Breastfeeding history and adenomyosis risk using a novel case-control study design

In this analytic sample, majority of study participants at the reference date were ages 45-49 years, white, and non-Hispanic. Adenomyosis cases more frequently reported lower educational attainment (some college or less) compared to population controls, but not hysterectomy controls ( Table 1 ). However, cases tended to report a lower household annual income (<$50,000) and a history of ever smoking cigarettes and have a BMI ≥ 30 kg/m 2 at the reference date, compared to both hysterectomy and population controls. Additionally, compared to both hysterectomy and population controls, cases more frequently reported an earlier age at menarche (≤ 11 years) and history of five or more pregnancies. We also compared sociodemographic and reproductive characteristics between those who had ever and never breastfed an infant at least eight weeks among population controls. Those who had a history of ever breastfeeding an infant at least eight weeks tended to report higher educational attainment (college graduate or more), higher household income (≥$70,000) and history of never smoking and have a lower BMI (<25 kg/m 2 ), compared to those who never breastfed an infant at least eight weeks ( Supplemental Table 2 ). Additionally, those who ever breastfed an infant at least eight weeks (vs. never) were more likely to report a history of four or more pregnancies. History of ever breastfeeding was common in this study population as most cases (85%), hysterectomy controls (88%), and population controls (91%) reported ever breastfeeding ( Table 2 ). We observed that participants who had ever breastfed an infant had a 30% decreased odds of adenomyosis, comparing cases to hysterectomy controls (adjusted odds ratio 0.7, 95% CI: 0.4, 1.2). Using the population controls, the association was stronger in magnitude with ever breastfeeding an infant being associated with a 40% decreased odds of adenomyosis (adjusted odds ratio 0.6, 95% CI: 0.3, 1.0). When we required at least eight weeks of breastfeeding an infant to be considered exposed, we observed similar associations with narrower confidence intervals (cases vs. hysterectomy controls: adjusted odds ratio 0.7, 95% CI: 0.5, 1.0; cases vs. population controls: adjusted odds ratio 0.6, 95% CI: 0.4, 0.8). Regarding lifetime months of breastfeeding, we observed a 40% decreased odds of adenomyosis with ≥ 12 months of lifetime breastfeeding (vs. <3 months), comparing cases to hysterectomy controls (adjusted odds ratio 0.6, 95% CI: 0.4, 0.9; P =0.068 for trend). Using the population controls, the association was stronger in magnitude: a 60% decreased odds of adenomyosis with ≥ 12 months of lifetime breastfeeding (vs. <3 months) (adjusted odds ratio 0.4, 95% CI: 0.2, 0.6; P <0,001 for trend). When we considered duration of lifetime exclusive breastfeeding, we observed the strongest association with 9-<12 months of lifetime exclusive breastfeeding (vs. <3 months) (cases vs. hysterectomy controls: adjusted odds ratio 0.4, 95% CI: 0.2, 0.8; . P =0.012 for trend; cases vs. population controls: adjusted odds ratio 0.3, 95% CI: 0.2, 0.6; P <0.001 for trend). In our sensitivity analyses, we observed results similar to those in the main analyses for the associations between breastfeeding history and adenomyosis after adjusting for gravidity instead of parity ( Supplemental Table 3 ), additionally adjusting for BMI at an earlier age (when participants were in their 20s) ( Supplemental Table 4 ), and considering exclusive breastfeeding of an infant for at least six months ( Supplemental Table 5 ). In our sensitivity analyses comparing cases to population controls restricted to those who would allow for a hysterectomy if warranted, we observed associations between breastfeeding history and adenomyosis generally comparable to results observed from the main analyses ( Supplemental Table 6 ).

We used data from a case-control study that used (cis)women-related terminology when the study was conducted and did not collect data on gender. Therefore, we used gender-inclusive language when describing the study population ( 10 ). The case-control study was conducted among pre- and postmenopausal enrollees ages 18-59 years of a large, integrated healthcare system, Kaiser Permanente Washington, in Washington State. At the time of the case-control study, the healthcare system was known as Group Health. Cases were individuals diagnosed for the first time with pathology-confirmed adenomyosis by hysterectomy between April 1, 2001 and March 31, 2006. Cases were identified by review of Group Health electronic databases of hospitalizations, inpatient and outpatient surgery, and medical visits for the ICD 9th revision diagnostic codes 617.0, “endometriosis of uterus.” Since this ICD-9 code includes diagnoses of endometriosis and adenomyosis, medical record review was conducted to identify individuals diagnosed with adenomyosis. We required the presence of adenomyosis be confirmed by pathology report, and individuals only diagnosed with endometriosis were excluded from the case group. The most common indications for hysterectomy among cases were abnormal uterine bleeding and pain ( Supplemental Table 1 ). This study design consisted of two control groups who were also ascertained via Group Health enrollment database. Hysterectomy controls were individuals who had confirmed absence of adenomyosis on pathology report by hysterectomy for benign disease during the same period that cases were diagnosed (years 2001-2006). In hysterectomy controls, the most common indications for hysterectomy were abnormal uterine bleeding and pain; additionally, uterine fibroids were the most common pathology finding ( Supplemental Table 1 ). Population controls were individuals with an intact uterus, who had no history of adenomyosis diagnosis, and had been enrolled in Group Health at some point between April 1, 2001 and March 31, 2006. Population controls were randomly selected from the health plan database and were frequency matched to cases by 5-year age groups. To be eligible, participants needed to be enrolled in Group Health for at least six months before the reference date. The reference date for cases was the date of first visit to Group Health for symptoms leading to adenomyosis diagnosis. The reference date for hysterectomy controls was the date of first visit to Group Health for symptoms leading to hysterectomy. Reference dates were assigned to population controls corresponding to the distribution of reference dates among cases. The potential cases and controls (598 cases, 431 hysterectomy controls, and 726 population cases) received a letter of invitation for the case-control study followed by a phone call from Group Health personnel for the individual’s name and telephone number that could be forwarded to Fred Hutchinson Cancer Research Center. Of those invited, 449 cases, 291 hysterectomy controls, and 707 population controls agreed to be contacted. After additional eligibility screening and completing the informed consent process, which included consenting to medical record review, 402 cases, 241 hysterectomy controls, and 354 population controls were enrolled. We restricted the study population to those for whom abstracted medical record data were available, including pathology confirmation of adenomyosis in cases or its absence in hysterectomy controls. Data for 386 cases, 233 hysterectomy controls, and 323 population controls were available for the present analyses. The Human Research Protection Program at Michigan State University determined that the present analyses using de-identified data did not involve human subjects. In the original case-control study, the main study activity was a structured, in-person interview conducted by a trained, female interviewer. Study participants were asked about a range of topics, from lifestyle behaviors to medical and pregnancy history, including history of live births and breastfeeding. For each live birth reported, participants were asked if they had breastfed this infant at all (no, yes) and if they had breastfed this infant for at least two weeks (no, yes). Participants were also asked the age (weeks or months) at which the infant began to take food, formula, or milk other than breast milk, regularly. Each participant was then asked the age (weeks or months) when the child stopped breastfeeding altogether. Information on duration of amenorrhea with breastfeeding was not collected during the in-person interview. Using this information, we created binary variables characterizing a history of ever breastfeeding (no, yes) and ever breastfeeding an infant at least eight weeks (no, yes). The duration of eight weeks aligns with the hypothesized minimum amount of time needed for lactation to induce lactational amenorrhea, as return to ovulation for non-lactating individuals may occur as early as six weeks postpartum ( 11 , 12 ). We estimated the total months of lifetime breastfeeding by summing the reported months of breastfeeding across all livebirths. For use in analyses, we categorized total months of lifetime breastfeeding as 0- <3, 3- <6, 6- <9, 9- <12, and ≥12 months. We also estimated the lifetime duration of exclusive breastfeeding, defined as duration of breastfeeding before the reported regular feeding of food, formula, or milk other than breast milk, by summing the reported time of exclusive breastfeeding across all livebirths. We categorized the lifetime duration of exclusive breastfeeding as 0- <3, 3- <6, 6- <9, 9- <12, and ≥12 months. Given our interest in capturing periods when infant feeding from lactation was the primary source of nutrition, we use the gendered terms breastfeeding and exclusive breastfeeding to characterize lactation history. Only participants who ever had a livebirth had the opportunity to breastfeed. For this reason, the final analytic sample comprised parous participants: 330 cases, 198 hysterectomy controls, and 246 population controls. We used descriptive statistics to compare sociodemographic and reproductive characteristics between the cases and two control groups. We examined the association between breastfeeding and adenomyosis risk using the breastfeeding variables of ever breastfed, ever breastfed an infant at least eight weeks, lifetime breastfeeding duration, and lifetime exclusive breastfeeding duration. We estimated the adjusted odds ratio and 95% confidence intervals (CI) using unconditional multivariable logistic regression, comparing cases to hysterectomy controls and population controls in separate analyses. We selected covariates a priori for adjustment based on associations with adenomyosis reported in the literature ( 3 ) and their sociological impact on breastfeeding ( 13 ). All analyses were adjusted for age at reference date (20-39, 40-44, 45-49, 50-59 years), cigarette smoking (never, current, former), education (high school graduate or general equivalency diploma, some college/vocational/technical college, college graduate, postgraduate), and parity (1, 2, 3, ≥4 livebirths). Since less than 1% of data on covariates were missing, we used available data in the analyses. As population controls were assigned a reference date based on the distribution of reference dates in cases, analyses comparing cases to population controls were additionally adjusted for reference year (continuous). We did not adjust for duration of hormonal contraceptive use, which primarily consisted of oral contraceptives prior to year 2001, due to the concern for reverse causation. Since oral contraceptives can lower estrogen levels to a consistently low state ( 14 , 15 ), oral contraceptives can be used as first line of treatment for menstrual pain and pelvic pain due to suspected adenomyosis ( 4 ). To test the trend across lifetime breastfeeding and exclusive breastfeeding duration categories, we created a continuous variable, assigning to cases and controls the median category value identified among controls in each category, and included that variable in the adjusted logistic regression model. Analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC). We conducted several sensitivity analyses. First, we repeated the analyses adjusting for gravidity instead of parity. This was done to evaluate potential residual confounding from adjusting for parity in the main analyses. Gravidity, rather than parity, may capture increased adenomyosis risk from trophoblast invasion of the inner myometrium that occurs in early pregnancy ( 16 ). Second, we repeated the analyses additionally adjusting for body mass index (BMI) which is positively associated with adenomyosis ( 3 ) and negatively associated with initiation and duration of breastfeeding ( 17 ). In this sensitivity analysis, we estimated the participant’s BMI at an earlier age – using the average weight participants reported when they were in their 20s – and height measured at the in-person interview to best approximate BMI prior to breastfeeding and adenomyosis development. Third, as data were not available on lactational amenorrhea, we conducted a sensitivity analysis comparing those who had ever and never exclusively breastfed an infant for at least six months; the majority of women (75%) who exclusively breastfeed are amenorrheic at six months postpartum ( 11 , 18 ). Lastly, for analyses comparing cases to population controls, we repeated the analyses after restricting population controls to those who reported they would “probably” or “definitely” allow for hysterectomy if it was recommended to them if they “developed severe menstrual bleeding, severe menstrual pain, or severe pelvic pain every month for six months or more” (n = 126). This sensitivity analysis sought to make the population controls more representative of the population that gave rise to cases given that the cases underwent hysterectomy.

In our analyses using data from a case-control study among enrollees of a large, integrated healthcare system in western Washington state, we observed that breastfeeding initiation was associated with a 40% decreased odds of adenomyosis, with the magnitude of association being stronger with longer duration of lifetime breastfeeding. Our findings agree with those from a previous large cohort study and offer further support that breastfeeding may modify the risk of adenomyosis - a condition associated with substantial pain symptoms that is only definitively cured by hysterectomy.

In the present analysis using data from a population-based case-control study of adenomyosis employing two control groups, we observed that breastfeeding was associated with decreased adenomyosis risk. The magnitude of the association was stronger with increasing lifetime breastfeeding and exclusive breastfeeding duration. The observed association between breastfeeding history and adenomyosis is biologically plausible. One theory of adenomyosis pathogenesis postulates that a hyperestrogenic state promotes increased endometrial basalis proliferation and tissue microtrauma in the junctional zone leading to the invagination of basalis endometrium into the myometrium ( 1 , 5 ). Evidence for the hyperestrogenic state includes the increased local and ovarian estrogen production among individuals with adenomyosis ( 19 - 21 ). Prolonged anovulation and the subsequent reduction of estradiol levels with breastfeeding could reduce the risk of adenomyosis. The infant suckling stimulus directly inhibits the pulsatile secretion of the gonadotropin-releasing hormone by the hypothalamus, leading to the inhibition of the luteinizing hormone and the follicle-stimulating hormone from the pituitary ( 9 ). The suppression of follicle-stimulating hormone prevents follicular recruitment, resulting in anovulation and low estrogen production by the granulosa cells ( 22 ). The infant suckling stimulus also suppresses the release of dopamine, an inhibitor of prolactin; the resulting prolonged increase in prolactin levels, hyperprolactinemia, inhibits the release of gonadotropin-releasing hormone, luteinizing hormone, and follicle-stimulating hormone and leads to hypoestrogenism ( 23 , 24 ). The dependence of the breastfeeding-induced hypoestrogenic state on duration and exclusivity of breastfeeding supports the dose-response association observed in our study; longer breastfeeding duration was associated with lower odds of adenomyosis. Our observation of an inverse association contrasts that observed in a cross-sectional study among pre- and post-menopausal women undergoing a transvaginal ultrasound examination for medical indication at a gynecologic clinic ( 7 ). That study reported the suggestion of a positive association between a history of breastfeeding for >6 months (vs. ≤6 months) and ultrasound-detected prevalent adenomyosis among parous women (unadjusted odds ratio 1.19, 95% CI: 0.76, 1.87). The reported association was not adjusted for confounding factors such as age and pregnancy history, which are positively associated with breastfeeding and adenomyosis and could increase the magnitude of the association. However, our finding is consistent with that observed in a large cohort study of over 80,000 female teachers in California that were followed for inpatient hospitalization with the diagnosis of adenomyosis ( 8 ). That study reported a 26% decrease in the prevalence of surgically-confirmed adenomyosis with a history of ever breastfeeding (prevalence odds ratio 0.74, 95% CI: 0.62, 0.88) among parous women. The discrepant results across studies are likely due to the differences in sampling frames, method of adenomyosis diagnosis, and adjustment for confounding. Our study had several limitations. First, we relied on recall to ascertain information on breastfeeding history, including duration, for each livebirth a participant reported. Data from several studies comparing the prospective collection of breastfeeding duration data to maternal recall one to two decades later suggest high accuracy of maternal recall. The measures of reliability ranged from 0.84-0.95 for 8-10 years of recall ( 25 , 26 ) and 0.82-0.86 for 20 years of recall ( 27 , 28 ). This suggests exposure misclassification for breastfeeding duration due to recall is minimal. Second, we also relied on recall for body weight ascertainment for when the participant was in their 20’s. Several studies comparing measured body weight to that recalled one to three decades later have observed high accuracy of recalled body weight ( 29 - 33 ). Third, information was not collected during the in-person interview on the timing of return to menses while breastfeeding. Information on return to menses after birth would have allowed for the estimation of lactational amenorrhea that is indicative of the hypoestrogenic state experienced with breastfeeding. We had available information on the timing of regular supplementary feeding to estimate the duration of exclusive breastfeeding. The time of supplementary feeding has been observed to be correlated with the time of menstrual resumption (Pearson correlation coefficient = 0.56, P < 0.01) in lactating postpartum women ( 34 ). In our sensitivity analysis considering exclusive breastfeeding of an infant for at least six months, we continued to observe a strong association between breastfeeding history and adenomyosis. Fourth, it is possible that some population controls may have undiagnosed adenomyosis as they did not have confirmation of disease absence through hysterectomy. Although the prevalence of adenomyosis is not known in the general population, Naftalin et al. (2012) observed an adenomyosis prevalence of 21% using transvaginal ultrasound imaging among women with a medical indication, including pelvic pain ( 7 ). Among the population controls in the present study (n = 246), 53 (22%) reported experiencing pelvic pain when not menstruating. Hence, if adenomyosis was present in 21% of these population controls with pelvic pain, then an estimated 5% of population controls could have undiagnosed adenomyosis. This suggests that outcome misclassification among the population controls is likely to be low. Lastly, there is potential for incomplete adjustment as breastfeeding is strongly associated with other behaviors and socioeconomic factors ( 13 , 35 ) and few individuals in our study population of parous health plan enrollees in the Pacific Northwest reported never breastfeeding. Although our observed high breastfeeding rates mirror what is known about breastfeeding frequency in the Pacific Northwest ( 13 ), the results from our study may not be generalizable to other parous populations. Conversely, due to the high frequency of breastfeeding history and collection of detailed breastfeeding data, we were able to investigate the relationship between adenomyosis and duration of breastfeeding. Our study benefitted from the case-control study being conducted within a population of health plan enrollees and employing two control groups. The gold standard for adenomyosis diagnosis, histopathologic confirmation of disease after hysterectomy, creates a substantial challenge in selecting a comparison group of non-cases that allows for a valid epidemiologic study design ( 3 ). Although the absence of histopathologic disease can be confirmed among individuals undergoing hysterectomy and hysterectomy controls are similar to cases on factors, including willingness, related to undergoing a hysterectomy, the medical indications warranting hysterectomy may be associated with breastfeeding history and introduce selection bias. For this reason, in the original case-control study, another control group was employed: population controls randomly selected from the underlying population that gave rise to cases. This was possible by using the sampling frame of health plan enrollees. Although this selection approach minimizes selection bias, population controls may differ from cases on factors related to undergoing hysterectomy and the results may be susceptible to confounding. However, results from our sensitivity analysis comparing cases to population controls restricted to those who would “probably” or “definitely” have a hysterectomy if needed were similar to those in the main analysis, suggesting that potential bias from this source of confounding was minimal. The use of hysterectomy and population controls allowed us to comprehensively evaluate the association between breastfeeding history and adenomyosis. Going forward, advancements in imaging technology now allow for the non-invasive detection of adenomyosis in the general population. By overcoming study design issues related to the selection of non-cases, screening of the general population would provide a fuller picture of the association between breastfeeding history and adenomyosis risk.

Adenomyosis is characterized by the presence of endometrial glands and stroma within the uterine myometrium ( 1 ). This condition is associated with substantial morbidity, including painful menstruation, chronic pelvic pain, abnormal uterine bleeding, and pain with urination ( 2 ). Definitive treatment of symptoms is primarily by hysterectomy ( 3 , 4 ). While the etiology of adenomyosis remains unknown, it is recognized that estrogen is central to disease pathogenesis; the presence of a hyperestrogenic state initiates endometrial proliferation, inflammation, and endometrial invagination of the myometrium ( 5 ). Since breastfeeding induces a hypoestrogenic state and suppresses ovulation ( 6 ), breastfeeding history may decrease the risk of adenomyosis. Only two previous studies have investigated the association between breastfeeding history and adenomyosis risk, and the studies reported associations in opposing directions ( 7 , 8 ). The discrepant results are likely due to the challenges in designing a valid epidemiologic study of adenomyosis, particularly the selection of non-cases ( 3 ). Adenomyosis has been historically diagnosed through histopathologic examination of uterine tissue post-hysterectomy ( 3 ). As such, hysterectomy controls are commonly selected for comparison. Although hysterectomy controls have confirmed absence of adenomyosis and are similar to cases on factors related to having a hysterectomy, they are not selected from the underlying population that gave rise to cases. Conversely, a control group comprising randomly selected individuals from the underlying source population would allow for controls to represent the frequency of exposure in the population that gave rise to cases. However, differences in measured and unmeasured factors between cases and population controls related to the cases’ willingness to undergo hysterectomy may result in confounding. The discrepant results across prior studies also may be due to variation in the extent of breastfeeding. The impact of breastfeeding on estrogen deficiency depends on the daily frequency and months of lactation, and whether breastfeeding is the only source of infant nutrition ( 9 ). Neither of the prior studies investigated breastfeeding duration, particularly exclusive breastfeeding duration, on adenomyosis risk. Given these challenges, the purpose of the present study was to investigate the association between breastfeeding history, including exclusive breastfeeding and duration of breastfeeding, and adenomyosis risk, improving upon previous studies by using two control groups. We hypothesized that among parous women, breastfeeding, specifically longer duration of breastfeeding, would decrease risk of adenomyosis.