The association between alcohol intake and fecundability during menstrual cycle phases

Abstract

STUDY QUESTION

Is increased alcohol intake in different phases of the menstrual cycle associated with fecundability in women?

SUMMARY ANSWER

Heavy intake (>6 drinks/week) of alcoholic beverages in the luteal phase and ovulatory subphase was associated with reduced odds of conception; moderate intake (3–6 drinks/week) during the luteal phase was also associated with reduced fecundability.

WHAT IS KNOWN ALREADY

Despite strong indications for increased risk of infertility among drinking women with intention to conceive, inconsistencies in previous results point to possible residual confounding, and have not thoroughly investigated timing of drinking and other drinking patterns during the menstrual cycle.

STUDY DESIGN, SIZE, DURATION

Participants in The Mount Sinai Study of Women Office Workers (MSSWOW), a prospective cohort study of fertility, were recruited and followed between 1990 and 1994, and completed daily diaries reporting their alcohol intake (type and number of drinks) for a maximum of 19 months of follow-up (N = 413).

PARTICIPANTS/MATERIALS, SETTING, METHODS

Participants were between 19 and 41 years of age. After completion of baseline surveys, they were asked to record their alcoholic beverage intake as number of drinks of beer, wine, and liquor per day, in addition to other exposures such as caffeine and smoking. Furthermore, they submitted urine samples each month to assess pregnancy. Menstrual cycle phases were calculated using the Knaus–Ognio approach. Discrete survival analysis methods were employed to estimate the association between categories of alcohol intake in each phase of menstrual cycle and fecundability.

MAIN RESULTS AND THE ROLE OF CHANCE

In the luteal phase, both moderate drinking (3–6 drinks/week, Fecundability Odds Ratio (FOR)=0.56, CI: 0.31, 0.98) and heavy drinking (>6 drinks/week, FOR = 0.51, CI: 0.29, 0.89) were associated with a reduction in fecundability, compared to non-drinkers. For the follicular phase, heavy drinking in the ovulatory sub-phase (FOR = 0.39, CI: 0.19, 0.72) was similarly associated with reduced fecundability, compared to non-drinkers. For the pre-ovulatory sub-phase, heavy drinking (>6 drinks/week, FOR = 0.54, CI: 0.29, 0.97) was associated with reduction in fecundability, but this association was inconsistent when subjected to sensitivity tests. Each extra day of binge drinking was associated with 19% (FOR = 0.81, CI: 0.63, 0.98), and 41% (FOR = 0.59, CI: 0.33, 0.93) reduction in fecundability for the luteal phase and ovulatory sub-phase respectively, but no association was observed in the pre-ovulatory sub-phase. No meaningful differences in fecundability between beverages were observed in any menstrual phase.

LIMITATIONS, REASONS FOR CAUTION

Patterns of alcohol intake in this cohort suggest a lower average alcohol intake compared to more recent national averages for the same demographic group. Sample sizes were small for some subgroups, resulting in limited power to examine specific beverage types in different phases of the menstrual cycle, or to assess interaction. In addition, the influence of male partner alcohol intake was not assessed, the data relied on self-report, and residual confounding (e.g. unmeasured behaviors correlated with alcohol intake) is a possibility.

WIDER IMPLICATIONS OF THE FINDINGS

Results suggest an inverse association between alcohol and fecundability, and support the relevance of menstrual cycle phases in this link. More specifically, moderate to heavy drinking during the luteal phase, and heavy drinking in the ovulatory window, could disturb the delicate sequence of hormonal events, affecting chances of a successful conception.

STUDY FUNDING/COMPETING INTEREST(S)

Authors declare no conflict of interest. This work was supported by the National Institutes of Health grant, R01-HD24618.

TRIAL REGISTRATION NUMBER

N/A

Introduction

Difficulty conceiving is a notable problem among women seeking pregnancy. In the USA, 12% of couples experience difficulty conceiving or impaired ability to have a live birth (Chandra et al., 2013). Fecundability is defined as the probability of conceiving during a single menstrual cycle (Dunselman et al., 2014), which is a more specific term than fertility, which refers to the biological capacity for reproduction (Louis, 2011). Among the plethora of fecundability risk factors (Baird and Beverly, 2000), alcohol consumption in particular has come under increased scrutiny (Gill, 2000) for being in widespread use, as 86% Americans ≥18 years of age reported consuming alcohol at least once in their lifetime, with >50% during the most recent month (National Survey on Drug Use and Health, 2006).

In comparison to the well-documented impact of alcohol on pregnancy outcomes (Patra et al., 2011), the effect of alcohol intake on fecundability is still debated largely owing to methodological concerns. Factors including bias in selection of reference groups, decrease in the level of drinking with progression of age, underestimation of exposure, absence of context or inadequate adjustment for relevant lifestyle, and confounding factors pose challenges to inferential utility of many alcohol-fecundability studies (Rossi et al., 2014).

While most studies point toward deleterious effects of alcohol (Fan et al., 2017), results from few studies were inconsistent. Among the group of studies reporting no tangible negative effect, a number of limitations were notable. For example, in the Danish cohort study, no consideration was given to the pattern of drinking (e.g. binge drinking) (Mikkelsen et al., 2016). The UK study did not account for a range of factors that could confound the association (Hassan and Killick, 2004). Most importantly, none of the prior studies assessed the temporal pattern of alcohol drinking, an important factor given the hormonal fluctuation during the menstrual cycle (Mumford et al., 2011). It is possible that alcohol may differentially interfere with one or more of the processes of ovulation (Mendelson et al., 1989), conception, or implantation (Rossi et al., 2011). Quantification of possible links between alcohol intake during specific menstrual cycle phase/sub-phases and fecundability may potentially help illuminate the underlying biological pathways.

Another particularly understudied area is whether the association may vary by alcoholic beverage type. An early cohort analysis concluded that wine consumers had slightly shorter time to pregnancy compared to nondrinkers (Juhl et al., 2003), but this was not supported by other studies (Mikkelsen et al., 2016).

To better elaborate, in this paper, we explored whether alcohol intake among premenopausal women would be associated with probability of conception, and the primary objective was to determine whether this association could be narrowed down to a specific phase of the menstrual cycle (pre-ovulatory, ovulatory, and luteal) that may confer greater risk to fecundability. Secondarily, we also examined the effect of binge drinking; whether such associations could be modified at any phase by other factors, including smoking and caffeine intake; and whether certain types of alcoholic beverages (liquor, wine, or beer) may be differently associated with probability of conception.

Materials and methods

Study population

The Mount Sinai Study of Women Office Workers (MSSWOW) was designed to evaluate the reproductive health of working women who were of childbearing age (ages 19–41 years). Women employed at 14 different private and public entities in the states of New York, New Jersey, and Massachusetts were recruited and followed up from 1990 to 1994 (Marcus and Gerr, 1996). A total of 4640 women completed a baseline self-administered questionnaire (Small et al., 2006). Participants were invited to participate in a follow-up study that involved a baseline interview, daily diaries and urine sample collection (details below). Women were excluded if they had not been at risk of pregnancy (i.e. not sexually active in the month leading to study or were consistently using birth control methods), were having fertility treatment, had a hysterectomy, been diagnosed with polycystic ovaries or infertility, had been trying to conceive for at least 12 months prior to study, or had partners with vasectomies. Additionally, they were excluded from the follow-up study analyses if they did not provide any urine samples. Of the remaining 470 women who completed daily diaries and were eligible to be included in the analysis, 57 participants did not provide information on alcohol intake at any point during the study or exhibited short empirical cycles (e.g. <22 days) and were thereby excluded from analyses; the study cohort for this analysis, thus, included 413 individuals. No participant was excluded based on pregnancy intention criteria. Supplementary Fig. S1 provides an illustration of study participants and the selection process. The original research protocol was approved by the Institutional Review Board at Mount Sinai School of Medicine, New York, NY, and Emory University, Atlanta, GA in 1990 and all participants provided written informed consent.

Data collection

Participants were interviewed at baseline for demographic, behavioral, anthropometric characteristics, and reproductive history (Marcus and Gerr, 1996). Information on age, BMI, race, marital status, age at menarche, income level, education, medical and pregnancy history, and intention to get pregnant was recorded. For the follow up period, the study required participants to complete daily diaries and urine collection at least 2 days at each menstrual cycle (Small et al., 2006). Urine samples were provided on Days 1 and 2 of each menstrual cycle (during the first 2 days of bleeding), or when a menstrual cycle was expected if bleeding did not occur, to perform a pregnancy test. Daily diaries contained observations on menstrual cycle characteristics, frequency and time of intercourse during the menstrual cycle, use of birth control methods (e.g. condom), caffeine (number of caffeinated drinks), smoking (number of cigarettes per day), self-reported stress level (on a scale from 1 to 4), physical exercise (number of days reported having exercise that was enough to break a sweat), and intake of drinks containing alcohol.

Assessment and definition of exposure(s)

The number of drinks and types of alcoholic beverages were recorded with daily diaries. Beverages were categorized broadly into three distinct categories: beer, wine, and liquor. Each drink consisted of either 12 ounces of beer, 5 ounces of wine, or 1.5 ounces of liquor. Total alcohol intake per day was calculated as sum of the number of drinks taken across all three categories by participant. To obtain average alcohol intake per day during the cycle, we divided the total number of drinks for the menstrual cycle over the sum of the number of days in which the participant recorded a value for alcohol intake (including zero), excluding days with no entry (i.e. missing days) from the denominator. We used a similar approach to obtain average daily intake of each beverage. To better explore a dose–response association, we created a categorical variable, splitting alcohol intake levels into four groups, which could vary by cycle: no alcohol intake for the cycle (reference); light drinking (1–2 drinks per week (i.e. >0 up to <2.5 drinks)); medium drinking (3–6 drink per week (i.e. 2.5 up to 6)); and heavy drinking (>6 drinks per week). Lastly, to assess the pattern of binge drinking with fecundability, we created a continuous variable where the number of days on which participants reported to have had a binge level alcohol intake (≥4 drinks/day) were tallied by cycle or menstrual phase, respectively.

Estimation of menstrual phases

A typical menstrual cycle consists of two successive phases: follicular and luteal. Of the two, the follicular phase consists of pre-ovulatory and ovulatory sub-phases. Natural variation in the length of menstrual cycles is primarily driven by the length of pre-ovulatory phase (Lenton et al., 1984); in contrast, both ovulatory and luteal phases are more consistent in length. Hence, given the stability of luteal phase, we principally used the ending day of each cycle as the reference point and estimated the menstrual phases backward for each cycle for non-conception cycles. In the absence of hormonal measurements, using the Knaus–Ognio approach (Knaus, 1929; Rötzer, 1968), we have assumed the luteal phase to include the final 14 days of each cycle (Days -14 to -1 before the onset of the next menses). We also assumed 5 days preceding the luteal phase should be the approximate interval for ovulatory phase (Days -19 to -15 prior to proceeding menses) (Holt and Basic, 2010), with remaining days at the beginning of the cycle counted toward pre-ovulatory sub-phase (Supplementary Fig. S2). Implantation likely occurs 6–12 days after fertilization, during the luteal phase (Wilcox et al., 1999). In cycles where pregnancy occurred, the empirical cycle lengths were expectedly long (>38 days). For these cycles, we substituted the empirical cycle lengths with the median length of prior cycles. If the participant had successful conception in the first complete observed cycle, we used their baseline reported ‘cycle length’ to estimate menstrual phases.

To assess the possibility of bias in estimate with our principle approach, we alternatively calculated the median cycle length per participant and used the median cycle length day as the marking point to estimate menstrual phases, regardless of the outcome status for the cycle.

Covariate assessment

After an extensive literature review, preliminary analyses, and using a literature-informed directed acyclic graph (DAG) (Supplementary Fig. S3) to help identify the relevant covariates in the exposure–outcome association, we chose to include BMI categories, age, sociodemographic factors (education categories, income levels, marriage status) (Rostad, 2016), caffeine, smoking categories, stress level (Lynch et al., 2014), physical activity, and parity as covariates in the analysis. We also adjusted for intention to get pregnant because of its association with drinking patterns (Terplan et al., 2014) and probability of conception (Chuang et al., 2009). Even though frequency of unprotected intercourse may partially mediate the association between alcohol and fecundability, we considered the variable as a covariate because we were interested in alternative biological mechanisms by which alcohol may affect fecundability. Additionally, unprotected intercourse could be a surrogate for unmeasured risk-taking behaviors that may affect both fecundability and alcohol intake. Frequency of unprotected intercourse was recorded for all participants irrespective of whether they reported that they were trying to get pregnant at baseline.

For covariates that were reported in the daily diary (caffeine, cigarettes per day, stress), a cycle-level mean was calculated by dividing the sum by the number of days of observations in a cycle, and could vary from cycle to cycle for an individual woman. Likewise, the number of acts of unprotected intercourse in the estimated ovulatory sub-phase could vary from cycle to cycle.

Outcome assessment

Cycles were counted using menstruation data from the daily diaries. One menstrual cycle was defined as the start of a woman’s menstrual period to the start of the next, regardless of the duration of the cycle. If the woman entered the study in the middle of her menstrual cycle, the days until her next period were not used in the analysis. Participants were followed until clinically pregnant or the end of the study.

We excluded suspect polymenorrhic cycles (i.e. shorter than 22 days) (Long, 1990) from analyses (66 out of 2971 cycles). For cycles longer than 38 days (suspect oligomenorrhea), we included 38 days from each cycle (total 24 cycles). (We explored two alternative approaches with long cycles: accounting for up to 65 days of each (max length of majority of long cycles), as well as excluding all long cycles; but point estimates and CIs were highly comparable to selected approach).

Pregnancies were ascertained by two consecutive days of hCG greater than 0.25 ng/ml in a 4-day window beginning either on the first day of menses, or at the expected onset of menses when menses did not occur and confirmed by physician diagnosis. This criterion was derived by analyzing daily urine samples collected from women with tubal ligations (Small et al., 2006). For the uncounted cycles when participants joined the study in the middle of their cycles, urine samples were still collected at the end of that cycle/beginning of the next. No-one had an elevated hCG in a first partial cycle. The Core Laboratory at the Irving Center for Clinical Research at Columbia University and the Center for Clinical Research at Mount Sinai School of Medicine measured hCG. Extensive quality control and split sample comparisons between the two laboratories ensured comparable results (Small et al., 2006).

Statistical approach

Distributions of all covariates were examined. Except for BMI and smoking that were classified using standard approaches, all covariates were treated either as continuous or converted to categorical variables based on empirical distributions. The dataset contained missing observations, which were imputed using a predictive matching mean method assuming missing at random (MAR). Imputation was performed under ‘mice’ package for R.

In longitudinal studies, where participants are followed prospectively with intermediate milestones identified as they occur, a proportion of participants may be enrolled who already passed a number of milestones. This is known as left censoring and has the potential to introduce bias in the outcome estimates (Cain et al., 2011). To account for this, the number of months at risk for pregnancy prior to enrollment to the study was used to reclassify the follow-up cycles. For instance, if a woman answered she had already been at risk for 6 cycles, then her Cycle 1 would be renamed as Cycle 7.

Discrete survival analysis was employed to determine whether alcohol intake was associated with fecundability during any of the menstrual phase/sub-phases or the entire menstrual cycle. This method is more efficient than logistic regression with dichotomous outcome (Baird et al., 1986) and more appropriate than a Cox proportional hazards model, which assumes a continuous measure of time. The discrete hazard obtained from this method is defined as the probability that a woman became pregnant in a menstrual cycle, given that she did not conceive in the previous cycle(s). The underlying probability of conception is assumed to be similar across cycles for the same individual. The fecundability odds ratio (FOR) obtained from the discrete survival analysis model represents the odds ratio for pregnancy per menstrual cycle for a given drinking category versus non-drinkers. The fully adjusted model was used as the base or starting formulation that included all relevant covariates, and then variables were removed which did not significantly confound the exposure–outcome association by ≥ 10% to achieve model parsimony and more precise estimates.

For the binge drinking analysis, the FOR represents the odds ratio for conception per extra day of binge drinking in the cycle or menstrual phase/sub-phase, respectively.

Sensitivity analyses

To examine the robustness of the results, we conducted a number of sensitivity analyses. We conducted subgroup analyses by intention to conceive and explored if the exposure–outcome association differed among those who were trying to get pregnant at baseline from those who indicated otherwise.

As reported in the menstrual phase estimation sub-section, we also estimated menstrual phases using median cycle lengths for each participant regardless of the cycle outcome, in comparison to our principle approach where we used empirical cycle length for non-conception cycles and median cycle length otherwise. Associations between alcohol intake levels and fecundability were assessed under this approach to assess if these differed from estimates obtained under the principle approach.

Given that at least one recent large observational study reported a slightly different length for luteal phase (Bull et al., 2019), in contrast to our approach which is based on 14 days length for luteal phase (Faust et al., 2019), we explored if scales of associations between alcohol intake and fecundability would substantially differ should intervals for menstrual phases be altered. To assess this, we allocated the interval between Days -1 and -12 preceding the next cycle to luteal phase, Days -13 to -17 to ovulatory, and less than -17 to pre-ovulatory sub-phases and ran the same survival analysis as in the main (i.e. select) model.

We also evaluated if long follow-up periods would affect observed associations by excluding follow-up cycles ≥12 from dataset.

To estimate how strong unmeasured confounder(s) would need to be to explain away the association between alcohol intake and fecundability, we calculated E values (VanderWeele and Peng, 2017) for the observed associations in the main model.

Finally, the associations were estimated using exclusively empirical data (where no imputation was carried to replace missing observations) and estimates from this analysis were presented as a supplementary table for comparison. All statistical analyses were completed using R version 3.5 (Team R. Core, 2013).

Results

Study population

A sizeable majority of the study participants were white, non-Hispanic, married women with at least some college education (Table I). A quarter of those included in this analysis reported that they were trying to get pregnant at enrollment. One-third of all participants achieved pregnancy during follow-up (N = 133 out of 413). The overall median follow-up time was 4 cycles (interquartile range, IQR = 2–7). The median reported intake level of alcohol during the study was 0.27 drink/day (IQR = 0.04–0.70), which equates to just under two drinks (1.9) per week (Supplementary Fig. S4); the maximum average daily observation was ∼10 drinks/day; but the maximum total beer and liquor drinks in a day were 20 drinks for each, and 12 for wine, observed in different individuals and days. The distributions of average daily alcohol intake in all menstrual cycle phases/sub-phases were comparable to one another (Supplementary Fig. S4). Those with high average of alcohol intake (>6 drinks per week, Table I) were less likely to conceive during follow-up (27% vs 41% in non-drinking group), and were more likely to be smokers; there was no statistically significant difference using χ² test in proportion of women who were trying to conceive at beginning of the follow-up between heavy drinkers and other groups (Table I).

Table I.

Descriptive characteristics of 413 women by mean alcohol intake level during the follow-up period.

Variable	Heavy Drinkers (Mean intake >6 drinks/ week) (N = 81) No. (%)	Medium Level Drinkers (Mean intake 3-6 drinks/ week) (N = 104) No. (%)	Light Level Drinkers (Mean intake 1-2 drinks/ week) (N = 182) No. (%)	Non-Drinkers (N = 46) No. (%)
Age (years)
19–24	13 (16.0)	5 (4.8)	16 (8.8)	3 (6.5)
25–29	15 (18.5)	43 (41.3)	60 (33.0)	16 (34.8)
30–34	37 (45.7)	33 (31.7)	58 (31.9)	12 (26.1)
35–41	16 (19.8)	23 (22.1)	48 (26.4)	15 (32.6)
Race
White	73 (90.1)	94 (90.4)	147 (80.8)	21 (45.7)
Black	7 (8.7)	7 (6.7)	19 (10.4)	15 (32.6)
Other	1 (1.2)	3 (2.9)	16 (8.8)	10 (21.7)
Marital Status
Married	43 (53.1)	74 (71.2)	124 (68.1)	28 (60.9)
Single	38 (46.9)	30 (28.8)	58 (31.9)	18 (39.1)
Highest education
High school or less	19 (23.5)	22 (21.2)	44 (24.2)	12 (26.1)
Some college	24 (29.6)	37 (35.6)	55 (30.2)	20 (43.5)
College or higher	38 (46.9)	45 (43.3)	83 (45.6)	14 (30.4)
BMI (kg/m2)
<20	15 (18.6)	26 (25.0)	28 (15.4)	4 (8.7)
20–<25	47 (58.0)	46 (44.2)	87 (47.8)	24 (52.2)
25–<30	10 (12.3)	22 (21.2)	39 (21.4)	10 (21.7)
≥30	9 (11.1)	10 (9.6)	28 (15.4)	8 (17.4)
Previous pregnancies
0	30 (37.0)	38 (36.5)	68 (37.4)	16 (34.8)
1	30 (37.0)	26 (25.0)	55 (30.2)	9 (19.6)
2	10 (12.3)	25 (24.0)	27 (14.8)	11 (23.9)
3≤	11 (13.6)	15 (14.4)	32 (17.6)	10 (21.7)
Menstrual cycle length* (Days), median [IQR]
	29.0 [27.0, 31.0]	29.0 [28.0, 32.0]	30.0 [28.0, 32.0]	30.0 [28.0, 33.0]
Months at risk of pregnancy prior to study
0	35 (43.2)	46 (44.2)	78 (42.9)	21 (45.7)
1–<6	25 (30.8)	36 (34.6)	53 (29.1)	9 (19.6)
≥6	21 (26.0)	22 (21.2)	51 (28.0)	16 (34.8)
Trying to get pregnant
Yes	22 (27.2)	21 (20.2)	46 (25.3)	13 (28.3)
No	59 (72.3)	83 (79.8)	136 (74.7)	33 (71.7)
Average frequency of unprotected intercourse during the estimated ovulatory window
<2	65 (80.2)	93 (89.4)	150 (82.4)	39 (84.8)
≥2	16 (19.3)	11 (10.6)	32 (17.6)	7 (15.2)
Conceived*
Yes	22 (27.2)	34 (32.7)	58 (31.9)	19 (41.3)
No	59 (72.8)	70 (67.3)	124 (68.1)	27 (58.7)
Avg cigarettes/day*
0	38 (46.9)	60 (57.7)	111 (61.0)	37 (80.4)
<1–10	28 (34.6)	27 (26.0)	51 (28.0)	5 (10.9)
>10	15 (18.5)	17 (16.3)	20 (11.0)	9 (19.6)
Avg. self-reported stress level* ¥
1.0–1.5	16 (19.8)	13 (12.5)	29 (15.9)	9 (19.6)
>1.5–2.5	54 (66.7)	75 (72.1)	127 (69.8)	39 (84.8)
>2.5–4	11 (13.6)	16 (15.4)	26 (14.3)	6 (13.0)
Avg. exercise level*∅
No exercise	8 (9.9)	3 (2.9)	14 (7.7)	9 (19.6)
Low exercise	28 (34.6)	51 (49.0)	87 (47.8)	20 (43.4)
Medium exercise	33 (40.7)	40 (38.5)	59 (32.4)	8 (17.4)
High exercise	12 (14.8)	10 (9.6)	22 (12.1)	9 (19.6)

^*During the follow-up for the study.

^¥Stress levels: 1 is the lowest, 4 the highest.

^∅Exercise levels: Proportion of days reported exercising during the cycle (Low = up to 20% of days, Medium = up to 50% of the days, High = more than 50% of the days during the cycle).

Effect of alcohol on fecundability

Associations between alcohol intake and fecundability were assessed under multivariable discrete survival analysis approaches (Table II). For the luteal phase, moderate drinking (3–6 drinks/week, FOR = 0.56, CI: 0.31, 0.98) and heavy drinking (>6 drinks/week, FOR = 0.51, CI: 0.29, 0.89) were associated with reduction in fecundability, compared to non-drinkers. For the ovulatory subphase, only heavy drinking (FOR = 0.39, CI: 0.19, 0.72) was associated with reduced fecundability, compared to non-drinkers. Like the ovulatory sub-phase, in the pre-ovulatory sub-phase, only heavy drinking was associated with fecundability (FOR= 0.54, CI: 0.29, 0.97) (Table II). Meanwhile, for the entirety of the cycle, we observed that heavy drinking was associated with a significant reduction in fecundability (FOR = 0.51, CI: 0.27, 0.95), compared to non-drinkers. There was no strong evidence for an association between light or moderate drinking with fecundability, when the entire cycle was considered (Table II).

Table II.

Fecundability odds ratios comparing different dosage groups during menstrual cycle windows.

Alcohol intake level for the interval (drinks/week)	Pre-Ovulatory			Ovulatory			Luteal			Cycle
	N women* (n cycles)	FOR**	95% CI	N women* (n cycles)	FOR**	95% CI	N women* (n cycles)	FOR**	95% CI	N women* (n cycles)	FOR**	95% CI
0	315 (1082)	1.00		344 (1443)	1.00		261 (903)	1.00		192 (628)	1.00
1-2	264 (657)	1.04	(0.64, 1.68)	182 (284)	1.28	(0.73, 2.18)	285 (884)	0.79	(0.50, 1.24)	278 (1054)	0.78	(0.48, 1.29)
3-6	240 (585)	1.14	(0.69, 1.85)	253 (634)	0.94	(0.59, 1.49)	205 (524)	0.56	(0.31, 0.98)	203 (634)	0.78	(0.45, 1.37)
>6	166 (581)	0.54	(0.29, 0.97)	182 (543)	0.39	(0.19, 0.72)	162 (592)	0.51	(0.29, 0.89)	139 (589)	0.51	(0.27, 0.95)

^*Individual women may contribute to multiple drinking categories, if they averaged different number of drinks per cycle during different cycles of the study.

^**Associations adjusted for age, BMI category, physical activity, parity, trying to get pregnant, unprotected sex during the ovulatory window, and months at risk of pregnancy prior to study. FOR: fecundability odds ratio.

By beverage, sub-group analysis suggested that moderate and heavy beer drinking was associated with considerable reduced odds of conception in luteal phase and ovulatory sub-phase (Table III), largely reflecting results from the overall analysis, though estimates for association with heavy drinking in luteal phase were imprecise due to reduced sample size. The direction of association suggested reduced fecundability in heavy beer drinking groups in the pre-ovulatory sub-phase (compared to non-drinkers) but the estimate was imprecise (FOR = 0.67, CI: 0.30, 1.35). For wine, although estimates suggested an inverse association between wine intake and fecundability in both luteal phase and ovulatory sub-phase, the estimates were imprecise largely due to smaller sample sizes (compared to the beer drinking sub-group), particularly for heavy wine drinking (Table III). Similarly imprecise estimates were also observed for liquor in the luteal phase and ovulatory sub-phase. Indeed, detailed assessment of the beverage drinking patterns revealed that while fewer individuals reported consuming beer during follow-up (N = 265) (compared to N = 311 for wine and N = 309 for liquor), the beer drinkers were more likely to be heavy drinkers. In addition, it was observed that the average number of beer binge drinking days per reporting cycle (where the beverage was consumed) was 3.5, over three times higher than wine (1.1 days) and liquor averages (1.2 days), respectively.

Table III.

The association of beer, wine, and liquor with fecundability, among women office workers.

Beverage	Beverage intake level for the window (drinks/week)	Pre-Ovulatory			Ovulatory			Luteal
		N women* (n cycles)	FOR**	95% CI	N women* (n cycles)	FOR**	95% CI	N women* (n cycles)	FOR**	95% CI
Beer	Non-drinkers	382 (1950)	1.00		388 (2222)	1.00		350 (1826)	1.00
	1–2	188 (423)	1.55	(0.96, 2.45)	112 (157)	1.42	(0.72, 2.62)	204 (580)	0.65	(0.39, 1.05)
	3–6	133 (283)	0.81	(0.40, 1.52)	144 (284)	0.55	(0.25, 1.08)	112 (258)	0.47	(0.21, 0.96)
	>6	87 (249)	0.67	(0.30, 1.35)	86 (241)	0.38	(0.13, 0.88)	79 (239)	0.61	(0.26, 1.25)
Wine	Non-drinkers	387 (1871)	1.00		394 (2131)	1.00		353 (1679)	1.00
	1–2	245 (600)	1.48	(0.97, 2.22)	133 (208)	0.65	(0.28, 1.13)	263 (842)	0.97	(0.64, 1.45)
	3–6	147 (288)	0.71	(0.32, 1.49)	183 (409)	0.86	(0.49, 1.43)	116 (244)	0.58	(0.23, 1.22)
	>6	52 (147)	1.13	(0.44, 2.50)	69 (156)	0.56	(0.16, 1.43)	48 (138)	0.88	(0.29, 2.16)
Liquor	Non-drinkers	385 (1994)	1.00		400 (2321)	1.00		379 (1848)	1.00
	1–2	248 (619)	0.92	(0.58, 1.42)	141 (212)	0.72	(0.29, 1.51)	266 (788)	0.63	(0.39, 0.98)
	3–6	118 (220)	1.12	(0.57, 2.04)	147 (278)	0.78	(0.38, 1.47)	100 (193)	0.54	(0.20, 1.20)
	>6	42 (72)	0.75	(0.20, 2.08)	59 (93)	0.93	(0.31, 2.29)	43 (74)	0.79	(0.24, 2.07)

^*Individual women may contribute to multiple drinking categories, if they averaged different number of drinks per cycle during different cycles of the study. Overall, N = 265 individuals reported consuming beer during follow-up compared to N = 311 for wine and N = 309 liquor.

^**Associations adjusted for age, BMI category, physical activity, parity, trying to get pregnant at baseline, and frequency of unprotected sex during the ovulatory window.

We also investigated the association between number of binge drinking days with fecundability in each phase/sub-phase and for the entirety of the cycle. In adjusted analyses, each extra day of binge drinking in a cycle was associated with a 9% reduction in fecundability (FOR = 0.91, CI: 0.81, 1.00). Each extra day of binge drinking was associated with 41% (FOR = 0.59, CI: 0.33, 0.93) and 19% (FOR = 0.81, CI: 0.63, 0.98) reduction in fecundability for the ovulatory sub-phase and luteal phase respectively, but not pre-ovulatory sub-phase (FOR = 0.93, CI: 0.76, 1.11).

Finally, we also explored the possibility of interaction between alcohol and other covariates, including smoking, caffeine, self-reported stress level, exercise, and BMI, but there were no robust or biologically meaningful results in either cycle-wise or phase-specific models.

Sensitivity analyses

In the sub-group analysis among those with and without intention to get pregnant at baseline, noticeable differences emerged (Supplementary Table SI). For those not intending to get pregnant, only heavy drinking had a noticeable, albeit imprecise association (given reduced sample size) with reduction in fecundability in the luteal phase (FOR = 0.55, CI: 0.21, 1.16). There was steeper inverse association with heavy drinking in ovulatory sub-phase (FOR = 0.29, CI: 0.11, 0.67), with imprecise evidence for an inverse association in the pre-ovulatory sub-phase (FOR = 0.54, CI: 0.23, 1.19), compared to non-drinkers. Among participants intending to get pregnant at baseline, any drinking in the luteal phase was associated with reduced fecundability compared to non-drinkers, in spite of an imprecise estimate for heavy drinking level due to the smaller sample size. There was also a suggestion of an effect of heavy drinking during the ovulatory subphase, but the estimate was imprecise (FOR = 0.46, CI: 0.14, 1.26); the sample size was relatively small (40 women, 114 cycles). There was no meaningful evidence for association in the pre-ovulatory sub-phase, although the sample size was reduced.

There were negligibly meaningful variations in scales of associations between alcohol intake levels and fecundability after excluding follow-up cycles exceeding 12 in comparison to estimates reported in Table II (Supplementary Table SII, Model a), except for pre-ovulatory sub-phase where there was no association with any level of drinking. Using different intervals for the menstrual phases, point estimates suggested a similar reduction in fecundability for heavy drinking in the luteal phase, and a precise but relatively attenuated association in the pre-ovulatory sub-phase (Model b, FOR = 0.43 vs FOR = 0.39 in Table II). Lastly, there was no meaningful difference between estimated scales of associations between alcohol intake levels and fecundability where menstrual phases were calculated based on median cycle length approach (Model c, Supplementary Table SII) compared to those reported in Table II, except the association between heavy drinking and fecundability in pre-ovulatory phase, which was marginally imprecise (FOR = 0.58, C: 0.31, 1.07).

The e-value for the effect of unmeasured confounder was 2.15 (lower confidence limit (CL): 1.31) for the luteal phase, 2.58 (lower CL: 1.64) for the ovulatory, 2.06 (lower CL: 1.14) for the pre-ovulatory sub-phases, and 2.15 (lower CL: 1.19) for the entire cycle (using the estimates for heaving drinking groups in Table II). These values indicate that for the observed associations to be explained away by an unobserved confounder, it would need to have a risk ratio of slightly over 2 beyond factors accounted for in the study.

The dataset included approximately 6% missing daily level observations where participants variably failed to report whether they had consumed at least one beverage anytime in a cycle. Individuals with more frequent missingness tended to be less active and smoked more. We imputed missing data points and incorporated them into our main analysis. However, associations in Table II were additionally re-estimated using only empirical data (Supplementary Table SIII), which did not meaningfully differ from results in Table II.

Discussion

These results demonstrate a consistent, clinically meaningful and significant association between heavy drinking during the middle to latter part of the menstrual cycle and fecundability. Results also highlight the importance of the timing element in the association between alcohol intake and fecundability in women. During the luteal phase, a dose–response effect was apparent, with drops in fecundability for even moderate drinkers. During the other phases, the effects for moderate drinking were less consistent and point estimates were not suggestive of linear trend; therefore, caution must be exercised against assuming a linear dose–response association.

Estimates from these analyses are comparable to the similarly sized study of a cohort of 430 couples (Jensen et al., 1998), in which the reported adjusted FORs for 6–10 drinks a week was 0.55, compared to non-drinkers. Also, in an assessment among women in New York and Vermont who were trying to conceive, the FOR among those with >7 drinks/week was 0.65 compared to those who did not drink (Hakim et al., 1998); even though the sample size for this study was substantially smaller (124 individuals).

In comparison, a recent meta-analysis of 19 papers on alcohol consumption and fecundability, involving close to 10 000 women, estimated a risk ratio of 0.77 (95% CI 0.61–0.94) for >7 drinks/week (Fan et al., 2017). But this assessment is a weighted estimate accounting for studies including the Danish cohort study (Mikkelsen et al., 2016) and The Nurses’ Health Study II (Chavarro et al., 2009) that did not find any discernible association between alcohol intake and reduced fecundability at moderate levels (<14 drinks/week).

By menstrual phase, estimates were suggestive of heavy drinking being associated with a markedly lower probability of conception in both the ovulatory sub-phase and luteal phase, compared to non-drinkers. Estimates were robust after excluding participants with long follow-up time, or using a different length for luteal phase, or using alternative methods for estimating menstrual phases. Although sub-group analyses did not suggest an association between light-moderate alcohol intake in the luteal phase and fecundability among participants without intention to get pregnant, the information on intention to conceive was recorded at baseline. It was likely that a proportion of individuals opted otherwise during the follow-up, and, hence, sub-group estimates may not be indicative and require further investigation with cycle-level information for intention to conceive.

We also explored the association between beverage type and fecundability. Analyses showed more individuals with a higher frequency of heavy drinking level with beer. Heavy beer drinking was associated with the lower fecundability in all phase/sub-phases. As noted, the study cohort exhibited a higher tendency for binge and heavy drinking with beer, a behavioral pattern observed in other studies as well (Naimi et al., 2007; Stern et al., 2017). It could be that the association between beer drinking and fecundability reflects the pattern of consumption rather than interaction with chemicals contained in beer.

While the analyses provide strong indications for an association between binge drinking and reduced fecundability in all menstrual phase/sub-phases, the scale of association was steeper in ovulatory sub-phase (FOR = 0.59, CI: 0.33, 0.93) compared to the luteal phase (FOR = 0.81, CI: 0.63, 0.98) per extra day of binge drinking during the window. It is noteworthy that no meaningful association could be observed between binge drinking in the pre-ovulatory sub-phase and fecundability (FOR = 0.93, CI: 0.76, 1.11). Although the number of cycles where binge drinking was reported in preovulatory phase was lower (n = 298) compared to cycles where binge drinking was reported during luteal phase (n = 372), it was higher than binge drinking cycles in ovulatory sub-phase (n = 275). However, it should be considered that while the luteal phase and the ovulatory sub-phase were assumed to have fixed lengths, the pre-ovulatory intervals’ lengths were variable between individuals, with some exhibiting lengthier pre-ovulatory windows. Hence, the discrepancy in estimates could be due to wider temporal distance between binge drinking days, or heavier drinking happening early in the cycle and hence less likely to affect hormonal processes related to ovulation.

The data provide further evidence for a strong effect of heavy drinking in the luteal phase and the ovulatory sub-phases. While intake of alcohol could well result in perturbation of hormonal levels at any time during the menstrual cycle, a sudden imbalance at a short, yet sensitive, hormonal window (e.g. just during or prior to ovulation, gamete transportation or implantation) may result in a significant reduction in fecundability.

Biologically, at least part of association between alcohol intake and fecundability may be mediated through variation in steroid hormones (Gill, 2000), particularly oestradiol (Martin, 1999). How a mistimed upsurge in estradiol levels could affect fecundability is the subject of speculation. However, at least two mechanisms are more frequently proposed: dysregulation of the ovulation process and/or interruption of the course of blastocyst implantation (Gill, 2000), which could overlap or occur independently depending on the context. Under the first scenario, it has been suggested that the sustained increase in oestradiol levels during the menstrual cycle reduces FSH, a gonadotrophin synthesized and released from the anterior pituitary gland; a reduction in FSH level then suppresses folliculogenesis, leading to anovulation (Mendelson et al., 1989; Hakim et al., 1998).

The second hypothesis proposes that alcohol-associated hormonal perturbation could affect fecundability through the implantation process. The implantation process initiates in the luteal phase, following the ovulatory sub-phase, when the ovum is released into the fallopian tube; after fertilization by the sperm, the blastocyst moves toward the uterine endometrium for eventual implantation. The luteal phase is divided into three windows: pre-receptive (first 7 days), receptive (8–10 days), and refractory intervals, respectively (Cha et al., 2012). For successful pregnancy, the conceptus needs to be implanted during the receptive window, and the probability of pregnancy loss increases when the receptive sub-window is shifted toward later days (Wang and Dey, 2006). The timing of these phases is regulated by hormonal feedback mechanisms, specifically the estradiol level (Wilcox et al., 1999; Ma et al., 2003). Oestradiol concentration is also suggested to affect embryonic adhesion to the endometrium such that an abnormally elevated level of hormone could lower implantation chance (Valbuena et al., 2001); however, a very low level of the same hormone can also be adversely associated with implantation probability (Farhi et al., 2000). It implies that successful implantation requires a delicate sequence of molecular events, disruption of which by a sustained high or very low levels of estradiol at the wrong time(s) could adversely affect fecundability (Rashid et al., 2011). Although the potential effects of alcohol intake on the pre-ovulation hormonal processes require further assessment, results from this study provide suggestive evidence for an effect of alcohol on both ovulation and implantation.

These data were collected between 1990 and 1994, and there has been a noticeable increase in alcohol consumption among women of reproductive age both in the USA (Grant et al., 2017) and worldwide (Manthey et al., 2019). Furthermore, the distribution of BMI, a strong risk factor for infertility (Zain and Norman, 2008) in this US subpopulation, points to a leaner cohort compared to a similar demographic group in recent times (Hales et al., 2017). Previous studies suggested that BMI may be an effect modifier for the alcohol–oestrogen association (Shin et al., 2015). Thus, it may be more important to investigate this study question with a current population and larger samples, given these secular trends.

A notable limitation in our assessment is the lack of information on male partners. Alcohol use can lead to decreased sperm production (Grover et al., 2014) and motility (Guthauser et al., 2014). Thus, if the women in this study have a similar drinking pattern to their male partners, the effect of alcohol on the males may be partially responsible for the associations observed.

Self-reporting of alcohol intake can also potentially bias results (Weinberg et al., 1994), particularly if the women who were trying to conceive were more likely to underreport alcohol intake. However, participants reported alcohol intake prospectively, before knowing whether they conceived, which should lessen the effect of underreporting due to social desirability. We also stratified by pregnancy intention, and the results were similar to the main analysis.

The possibility of reverse causation, whereby those who had been trying to conceive for longer at study entry were more or less likely to consume alcohol, could not be directly assessed because information on how long they were trying to get pregnant at study entry was not available. However, the similar distribution of alcohol intake levels among those intending to conceive at the baseline (Table I), compared to other groups, suggested this was less likely.

Since direct measurement of hormones was not available, the method employed for estimation of menstrual phases (Knaus, 1929; Rötzer, 1968) could have led to some misclassification of the phases, particularly during cycles of conception. However, we conducted a sensitivity analysis where a slightly different definition was employed to estimate menstrual phases during the cycle of conception; the trends were similar using both methods, providing more evidence for the stability of the observed associations.

In the absence of direct hormonal measurement, bias due to alcohol use affecting follicular phase length could not be ruled out. However, analysis of cycle lengths comparing groups with different average alcohol intake levels (Table I) suggested similar distributions, which would minimized the scale of bias.

Finally, there is the possibility of residual confounding, whereby other unmeasured behaviors correlated with alcohol intake may have affected fecundability. However, the average e-value for observed estimates among heavy drinkers was >2 (with lower limits outside 1), which suggested an unobserved confounder shall have a risk ratio of 2 of more beyond factors accounted for in the study to explain away associations, adding to the evidence for robustness of the observed associations.

Study strengths include the longitudinal design that allowed a daily assessment of alcohol intake for a period of up to 19 cycles. Since observations were recorded daily, it was possible to assess this association in specific phases of the menstrual cycle. We also were able to adjust our models for a sizeable number of covariates that could have confounded the association, many of which were also assessed prospectively in the daily diary. We were also able to account for left-truncation and late entry via recalibration of risk cohorts across follow-up cycles using the months at risk for pregnancy prior to enrollment variable. Sensitivity analyses, which included removing long follow-up cycles from the study, using varied formulations to calculate menstrual phases, alternative and comparative estimation of menstrual phases in conception cycles, and sub-group analysis by intention to get pregnant, made it possible to assess robustness of the observed association in our study. Finally, in contrast to recent reports, which found a decreased level of drinking among women with intention to conceive (Tough et al., 2006; Pryor et al., 2017), the proportion of women who were heavy drinkers in this study was similar among those trying to get pregnant and those who were not, which increased the power to examine fecundability among the heavy drinkers.

In conclusion, results from these analyses demonstrated an inverse association between alcohol intake and fecundability and support the relevance of menstrual cycle phases. Luteal phase and ovulatory sub-phase heavy drinking exhibited the most consistent inverse association with fecundability irrespective of analytical approach. This is consistent with an impact of alcohol on the implantation and ovulation processes. In some analyses, the pre-ovulatory sub-phase heavy drinking was also negatively associated with probability of conception; however, estimates were uncertain when subjected to sensitivity tests. These results support previous studies that showed heavy and/or binge drinking were predictors for lower probability of conception, and also suggests that even modest drinking levels may decrease fecundability if consumed during critical physiologic intervals of the menstrual cycle.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Authors’ roles

M.Y.A. and K.C.T. envisioned the study and completed the study design. M.Y.A. coded and carried out the statistical analysis and wrote the draft. M.M. reviewed the contents, suggested feedback, provided technical insights, and helped revise the manuscript. K.C.T. supervised the study, contributed to data analysis, and finalized the draft. All authors read and approved the final manuscript.

Funding

This work was supported by the National Institutes of Health grant, R01-HD24618.

Conflict of interest

Authors declare no conflict of interest.