of women smokers, greater proportions were younger, were less educated, were heavier,
were Caucasian, consumed alcohol, and reported more physical activity than nonsmokers not
passively exposed to cigarette smoke or women passively exposed to smoke (table 1).

When the linear mixed model was used, estimates were similar based on either maximum
likelihood or restricted maximum likelihood methods. The first-order autoregressive structure
provided the smallest Akaike Information Criterion values (better model fit). Hence, we
reported the restricted maximum likelihood estimates by using the first-order autoregressive
structure (tables 2, 3, 4, and 5). The only exception for use of the first-order autoregressive
structure was for the model used to determine risk factors for luteal phase length, which also
included length of the prior luteal phase as a predictor. Based on 943 cycles, the
autoregressive parameter () in the first-order autoregressive structure was statistically
significant (p < 0.01) for cycle length but not for follicular phase or luteal phase lengths.
Similar estimates were obtained for these autoregressive parameters based on the 634 cycles
from women who provided data for more than one cycle.

Continuous measures

Adjusted effects of factors on mean cycle, follicular phase, and luteal phase lengths that were
modeled by using three different linear mixed models are shown in tables 2, 3, 4, and 5
(tables 3 and 4 both address follicular phase lengths), respectively. Model 1 was fitted on the
basis of 943 cycles, and models 2 and 3 were based on 634 cycles. Models 1 and 2 included
the same set of covariates. Model 3 added length of the prior luteal phase as a predictor. Our
results indicated that estimates were consistent across models.

Mean cycle length did not vary significantly by BMI, smoking, or physical activity in all
three models. Compared with women aged 34 years or younger, women aged 35 years or
older had a significantly decreased (0.94 days, 95 percent confidence interval (CI): 1.83,
0.05) adjusted mean cycle length (model 1). Asian women had a significantly increased (1.65
days, 95 percent CI: 0.54, 2.76) adjusted mean cycle length compared with Caucasian women
because of a significantly increased (2.05 days, 95 percent CI: 0.53, 3.56) follicular phase
length for those less than age 35 years (model 1). Women who consumed one or more
alcoholic drinks per week had a significantly decreased (1.26 days, 95 percent CI: 2.21,
0.31) adjusted mean cycle length because of a significantly decreased follicular phase length
(1.73 days, 95 percent CI: 3.19, 0.28) for those aged 35 years or older (model 1). Mean
cycle length was significantly inversely associated with the prior luteal phase length (0.18
days, 95 percent CI: 0.36, 0.00; model 3).

Mean follicular phase length did not differ significantly by BMI in all three models, and age
modified the effects of other factors. Compared with women nonsmokers aged 35 years or
older not passively exposed to cigarette smoke, current smokers in the same age group had a
significantly decreased (2.17 days, 95 percent CI: 3.97, 0.37) mean follicular phase length
(model 1). In contrast, follicular phase length was not significantly influenced by smoking for
women less than age 35 years. For women aged less than 35 years, physical activity of 4 or
more hours per week was associated with a significantly increased (2.26 days, 95 percent CI:
0.29, 4.23) adjusted mean follicular phase length (model 3), but no such association was
observed for women aged 35 years or older. In addition, each 1-day increase in the length of
the prior luteal phase was associated with a significant decreased adjusted mean follicular
phase length in the subsequent menstrual cycles for both age groups (model 3).

Mean luteal phase length was significantly associated with length of the prior luteal phase.
Each 1-day increase in the length of the prior luteal phase was associated with an increase of
0.18 days (95 percent CI: 0.10, 0.26) in the subsequent adjusted mean luteal phase length.
Mean luteal phase length was not significantly associated with any other factors in all three
models.

Dichotomous measures

The likelihood of a menstrual cycle shorter than 25 days, based on the marginal logistic
regression model, was not appreciably associated with any of the risk factors examined (table
6). However, women aged 35 years or older were less likely than younger women to have
long cycles (>35 days) (adjusted odds ratio (AOR) = 0.53, 95 percent CI: 0.28, 1.01).
Compared with Caucasian women, Asian women were significantly more likely to have long
cycles (AOR = 2.60, 95 percent CI: 1.20, 5.65) and follicular phases longer than 23 days
(AOR = 2.09, 95 percent CI: 0.97, 4.49). Women who consumed one or more alcoholic
drinks per week were significantly less likely to have long cycles (AOR = 0.38, 95 percent
CI: 0.20, 0.69) or long follicular phases (AOR = 0.39, 95 percent CI: 0.21, 0.72). Physical
activity of 4 hours or more per week compared with no activity was significantly negatively
associated with short follicular phases (<13 days) (AOR = 0.36, 95 percent CI: 0.18, 0.74).
Increasing physical activity was also moderately positively associated with long cycles and
long follicular phases. Smoking status and BMI were not significantly associated with any of
the dichotomous endpoints.

DISCUSSION
Our prospective study is one of the largest to include use of daily hormone measures to assess
menstrual cycle outcomes. Our results show that demographic and lifestyle factors are
associated with menstrual cycle characteristics and that the patterns of effects are similar
whether outcomes are examined as continuous or dichotomous variables.

Previous data on ethnic differences in menstrual function are somewhat limited. Other than
genetic heterogeneity, such racial differences may also be due to diet and other cultural
factors (30). However, the random intercept term in the linear mixed model in the present
study might have accounted for these factors if we assume that they did not vary across
menstrual cycles.

The present results indicated that current smoking was associated with a significant decrease
in mean follicular phase length for women over the age of 35 years. An earlier study of
largely Caucasian women (11) reported that smoking was associated with a decreased mean
menstrual segment length and an increased risk of a short menstrual segment. This finding is
consistent with an effect of smoking on the progression of follicular recruitment and
maturation but not on ovulation or luteal function. This effect is likely one of accelerated
follicular maturation, possibly due to increased pituitary drive, since rate of follicular
development is a direct result of stimulation by follicle-stimulating hormone (FSH) (31).
Increased FSH production could result from perturbation of inhibin production, increased
metabolism of circulating estrogen, or increased hypothalamic drive through interference
with the estrogen negative feedback (31, 32). Thus, we can speculate that exposure to
cigarette smoke may affect FSH production by different mechanisms depending on the
quality or quantity of the exposure. At high, direct exposures, FSH drive increases and
truncates the follicular phase by stimulating ovarian follicles to develop faster (32, 33). The
current study did not provide any direct information as to which mechanism was operating,
however.

While largely speculative, the concept that cigarette smoke affects FSH is consistent with
mechanisms for other related observations. For example, the apparent protective effect of
smoking on breast and endometrial cancer (3, 5, 34, 35) could be a likely result of shorter
periods of exposure to unopposed estrogens due to truncated follicular phases. In addition,
the earlier age at menopause associated with smoking (36) would be consistent with short and
more frequent ovulation and an earlier depletion of oocytes.

In the present study, physical activity of 4 or more hours per week was associated with an
increased cycle length, which could be due to a dampening of FSH pulses during the luteal-
follicular transition, leading to delayed maturation of the next cohort of follicles (26, 31).
Increased cycle length is associated with delayed ovulation and increased follicular phase
length, since luteal phases are self-limited to 14 days (33).

Cycle length has been negatively associated with age because of shortening of the follicular
phase (2, 37), consistent with our results. Harlow et al. (7) demonstrated that overweight is
associated with the probability of long cycles in college women, but we found no association
of BMI with any outcomes in our older and more ethnically diverse population. In addition,
just as we found, alcohol consumption has been reported to be associated with a reduction in
long cycles in young women (38) and changes in hormone dynamics (39, 40).

We modeled the between-woman and within-woman variances, as well as the fixed effects
for risk factors, simultaneously. For example, when the first-order autoregressive within-
woman covariance structure was used, the between-woman standard deviation was 3.45 days
and the within-woman standard deviation was 2.84 days for mean cycle length, based on 943
cycles. This finding confirms the expected greater variability in menstrual cycle length
between women than within women. The autoregressive parameter for cycle length was
0.22 (p = 0.0006), suggesting that cycle length may vary between consecutive ovulatory
menstrual cycles: a pattern of a longer cycle followed by a shorter cycle, and vice versa. The
intrawoman correlation for cycle length was 0.51 between any adjacent pair of observations.

The target population in the present study was women who were experiencing menstrual
cycles and were ovulating. Our results show that excluding women who contributed data on
only one ovulatory cycle did not appreciably change the estimates. Bias was possible because
we excluded these 29 women for having anovulatory cycles only; they might have had
unobserved ovulatory cycles. However, this small subset of the study population was unlikely
to bias the results greatly. There is no issue of imputation in our paper, because 1) excluding
cycles for which data were incomplete was unlikely to cause bias if it did not result in
excluding women, and 2) the missingness mechanism would have to be greatly simplified so
that any imputation of menstrual cycles was doable. It is entirely possible that this
mechanism depends on many factors, including the dropout mechanism and correlation
pattern within women. Therefore, reanalysis with only simple imputation that does not take
such factors into account may result in false security.

One of our strongest findings was that mean cycle (or fol-licular phase) length for ovulatory
cycles was highly predicted by length of the prior luteal phase. This noteworthy finding needs
to be considered in future studies evaluating menstrual cycle and hormonal patterns, since it
suggests that cycle length and dynamics might well be correlated with prior luteal phase
length.

In conclusion, we examined the association of demographic and lifestyle factors with

menstrual cycle characteristics in a large sample of women aged 2044 years by using both
continuous and categorical outcomes and multivariable, repeated-measures statistical
techniques. We found our results to be both internally consistent and largely consistent with
prior studies that did not include data based on daily urinary hormone metabolites to assess
ovarian function. These results suggest that nonmodifiable host factors, such as ethnicity, and
potentially modifiable risk factors, such as smoking, physical activity, and alcohol
consumption, may affect menstrual cycle outcomes. Therefore, both genetic and
environmental factors may influence these characteristics, which in turn are related to long-
term disease risk.

ACKNOWLEDGMENTS
Supported by research grant 88K0962 from the Semiconductor Industry Association and a
Research Career Development Award (5-K04-ES000202) to Dr. Ellen Gold from the
National Institute of Environmental Health Sciences. Also supported by research grants 1RT-
0480 and 8DT-0172 from the Tobacco Related Disease Research Program (TRDRP).

The authors thank Dr. K. O. Waller for sharing her SAS code to assess ovulatory status and
day of ovulation.

Correspondence to Dr. Yan Liu, Center for Family Studies, Department of Psychiatry and
Behavioral Sciences, School of Medicine, University of Miami, 1425 NW 10th Avenue,
Sieron Building, 3rd Floor, Miami, FL 33136 (email: yliu@med.miami.edu).

View largeDownload slide

FIGURE 1. Model building to assess factors affecting menstrual cycle characteristics: the
linear mixed model, California and Utah, 19891991. AR(1), first-order autoregressive; VC,
variance components; CS, compound symmetry; ML, maximum likelihood; REML, restricted
maximum likelihood.

TABLE 1.

Selected baseline characteristics of participants studied regarding factors affecting menstrual

cycle characteristics, by smoking status, California and Utah, 19891991

Passively exposed
NSNPE* Smoker Total
to smoke
(n = (n = (n =
(n = 88)
169) 52) 309)
No. % No. % No. % No. %
Age (years)
29 32 19 11 21 12 14 55 18
3034 59 35 18 35 36 41 113 36
3539 54 32 18 35 21 24 93 30
40 24 14 5 9 19 21 48 16
Ethnicity
Caucasian 74 44 33 64 35 40 142 46
Asian 53 31 8 15 30 34 91 29
Other 42 25 11 21 23 26 76 25
Education (years)
12 38 23 26 50 31 35 95 31
1315 66 39 20 38 35 40 121 39
16 65 38 6 12 22 25 93 30
Body mass index (kg/m2)
<19.0 19 11 6 12 1 1 26 8
19.025.7 107 63 26 50 58 66 191 62
>25.7 43 26 20 38 29 33 92 30
Physical activity (no. of
hours/week)
0 37 22 8 15 17 19 62 20
1 3 90 53 25 48 41 47 156 51
4 42 25 19 37 30 34 91 29
Alcohol consumption (no.
of drinks/week)
0 85 50 16 41 48 55 149 48
1 84 50 36 59 40 45 160 52

* NSNPE, nonsmoker not passively exposed to cigarette smoke.

View Large
TABLE 2.

Factors* affecting mean menstrual cycle length modeled by using different multiple linear
mixed models, California and Utah, 19891991

Adjusted effects
Model 1 (309 Model 2 (239 Model 3 (239
women, 943 women, 634 women, 634
cycles) cycles) cycles)
No. of No. of No. of
95% CI 95% CI 95% CI
days days days
1.83, 1.72, 1.77,
Age (years): 35 0.94 0.80 0.83
0.05 0.11 0.11
Ethnicity#
0.54, 0.84, 0.89,
Asian 1.65 1.98 2.06
2.76 3.13 3.24
0.55, 0.69, 0.66,
Other 0.59 0.53 0.58
1.73 1.74 1.82
Body mass index
(kg/m2)**
2.11, 1.82, 1.87,
<19.0 0.47 0.17 0.18
1.18 1.49 1.51
0.57, 0.75, 0.83,
25.7 0.50 0.39 0.34
1.57 1.53 1.50
Smoking
0.68, 0.66, 0.66,
Passively exposed 0.34 0.40 0.42
1.37 1.46 1.50
1.43, 1.24, 1.18,
Smokers 0.18 0.07 0.16
1.08 1.38 1.50
Alcohol consumption (no. 2.21, 1.77, 1.77,
1.26 0.78 0.75
of drinks/week): 1 0.31 0.21 0.26
Physical activity (no. of
hours/week)
0.70, 0.50, 0.52,
13 0.46 0.67 0.67
1.62 1.84 1.87
0.32, 0.29, 0.33,
4 1.00 1.08 1.06
2.32 2.45 2.46

* Each factor was simultaneously adjusted for all other factors.

For participants who contributed at least two cycles (ni 2).

For participants who contributed at least two cycles (ni 2); in addition to other covariates,
length of the prior luteal phase was also included as a predictor.

CI, confidence interval.

Compared with age 34 years or younger.

# Compared with Caucasian.

** Compared with a body mass index of 19.025.7.

Compared with nonsmokers not passively exposed.

Compared with none.

View Large
TABLE 3.

Factors* affecting mean follicular phase length modeled by using different multiple linear
mixed models (age: 34 years), California and Utah, 19891991

Adjusted effects
Model 1 (168 Model 2 (120 Model 3 (120
women, 473 women, 305 women, 305
cycles) cycles) cycles)
No. of 95% No. of No. of
95% CI 95% CI
days CI days days
Ethnicity
0.53, 0.33, 0.49,
Asian 2.05 1.89 2.06
3.56 3.45 3.63
0.81, 0.96, 0.86,
Other 0.87 0.91 1.02
2.54 2.78 2.89
Body mass index (kg/m2)#
2.94, 2.24, 2.20,
<19.0 0.89 0.22 0.18
1.16 1.79 1.84
1.09, 0.82, 0.91,
25.7 0.46 0.85 0.76
2.02 2.52 2.43
Smoking**
2.11, 2.26, 2.22,
Passively exposed 0.68 0.76 0.72
0.75 0.74 0.78
1.15, 1.30, 1.30,
Smokers 0.57 0.50 0.50
2.29 2.29 2.30
Alcohol consumption (no. 2.29, 2.18, 2.14,
1.01 0.84 0.80
of drinks/week): 1 0.28 0.50 0.53
Physical activity (no. of
hours/week)
0.66, 0.22, 0.20,
13 1.02 1.49 1.51
2.70 3.21 3.23
0.13, 0.27, 0.29,
4 1.75 2.24 2.26
3.63 4.21 4.23

* Each factor was simultaneously adjusted for all other factors.

For participants who contributed at least two cycles (ni 2).

For participants who contributed at least two cycles (ni 2); in addition to other covariates,
length of the prior luteal phase was also included as a predictor.

CI, confidence interval.

Compared with Caucasian.

# Compared with a body mass index of 19.025.7.

** Compared with nonsmokers not passively exposed.

Compared with none.

View Large
TABLE 4.

Factors* affecting mean follicular phase length modeled by using different multiple linear
mixed models (age: 35 years), California and Utah, 19891991

Adjusted effects
Model 1 (141 Model 2 (119 Model 3 (119
women, 470 women, 329 women, 329
cycles) cycles) cycles)
No. of No. of No. of
95% CI 95% CI 95% CI
days days days
Ethnicity
1.07, 0.21, 0.17,
Asian 0.56 1.56 1.56
2.19 3.33 3.29
Other 0.00 1.60, 0.84 0.92, 0.82 0.90,
 1.59 2.60 2.54
2
Body mass index (kg/m )#
2.66, 2.49, 2.56,
<19.0 0.07 0.45 0.33
2.80 3.40 3.21
0.49, 1.27, 1.30,
25.7 1.00 0.39 0.32
2.49 2.05 1.94
Smoking**
0.01, 0.90, 0.89,
Passively exposed 1.51 0.73 0.71
3.03 2.36 2.30
3.97, 3.40, 3.19,
Smokers 2.17 1.42 1.24
0.37 0.57 0.71
Alcohol consumption (no. 3.19, 3.23, 3.17,
1.73 1.63 1.60
of drinks/week): 1 0.28 0.03 0.04
Physical activity (no. of
hours/week)
0.95, 0.93, 0.94,
13 0.71 0.86 0.81
2.37 2.66 2.57
1.07, 1.32, 1.27,
4 0.81 0.76 0.77
2.69 2.84 2.80

* Each factor was simultaneously adjusted for all other factors.

For participants who contributed at least two cycles (ni 2).

For participants who contributed at least two cycles (ni 2); in addition to other covariates,
length of the prior luteal phase was also included as a predictor.

CI, confidence interval.

Compared with Caucasian.

# Compared with a body mass index of 19.025.7.

** Compared with nonsmokers not passively exposed.

Compared with none.

View Large
TABLE 5.

Factors* affecting mean luteal phase length modeled by using different multiple linear mixed
models, California and Utah, 19891991

Adjusted effects
Model 1 (309 Model 2 (239 Model 3 (239
women, 943 women, 634 women, 634
cycles) cycles) cycles)
 No. of 95% No. of No. of
95% CI 95% CI
days CI days days
0.45, 0.55, 0.50,
Age (years): 35 0.04 0.06 0.06
0.37 0.43 0.37
Ethnicity#
0.22, 0.35, 0.35,
Asian 0.29 0.26 0.20
0.80 0.88 0.75
0.31, 0.68, 0.67,
Other 0.22 0.03 0.09
0.75 0.62 0.48
Body mass index
(kg/m2)**
0.60, 0.96, 0.88,
<19.0 0.16 0.07 0.09
0.91 0.82 0.70
0.71, 0.81, 0.67,
25.7 0.22 0.20 0.12
0.28 0.41 0.42
Smoking
0.45, 0.26, 0.23,
Passively exposed 0.02 0.31 0.27
0.50 0.88 0.78
0.11, 0.26, 0.25,
Smokers 0.46 0.44 0.37
1.04 1.14 0.99
Alcohol consumption (no. 0.22, 0.07, 0.05,
0.21 0.46 0.43
of drinks/week): 1 0.65 0.99 0.90
Physical activity (no. of
hours/week)
0.67, 0.81, 0.69,
13 0.14 0.18 0.13
0.39 0.45 0.42
0.78, 1.01, 0.89,
4 0.17 0.28 0.24
0.43 0.45 0.41

* Each factor was simultaneously adjusted for all other factors.

For participants who contributed at least two cycles (ni 2).

For participants who contributed at least two cycles (ni 2); in addition to other covariates,
length of the prior luteal phase was also included as a predictor.

CI, confidence interval.

Compared with age 34 years or younger.

# Compared with Caucasian.

** Compared with a body mass index of 19.025.7.

Compared with nonsmokers not passively exposed.

Compared with none.

View Large
TABLE 6.

Adjusted odds ratios and 95% confidence intervals for dichotomous menstrual endpoints by
risk factors,* California and Utah, 19891991

Dichotomous menstrual endpoints

Short Long Short Long
Short cycle Long
follicular follicular luteal luteal
(<25 cycle (>35
phase (<13 phase (>23 phase (<11 phase (>15
days) days)
days) days) days) days)
95% 95% 95% 95% 95% 95%
OR OR OR OR OR OR
CI CI CI CI CI CI
Age (years): 0.67, 0.28, 0.73, 0.34, 0.74, 0.50,
1.14 0.53 1.19 0.62 1.15 0.97
35 1.94 1.01 1.92 1.14 1.81 1.87
Ethnicity
0.38, 1.20, 0.51, 0.97, 0.38, 0.34,
Asian 0.77 2.60 0.94 2.09 0.68 0.81
1.53 5.65 1.71 4.49 1.20 1.92
0.45, 0.26, 0.38, 0.45, 0.78, 0.86,
Other 0.84 0.69 0.68 1.04 1.32 1.98
1.55 1.79 1.23 2.44 2.25 4.56
Body mass
index (kg/m2)
0.33, 0.15, 0.70, 0.17, 0.24, 0.03,
<19.0 0.90 0.63 1.52 0.62 0.66 0.20
2.46 2.58 3.30 2.27 1.80 1.54
0.54, 0.44, 0.72, 0.50, 0.59, 0.28,
25.7 1.00 1.09 1.24 1.12 0.96 0.62
1.83 2.69 2.16 2.51 1.57 1.35
Smoking#
Passively 0.59, 0.90, 0.60, 0.89, 0.68, 0.70,
1.13 1.79 1.05 1.67 1.17 1.51
exposed 2.18 3.54 1.86 3.11 2.04 3.25
0.54, 0.64, 0.68, 0.55, 0.43, 0.82,
Smokers 1.05 1.52 1.18 1.30 0.78 1.86
2.07 3.66 2.14 3.09 1.41 4.20
Alcohol
consumption
0.45, 0.20, 0.70, 0.21, 0.49, 0.49,
(no. of 0.81 0.38 1.19 0.39 0.78 1.06
1.46 0.69 2.03 0.72 1.24 2.28
drinks/week):
1**
Physical
activity (no. of
hours/week)**
0.57, 0.56, 0.49, 0.63, 0.63, 0.35,
13 1.08 1.31 0.85 1.33 1.19 0.74
2.05 3.08 1.47 2.82 2.25 1.57
0.37, 0.59, 0.18, 0.57, 0.64, 0.32,
4 0.77 1.75 0.36 1.54 1.31 0.76
1.63 5.24 0.74 4.14 2.67 1.79

* Each factor was simultaneously adjusted for all other factors.

 OR, odds ratio; CI, confidence interval.

Compared with age 34 years or younger.

Compared with Caucasian.

Compared with a body mass index of 19.025.7.

# Compared with nonsmokers not passively exposed.

** Compared with none.

View Large

References
1.
Kato I, Toniolo P, Koenig KL, et al. Epidemiologic correlates with menstrual cycle length in
middle aged women.
Eur J Epidemiol

1999
;
15
:
809
14.
Google Scholar
2.
Harlow SD, Ephross SA. Epidemiology of menstruation and its relevance to womens health.
Epidemiol Rev

1995
;
17
:
265
86.
Google Scholar
3.
Henderson BE, Ross RK, Judd HL, et al. Do regular ovulatory cycles increase breast cancer
risk?
Cancer

1985
;
56
:
1206
8.
Google Scholar
4.
Landgren BM, Unden AL, Diczfalusy E. Hormonal profile of the cycle in 68 normally
menstruating women.
Acta Endocrinol

1980
;
94
:
89
98.
Google Scholar
5.
Whelan EA, Sandler DP, Root JL, et al. Menstrual cycle patterns and risk of breast cancer.
Am J Epidemiol

1994
;
140
:
1081
90.
Google Scholar
6.
Harlow SD, Zeger SL. An application of longitudinal methods to the analysis of menstrual
diary data.
J Clin Epidemiol

1991
;
44
:
1015
25.
Google Scholar
7.
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.
Google Scholar
8.
Lasley BL, Lohstroh P, Kuo A, et al. Laboratory methods for evaluating early pregnancy loss
in an industry-based population.
Am J Ind Med

1995
;
28
:
771
81.
Google Scholar
9.
Eskenazi B, Gold EB, Samuels SJ, et al. Prospective assessment of fecundability of female
semiconductor workers.
Am J Ind Med

1995
;
28
:
817
31.
Google Scholar
10.
Fenster L, Waller K, Chen J, et al. Psychological stress in the workplace and menstrual
function.
Am J Epidemiol

1999
;
149
:
127
34.
Google Scholar
11.
Fenster L, Quale C, Waller K, et al. Caffeine consumption and menstrual function.
Am J Epidemiol

1999
;
149
:
550
7.
Google Scholar
12.
Windham GC, Elkin EP, Swan SH, et al. Cigarette smoking and effects on menstrual
function.
Obstet Gynecol
1999
;
93
:
59
65.
Google Scholar
13.
Gold EB, Eskenazi B, Hammond SK, et al. Prospectively assessed menstrual cycle
characteristics in female wafer-fabrication and nonfabrication semiconductor employees.
Am J Ind Med

1995
;
28
:
799
815.
Google Scholar
14.
Bogumil RJ, Ferin M, Rootenberg J, et al. Mathematical studies of the human menstrual
cycle. I. Formulation of a mathematical model.
J Clin Endocrinol Metab

1972
;
35
:
126
43.
Google Scholar
15.
Shack WJ, Tam PY, Lardner TJ. A mathematical model of the human menstrual cycle.
Biophys J

1971
;
11
:
835
48.
Google Scholar
16.
Marshall J. Predicting length of menstrual cycle.
Lancet

1965
;
191
:
263
5.
Google Scholar
17.
Murphy SA, Bentley GR, OHanesian MA. An analysis for menstrual data with time-varying
covariates.
Stat Med

1995
;
14
:
1843
57.
Google Scholar
18.
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.
Google Scholar
19.
Munro CJ, Stabenfeldt GH, Cragun JR, et al. Relationship of serum estradiol and
progesterone concentrations to the excretion profiles of their major urinary metabolites as
measured by enzyme immunoassay and radioimmunoassay.
Clin Chem

1991
;
37
:
838
44.
Google Scholar
20.
Waller K, Swan SH, Windham GC, et al. Use of urine biomarkers to evaluate menstrual
function in healthy premenopausal women.
Am J Epidemiol

1998
;
147
:
1071
80.
Google Scholar
21.
Laird NM, Ware JH. Random effects models for longitudinal data: an overview of recent
results.
Biometrics

1982
;
38
:
963
74.
Google Scholar
22.
Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models.
Biometrika

1986
;
73
:
13
22.
Google Scholar
23.
Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes.
Biometrika

1986
;
42
:
121
30.
Google Scholar
24.
SAS Institute, Inc. SAS/STAT users guide, version 8. Cary, NC: SAS Institute, Inc, 1999.
Google Scholar
25.
Thibaut R, Jacqmin-Gadda H, Chne G, et al. Bivariate linear mixed models using SAS proc
MIXED.
Comput Methods Programs Biomed

2002
;
69
:
249
56.
Google Scholar
26.
Akaike H. A new look at the statistical model identification.
IEEE Trans Automat Control

1974
;
19
:
716
23.
Google Scholar
27.
Sternfeld B, Jacobs MK, Quesenberry CP Jr, et al. Physical activity and menstrual cycle
characteristics in two prospective cohorts.
Am J Epidemiol

2002
;
156
:
402
9.
Google Scholar
28.
Little RJA, Rubin DB. Statistical analysis with missing data. New York, NY: John Wiley &
Sons, 1987.
Google Scholar
29.
Lipstiz SR, Molenberghs G, Fitzmaurice GM, et al. GEE with Gaussian estimation of the
correlations when data are incomplete.
Biometrics

2000
;
56
:
528
36.
Google Scholar
30.
Bagga D, Ashley JM, Geffrey SP, et al. Effects of a very low fat, high fiber diet on serum
hormones and menstrual function. Implications for breast cancer prevention.
Cancer

1995
;
76
:
2491
6.
Google Scholar
31.
De Souza MJ, Miller BE, Sequenzia LC, et al. Bone health is not affected by luteal phase
abnormalities and decreased ovarian progesterone production in female runners.
J Clin Endocrinol Metab

1997
;
82
:
2867
76.
Google Scholar
32.
Shideler SE, DeVane GW, Kalra PS, et al. Ovarian-pituitary hormone interactions during the
perimenopause.
Maturitas

1989
;
11
:
331
9.
Google Scholar
33.
Vollman RF. The menstrual cycle. Philadelphia, PA: W B Saunders, 1977.
Google Scholar
34.
Persson I. Estrogens in the causation of breast, endometrial and ovarian cancersevidence
and hypotheses from epidemiological findings.
J Steroid Biochem Mol Biol

2000
;
74
:
357
64.
Google Scholar
35.
Terry PD, Miller AB, Rohan TE. A prospective cohort study of cigarette smoking and the
risk of endometrial cancer.
Br J Cancer

2002
;
86
:
1430
5.
Google Scholar
36.
Gold EB, Bromberger J, Crawford S, et al. Factors associated with age at natural menopause
in a multiethnic sample of midlife women.
Am J Epidemiol

2001
;
153
:
865
74.
Google Scholar
37.
Dennerstein L, Gotts G, Brown JB. Effects of age and non-hormonal contraception on
menstrual cycle characteristics.
Gynecol Endocrinol

1997
;
11
:
127
33.
Google Scholar
38.
Cooper GS, Sandler DP, Whelen EA, et al. Association of physical and behavioral
characteristics with menstrual cycle patterns in women age 2931 years.
Epidemiology

1996
;
7
:
624
8.
Google Scholar
39.
Reichman ME, Judd JT, Longcope C, et al. Effects of alcohol consumption on plasma and
urinary hormone concentrations in premenopausal women.
J Natl Cancer Inst

1993
;
85
:
722
7.
Google Scholar
40.
Mendelson JH, Mello NK. Chronic alcohol effects on anterior pituitary and ovarian hormones
in healthy women.
J Pharmacol Exp Ther

1988
;
245
:
407
12.
Google Scholar

View Metrics

Email alerts

New issue alert

Advance article alerts
Article activity alert
Receive exclusive offers and updates from Oxford Academic

More on this topic

Meta- and Pooled Analysis of GSTT1 and Lung Cancer: A HuGE-GSEC Review
Reliability of the Behavioral Risk Factor Survey in a Triethnic Population
Use of Urine Biomarkers to Evaluate Menstrual Function in Healthy Premenopausal Women
Ethnic Variations in Mammographic Density: A British Multiethnic Longitudinal Study

Related articles in

Web of Science
Google Scholar

Related articles in PubMed

Effects of Combined Aerobic-Strength Training vs Fitness Education Program in COPD

Patients.
Impact of anatomical placement of an accelerometer on prediction of physical activity energy
expenditure in lower-limb amputees.
Weight loss from 20 years of age is associated with cognitive impairment in middle-aged and
elderly individuals.
Induced gamma band activity from EEG as a possible index of training-related brain
plasticity in motor tasks.

Citing articles via

Web of Science (54)

Google Scholar
CrossRef

Latest
Most Read
Most Cited

Fish intake, genetic predisposition to alzheimer's disease and decline in global cognition and
memory in five cohorts of older persons
Relationship of lipids and lipid-lowering medications with cognitive function: The Multi-
Ethnic Study of Atherosclerosis
Mood Disorders and Risk of Herpes Zoster in Two Population-Based Case-Control Studies in
Denmark and the United Kingdom
Exercise for the Prevention of Low Back Pain: Systematic Review and Meta-Analysis of
Controlled Trials
Characterizing and Comparing the Seasonality of Influenza-Like Illnesses and Invasive
Pneumococcal Diseases Using Seasonal Waveforms

About American Journal of Epidemiology

Editorial Board
Author Guidelines
Facebook
Twitter

LinkedIn
Purchase
Recommend to your Library
Advertising and Corporate Services
Journals Career Network

Online ISSN 1476-6256

Print ISSN 0002-9262
Copyright 2017 Johns Hopkins Bloomberg School of Public Health

About Us
Contact Us
Careers
Help
Access & Purchase
Rights & Permissions
Open Access
Connect

Join Our Mailing List

OUPblog
Twitter
Facebook
YouTube
Tumblr

Resources

Authors
Librarians
Societies
Sponsors & Advertisers
Press & Media
Agents

Explore

Shop OUP Academic

Oxford Dictionaries
Oxford Index
Epigeum
OUP Worldwide
University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the

University's objective of excellence in research, scholarship, and education by publishing
worldwide

Copyright 2017 Oxford University Press

Privacy Policy
Cookie Policy
Legal Notices
Site Map
Accessibility
Get Adobe Reader