Research Portfolio

Funding Opportunities

Join our Mailing List
Join our mailing list to be notified of new funding opportunities.

Your Email

To receive information about funding opportunities, events, and program updates.

Modeling the relation of smoking to the ovarian function

Institution: University of California, Davis
Investigator(s): Yan Liu, M.D.
Award Cycle: 1999 (Cycle 8) Grant #: 8DT-0172 Award: $53,448
Subject Area: Epidemiology
Award Type: Dissertation Awards

Initial Award Abstract
Other than its well-known carcinogenic effect, tobacco smoke might also act as a toxin to the reproductive endocrine environment in women. Menstruation and ovarian steroid patterns are primary markers of human female reproductive endocrinologic and ovarian function. Researches only involve laboratory animals or a few women have show that tobacco smoke could influence ovarian sex hormone biosythesis. Investigation of menstruation and its hormonal bases based on large-sample-size data is limited. In fact, major difficulties in analysis and interpretation of cyclic menstrual and hormone data arise when much of the observed variability in menstrual characteristics stems from changes within women as opposed to differences between women and steroid data are curved over an entire menstrual cycle.

The improvement of less-cost biological assay techniques, rapid development of computer hardware/software and considerable exploration of modern statistical inference methods allow us to model human ovarian function based on cycle-to-cycle daily diary and urine collection now. The primary goal of the research proposed is to explore methodology for investigating the influence of cigarette smoking in the presence of other host and environmental factors on the ovarian function of women based on daily diary and urinary assay data. The study data were collected previously. The importance of this existing database is that it contains information on urine specimens collected not only over an entire menstrual cycle for each of 402 women but also over several successive cycles for most of those women. The effects of cigarette smoking on ovarian steroid secretion and metabolic clearance rate (MCR) of progesterone, in premenopausal women will be assessed with control of variables we collected, such as age, race, parity, body mass index (BMI), alcohol use, caffeine consumption and physical activity. We also want to evaluate the relation of menstrual cycle length and follicular and luteal phase lengths with tobacco exposure.

The response variables from an individual are usually correlated since they were measured repeatedly over time. In the present proposed study, appropriate models that allow for such correlation is discussed. Menstrual characteristics will be modeled as linear functions of the smoking and other variables using the general linear mixed model (GLMM). But typical menstrual cycle measures may not be sensitive indicators of changes or disturbances in hormonal function resulting from cigarette smoke exposure. We will model curves which reflect hormone profiles over time using variants of the GLMM, which are expected to be more sensitive to such changes. The above analyses rely on the assumption of normality of the data and on having moderate to large sample sizes. We will perform a special technique, bootstrap analysis, in order to check on these assumptions and to provide inferences that are less dependent on these assumptions.

The present proposed study should shed light on the mechanisms by which cigarette smoke adversely affects women's endocrinologic and menstrual function and provide new approaches for scientists in their efforts to detect more sensitively the tobacco-related health effects with respect to reproduction in premenopausal women.

Final Report
The main purpose of this study was to make accurate estimates for the effect of smoking on the ovarian function, using appropriate statistical models and algorithms for assessing ovulatory status and timing of ovulation applied to daily urinary hormone measures.

We used the Kassam et al. algorithm to detect anovulatory cycles in our data. Since assessment of the outcome, anovulation, was imperfect, we performed a logistic regression analysis that allowed for an imperfect outcome determination for some menstrual cycles, to estimate the adjusted effect of smoking on anovulation, based on the ovulatory status determined by the Kassam algorithm. This special logistic regression model based on one menstrual cycle randomly selected from each woman indicated smoking to be associated with a moderately increased risk of anovulation, and passive smoke exposure was not associated with anovulation.

We examined the effect of smoking on menstrual cycle characteristics based on the multiple correlated cycles on the same woman. The day of ovulation for each ovulatory cycle was determined based on the Waller et al. algorithm. Linear mixed effects models were used to assess continuous menstrual characteristics to account for the potential correlation of within-woman measurement. Adjusted mean cycle length and mean follicular phase length did not significantly vary by smoking status, but interaction of smoking with age was observed for mean follicular phase length. Adjusted mean luteal phase length was not related to smoking. Random effects logistic regression using generalized estimation equations was also used for binary menstrual cycle outcomes. Second-hand smoke exposure was related to an increased risk of long menstrual cycles (>35 days) and long follicular phases (>23 days). Current smoking was not significantly related to any of the dichotomous endpoints.

We modeled the mean curves for women’s urinary estrogen and progesterone metabolite profiles as well as examined the effect of smoking on those hormone levels in the time course, respectively, based on hormone measures assayed in the daily urine samples in one randomly selected menstrual cycle from each participant, using the semiparametric stochastic mixed model developed by Zhang et al. Two particular approaches were employed using Zhang’s model. We first modeled the random curves across smoking status with the continuous variate age and categorical variable body mass index as the fixed effects. Our goal was to compare differences in the shapes of mean hormone profiles across smoking status. We then fitted Zhang’s model with fixed effects for categorical variables for body mass index, ethnicity, smoking, alcohol consumption and physical activity as well as the continuous variate age to evaluate factors that may affect hormone excretion. We applied these two approaches to the entire menstrual cycles and separately to the follicular and luteal phases, respectively. Results obtained from the first approach indicated that smokers exhibited a less pronounced estrogen metabolite excretion at the midcycle as well as in the luteal phase. Smokers also had a more pronounced progesterone metabolite excretion level during the entire menstrual cycle. Women who passively exposed to smoke did not differ in either estrogen or progesterone metabolite excretion compared with non-smokers not passively exposed. Results obtained from the second approach were consistent with the first approach. Smoking was associated with a significantly increased level of progesterone metabolite excretion in both follicular and luteal phases as well as the entire cycle.

To conclude, work to date followed the study protocol well. We have made considerable progresses on all our primary goals. Manuscripts to address each sub-study have been completed to submit to scientific journals.