Systematic exclusion at study commencement masks earlier menopause for Black women in the Study of Women’s Health Across the Nation (SWAN)

Abstract

Background

Shorter average lifespans for minoritized populations are hypothesized to stem from ‘weathering’ or accelerated health declines among minoritized individuals due to systemic marginalization. However, evidence is mixed on whether racial/ethnic differences exist in reproductive ageing, potentially due to selection biases in cohort studies that may systematically exclude ‘weathered’ participants. This study examines racial/ethnic disparities in the age of menopause after accounting for differential selection ‘into’ (left truncation) and ‘out of’ (right censoring) a cohort of midlife women.

Methods

Using data from the Study of Women’s Health Across the Nation (SWAN) cross-sectional screener (N = 15 695) and accompanying ∼20-year longitudinal cohort (N = 3302) (1995–2016), we adjusted for potential selection bias using inverse probability weighting (left truncation) to account for socio-demographic/health differences between the screening and cohort study, and multiple imputation (right censoring) to estimate racial/ethnic differences in age at menopause (natural and surgical).

Results

Unadjusted for selection, no Black/White differences in menopausal timing [hazard ratio (HR)=0.98 (0.86, 1.11)] were observed. After adjustment, Black women had an earlier natural [HR = 1.13 (1.00, 1.26)] and surgical [HR= 3.21 (2.80, 3.62)] menopause than White women with natural menopause—corresponding to a 1.2-year Black/White difference in menopause timing overall.

Conclusions

Failure to account for multiple forms of selection bias masked racial/ethnic disparities in the timing of menopause in SWAN. Results suggest that there may be racial differences in age at menopause and that selection particularly affected the estimated menopausal age for women who experienced earlier menopause. Cohorts should consider incorporating methods to account for all selection biases, including left truncation, as they impact our understanding of health in ‘weathered’ populations.

Introduction

There are excess cardiometabolic morbidity and mortality among minoritized populations, particularly Black populations in the USA.^1,2 Potentially attributable to ‘weathering’, or early health deterioration due to the cumulative impact of social/economic/political marginalization,^3,4 studies have found that cardiometabolic risk factors tend to accumulate earlier in Black populations with the Black/White disparity widening in early to mid-older age especially for women.^4–6 The midlife is when women typically experience menopause, signalled by the occurrence of a women’s final menstrual period (FMP). The FMP can naturally occur over a transition period^7,8 or can be prematurely induced via reproductive surgeries (hysterectomy and/or bilateral oophorectomy) prior to natural FMP.9–11 The FMP may be a vital marker for women’s overall and cardiometabolic health,^12,13 as an earlier FMP has been associated with increased risk of low high density lipoprotein (HDL), high waist circumference and hypertension¹⁴ as well as cardiovascular disease and stroke.^13,15–17 An earlier FMP is hypothesized to prolong the time in which a woman experiences low reproductive steroid hormone production causing earlier onset of cardiometabolic disorders.^12,13,18 However, evidence is inconsistent as to whether racial/ethnic differences exist in age of FMP,^19–27 which must necessarily precede clarification of the contribution of FMP timing to racial disparities in cardiometabolic risk and life expectancy.

Inconsistent findings on racial/ethnic differences in FMP^19–27 could be partially due to systematic exclusion or differential selection in cohort studies that have been used to study age at FMP. Selection bias arises from differential probability of ‘selection into’ and ‘out of’ the study sample, which may distort study estimates and/or limit generalizability to the populations of interest.²⁸ Left truncation, caused by experiencing the outcome of interest prior to recruitment into a study and right censoring, or being censored in analysis due to missing data, loss to follow-up and/or non-occurrence of the outcome, are specific mechanisms of selection into and out of a cohort that may cause selection bias if the likelihood of either differs by population subgroups of interest, such as racial/ethnic group.²³

Selection bias due to left truncation may be a consequence of the process of ‘weathering’. As cohorts are designed to observe the occurrence of the outcome of interest, in this case the menopause or FMP, they recruit outcome-free participants at an age range before the typical onset of the outcome.^28,29 However, if certain populations are ‘weathering’ and experiencing the outcome earlier in life, the cohort recruitment age could differentially miss these high-risk subgroups, potentially leading to an overestimation of the average age of menopause²⁹ and underestimation of racial/ethnic differences in the resulting sample.

Selection bias can also be induced in follow-up by differential rates of right censoring or restriction of the study sample in analyses. In studies of menopause, it is common to either right censor or exclude women who experience surgical menopause.^{10,19,20,22,25,26,30} However, previous work has shown that Black and Hispanic women often have higher rates of surgical menopause than White women,^10,31,32 so right censoring surgical menopause cases could be informative or non-independent from age of natural menopause,³³ causing a bias in the estimation of racial/ethnic differences in FMP age.

Although right censoring and loss to follow-up are selection biases addressed commonly in cohorts of ageing,^33,34 left-truncation bias is often relegated to a sentence in the limitations section. This study uses the Study of Women’s Health Across the Nation (SWAN), which is a multi-ethnic cohort of midlife women that previously found no significant racial/ethnic differences in the timing of FMP after adjustment,²⁶ to evaluate the impact of left truncation and right censoring on racial/ethnic differences in age at FMP. Exploiting the unique features of SWAN that allow precise computation of the probability of being selected into the cohort study, this analysis employs a combination of inverse probability weighting³⁵ and imputation³⁶ techniques to simultaneously correct for multiple forms of potential selection bias and re-estimate racial/ethnic differences in FMP timing.

Methods

Data were from SWAN, a multi-site, multi-ethnic longitudinal cohort (N = 3302, 1995/1997–2015/2016) of midlife women and the screening study used to recruit the cohort (N = 15 695, 1995–1997).³⁷ Seven sites recruited White women, four sites Black women (Detroit, MI; Boston, MA; Chicago, IL; Pittsburgh, PA), one site Chinese women (Oakland, CA; Cantonese), one site Hispanic women (Newark, NJ; Spanish) and one site Japanese women (Los Angeles, CA; Japanese). The eligibility criteria for the community-based screening study were: 40–55 years old, self-identified as the racial/ethnic group for the designated site, spoke either English or the language for the designated site and had a primary residence within the geographic area for each site.^25,37 Women were interviewed to determine whether they met the eligibility criteria for the cohort: 42–52 years old, self-identified as the racial/ethnic group for the designated site, primary residence near the site, not on hormone therapy (HT) (birth control, fertility drugs, estrogens or progestins, hormone patches or creams, hormone injections or post-menopausal hormones), not currently pregnant, have not had a hysterectomy and/or bilateral oophorectomy and at least pre- to perimenopausal (had a period at least in the last 3 months). Women in the screening study were asked whether they had had either a hysterectomy and/or bilateral oophorectomy. If not, they were asked whether they had had their menses at least once in the last 3 months. The cohort included in-person clinic visits and interview questions approximately yearly. The Institutional Review Boards at each study site approved the protocol and participants provided informed consent for the screening and all cohort visits.

FMP

The outcome was age at FMP, including surgical and natural FMP. Both were determined from annual cohort interviews asking whether a woman had experienced amenorrhea for at least 12 months (without HT use) or had had reproductive surgery (bilateral oophorectomy, hysterectomy and/or unilateral oophorectomy). The type of FMP was collected and considered as a moderator in the association between racial/ethnic group and FMP age.

Demographic and health characteristics

Self-reported primary racial/ethnic group was collected at screening. Educational level was self-reported by indicating the highest level of education achieved (18 categories from did not go to school to doctoral degree); answers were collapsed into three categories (≤ high-school education, some college or an associate degree, ≥ a college degree). Participants were asked: ‘How hard is it for you to pay for the very basics like food, housing, medical care, and heating?’ Responses were ‘very hard’, ‘somewhat hard’ and ‘not very hard at all’. Participants were asked to rate their general health on a five-point scale from poor to excellent; answers were collapsed into a four-level variable from excellent to fair/poor. Waist circumference was measured using a measuring tape placed horizontally at the level of the natural waist or the narrowest part of the torso. Body mass index was calculated as weight in kilograms divided by measured height in metres squared. Self-reported alcohol intake (drinks per week) was collapsed into categories ‘none to low’ (<2 drinks/week), ‘moderate’ (2–7 drinks/week) and ‘high’ (>7 drinks/week). Physical activity was assessed using a modified Baecke questionnaire and included sport and exercise, leisure and household/childcare activity; scores ranged from 3 to 15.³⁸

Weighting for selection into the cohort

The cross-sectional screening study (ages 42–52 years only) in conjunction with the cohort was used to identify selection mechanisms that included eligibility for the study and participation (Supplementary Figure S1, available as Supplementary data at IJE online). Both types of selection were determined to be influenced by factors such as age, education, fibroid diagnoses and overall health that would also influence age of FMP causing collider bias,^39–42 threatening the internal validity of estimations of racial/ethnic differences in FMP. Three inverse probability weights (IPW) were used to account for selection into the cohort to: (i) up-weight women in the cohort who were representative of women (ages 42–52 years only) who did not make it into the cohort from the cross-sectional screening due to ineligibility (eligibility weight), (ii) up-weight women who were eligible but unable/unwilling to participate in the cohort (participation weight) and (iii) given the age range of 42–52 years for recruitment into the study, weight the cohort to represent the cohort that would have been retained had everyone been recruited at age 42 years old (study design weight).

Predictors tested in the logistic regression (propensity score) models for inclusion in the eligibility and participation weights included demographic, reproductive and health-related factors; additionally interactions between race/ethnicity and demographic predictors were considered. Predictors in the final models were selected a priori, with forward–backward selection used only to select between variables measuring the same construct (i.e. socio-economic variables such as education, financial hardship and employment). Age was excluded as a predictor in the eligibility weight to avoid double counting of the effect of age between the study design and eligibility weight. Prediction models, using the internal cross-sectional screening data, were compared using the area under the curve to achieve the combination of the best fit and most parsimonious models. Final predictors for each model are provided in Supplementary Table 1 (available as Supplementary data at IJE online). Density curves were used to assess the overlap in propensity scores between the cross-sectional screening and cohort internal data sets. To evaluate weight performance, covariate balance was assessed between the SWAN cross-sectional and cohort internal data using standardized mean differences, for the unweighted cohort and with each of the successive weights.⁴³ Differences of <0.25 represented adequate covariate balance.⁴³

The study design weight is designed to account for the differential probability of entry into the cohort given the various ages that women were recruited (ages 42–52 years in SWAN) by exploiting the known probability of entry into the cohort based on age of observed/imputed FMP. This weight was calculated as the inverse probability of the time (in years) each woman was eligible over the 10-year period between 42 and 52 years. This is calculated regardless of age of entry into the cohort to effectively weight the cohort to represent the cohort that would have been retained if every woman entered the cohort at age 42 years. In a simple example, the 10-year probability of inclusion for a woman with an observed surgical FMP at age 50 years would be: 50 years old – 42 years old = 8 years of possible inclusion/10-year range of age-eligibility = 0.80 10-year probability of inclusion, giving a weight of 1.25 (1/0.80). Women with earlier-occurring FMPs receive higher weights, as such women were less likely to be included in the study than women with later-occurring FMPs, as illustrated in Supplementary Table S3 (available as Supplementary data at IJE online). For women who had had a hysterectomy/bilateral oophorectomy, the upper age bound was the age of surgery. For women with natural menopause, the upper bound was age reaching late perimenopause (>3 months amenorrhea) as per the initial SWAN cohort exclusion criteria. Additionally, observed years of HT prior to these end points were deducted from each participant’s 10 years of potential inclusion. If any end points were missing, imputed values were used. The weight assumes that recruitment age is random and women did not have any additional exclusion events (HT and/or pregnancy) during the unobserved years (age 42 years to age of entry into the study). We believe that the assumption of random age of recruitment is reasonable and that the impact of additional exclusion events to be minor.

All weights were multiplied to simultaneously account for each selection mechanism.⁴⁰ Descriptive statistics for the resulting weights are provided in Supplementary Table S2 (available as Supplementary data at IJE online).

Multiple imputation for right censoring

Multiple imputation using chained equations via the Gibbs-Like algorithm (default for STATA) was used to impute missing information on the type of FMP (natural or surgical), the age at FMP and covariates/mediators (fraction of missing information $≤$50%) in 10 imputation sets. Predictors for imputed variables included demographic and reproductive characteristics at baseline, diagnoses of reproductive and/or chronic conditions prior to FMP and menopausal symptoms, reproductive hormone levels, lifestyle, demographic, mental health and stress factors at the last known value while still menstruating. Participants were missing FMP age due to interval censoring (4.91%), HT use (16.11%), missing surgery date (0.24%) and loss to follow-up (16.38%). In SWAN, there was a modified missing data pattern for FMP type and age, as assessment of FMP type preceded assessment of FMP age. Thus, a two-step imputation was used to assign: (i) FMP type (natural or surgical) and then (ii) within each set, FMP age using FMP type as an additional predictor and interaction variable. Truncated linear regression was used to impute FMP age where the left bound was the last observed menstrual flow without HT use (with 3-month HT use washout period) and right bound was the first report of 12 months of amenorrhea, first report of stop in HT use (with 3-month washout period), first report of surgical menopause or age 60 years (99.7% of observed women in SWAN reach FMP by age 60 years).

Statistical analysis

Mean, proportions and standard errors were calculated and pooled across the imputation sets using Rubin’s rules⁴⁴ for all covariates in the cross-sectional screening sample, among eligible women only, and in the resulting cohort sample across race/ethnic groups. Bivariate associations were calculated between race/ethnic group and covariates.

Cox proportional hazard models with age as the timescale were used to examine the effect of each form of selection on racial/ethnic differences in age at FMP. To model the full FMP experience, age of FMP was the outcome and type of FMP (natural or surgical) was a moderator of racial/ethnic differences in FMP age. As a sensitivity analysis, all Cox proportional hazard models and associated predicted median ages were estimated with further adjustment for potential mediators including baseline and time-varying (last known prior to FMP): education, self-reported health, waist circumference, smoking, alcohol use and physical activity. Lastly, there were no deaths prior to FMP in observed data but a small number of women (∼22 in each imputed data set) had imputed FMPs after their age at death (<1-year difference). As a sensitivity analysis, all models were re-estimated excluding women who died prior to imputed FMP. The last sensitivity analyses examined the association between racial/ethnic group and natural menopause, censoring for surgical menopause in models not accounting and accounting for right censoring due to loss to follow-up and left truncation. All models were fitted into each of the 10 imputed data sets and estimates pooled using Rubin’s rules.⁴⁴

Results

Nearly half (41.6%) of screening participants were eligible for the cohort study, 50.65% of whom participated in the cohort. Eligibility and participation varied by racial/ethnic group: Japanese women had the highest proportion eligible (54.4%) and Black women the lowest (38.9%). Low eligibility for Black women was largely due to the higher prevalence of surgical menopause (30.9%) compared with other groups (5.7–17.3%). Surgical menopause was the leading cause of ineligibility for Black women only. Once eligible, Japanese women had the highest participation rate (69.1%) and Chinese women the lowest (35.7%) (Table 1).

Compared with the cross-sectional screening sample, women in the cohort are more educated, less likely to have fibroid diagnosis and in better overall health whereas the proportions of each racial/ethnic group and educational levels represented stayed relatively stable (Supplementary Table S1, available as Supplementary data at IJE online).

In the cohort, 47% were White and 28.3% were Black women whereas <10% were Hispanic, Japanese and Chinese women (7.6–8.7%). Most of the cohort had a college degree or greater (42.6%), had never smoked (58.0%) and consumed fewer than two drinks per week (49.9%). Black and Hispanic women had lower educational levels and were more likely to smoke, and White women were more likely to consume at least seven drinks per week. In follow-up, most of the cohort had a natural menopause (89.9%). Black and Hispanic women (13.8% and 10.6%) had the highest proportions of surgical menopause compared with all other groups (4.8–9.5%) (Table 2).

Propensity score distributions for eligibility and participation in the cross-sectional screening and cohort overlapped (Supplemental Figure S2, available as Supplementary data at IJE online). After applying successive weights to the cohort, the overall and race/ethnic-specific covariate balance improved for most variables, including education, financial hardship, fibroid diagnoses, self-rated health, diabetes diagnoses, osteoporosis diagnoses, cancer diagnoses and current smoking. The covariate balances for body mass index were similar when unweighted and weighted (–0.04 unweighted vs 0.03 weighted) whereas for heart attack/angina diagnoses, covariate imbalance slightly increased (–0.31 unweighted vs 0.43 weighted) (Figure 1).

In the Cox proportional hazard model unadjusted for selection (Model 1), all racial/ethnic groups had a higher hazard and earlier natural FMP than White women [range HR_Japanese = 1.02 (0.88–1.19)–HR_Hispanic = 1.27 (1.03–1.57)]. In Model 2, incorporating FMP type, all the hazard ratios for natural FMP are lower whereas surgical FMP is earlier for all groups, other than Chinese women vs White women with a natural FMP. In Model 3, which additionally corrected for right censoring, the hazard ratios for natural FMP increased for Black and Japanese women and were slightly lower for Chinese and Hispanic women whereas surgical FMP hazard ratios increased for White, Chinese and Hispanic women and decreased for Black and Japanese women. In Model 4, which additionally corrected for left truncation, all racial/ethnic groups had a higher hazard of earlier natural FMP compared with White women except Japanese women [HR_Black = 1.15 (1.04–1.27), HR_Chinese = 1.02 (0.87–1.16), HR_Hispanic = 1.18 (0.97–1.38), HR_Japanese = 0.97 (0.83–1.10)]. All racial/ethnic groups had a higher hazard of earlier surgical FMP compared with White women with natural FMP [range HR_Japanese = 1.57 (0.99–1.87)–HR_Black = 3.02 (2.58–3.45)] (Table 3).

After adjustment of Model 1 (unadjusted for selection) for covariates, including baseline and time-varying education, self-reported health, waist circumference, smoking status, alcohol use and physical activity score, all hazard ratio confidence intervals included the null [range HR_Chinese = 0.88 (0.73–1.07)–HR_Hispanic = 1.04 (0.81–1.33)] except for Japanese women compared with White women [HR_Japanese = 0.84 (0.71–1.00)]. After adjustment of Model 4 (adjusted for all forms of selection) for the same covariates, Black women had earlier and Japanese women later natural FMP compared with White women [HR_Black = 1.13 (1.00–1.26), HR_Japanese = 0.83 (0.69–0.98)]. For surgical FMP, all groups other than Chinese women (n_events = 13) had an earlier surgical FMP than White women with natural FMP [range HR_Japanese = 1.43 (1.02–1.85)–HR_Black = 3.21 (2.80–3.62)].

In a sensitivity analysis excluding women who died prior to imputed FMP (∼22 women), all hazard ratios in Model 1 (unadjusted for selection) included the null for racial/ethnic differences in natural menopause [range HR_Japanese = 1.02 9 (0.88–1.19)–HR_Black = 1.09 (0.97–1.22)] except for Hispanic vs White women [HR_Hispanic = 1.27 (1.03–1.57)]. After accounting for all forms of selection (Model 4), Black women had earlier natural FMP than White women [HR_Black = 1.16 (1.04–1.28), range HR_Japanese = 0.97 (0.83–1.10)–HR_Hispanic = 1.18 (0.97–1.38)]. And White, Black, Hispanic and Japanese women had earlier surgical menopause compared with White women with natural menopause (range HR_Japanese = 1.43 (0.99–1.87)–HR_Black = 3.10 (2.67–3.52)] (Supplementary Table S4, available as Supplementary data at IJE online).

In a sensitivity analysis modelling the racial/ethnic differences in natural menopause, censoring for surgical menopause (n_events ∼ 2498), all hazard ratios in Model 4 (adjusted for selection) included the null for racial/ethnic differences in natural menopause (range HR_Hispanic = 0.94 (0.75–1.13)–HR_Chinese = 1.15 (0.90–1.39)].

Without adjustment for selection (Model 1), there are no racial/ethnic differences in the predicted median age of natural FMP [range (in years) median_Hispanic = 51.7 (51.3–52.8)– median_Japanese = 52.7 (52.2–53.2)]. Adjusted for selection (Model 4), Black women had earlier [51.4 (51.1–51.7)] and Japanese women later [51.9 (51.8–52.9)] predicted median age of natural FMP than White women [52.0 (51.7–52.3)]. Furthermore, Black women had earlier [47.1 (45.6–48.8)] and Japanese women later [51.4 (50.5–53.0)] predicted median age of surgical FMP vs White women [48.9 (47.7, 50.8), Figure 2].

Discussion

This study is among the first to estimate racial differences in age at menopause after accounting for selection bias related to left truncation and right censoring. Results demonstrate that ignoring selection biases led to falsely high estimations of the average age of menopause and underestimated race/ethnic differences in menopausal timing. These biases particularly affected Black women whose age of natural FMP was overestimated in the uncorrected model by an average of 1.1 years. Correcting for selection bias, the predicted median age of menopause in Black women was 1.20 years earlier than in White women—0.60 years earlier for natural menopause and 1.80 years earlier for surgical menopause. Black/White differences remained after controlling for individual-level socio-economic indicators, overall health and health behaviours signalling that structural factors contributing to ‘weathering’^3,4 and reproductive ageing require continued and renewed attention.

Results show that Black women had an earlier natural menopause than White women robust to adjustment for potential mediating factors, with similar trends for Hispanic women. Other studies conducted in younger women, most of which censored or excluded women with surgical menopause, have found as large as 1 year of difference in the median age of natural FMP^19,21,25,45 in Black/Hispanic women compared with White women.^{19–22,25,26,30,45} In contrast, longitudinal analysis in SWAN reported no Black/White differences in the timing of FMP after adjustment.²⁰ After accounting for selection biases, Black women had an earlier natural FMP compared with White women by 0.4 years and, although confidence intervals included the null, Hispanic women followed a similar trend of an earlier natural FMP by 0.5 years.

Black and Hispanic women had a higher prevalence and earlier ages of surgical menopause vs other groups in SWAN. Further, the largest proportion of left truncation in the SWAN study stemmed from Black women excluded due to surgical menopause. Previous studies have highlighted the high prevalence of surgical menopause among Black and Hispanic women, especially Black women.^10,31,32 The differential risk of hysterectomy/oophorectomy could be partially due to the higher prevalence of reproductive morbidities (such as uterine fibroids) earlier in life for Black women, which are often treated with surgery.⁴⁶ In a 2005 study, Powell et al. also documented the increased risk of surgical menopause for Black and Hispanic women, noting that increased risk independently of known risk factors is consistent with an ‘overuse of elective hysterectomy’ in these populations.³¹

Selection bias led racial/ethnic differences in FMP to be underestimated. In a simulation analysis by Cain et al. that estimated natural FMP in a cohort with the same recruitment age (42.5–52.5 years old), the effect of left truncation on estimations of FMP age resulted in a positive bias of 1.29 years.²⁹ In the present study, there was an average positive bias of 0.72 years between the ‘biased’ and ‘unbiased’ models, although the magnitude of bias varied by racial/ethnic group where Black women had the highest (+1.10 years) and White women had the lowest (+0.50 years) bias. The lower bias overall in this study vs the Cain et al. results²⁹ could be partially due to the difference in outcome (natural FMP only vs natural and surgical FMP in this study) yet the amount of bias was similar for Black women. This finding extends Cain et al.’s results suggesting that there can be heterogeneity in the effects of left truncation across subgroups. The larger bias for Black women extends results of another simulation by Mayeda et al. in which researchers found that survival bias, a type of left truncation caused by mortality prior to study commencement, underestimates the association between education and cognitive decline.⁴² As in Mayeda et al., the present study shows that racial/ethnic differences in FMP were also underestimated, as Black women had higher levels of left truncation and thus a larger bias. The higher levels of left truncation among Black women in SWAN are indicative of potential ‘weathering’ in this population leading to earlier FMP and exclusion from the cohort.

Further, it is common in reproductive ageing studies to censor or exclude women who have had surgical menopause,^{10,19,20,22,25,26,30} which can cause bias in the estimation of racial differences in menopause despite correction for other selection forces as shown in our sensitivity analysis (Supplementary Table 5, available as Supplementary data at IJE online). We incorporated surgical menopause as an outcome and used a two-step approach to multiple imputation to avoid informative censoring by surgical menopause. Censoring surgical menopause caused the hazard ratio for Black and Hispanic women to be higher than in models that incorporated surgical menopause into the outcome, signalling that surgical FMP may be non-independent from natural FMP³⁴ and differential by racial/ethnic group, causing bias in estimation.

This study has some limitations. The low prevalence of surgical FMPs in Chinese, Hispanic and Japanese women produced imprecise estimates of age at surgical menopause for these groups. Additionally, the hazard ratios in our ‘biased’ model differ slightly from the hazard ratios in previous SWAN longitudinal analyses of natural FMP²⁶ that were conducted in 2007 and included fewer observed FMPs (1403 natural FMP events in original analyses²⁶ vs 1804 in this analyses). Multiple imputation assumes that right-censored data are missing at random,³⁶ although Hispanic women were more likely to be right censored than other racial/ethnic groups in SWAN. Therefore, imputation models used racial/ethnic group as a predictor. Regardless, there may be some additional bias in estimates that may cause racial/ethnic disparities in FMP age to be underestimated. Although results are similar to previous simulation results,^29,42 the eligibility and participation weights assume the cohort is exchangeable with cross-sectional participants given covariates. There was notable overlap in the propensity scores for both the eligibility and participation weights (Supplementary Figure S2, available as Supplementary data at IJE online) but the weights were smaller than those in the cross-sectional study (Supplementary Table S2, available as Supplementary data at IJE online). Taken together, some women with very low odds of inclusion and participation may not be represented in the cohort to be upweighted. This highlights the novelty of the third study design weight developed for addressing left truncation. The study design weight uses the known probability of exclusion via observed age of FMP from women within the cohort. Nonetheless, although the weights have corrected for some bias induced by left truncation, there may still be additional bias unaccounted for in our estimations. Importantly, left-truncation bias stemming from ineligibility prior to age 42 years is unaccounted for.^24,44

In conclusion, although mitigation of bias was imperfect, results suggest that censoring for surgical menopause and failure to consider left truncation biased estimates of racial disparities in age at menopause. After adjustment, Black women had an earlier natural and surgical FMP compared with White women, independently of known risk factors—a finding with potentially important implications for Black women’s ageing and cardiometabolic health. Renewed research and attention on the mitigation of all forms of selection bias especially left truncation are warranted to allow a broader understanding of health and ageing in ‘weathered’ populations.

Ethics approval

Each of the SWAN study sites received institutional review board approval from their respective institutions (University of Michigan, Massachusetts General Hospital, Rush University, University of California Davis and Kaiser Health, University of California Los Angeles, University of Medicine and Dentistry—New Jersey Medical School, Albert Einstein College of Medicine and University of Pittsburgh). All participants gave written informed consent before enrolment and each year of the study, which was conducted in accordance with the principals of the Declaration of Helsinki.

Data availability

The data underlying this article are available in the NIH Aging Research Biobank, at https://agingresearchbiobank.nia.nih.gov/studies/swan.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

A.R. and S.D.H. conceived the research questions for the study. A.R., M.R.E. and S.D.H. designed the overall analytic strategy. M.R.E. conceived the study design weight used in analysis. S.D.H. and C.K.G. conceived and contributed to the design of SWAN and acquisition of the data. A.R. conducted the analysis and wrote the manuscript. All authors contributed to the interpretation of the data and critical revision of the manuscript for intellectual content. All authors have read and approved the final manuscript. A.R., M.R.E. and S.D.H. are guarantors of the work.

Funding

SWAN has grant support from the National Institutes of Health (NIH), DHHS, through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR) and the NIH Office of Research on Women’s Health (ORWH) (Grants U01NR004061; U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, U01AG012553, U01AG012554, U01AG012495 and U19AG063720). The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIA, NINR, ORWH or the NIH. A.R. received grant support from the National Institutes of Health, National Institution on Aging’s Ruth L. Kirschstein National Research Service Award (NRSA) Individual Pre-doctoral Fellowship to Promote Diversity in Health-Related Research (1F31AG064856-01). Clinical Centers: University of Michigan, Ann Arbor—Carrie Karvonen-Gutierrez, PI 2021–present, MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA—Sherri-Ann Burnett-Bowie, PI 2020–present; Joel Finkelstein, PI 1999–2020; Robert Neer, PI 1994–99; Rush University, Rush University Medical Center, Chicago, IL—Imke Janssen, PI 2020–present; Howard Kravitz, PI 2009–20; Lynda Powell, PI 1994–2009; University of California, Davis/Kaiser—Elaine Waetjen and Monique Hedderson, PIs 2020–present; Ellen Gold, PI 1994–2020; University of California, Los Angeles—Arun Karlamangla, PI 2020–present; Gail Greendale, PI 1994–2020; Albert Einstein College of Medicine, Bronx, NY—Carol Derby, PI 2011–present; Rachel Wildman, PI 2010–11; Nanette Santoro, PI 2004–10; University of Medicine and Dentistry—New Jersey Medical School, Newark—Gerson Weiss, PI 1994–2004; and the University of Pittsburgh, Pittsburgh, PA—Rebecca Thurston, PI 2020–present; Karen Matthews, PI 1994–2020. NIH Program Office: National Institute on Aging, Bethesda, MD—Rosaly Correa-de-Araujo 2020–present; Chhanda Dutta 2016–present; Winifred Rossi 2012–16; Sherry Sherman 1994–2012; Marcia Ory 1994–2001; National Institute of Nursing Research, Bethesda, MD—Program Officers. Central Laboratory: University of Michigan, Ann Arbor—Daniel McConnell (Central Ligand Assay Satellite Services). Coordinating Center: University of Pittsburgh, Pittsburgh, PA—Maria Mori Brooks, PI 2012–present; Kim Sutton-Tyrrell, PI 2001–12; New England Research Institutes, Watertown, MA—Sonja McKinlay, PI 1995–2001. Steering Committee: Susan Johnson, Current Chair; Chris Gallagher, Former Chair.

Acknowledgements

We thank the study staff at each site and all the women who screened for and participated in SWAN. We also thank Sybil Crawford PhD and John Randolph MD for their guidance and insight during the analysis, and Michelle Odden PhD for critical feedback on manuscript drafts.

Conflict of interest

None declared.

References