Study participants
We used data from the mothers of the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort. Full details of the study have been previously reported [17, 18]. ALSPAC enrolled 14,541 pregnancies in the South West of England (around the city of Bristol) with an expected delivery date between 1st April 1991 and 31st December 1992. The participating families have been followed up through to the current day [17]. Please note that the study website (http://www.bristol.ac.uk/alspac/researchers/our-data/) contains details of all the data and interview guides that are available through a fully searchable data dictionary and variable search tool.
In 2009–2011, all mothers still engaged with the study (N = 11,264) were invited to a follow up assessment clinic, with 4834 (43%) of invited women attending. The participating women were older and more educated than the original sample recruited in pregnancy [18]. A further three follow-up assessment clinics, each successively 1 to 2 years apart, were undertaken focusing on women who were pre-menopausal in the initial clinic and therefore likely to go through the menopausal transition during the subsequent three assessments, reflecting the aim to explore social, lifestyle, health and biological changes as women go through the menopausal transition [19]. This study is restricted to these three later clinics in which cognitive function tests were administered. Figure 1 describes the participant flow into the analyses. Women were included irrespective of whether they changed through one or all three of the menopausal stages of pre-, peri- and post-menopause as our primary exposures were not these categories but reproductive age and hormones. Women who had undergone surgical menopause at baseline or follow up were excluded, as were women reporting using hormone replacement therapy (HRT) or hormonal contraception at baseline, so that the focus was on changes occurring across a natural menopause. Observations for women who reported using HRT or hormonal contraception in follow up were also censored at the last point before reported use. The analysis sample consisted of 2411 women with 1386 women participating in all three assessment clinics. A majority of the participants (97%) were White British.
Participant flow into eligible and analysis groups, 2011–2015
Reproductive age
Women were asked a detailed set of questions about the date of their last menstrual period and the regularity of their menses by interview at each assessment clinic. These questions were designed to be able to categorise participants into Stages of Reproductive Aging Workshop (STRAW) categories [15]. FMP could be identified when at least 1 year of amenorrhea had occurred since the date of the last menstrual period. Using this information, reproductive age was calculated retrospectively using years since FMP and coded as zero when women were pre-FMP. Reproductive age could not be measured before FMP due to the relatively small number of women having their FMP during the study follow up. A binary variable on whether the woman had reached their FMP was also determined for each assessment clinic.
Hormones
Levels of FSH, LH, and AMH were assessed from fasting samples in women at the three assessment clinics without restrictions on which day in the menstrual cycle the participants were at the time of blood sampling. Women were instructed to fast overnight or for at least 8 h before the clinic visit, and the blood samples were processed within 4 h and stored at − 80 °C until thawed for hormonal analyses (with no previous thaw-freeze cycles). Serum FSH, LH and AMH were measured with a Roche Elecsys modular analytics Cobas e411 using an electrochemiluminescence immunoassay. The AMH assay used was the fully automated Elecsys AMH Plus immunoassay from Roche Diagnostics [20].
Menopausal stage
STRAW criteria [15], using the date of the last menstrual period and the regularity of menses, were used to categorise women into menopausal stages [19]. In this study, we condensed the more detailed categories into (i) pre-menopausal (reproductive, STRAW categories − 5 to -3a), (ii) peri-menopausal (menopause transition and first year post-menopause, STRAW − 2, − 1 and + 1a) and (iii) post-menopausal (from second year post-menopause, STRAW +1b to + 2).
Cognitive tests
Six different cognitive tests were administered at each of the three assessment clinics according to a standardised protocol to assess specific domains of cognitive function (see Table 1). Higher scores on each test reflect better cognitive function.
Confounders
We adjusted for (1) educational attainment, as defined by the highest attained qualification (i) Certificate of Secondary Education (CSE), ordinary- (O-) level or vocational certificate (qualifications usually obtained at age 16, the UK minimum school leaving age when these women were at school), (ii) Advanced A-level (usually taken at 18 years) or (iii) university degree, and (2) age at first pregnancy. Information on both were obtained by questionnaire when the women were first recruited.
As the period between each of the assessments was 1 to 2 years, practice effects may have occurred in cognitive test performance. That is, performance may have improved, or an age-related decline be somewhat masked, as a result of familiarity with the test. We accounted for this in our analyses with a (3) time-varying continuous variable detailing the number of previous testing occasions. In addition, we adjusted for (4) the fieldworker who had administered the test to reduce any potential variation in performance related to how the tests were administered.
Statistical methods
Descriptive statistics were calculated and cognitive test scores at the first assessment clinic were examined by menopausal stage using analysis of variance.
Main analyses
Full details of the strategy for the main analyses, including details of all multilevel models, are provided in Supplementary Text (Additional file 1). Briefly, we used multilevel linear regression models to examine: (i) change in cognitive function domains by reproductive age (years since FMP) and chronological age and compare the contributions of each of these and (ii) the association of standardised LH, FSH and AMH levels (using mean and standard deviation (SD) from first assessment clinic, having replaced undetectable LH and AMH levels with 0.1 mIU/ml and 0.01 ng/ml respectively) with cognitive function. Multilevel models allow all women with at least one cognitive function assessment to be included in analyses under a missing-at-random (MAR) assumption and take account of the correlation between repeated measurements. As we only had up to three measurements in each woman, we had to assume any change with reproductive or chronological age or association with hormones were linear. We modelled each cognitive function domain in SD units, using the mean from the first assessment clinic and the estimated between-individual SD derived from the fully adjusted model.
The Bayesian Information Criteria (BIC) was used to assess and compare how reproductive and chronological age explained variation in cognitive function. The main models were adjusted for fieldworker effects, practice effects, chronological age, education and age at first pregnancy. To assess associations of reproductive hormones (FSH, LH and AMH) with cognitive function, each was included as a time-varying exposure in separate models, with the results reflecting the difference in cognitive function between women with one SD difference in hormone level at any given age.
Lastly, we studied differences in the extent of improvement in cognitive function by practice at pre-, peri- and post-menopause. We tested whether the interaction between practice effects (with a random slope) and menopausal stage improved model fit in a model including chronological age, education and age at first pregnancy using log likelihood tests.
Sensitivity analyses
We compared baseline cognitive function scores by the duration of follow-up time available to examine whether results may have been biased by loss to follow-up. We also repeated the main analyses in a sample restricted to women who participated in all three clinics. All analyses were conducted in Stata 15.1 (StataCorp, Texas, US) and MLwIN version 3.01 using command ‘runmlwin’ [21].
Ethical approval
Ethical approval for the data collection was obtained from the ALSPAC Ethics and Law Committee and the Local National Health Service Research Ethics Committees. Informed written consent for the use of data collected via questionnaires and clinics was obtained from participants. Consent for biological samples has been collected in accordance with the Human Tissue Act (2004).
