Observational analyses
Data from the UK Biobank, EPIC (European Prospective Investigation into Cancer and Nutrition) and 12 other studies included in the EHBCCG (Endogenous Hormones and Breast Cancer Collaborative Group) consortium were used.
UK Biobank
This is a prospective cohort study involving about 500 000 adults, including over 270 000 women, aged 40–69 years when recruited between 2006 and 2010. At the initial assessment visit, usual alcohol intake was assessed using a touchscreen questionnaire, and blood samples were collected from which serum was prepared and concentrations of hormones and SHBG were measured using chemiluminescent immunoassays. The current analysis included pre-menopausal women, who reported they had not had their menopause (i.e., periods had not stopped), and were younger than 50 years of age, and post-menopausal women, who reported they had gone through menopause, or were 55 years or older, or reported a bilateral oophorectomy; those who had a prior history of cancer ((except for non-melanoma skin cancer) or reported currently using hormone therapy (hormone replacement therapy (HRT) and/or oral contraceptives (OCs)) were excluded. Detailed information on the study design and methodology [19], calculation of alcohol intake in grams per day [18] and the assay data [20] has been reported elsewhere.
EPIC
This is a prospective cohort study involving about 520 000 adults, including over 360 000 women, aged 25–70 years when recruited from 23 centres across 10 European countries between 1992 and 2000. Diet, including usual alcohol intake, was measured by country-specific questionnaires that were validated against reference measurements based on twelve 24-hour diet recall interviews [21]. Blood samples were collected from about 74% of the participants. The current analysis included pre- and post-menopausal women from nested case-control studies on breast, ovarian, endometrial, cervical, liver and thyroid cancer risk for whom serum (in most of these studies) or plasma concentrations of sex hormones and SHBG were measured. Both pre-cases (women who were cancer-free at the time of blood collection but were subsequently diagnosed with the cancer of interest during follow-up) and controls were included, except for the liver cancer study where only controls were included. Participants were categorised as pre-menopausal if they reported regular menstrual cycles over the 12 months prior to blood collection or were younger than 42 years at recruitment, and as post-menopausal if they reported having had no menses over the past 12 months, were older than 55 years, or reported a bilateral oophorectomy. Women who reported currently using hormone therapy (HRT and/or OCs) were excluded, as well as those from Greece (due to a restriction concerning information governance). Detailed information on the study design and methodology [22], calculation of alcohol intake in grams per day and the assay data [23] has been reported elsewhere.
The EPIC study data for breast cancer were included in the EHBCCG but the EPIC data were analysed separately here because, since the publication of the collaborative analyses, more nested case-control studies of other cancer sites have been conducted and hormone assay data are now available for a larger sample of women.
EHBCCG: Of the seven prospective studies of pre-menopausal women included in the collaborative analysis,[16] three with available information on usual alcohol intake were included: Nurses’ Health Study II (NHS-II), USA; New York University Women’s Health Study (NYU WHS), USA; and the Study of Hormones and Diet in the Etiology of Breast Tumors (ORDET), Italy. Of the 18 studies of post-menopausal women, 11 were included: Cancer Prevention Study-II Nutrition Cohort (CPS-II Nutrition Cohort), USA; Malmö/Umeå, Sweden; the Melbourne Collaborative Cohort Study (MCCS), Australia; the Multiethnic Cohort (MEC), USA; Nurses’ Health Study (NHS I), USA; NYU WHS, USA; ORDET, Italy; Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial cohort (PLCO), USA; Study of Osteoporotic Fractures (SOF), USA; United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS), UK; and the Women’s Health Initiative, Observational Study (WHI-OS), USA. Women who reported currently using hormone therapy (HRT and/or OCs) were excluded.
Details of the individual studies included in this analysis are presented in the Supplementary Materials. These include: references for component studies within EPIC and EHBCCG (Supplementary Table S1), number of women who contributed to each hormone analysis (Supplementary Table S2), measurement of usual alcohol intake (Supplementary Table S3), and blood sample (serum vs. plasma), type of assay and coefficients of variation for the measured hormones and SHBG (Supplementary Table S4). In all studies, concentrations of free oestradiol and testosterone were calculated from those of total oestradiol and testosterone respectively and of SHBG, assuming that the binding of these hormones to serum SHBG and albumin follows the law of mass action [24]. As albumin concentration was not measured in EPIC and EHBCCG, it was assumed to be constant at 40 g/L [25].
Statistical analysis
Analyses were undertaken separately for pre- and post-menopausal women in UK Biobank, EPIC and EHBCCG. STATA 17 (StataCorp, College Station, Texas) was used for all analyses.
Hormone concentrations were logarithmically transformed. In pre-menopausal women, concentrations were standardised for phase of the menstrual cycle (early follicular, late follicular, mid-cycle, early luteal, mid-luteal and late luteal) with residuals from the mean for each cycle phase. The cycle phase was determined using forward dating (UK Biobank [18]), or both forward and backward dating with the latter used where possible (EPIC [26] and EHBCCG [16]).
For each study, hormone concentrations and 95% confidence intervals (CIs) per 10 g/day (approximately one standard drink/day) increment in alcohol intake were estimated using multivariable linear regression models, adjusting for individual component studies (EPIC and EHBCCG), case-control status (EPIC and EHBCCG), age at blood collection (in 2-year categories for pre-menopausal women and 5-year categories for post-menopausal women), previous alcohol use among non-current drinkers (UK Biobank and EPIC), smoking (never, former, current), body mass index (BMI) (< 22.5 kg/m2, 22.5–24.9 kg/m2, 25-27.4 kg/m2, 27.5–29.9 kg/m2, 30-34.9 kg/m2, ≥ 35 kg/m2), number of full-term pregnancies (0, 1, 2, 3, 4+), past use of hormone therapy (HRT and/or OCs; yes/no), age at menopause (in 3-year categories; post-menopausal women only) and menopausal type (natural, surgical; post-menopausal women only). The study-specific results were then pooled using fixed-effect meta-analysis. Potential differences in the estimates by menopausal status were assessed using the Chi-square test for heterogeneity.
In pre-menopausal women, subgroup analyses were undertaken for total oestradiol, oestrone, progesterone and total testosterone by phase of the menstrual cycle (follicular, mid-cycle and luteal). In both pre- and post-menopausal women, subgroup analyses were undertaken for total oestradiol, oestrone and total testosterone by type of the assay used (direct, extraction and mass spectrometry); the individual studies that contributed to each assay type are presented in Supplementary Table S5. Sensitivity analyses were undertaken by restricting the sample to those who reported alcohol intake of < 15 g/day, to those who reported intake of < 30 g/day (i.e. excluding heavy drinkers), and also to those whose blood samples were collected during an ovulatory cycle (progesterone concentrations measured in the mid-luteal phase ≥ 12.72 nmol/L (~ 400 ng/dL) [27].
MR and colocalisation analyses
Data on alcohol intake
A genetic instrument in the ADH1B (Alcohol Dehydrogenase 1B) gene (rs1229984) for self-reported alcohol intake (number of drinks per week) was extracted from a GWAS (genome-wide association study) meta-analysis undertaken by the GWAS and Sequencing Consortium of Alcohol and Nicotine Use (GSCAN) [28]. This variant was used due to its highly biologically plausible association with alcohol intake [29]. The minor A allele of this variant increases the activity of ADH1B that oxidises ethanol to acetaldehyde, resulting in unpleasant reactions and limiting further drinking [30]. While this polymorphism is less common in people of white European ancestry with a frequency of < 5% (cf. 90% in East Asians), it is nonetheless a strong genetic predictor of alcohol intake in this population [30]. Estimates were available per one SD (approximately 9 drinks/week) increment in alcohol intake and extracted from the GWAS meta-analysis excluding the UK Biobank (n = 226 223) to avoid sample overlap between the GWAS for alcohol intake and that for hormone concentrations. The ADH1B variant explains 0.19% of the variance in alcohol intake.
Data on testosterone and SHBG: Summary statistics for the association of rs1229984 with SD increments in the concentrations of hormones and SHBG were obtained from a publicly available GWAS of all women, irrespective of menopausal status, from the UK Biobank, extracted from the OpenGWAS platform [31] (dataset used for total testosterone: ieu-b-4864 involving 199 569 women; free testosterone: ieu-b-4869 involving 180 386 women; and SHBG: ieu-b-4870 involving 214 989 women). Data on oestradiol were available but were not used due to the potential limitations related to measurement of this hormone in the UK Biobank (see details in the Discussion); data on the other sex hormones were not available.
MR analyses
MR assesses the associations between exposure(s) and outcome(s) using genetic variants associated with the exposure of interest as instrumental variables. A Wald ratio was calculated using the “TwoSampleMR” [32] package in R. To be able to present the MR results in a way which is directly comparable to the observational results, assuming that one standard drink contains 10 g of alcohol, the β estimates generated from the Wald ratio (per one SD increment in alcohol intake) were converted to the estimates per 10 g/day increment. The results were then multiplied by 0.341 (assuming that, for a normal distribution, one SD is 34.1% of the range) to convert the difference in hormone concentrations from units expressed as SD to percentages.
Colocalisation analyses: Colocalisation assesses the probability that two traits are affected by the same genetic variants at a given locus. Using the ADH1B variant, colocalisation analyses were conducted to identify the presence of a shared causal locus between alcohol intake and concentrations of testosterone and SHBG where a conventionally significant association was observed in MR analyses. The “coloc” package [33] in R was used to estimate the posterior probability for two traits sharing the same causal variant (PP4) in a 150 kb LD (linkage disequilibrium) window centred on rs1229984, with PP4 > 0.70 corresponding to strong evidence of colocalisation [34]. Priors chosen were: p1 = 10− 3, p2 = 10− 4, and p12 = 10− 5, or approximately a 75% prior belief that a signal will only be observed in the GSCAN GWAS and < 0.01% prior belief in favour of colocalisation between the two traits at a given locus [35].