Early-Life Exposures and Odds of Adenomyosis: A Population-Based Case-Control Study

M.S.H. and K.U. conceived and designed the study. M.S.H. conducted the data analysis and drafted the manuscript. All authors substantially contributed to interpretation of the data, review and approval of the final version of the manuscript, and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the research were appropriately investigated and resolved.

The present analyses using de‐identified data were determined to not involve human subjects by the Human Research Protection Program at Michigan State University.

In the present analysis on early‐life factors and odds of adenomyosis in adulthood, our data suggested increased odds of adenomyosis with younger maternal age at birth and maternal cigarette smoking. Using population controls, we also observed increased odds of adenomyosis with later birth order. The key strengths were the population‐based sampling design of health plan enrollees and the use of two control groups to thoroughly investigate associations between early‐life factors and adenomyosis. Use of the health plan enrolment database allowed for selection of population controls from the underlying population that gave rise to adenomyosis cases. For instance, the frequency of demographic characteristics among the population controls mirrored the distribution among female health plan enrollees and the general population in that region [ 22 ]. Thus, population controls represented the frequency of early‐life exposures in the population and provide for a valid epidemiologic study design [ 1 ]. Although bias can still arise from factors associated with the willingness to undergo hysterectomy among the cases, results from our sensitivity analyses restricting population controls to those who would have a hysterectomy if warranted suggest this bias may be minimal. Alternatively, the selection of hysterectomy controls is convenient, allows for confirmation of disease absence and forms a control group similar to cases on factors related to undergoing hysterectomy. However, the selection of individuals undergoing hysterectomy may introduce bias. The medical indications warranting a major surgical procedure such as hysterectomy can also be associated with early‐life factors, affecting the frequency of exposure in this control group. Although neither control group provides a bias‐free estimate, we observed the same direction of association in analyses, except birth order, using population controls or hysterectomy controls. The similar results yielded for several early‐life exposures—regardless of the control group used—lend support that select early‐life factors may contribute to subsequent adenomyosis development in adulthood. First, we relied on participant recall to ascertain information on early‐life exposures. Less than 50% of the study participants reported receiving help from their mother or other knowledgeable family member on the family history questionnaire. Despite the limited availability of mother's input in the present study, a high to moderate correlation has been suggested between a mother's and daughter's recall of daughter's intrauterine exposures, specifically birthweight, birth order and maternal smoking [ 19 ]. Previous studies comparing participant recall to birth certificate data also suggest high accuracy of recall [ 19 , 23 , 24 ]. In our sensitivity analysis restricting the study population to participants that received mother's help, we observed similar results to the main analysis, for the three exposures we could evaluate. Together, these data suggest exposure misclassification from participant recall of early‐life exposures is likely minimal. Second, although we sought to explore a panel of early‐life factors, we were limited by the sample size for rarer exposures. For example, few cases and controls were born outside the United States. Further, exposures known to be associated with a higher oestrogen profile, such as those who had mothers who had taken DES while pregnant, were a multiple gestation and were fed soy formula as infants, were rare in this population. Further studies are warranted to investigate these early‐life factors. Third, the potential for confounding from the lack of data on maternal sociodemographic characteristics is possible. Data on maternal education or income at the time of participant's birth were not collected. Fourth, we were not able to confirm absence of adenomyosis in population controls. At the time this case–control study was conducted, advancements in imaging technology were not yet available to screen population controls; consensus on terminology to describe features of adenomyosis from ultrasound‐based images has only been reached in the past decade, with a lack of consensus on imaging criteria for the diagnosis of adenomyosis remaining [ 25 ]. Thus, undiagnosed adenomyosis among population controls is possible. However, we previously estimated that the prevalence of undiagnosed adenomyosis was likely low in our study population, possibly around 4% in population controls, and therefore may not have altered our findings [ 11 ]. For this calculation, we used the prevalence of pelvic pain when not menstruating (20%) in population controls and the adenomyosis prevalence of 21% in a cross‐sectional transvaginal ultrasound study of pre‐ and post‐menopausal gynaecologic clinic patients with medical indications for imaging, including pelvic pain [ 26 ]. The defining characteristic of adenomyosis, the presence of endometrial glands and stroma within the uterine myometrium, could plausibly originate from the disruption of uterine tissue development in utero, resulting in altered morphology or weakened endometrial‐myometrial border. Gestation is a period of substantial uterine tissue development, with the myometrium being well‐differentiated and endometrial glands being present by mid‐gestation at 20–22 weeks [ 8 , 27 ]. Additionally, given that oestrogen has a central role in the pathogenesis of adenomyosis, an altered estrogenic hormonal milieu from the disruption of the HPO‐axis during the intrauterine period could also contribute to subsequent adenomyosis development. It has been postulated that a hyperestrogenic state promotes a series of events involving inflammation, proliferation and local oestrogen production that contributes to invagination of endometrial cells into the myometrium [ 4 ]. The HPO‐axis is active during gestation and is susceptible to disruption; follicle‐stimulating hormone and luteinizing hormone have been detected in foetal sera as early as 8 weeks of gestation and towards late gestation, the HPO‐axis responds to high maternal oestrogen exposure with the suppression of activity [ 5 , 6 , 7 , 28 ]. Further, early‐life factors may also operate through other pathways, including inflammation, endocrine dysregulation and dysfunctional uterine contractility, given the possible mechanisms of adenomyosis development [ 29 ]. In support of the developmental origins of adenomyosis development, associations have been observed between early‐life exposures and other gynaecologic conditions in adulthood that commonly co‐occur with adenomyosis, including endometriosis and uterine fibroids [ 15 , 16 , 17 , 30 , 31 , 32 , 33 ]. Our data suggested participants born to young mothers (≤ 19 years) had an increased odds of adenomyosis, using either control group. When comparing cases to population controls, those who were the fourth or later birth had an increased odds of adenomyosis; the association was strongest with almost twice the odds of adenomyosis for participants who were the third or later birth to younger mothers (< 25 years). Larger family size and younger maternal age when the participant's mother was pregnant with the participant may indicate aspects of family structure associated with increased maternal stress during the participant's foetal development. Such stress could arise from financial burdens, physical hardships and emotional stressors. Our results contrast with the one prior study examining maternal age and adenomyosis risk [ 9 ]. That study, conducted in a large, multigenerational cohort in Sweden, observed no association between maternal age and adenomyosis incidence (adjusted hazards ratio: 1.00 per year, 95% CI 0.92–1.10), although that study included only 24 adenomyosis cases. Our study suggested, among participants who were never‐smokers, an association between maternal cigarette smoking and increased odds of adenomyosis in analyses using either control group. Cigarette smoke is a complex mixture that exposes users to thousands of contaminants [ 34 ]. Prenatal exposure to cigarette smoke may disrupt gonadotropin production and alter ovarian function. Our research group previously reported participant cigarette smoking in adolescence and adulthood was associated with an increased odds of adenomyosis, regardless of the control group employed [ 12 ]. The results of our prior study with that of the present analyses suggest that exposure to cigarette smoking during multiple windows, intrauterine, adolescent and adulthood may increase the odds of adenomyosis. When investigating birthweight and subsequent odds of adenomyosis in adulthood, we observed the general pattern of an increased odds of adenomyosis with greater birthweight. Comparing cases to population controls, our data suggested a 40% increased odds of adenomyosis with higher birthweight (> 4080 g or 9 pounds); when comparing cases to hysterectomy controls, low birthweight (< 2500 g or < 5.5 pounds) was associated with a 40% decreased odds of adenomyosis. Higher birthweight may indicate high intrauterine exposure to oestrogen [ 35 ]. However, in our post hoc analysis among participants born singleton and not preterm, the association only persisted when comparing cases to population controls; the association comparing cases to hysterectomy controls was null. The two prior investigations of birthweight also observed no association with adenomyosis [ 9 , 10 ]. Although those two prior studies were large, population‐based studies of adenomyosis that collected data on participant birthweight from parental recall or birth records, both studies relied on ICD codes (ICD‐8, ICD‐9 and ICD‐10) to identify adenomyosis cases. These earlier ICD versions did not have a diagnostic code unique to adenomyosis that may result in misclassification [ 1 ]. Additionally, one study only detected a small number ( n  = 24) of adenomyosis cases [ 9 ], and the other study had limited data on potential confounding factors for adjustment [ 10 ].

We used data from a population‐based case–control study on adenomyosis; the study population and methods have been previously described [ 11 , 12 ]. As the case–control study used (cis)women‐related terminology and did not collect data on gender, we used gender‐inclusive language to describe the study population [ 13 ]. Briefly, the case–control study was conducted among pre‐ and post‐menopausal enrollees ages 18–59 years of a large, integrated healthcare system, in western Washington State. The healthcare system, Kaiser Permanente, was known as Group Health at the time the study was conducted. The use of the Group Health enrollee population allowed for the identification of cases within a defined health plan enrollee population and the selection of two control groups, hysterectomy controls and population controls. Cases were health plan enrollees who received a diagnosis of pathology‐confirmed adenomyosis by hysterectomy for the first time between April 2001, and March 2006. Cases were identified by medical record review and included individuals who had pathology confirmation of adenomyosis using uterine specimens obtained at the time of hysterectomy [ 11 , 12 ]. Hysterectomy controls were individuals who underwent hysterectomy for benign disease over the same period cases were diagnosed, and for whom the absence of adenomyosis was confirmed by pathology report review. The main indications for hysterectomy in cases and hysterectomy controls were abnormal uterine bleeding and pain [ 11 , 12 ]. Population controls were individuals with an intact uterus, no history of adenomyosis diagnosis, and enrolled in Group Health any time between 1 April 2001, and 31 March 2006. Population controls were randomly selected from the Group Health database and were frequency matched to cases by 5‐year age groups. Both hysterectomy and population controls needed to have medical record data available for abstraction and review, just as for the cases. In addition, cases and controls were required to be enrolled in Group Health for at least 6 months prior to the reference date. For cases, the reference date was the date of first visit within the healthcare system for symptoms that lead to a diagnosis of adenomyosis. For hysterectomy controls, the reference date was the date of the, first visit within Group Health for symptoms that lead to a hysterectomy. Each population control was assigned a reference date based on the distribution of reference dates among the adenomyosis cases. For the present analyses, data were available for 386 cases, 233 hysterectomy controls and 323 population controls. As previously described [ 11 , 12 ], the main study activity in the case–control study was a structured, in‐person interview that collected data on a range of topics from lifestyle behaviours to medical and pregnancy history. The structured interview was conducted by a trained female interviewer at the Fred Hutchinson Cancer Center or at the participant's home. Prior to the in‐person interview, study participants were mailed a family history questionnaire. The family history questionnaire collected information on mother's age at participant's birth, participant's birth order, mother's smoking status when pregnant with participant, participant's birthweight, preterm birth, foetal number, maternal diethylstilbestrol (DES) use and soy formula feeding as an infant (Text  S1 ). During the structured interview, data were also collected on country of birth. Informed by prior studies of early‐life factors and another gynaecologic condition, uterine fibroids, we evaluated the joint exposure to maternal age and birth order [ 14 , 15 ]. When completing the family history questionnaire, participants were instructed to confirm responses with their mother or other knowledgeable family member and to return the questionnaire at the in‐person interview. The family history questionnaire was completed and returned by 357 (92.5%) cases, 219 (94.0%) hysterectomy controls and 297 (92.0%) population controls. We conducted unconditional logistic regression to estimate adjusted odds ratios (aORs) and 95% confidence intervals between early‐life factors and adenomyosis. Cases were compared to population and hysterectomy controls in separate analyses as the two control groups represent different sampling frames. The population controls were selected from the overall health plan population and represent the frequency of early‐life exposures in the population that gave rise to cases. Hysterectomy controls represent the population of individuals undergoing hysterectomy, all of whom had a medical indication necessitating a major surgical procedure. We adjusted for potential confounding factors informed by prior studies of early‐life factors and endometriosis given the paucity of literature for adenomyosis [ 16 , 17 , 18 ]; all analyses were adjusted for participant age at the reference date (20–39, 40–44, 45–49 and 50–59 years), mother's age at participant's birth, mother's smoking status when pregnant with the participant, firstborn status (no, yes) and participant's birthweight. For the analyses comparing cases to population controls, analyses were additionally adjusted for the frequency matching variable of reference year (continuous). Analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC). We conducted a complete‐case analysis as the proportion of missing data was minimal (≤ 3%). We conducted a series of sensitivity analyses to evaluate the robustness of our results. First, as confounding may occur from the willingness to undergo hysterectomy, a procedure experienced by cases but not population controls, we repeated the analyses restricting the population controls to those who responded they would ‘probably’ or ‘definitely’ allow a hysterectomy to be performed if recommended ( n  = 158). This information was ascertained by asking population controls the following question, ‘If, on the (reference date), you developed severe menstrual bleeding, severe menstrual pain, or severe pelvic pain every month for six months or more, how likely would you be to allow a hysterectomy to be performed if recommended?’. Second, as the association between maternal cigarette smoking and the odds of adenomyosis could be confounded by the participant's own history of smoking [ 12 ], we examined this association among participants who were never‐smokers. Third, recognising exposure misclassification could be introduced when relying on participant self‐report for early‐life factors and that the participant's mother may accurately recall the participant's birth and early‐life history [ 19 ], we repeated the analyses only among participants who received help from mothers on the family history questionnaire. The present analyses using de‐identified data were determined to not involve human subjects by the Human Research Protection Program at Michigan State University.

The study population primarily comprised non‐Hispanic, white participants, the majority of whom were ages 40–49 years at the reference date (Table  1 ). Adenomyosis cases more frequently reported lower educational attainment (high school completion or less), lower household income (< $50,000), history of ever smoking, earlier age at menarche (≤ 10 years) and greater number of pregnancies (≥ 3 pregnancies) compared to either hysterectomy or population controls. Cases also tended to have a body mass index of 25 kg/m 2 or greater compared to population controls. We observed a similar pattern of demographic and reproductive characteristics by case status among participants who completed the family history questionnaire (Table  S1 ). Participant characteristics by case status, Kaiser Permanente Washington, 2001–2006. Cases ( n  = 386) n (%) b Hysterectomy controls ( n  = 233) n (%) b Population controls ( n  = 323) n (%) b Abbreviations: BMI, body mass index; GED, general equivalency diploma; HS, high school. At reference date. May not add to column total due to missing data. Using height measured at structured interview and self‐reported weight at reference date. Not including ectopic pregnancies given the absence of uterine implantation. Among the nine early‐life factors, four factors (maternal DES use, foetal number, born outside the United States, and soy formula feeding as an infant) had fewer than 15 cases or 15 controls (either hysterectomy or population controls) that were exposed (Table  S2 ); associations for these factors are not reported due to statistical instability. For the remaining five early‐life factors, comparing cases to population controls, our data suggested that younger maternal age at participant's birth (ages ≤ 19 years vs. ages 25–29 years) was associated with an 80% increased odds of adenomyosis (aOR 1.81, 95% CI 0.94, 3.50) and participants who were the fourth or later live birth (vs. firstborn) had a 50% increased odds of adenomyosis (aOR 1.51, 95% CI 0.88, 2.59) (Table  2 ). When we considered the combination of maternal age at delivery and birth order, our data suggested participants whose mothers were ages < 25 years and whose birth order was third or later born (vs. ≥ 25 and firstborn) had almost 90% increased odds of adenomyosis (aOR 1.88, 95% CI 0.84, 4.21). We observed only a modest increased odds of adenomyosis with maternal cigarette smoking and a 40% increased odds of adenomyosis with a birthweight of > 4080 g (vs. 2500–4080 g) (aOR: 1.44, 95% CI 0.74, 2.81). For the association between preterm birth and adenomyosis odds, both the magnitude of the effect size and width of confidence intervals based on the confidence limit ratio precluded its interpretation [ 20 , 21 ]. Adjusted odds ratios (aOR) and 95% CI comparing adenomyosis cases with hysterectomy controls and population controls in relation to early‐life factors, Kaiser Permanente Washington, 2001–2006 a . Among total number of participants who completed the family history questionnaire. May not add to column total due to missing data. Adjusted for age at reference date (20–39, 40–44, 45–49, 50–59 years), maternal age (≤ 19, 20–24, 25–29, 30–34, ≥ 35 years), maternal smoking (no, yes), firstborn (no, yes) and birthweight ( 4080 g). Adjusted for age at reference date (20–39, 40–44, 45–49, 50–59 years), reference year (continuous), maternal age (≤ 19, 20–24, 25–29, 30–34, ≥ 35 years), maternal smoking (no, yes), firstborn (no, yes) and birthweight ( 4080 g). When we repeated the analyses comparing cases and hysterectomy controls, we observed attenuated or null associations. Additionally, we observed lower birthweight (< 2500 g) was associated with a 40% decreased odds of adenomyosis (aOR 0.56, 95% CI 0.32, 0.98). We observed results generally similar to those of the main analyses when we compared cases to population controls who would allow for a hysterectomy if warranted (Table  S3 ). We observed a stronger association between maternal smoking and adenomyosis among never‐smokers (using hysterectomy controls aOR 1.30, 95% CI 0.75, 2.23; using population controls aOR 1.50, 95% CI 0.92, 2.46) (Table  3 ). Adjusted odds ratios (aOR) and 95% CI for the association between maternal cigarette smoking and adenomyosis among those who have never smoked, Kaiser Permanente Washington, 2001–2006 a . Among participants who completed the family history questionnaire. May not add to column total due to missing data. Adjusted for age at reference date (20–39, 40–44, 45–49, 50–59 years), maternal age (≤ 19, 20–24, 25–29, 30–34, ≥ 35 years), firstborn (no, yes) and birthweight ( 4080 g). Adjusted for age at reference date (20–39, 40–44, 45–49, 50–59 years), reference year (continuous), maternal age (≤ 19, 20–24, 25–29, 30–34, ≥ 35 years), firstborn (no, yes) and birthweight ( 4080 g). When we restricted the study population to participants who received their mother's help on the family history questionnaire, the sensitivity analyses included only 97 (27.2%) cases, 74 (33.8%) hysterectomy controls and 99 (33.6%) population controls. This was due to less than half of the study participants receiving help from their mother or other knowledgeable family member when completing the family history questionnaire (35.4% cases, 43.4% hysterectomy controls, 39.7% population controls). Participants who received help tended to be younger (< 44 years), premenopausal, and to report a lower household income (< $50,000), compared to those who did not receive mother's help on the questionnaire (Table  S4 ). Given the small sample size, we were only able to evaluate the associations for maternal age, participant birth order and maternal cigarette smoking using data collected with the mother's assistance. We observed similar patterns of associations, although accompanied by wider confidence intervals, to those reported in the main analysis (Table  S5 ). We conducted a post hoc analysis restricting the study population to those born singleton and not preterm given the small number of participants born as a twin, triplet or preterm. The results were similar to those of the main analyses (Table  S6 ).

Adenomyosis is characterised by the presence of endometrial glands and stroma within the myometrium of the uterus. This gynaecologic condition is associated with life‐altering symptoms, including chronic pelvic pain, painful menstruation and abnormal uterine bleeding. As the gold standard for adenomyosis requires histopathological examination of the uterus after hysterectomy, the true prevalence of adenomyosis remains unknown. Historic reliance on hysterectomy for diagnosis and the associated challenges in designing a valid epidemiologic study have also precluded firm conclusions about risk factors [ 1 , 2 ]. In addition, the aetiology of adenomyosis remains unknown. One theory of pathogenesis postulates that adenomyosis arises from the invagination of the endometrium into the myometrium, with oestrogen being central to disease development; oestrogen contributes to tissue injury and local oestrogen production that weakens the endometrial‐myometrial border and myometrial smooth muscle fibres [ 3 , 4 ]. Thus, it is plausible that adenomyosis development in adulthood could originate from disruption of the hypothalamus–pituitary–ovarian (HPO)‐axis and uterine tissue in utero. The HPO‐axis regulates ovarian production of oestrogen through the release of gonadotropins from the hypothalamus and anterior pituitary. The HPO‐axis begins to develop early in gestation with gonadotropins detected in foetal sera between 8 and 12 weeks of gestation [ 5 , 6 ]. The foetal HPO‐axis is active during gestation. For example, the HPO‐axis responds to the increase in maternal oestrogen production towards the end of gestation by suppressing HPO‐axis activity [ 7 ]. Uterine tissue also develops during gestation, with differentiation of the myometrium and presence of uterine endometrial glands occurring by approximately 20–22 weeks of gestation [ 8 ]. Only two studies have investigated the association between early‐life factors and adenomyosis [ 9 , 10 ]; both studies examined a limited number of early‐life factors, including birthweight, maternal age and gestational age at birth, and did not observe an association. Those studies were conducted using data linkages to nationwide administrative databases that relied on the International Classification of Disease (ICD) coding, specifically ICD‐8, ICD‐9 and ICD‐10, to identify cases. One study identified only 24 adenomyosis cases, and the other study had limited data on confounding factors. These limitations highlight the challenge in designing epidemiologic studies for adenomyosis. Thus, the purpose of the present study was to explore a range of nine early‐life factors in relation to the odds of adenomyosis using data from a case–control study of adenomyosis, conducted in a large, integrated health plan population.

The authors have nothing to report.

As‐Sanie served as a consultant for Sumitomo (formerly Myovant Sciences‐Pfizer), Bayer, and Organon; received author royalties from UpToDate, and grant funding from Sumitomo. The other authors declare no conflicts of interest.

Data S1. Tables S1–S6.