Abstract
Background
Phytoestrogen exposure from soy formula feeding during infancy may disrupt reproductive system development, resulting in altered menstrual bleeding in adulthood.
Methods
We investigated this relationship in a cohort of 1,696 young African American women using enrollment data from the Study of Environment, Lifestyle & Fibroids (2010-2012). Questionnaire data on soy formula feeding were available for 1,553 participants, 89% based on mother's report. Menstrual bleeding outcomes including those indicative of heavy menstrual bleeding were ascertained by interview. We estimated relative risks (RR) and 95% confidence intervals (CI) for associations between soy formula feeding and menstrual bleeding outcomes using log-binomial regression and log-multinomial regression, comparing participants ever fed and never fed soy formula.
Results
We observed associations between soy formula feeding and variables indicating a history of heavy menstrual bleeding, including ever experiencing heavy, gushing-type bleeding (RR 1.2, 95% CI: 1.0-1.4), ever use of non-steroidal anti-inflammatory drugs for heavy bleeding (RR 1.3, 95% CI: 1.0-1.6), and ever use of a contraceptive method for heavy bleeding (RR 1.2, 95% CI: 0.9-1.6). Among the subset of participants with menses in the past year who did not use medication that may alter menstrual flow (n=762), our data suggested that soy formula feeding was associated with heavier flow and the adverse impact of menstrual bleeding on quality of life, but confidence intervals were wide.
Conclusions
Our data suggested that soy formula feeding is associated with heavy menstrual bleeding. Our results support the idea that infancy is a susceptible developmental window for female reproductive function.
Introduction
Heavy menstrual bleeding is defined as excessive menstrual bleeding that interferes with physical, emotional, social, and economic quality of life.1 This condition is common; a recent survey of African American women reported that 39% of women had seen a doctor for heavy menstrual bleeding.2 Women have reported extraordinary strategies to cope with such bleeding, including missing work and avoiding activities that require leaving the home.3-5 As such, U.S. women whose menstrual flow has increased over the past year are 28% less likely to be at work on any given week than women with lighter or unchanged menstrual flow.6 In addition, excessive menstruation not only results in hospitalization, it is the most frequent diagnosis among hospitalizations for menstrual disorders; menstrual disorders account for 15% of gynecologic hospitalizations among U.S. reproductive-age women.7
Menstrual bleeding is regulated by local endometrial shedding and repair mechanisms, many of which are driven by ovarian steroids whose production are governed by the hypothalamus-pituitary-gonadal (HPG) axis during the menstrual cycle.8-11 Infancy may be a critical window for uterine and HPG-axis development as uterine tissue continues to differentiate, and some studies suggest that the HPG-axis may be active during infancy.12-16 Thus, exposure to exogenous hormones during infancy may disrupt this development, resulting in persistent changes in the uterus and HPG-axis and altered menstrual bleeding in adulthood.
We hypothesized that soy formula feeding during infancy may be one such exogenous hormonal exposure. Soy formula contains phytoestrogens, predominantly genistein and daidzein, that are structurally similar to estradiol.17 Infants fed soy formula can be highly exposed to these phytoestrogens,18 particularly when soy formula is the main source of nutrition. Exposure to genistein during the early neonatal period has been shown in rodent animal models to change the structure and function of the adult reproductive tract and HPG-axis.19-22
To our knowledge, only one prior study has examined the relationship between soy formula feeding in infancy and menstrual bleeding characteristics in adulthood.23 That study evaluated a broad range of reproductive outcomes. With regard to menstrual bleeding characteristics, the authors reported that soy formula feeding was associated with slightly longer bleeding duration but not with heavy menstrual flow.23 The purpose of the present analyses was to investigate the associations between soy formula feeding in infancy and several menstrual bleeding outcomes, including those indicative of heavy menstrual bleeding, among a cohort of young African American women.
Methods
Study population
We used enrollment data from the National Institute of Environmental Health Sciences (NIEHS) Study of Environment, Lifestyle & Fibroids (SELF), an ongoing cohort study designed to identify risk factors for fibroid incidence and growth. As previously described, SELF consists of 1,696 African American women who were recruited from the Detroit, Michigan area and who were ages 23 to 34 years at enrollment.24 The primary eligibility criteria for SELF included having an intact uterus and no prior diagnosis of uterine fibroids. The participants in SELF participated in several study activities at enrollment, including attending a clinic visit and completing computer-assisted telephone and web-based interviews and self-administered questionnaires. The conduct of SELF was approved by the Institutional Review Boards at NIEHS and at the collaborating healthcare system for this project, the Henry Ford Health System. Each participant in SELF provided informed consent before enrollment.
Exposure ascertainment
The exposure of interest in the present analyses, soy formula feeding during infancy, was ascertained using the Early Life Questionnaire that was given to SELF participants at the enrollment clinic visit and returned by mail. The questionnaire was designed so that the participant could interview her mother, other relatives, or her mother's friends to obtain the requested information. The information collected in the questionnaire included whether the participant was ever fed soy-based infant formula (yes, no), the duration of soy formula feeding (<1 month, 1-3 months, 4-6 months, >6 months) and whether the feeding was initiated within the first two months after birth (yes, no). The questionnaire was returned by 1,598 SELF participants (94%) and data on soy formula feeding was available for 1,553 participants (198 ever fed soy formula, 1355 never fed soy formula), comprising the study sample for the present analyses. Eighty-nine percent of these participants had the assistance of their mothers when completing the questionnaire.
Outcome ascertainment
We considered eight menstrual bleeding outcomes, the majority of which were focused on heavy menstrual bleeding. Three outcomes regarding any past experience with heavy menstrual bleeding were evaluated among the entire study sample of 1,553 participants. These three outcomes were collected during the computer-assisted telephone interview: 1) ever having menstrual periods with heavy, gushing-type bleeding defined as bleeding that was too much for pads or tampons to absorb, even when changed frequently (yes, no); 2) ever use of non-steroidal anti-inflammatory drugs (NSAIDs) such as ibuprofen or Motrin for heavy menstrual bleeding (yes, no); and 3) ever use of a hormonal method of birth control (birth controls pills, hormonal implant, patch, vaginal ring, or injections) or an intrauterine device (IUD) for heavy menstrual bleeding (yes, no).
The five other menstrual bleeding outcomes were based on information about the last menstrual period (LMP) as well as menstrual bleeding during the 12 months preceding the interview date. We evaluated these outcomes among a subset of participants. Starting with the study sample of 1,553 participants, we excluded women who reported no menses in the past year (primarily due to contraceptive method use) (n=124), women who reported menses in the past year but who either had no menses in the year before LMP (n=22) or used hormonal birth control or medication that alters the frequency of menses or induces cessation of menses at any time during the year before LMP (n=469). Among the remaining menstruating women, we additionally excluded women who were currently using NSAIDs for heavy menstrual bleeding (n=175) or whose NSAID use was not known (n=1). In this paper, we refer to this subset of participants (n=762; 89 exposed, 673 unexposed) as women having natural menses.
Of the five menstrual bleeding outcomes that we evaluated among women with natural menses, two were focused on menstrual bleeding characteristics, heaviness and duration, at the LMP. Menstrual flow heaviness was determined using information on pad and tampon product use on the heaviest day of bleeding during the LMP. We estimated the menstrual flow heaviness (ml of blood) by multiplying the absorbency of the most absorbent pad or tampon product used, total number of pads and tampons used during the entire day, and the frequency that the pads or tampons were fully saturated when changed (eTables 1-2, eFigure 1). We used a scoring system similar to that proposed by Wyatt et al.25 to assign the pad and tampon product absorbency in terms of menstrual blood (ml). The estimated menstrual flow was categorized as light, medium, and heavy (eTable 3). Duration of menstrual bleeding was defined as the number of days of bleeding at LMP, not including days of spotting. We categorized duration as ≤3, 4-5, and ≥6 days. The three other outcomes concerned the impact of menstrual bleeding on quality of life; these outcomes were collected in the computer-assisted web-interview and were assessed in relation to the 12 months before interview: 1) the extent to which menstrual bleeding was a problem (not a problem, a small problem, a medium problem, a big problem); 2) the extent to which menstrual bleeding interfered with being able to do things that the participant likes to do or has to do (not at all, a little, some/a lot); and 3) the number of days that the participant missed work or stopped home activities because of menstrual bleeding that we categorized as <3 and ≥3 days.
The majority of menstrual bleeding outcomes collected in SELF were adapted from questions used in the NIEHS Uterine Fibroid Study26-28 to allow for comparison of findings across studies. SELF improved upon the NIEHS Uterine Fibroid Study with the inclusion of additional questions on product absorbency and saturation to allow for the estimation of menstrual flow heaviness in a manner that is similar to an approach that uses a pictorial questionnaire.25
Ascertainment of participant and maternal characteristics
We considered the characteristics of the participant at enrollment and as an infant as well as the characteristics of her mother when she was pregnant with the participant. In SELF, these characteristics were primarily collected by questionnaire and interview with most of the infant and maternal characteristics collected using the Early Life Questionnaire; participant height and weight were measured at the clinic visit.
Statistical analyses
We directly estimated the relative risk (RR) and 95% confidence intervals (CI) for the association between soy formula feeding and menstrual bleeding outcomes using log-binomial regression, comparing participants who were ever and never fed soy formula as infants. For menstrual bleeding outcomes with more than two categories, we used log-multinomial regression (Leigh Blizzard, 12/19/2014, STATA command documentation).29 Log-multinomial regression also directly estimates the RR and 95% CI and compares the risk of a given menstrual bleeding outcome category between participants ever and never fed soy formula. In log-multinomial regression, one outcome category is not fitted in the modelling process29 and we always chose the outcome category representing the least bleeding or the least impact on quality of life as the non-fitted category. We present data only for our analyses of the dichotomous exposure of soy formula feeding in infancy (ever vs. never); the additional categorization of exposure by soy formula duration and timing of initiation (within two months of birth or after) resulted in small numbers and statistical instability.
In all analyses, we adjusted for participant age (continuous) given its association with heavy menstrual bleeding and its association with infant feeding practice when the participants were born.30 We also adjusted for maternal education (≤high school diploma or GED (general equivalency diploma), some college, associate's degree or higher degree); maternal education may be associated with reporting of soy formula feeding as well as unmeasured early life exposures that may be associated with soy formula feeding and heavy menstrual bleeding.
We conducted several sensitivity analyses. First, we repeated the analyses adjusting for exposure to breastfeeding during infancy (ever/never) as well as maternal hypertension/preeclampsia (no/yes). Secondly, we employed a stricter exposure definition, with exposure being defined as at least one month of soy formula feeding (versus never being exposed), excluding 24 participants fed soy formula for less than a month (10 participants were excluded among the subset with natural menses). Finally, we repeated the analyses restricting the study population to women who did not report a prior diagnosis of endometriosis, cervical polyps, or uterine polyps, and were not found to have prevalent fibroids on ultrasound at the enrollment visit (restricted study sample (n=1174): 155 exposed and 1019 unexposed; restricted subset with natural menses (n=576): 72 exposed, 504 unexposed). This was done to understand the relationships between soy formula feeding and menstrual bleeding outcomes not mediated by uterine conditions that are associated with heavy menstrual bleeding.11
Given that the ascertainment of soy formula feeding relied on the recall of the participant's mother or other relatives over 23-34 years, we conducted probabilistic bias analysis31, 32 to explore the impact of exposure misclassification on the associations between soy formula feeding and the three binary menstrual bleeding outcomes using the episensi command in STATA.33 We provide details of this analysis in the eAppendix, Description of probabilistic bias analysis. Briefly, we specified trapezoidal distributions for the sensitivity (minimum 0.80, modes 0.85 and 0.95, maximum 1.0) and specificity (minimum 0.92, modes 0.94 and 0.96, maximum 0.98) of exposure classification; and we used 50,000 random draws from these distributions in our analysis. We repeated the entire bias analysis under other distributional specifications to explore how they might influence the results (eTable 4).
The statistical analyses were conducted using STATA version 13.1 (StataCorp, College Station, TX) and we considered the significance level as α=0.05 in all analyses. The log-binomial regression estimates were similar to those obtained using Poisson regression with robust variance. We also provide estimates of the relative risk ratios and 95% CIs using multinomial logistic regression for our analyses involving more than two outcome categories (eTable 5).
Results
In our study sample of 1,553 participants, 13% reported ever being fed soy formula as infants. More than half of these participants were fed soy formula for more than six months (53%) and feeding was initiated within the first two months of birth (58%). Women who were fed soy formula as infants were generally similar demographically to women who were not fed soy formula, although a slightly greater percentage of exposed women were younger at enrollment and had a body mass index ≥25.0 kg/m2 (Table 1). Among the subset of participants with natural menses (n=762), we additionally observed that exposed women were less likely to be nulliparous and more likely have had menarche before age 13 years and have some college education compared to unexposed women.
Table 1.
Characteristics of participants by infant soy formula feeding exposure in the Study of Environment, Lifestyle & Fibroids (SELF), 2010-2012.
| Characteristic | All participants (n=1553) | Subset of participants with natural menses data (n=762)a | ||
|---|---|---|---|---|
|
| ||||
| Ever fed soy formula (n=198) | Never fed soy formula (n=1355) | Ever fed soy formula (n=89) | Never fed soy formula (n=673) | |
| n (%) | n (%) | n (%) | n (%) | |
| Participant age at enrollment (years) | ||||
| 23-25 | 56 (28) | 298 (22) | 20 (22) | 133 (20) |
| 26-28 | 50 (25) | 333 (25) | 20 (22) | 158 (23) |
| 29-31 | 51 (26) | 370 (27) | 25 (28) | 190 (28) |
| 32-35 | 41 (21) | 354 (26) | 24 (27) | 192 (29) |
| Parity (number of births) | ||||
| 0 | 85 (43) | 531 (39) | 31 (35) | 281 (42) |
| 1 | 47 (24) | 350 (26) | 23 (26) | 155 (23) |
| 2 | 35 (18) | 241 (18) | 15 (17) | 110 (16) |
| ≥3 | 31 (16) | 233 (17) | 20 (22) | 127 (19) |
| Age at menarche (years) | ||||
| ≤10 | 38 (19) | 248 (18) | 16 (18) | 124 (18) |
| 11 | 46 (23) | 260 (19) | 17 (19) | 134 (20) |
| 12 | 52 (26) | 361 (27) | 30 (34) | 172 (26) |
| 13 | 33 (17) | 230 (17) | 14 (16) | 115 (17) |
| ≥14 | 29 (15) | 256 (19) | 12 (13) | 128 (19) |
| Education | ||||
| ≤HS/GED | 36 (18) | 299 (22) | 16 (18) | 162 (24) |
| Some college or Associate/technical degree | 100 (51) | 676 (50) | 51 (57) | 336 (50) |
| Bachelor's, Master's or doctoral degree | 62 (31) | 379 (28) | 22 (25) | 174 (26) |
| Missing | 0 | 1 | 0 | 1 |
| Income (US$) | ||||
| <20,000 | 82 (42) | 621 (46) | 43 (49) | 326 (49) |
| 20,000-50,000 | 73 (37) | 501 (37) | 28 (32) | 241 (36) |
| >50,000 | 41 (21) | 224 (17) | 17 (19) | 102 (15) |
| Missing | 2 | 9 | 1 | 4 |
| Smoking status | ||||
| Never | 152 (77) | 994 (73) | 67 (75) | 494 (73) |
| Former | 16 (8) | 99 (7) | 7 (8) | 45 (7) |
| Current | 30 (15) | 262 (19) | 15 (17) | 134 (20) |
| Alcohol use at most | ||||
| Low/never | 52 (26) | 359 (26) | 27 (30) | 209 (31) |
| Moderate | 61 (31) | 454 (34) | 25 (28) | 207 (31) |
| Heavy | 85 (43) | 542 (40) | 37 (42) | 257 (38) |
| Body mass index (kg/m2) | ||||
| <18.5 | 4 (2) | 9 (1) | 2 (2) | 6 (1) |
| 18.5 to <25.0 | 28 (14) | 263 (19) | 13 (15) | 139 (21) |
| 25.0 to <30.0 | 45 (23) | 273 (20) | 25 (28) | 119 (18) |
| 30.0 to <35.0 | 40 (20) | 261 (19) | 14 (16) | 129 (19) |
| ≥35.0 | 81 (41) | 549 (41) | 35 (39) | 280 (42) |
Abbreviations: HS=high school degree, GED=general equivalency diploma.
Subset of participants with natural menses, defined as women who reported menstrual periods in the year before the last menstrual period and study interview date and who reported not using medications that may alter menses (hormonal birth control or other medications during the year before the last menstrual period or current use of non-steroidal anti-inflammatory drugs for heavy menstrual bleeding).
Women fed soy formula as infants were more likely than unexposed women to have mothers with an older age at delivery, pregnancy-related hypertension or preeclampsia when pregnant with the participant, and at least some college education. In addition, exposed women were more likely to weigh more at birth and to have been breastfed than unexposed women (Table 2). A similar pattern was also observed among the subset of exposed and unexposed women with natural menses.
Table 2.
Characteristics of participants' mothers (maternal) and early life of participants (infant) by infant soy formula feeding exposure in the Study of Environment, Lifestyle & Fibroids (SELF), 2010-2012.
| Characteristic | All participants (n=1553) | Subset of participants with natural menses data (n=762)a | ||
|---|---|---|---|---|
|
| ||||
| Ever fed soy formula (n=198) | Never fed soy formula (n=1355) | Ever fed soy formula (n=89) | Never fed soy formula (n=673) | |
| n (%) | n (%) | n (%) | n (%) | |
| Maternal | ||||
| Age at delivery (years) | ||||
| 12-19 | 35 (18) | 312 (23) | 19 (21) | 158 (23) |
| 20-24 | 63 (32) | 439 (32) | 26 (29) | 224 (33) |
| 25-29 | 52 (26) | 362 (27) | 23 (26) | 170 (25) |
| 30-34 | 35 (18) | 174 (13) | 18 (20) | 87 (13) |
| 35-44 | 13 (7) | 68 (5) | 3 (3) | 34 (5) |
| Maternal smoking | ||||
| No | 147 (74) | 994 (73) | 63 (71) | 479 (71) |
| Yes | 51 (26) | 361 (27) | 26 (29) | 194 (29) |
| Gestational or pre-existing diabetes | ||||
| No | 181 (93) | 1274 (95) | 81 (92) | 638 (96) |
| Yes | 14 (7) | 65 (5) | 7 (8) | 30 (4) |
| Missing | 3 | 16 | 1 | 5 |
| Pregnancy-related hypertension or pre-eclampsia | ||||
| No | 155 (79) | 1175 (89) | 72 (81) | 582 (89) |
| Yes | 40 (21) | 148 (11) | 17 (19) | 73 (11) |
| Missing | 3 | 32 | 0 | 18 |
| Highest educational level of mother or primary caregiver when participant was age 10 years | ||||
| ≤HS or GED | 72 (36) | 643 (48) | 31 (35) | 340 (51) |
| Some college orAssociate/technical degree | 93 (47) | 557 (41) | 49 (55) | 267 (40) |
| Bachelor's/Master's/doctoral degree | 33 (17) | 153 (11) | 9 (10) | 65 (10) |
| Missing | 0 | 2 | 0 | 1 |
| Economic status during participant's childhood | ||||
| Poor | 2 (1) | 69 (5) | 1 (1) | 36 (5) |
| Low income | 72 (36) | 480 (35) | 34 (38) | 256 (38) |
| Middle income | 110 (56) | 702 (52) | 51 (57) | 337 (50) |
| Well off | 14 (7) | 103 (8) | 3 (3) | 44 (7) |
| Missing | 0 | 1 | 0 | 0 |
| Infant | ||||
| Birth weight (g) | ||||
| <2500 | 29 (15) | 181 (13) | 10 (11) | 95 (14) |
| 2500 to <3500 | 113 (57) | 853 (63) | 51 (57) | 429 (64) |
| ≥3500 | 55 (28) | 311 (23) | 28 (31) | 146 (22) |
| Missing | 1 | 10 | 0 | 3 |
| Born a week or more before due date | ||||
| No | 140 (72) | 984 (75) | 69 (78) | 484 (75) |
| Yes | 54 (28) | 325 (25) | 19 (22) | 165 (25) |
| Yes, born 1-2 weeks early | 25 (13) | 186 (14) | 8 (9) | 83 (12) |
| Yes, born ≥3weeks early | 26 (13) | 121 (9) | 11 (12) | 73 (11) |
| Yes, unknown number of weeks born early | 3 (2) | 18 (1) | 0 (0) | 9 (1) |
| Missing | 4 | 46 | 1 | 24 |
| Multiple gestation | ||||
| No | 192 (97) | 1314 (97) | 86 (97) | 650 (97) |
| Yes | 6 (3) | 41 (3) | 3 (3) | 23 (3) |
| First born | ||||
| No | 111 (57) | 776 (58) | 47 (54) | 387 (58) |
| Yes | 84 (43) | 563 (42) | 40 (46) | 277 (42) |
| Missing | 3 | 16 | 2 | 9 |
| Breastfeeding duration (months) | ||||
| Never breastfed | 105 (53) | 950 (71) | 50 (56) | 470 (70) |
| Any breastfeeding | 93 (47) | 396 (29) | 39 (44) | 200 (30) |
| <1 | 18 (9) | 64 (5) | 9 (10) | 35 (5) |
| 1-3 | 34 (17) | 105 (8) | 18 (20) | 48 (7) |
| 4-6 | 21 (11) | 78 (6) | 7 (8) | 44 (7) |
| >6 | 16 (8) | 121 (9) | 5 (6) | 58 (9) |
| Duration unknown | 4 (2) | 28 (2) | 0 (0) | 15 (2) |
| Missing | 0 | 9 | 0 | 3 |
Abbreviations: HS=high school degree; GED=general equivalency diploma.
Subset of participants with natural menses, defined as women who reported menstrual periods in the year before the last menstrual period and study interview date and who reported not using medications that may alter menses (hormonal birth control or other medications during the year before the last menstrual period or current use of non-steroidal anti-inflammatory drugs for heavy menstrual bleeding).
The frequency of adverse menstrual bleeding outcomes ranged from 9% to 37%, with past history of menstrual periods with heavy, gushing-type bleeding (37%) and ever use of NSAIDs for heavy menstrual bleeding (25%) being the most commonly reported outcomes. Comparing women who were ever and never fed soy formula as infants, we observed an increased risk of ever having menstrual periods with heavy, gushing-type bleeding (RR 1.2, 95% CI: 1.0-1.4) and ever use of NSAIDs for heavy menstrual bleeding (RR 1.3, 95% CI: 1.0-1.6) in relation to soy formula feeding, adjusting for participant age and maternal education (Table 3). There was weaker evidence of an association between infant soy formula feeding and ever use of a contraceptive method for heavy menstrual bleeding (RR 1.2, 95% CI: 0.9-1.6).
Table 3.
Relative risk and 95% CI for the association between infant soy formula feeding and menstrual bleeding outcomes, Study of Environment, Lifestyle & Fibroids (SELF), 2010-2012.
| Ever fed soy formula (n=198)a | Never fed soy formula (n=1355)a | ||
|---|---|---|---|
| Characteristics | n (%) | n (%) | RR (95% CI)b |
| During lifetime | |||
| Ever heavy, gushing-type bleedingc | |||
| No | 113 (57) | 869 (64) | |
| Yes | 85 (43) | 486 (36) | 1.2 (1.0-1.4) |
| Ever use of NSAIDs for heavy bleeding | |||
| No | 137 (69) | 1029 (76) | |
| Yes | 61 (31) | 325 (24) | 1.3 (1.0-1.6) |
| Ever use of a contraceptive method for heavy bleeding | |||
| No | 153 (77) | 1104 (81) | |
| Yes | 45 (23) | 251 (19) | 1.2 (0.9-1.6) |
Abbreviations: RR=relative risk; CI=confidence interval; NSAIDs=non-steroidal anti-inflammatory drugs.
Numbers may not add to column total due to missing data.
Adjusted for age and maternal education.
Gushing-type menstrual bleeding was defined as bleeding that was too much for pads or tampons to absorb, even when changed frequently.
Among the subset of participants with natural menses, the duration of bleeding at the LMP did not differ by exposure status (Table 4). Our data suggested that soy formula feeding was associated with a modest increased risk of heavier menstrual flow at LMP and adverse impact of menstrual bleeding on the quality of daily life in the past 12 months, although the confidence intervals were wide or failed to show the expected monotonic increase.
Table 4.
Relative risk and 95% CI for the association between infant soy formula feeding and menstrual bleeding outcomes among participants with natural menses,a Study of Environment, Lifestyle & Fibroids (SELF), 2010-2012.
| Ever fed soy formula (n=89)b | Never fed soy formula (n=673)b | ||
|---|---|---|---|
| Characteristics | n (%) | n (%) | RR (95% CI)c |
| Last menstrual period | |||
| Menstrual bleeding duration (days)d | |||
| ≤3 | 25 (28) | 184 (28) | |
| 4-5 | 49 (55) | 367 (55) | 1.0 (0.8-1.2) |
| ≥6 | 15 (17) | 117 (18) | 1.0 (0.6-1.6) |
| Menstrual bleeding heavinesse | |||
| Light | 33 (38) | 284 (43) | |
| Medium | 33 (38) | 250 (38) | 1.0 (0.8-1.3) |
| Heavy | 21 (24) | 127 (19) | 1.2 (0.8-1.8) |
| Past 12 months | |||
| Problem with menstrual bleeding | |||
| Not a problem | 40 (45) | 338 (50) | |
| Small problem | 19 (22) | 126 (19) | 1.2 (0.8-1.8) |
| Medium problem | 17 (19) | 131 (19) | 1.0 (0.6-1.6) |
| Big problem | 12 (14) | 78 (12) | 1.2 (0.7-2.1) |
| Interference of menstrual bleeding with daily activities | |||
| Not at all | 45 (51) | 425 (63) | |
| A little | 29 (33) | 144 (21) | 1.6 (1.1-2.2) |
| Some/A lot | 15 (17) | 104 (15) | 1.1 (0.6-1.8) |
| Number of days missed work or stopped home activities due to menstrual bleeding | |||
| <3 | 78 (88) | 614 (91) | |
| ≥3 | 11 (12) | 59 (9) | 1.4 (0.8-2.5) |
Abbreviations: RR=relative risk; CI=confidence interval.
Restricted to the subset of participants with natural menses (n=762, 89 exposed, 673 unexposed), defined as women who reported menstrual periods in the year before the last menstrual period and study interview date and who reported not using medications that may alter menses (hormonal birth control or other medications during the year before the last menstrual period or current use of non-steroidal anti-inflammatory drugs for heavy menstrual bleeding).
Numbers may not add to column total due to missing data.
Adjusted for age and maternal education.
Excluding women who reported only spotting and not bleeding at last menstrual period.
Bleeding heaviness determined using information on number, absorbency, and saturation of pad and tampon products on heaviest day of bleeding during last menstrual period.
Our results did not change after further adjustment for breastfeeding or maternal hypertension/preeclampsia, requiring a minimum of at least one month of soy formula feeding to be considered exposed, or restricting the study population to women without other uterine conditions associated with heavy menstrual bleeding (eTables 6-13).
Under the assumptions embodied in our probabilistic bias analysis to correct for exposure misclassification, the associations between soy formula feeding and the outcomes of heavy, gushing-type bleeding, NSAID and contraceptive method use for heavy menstrual bleeding were generally stronger in magnitude than those of the conventional analyses (eTable 4).
Discussion
Our data suggested that infant soy formula feeding is associated with an increased risk of heavy menstrual bleeding.
Heavy menstrual bleeding can occur from alterations in the HPG axis and the local endometrial mechanisms that regulate endometrial shedding and repair.8-11 Local endometrial mechanisms, including vasoconstriction, hemostasis, endometrial epithelial growth, inflammation, and angiogenesis, limit menstrual blood loss and repair and regenerate the endometrium.8, 9 Many of these local endometrial mechanisms are driven by ovarian steroids.10 Within the HPG-axis, the production of ovarian steroids are governed by the release of gonadotropins—luteinizing hormone (LH) and follicular stimulating hormone (FSH)—from the anterior pituitary as well as gonadotrophin releasing hormone (GnRH) from the hypothalamus. Hence, given the tightly-linked relationship between the uterus and the organs of the HPG-axis, heavy menstrual bleeding can result from dysfunction within any one of these organs and this dysfunction could occur from disruption during early life development.
Infancy is a critical early life period of HPG-axis development12, 13, 15, 16 and uterine tissue differentiation.14 There is some evidence for the occurrence of a “mini-puberty” in infancy characterized by the activation of the hypothalamus and pituitary and the release of gonadotropins15 that result in ovarian follicle development with concomitant estrogen production.12, 13, 15, 16 This secreted estrogen could then stimulate target tissue including the uterus.12 Although the significance of this “mini-puberty” in infant females is not well understood, the postnatal hypothalamus, pituitary and ovarian actions may have a role in the maturation of reproductive organs and the development of feedback loop mechanisms.12, 16 Thus, the differentiation of the reproductive system and development of steroid feedback mechanisms during early postnatal life may be susceptible to disruption from exogenous hormones.
We hypothesized that soy formula feeding is one such exogenous hormonal exposure that disrupts development during infancy, resulting in altered menstrual bleeding in adulthood. Soy formula contains the isoflavonoid phytoestrogens genistein and daidzein, which are structurally similar to estradiol.17 Although these phytoestrogens interact weakly with estrogen receptors compared to estradiol (as reviewed in Jefferson et al.34 and McCarver et al.18), they appear to have a higher binding affinity than several man-made endocrine disrupting chemicals.35-37 Infants can be highly exposed to these phytoestrogens;18 soy formula may be the primary, if not the exclusive, source of nutrition and infants are able to absorb, digest, and excrete genistein and daidzein.38 In addition, infant girls exclusively fed soy formula have been shown to have greater estrogenization of the vaginal tissue compared to those fed cow's milk formula or breastmilk.39
The administration of genistein during the early neonatal period in rodent animal models has demonstrated changes to the structure and function of the reproductive tract and HPG-axis in adulthood.19-22, 40, 41 Adult uterine changes from early life genistein exposure include altered morphology such as hypertrophy of the uterine epithelial cells and myometrium.22 In the ovaries, neonatal genistein administration was observed to induce ovarian dysfunction that was dose-dependent, with a greater number of corpora lutea at the lowest doses of genistein and anovulation at the highest dose.21 A similar dose-dependent pattern was also observed in the pituitary; the lowest genistein dose was associated with an increased pituitary response to GnRH and higher LH secretion whereas higher genistein doses were associated with decreased pituitary response and lower LH secretion.20 Consistent with the ovarian and pituitary changes, neonatal genistein exposure was shown to impair adult GnRH activation in the hypothalamus.19 Two epigenetic studies have provided additional evidence of the long-term effects of environmentally relevant levels of genistein during the early postnatal period.40, 41 Both studies demonstrated that genistein reprogrammed rodent uterine gene expression, or induced epigenetic changes that persisted in adulthood; 37, 38 one of the studies showed that the changes in uterine gene expression increased uterine tissue sensitivity to estrogen.40
One prior human study evaluated soy formula feeding in relation to a broad range of reproductive and endocrine-related outcomes in adulthood.23 That study was conducted among young adults who participated as infants in a formula feeding study and were assigned to either soy or cow's milk formula.23 With regard to menstrual bleeding, the investigators reported that soy formula feeding was associated with slightly longer bleeding duration but not with heavy menstrual flow. However, no information was provided about how menstrual bleeding was assessed and whether participants on medications that may alter menstrual flow were included. Despite these omissions, the strength of that study was its detailed records from infancy on formula feeding exposure.
Unlike that study, we were limited by relying on retrospective report of soy formula feeding for participants ages 23-34 years. Nevertheless, nearly 90% of participants had the assistance of their mothers when completing the early life questionnaire. Based on limited studies conducted around the time the SELF cohort was born (years 1975-1989),42, 43 mothers would be expected to be good sources of information regarding infant feeding. One small study conducted in a suburb of Detroit, Michigan (where SELF is based) suggested that mothers, as opposed to healthcare providers, frequently initiate the decision to change formula.43 A second study reported that mothers who changed to a special formula, such as soy formula, most frequently did so for the symptoms of colic and excessive crying.42 Hence, the recall of mothers in SELF with regard to the specific feeding of soy formula may have been aided by their role in the feeding decision-making process and the historic reasons for soy formula use that may be memorable. Despite the potential for exposure misclassification given the reliance on maternal recall over 23-34 years, the similarity in frequency of any exposure to soy formula in the present analyses (13%) to that estimated using data on formula feeding from the 1980s (∼11%)44 suggests that the error in reporting of exposure may have been minimal. In addition, the results of our probabilistic bias analysis suggest that exposure misclassification may not account for our observed results; assuming the distributions of bias parameters that we used were realistic, correction for exposure misclassification resulted in relative risks that were stronger in magnitude and had comparable precision to those of the conventional analyses.
In SELF, participants ages 23-34 years retrospectively recalled information on menstrual bleeding outcomes. The use of recalled information may have resulted in outcome misclassification; however, an earlier study of menstrual bleeding patterns reported high accuracy in recall of bleeding duration.45 In addition, retrospective recall allowed us to obtain information on several indicators of heavy menstrual bleeding that would have otherwise been unavailable. Indicators of heavy menstrual bleeding experienced over the participants' reproductive lifetime, such as ever having heavy, gushing type menstrual bleeding and ever use of NSAIDs for heavy menstrual bleeding, are not routinely recorded in medical records.
The availability of lifetime indicators of heavy menstrual bleeding strengthened our study as it allowed us to investigate the relationship between infant soy formula and heavy menstrual bleeding among the entire study sample. This is important as the entire study sample included participants who are affected by heavy menstrual bleeding but, due to pharmacologic treatment for the condition, were not included in the analyses of the subset of participants with natural menses; these treated participants are likely to be the most severely affected by heavy menstrual bleeding.
Our study was also strengthened by the collection of data on the impact of menstrual bleeding on the participant's quality of daily life. This assessment is in alignment with current clinical guidelines that recommend that heavy menstrual bleeding be determined by the woman's perception of bleeding that is excessive and its interference with physical, emotional, social, and financial quality of life. 1, 11, 46
The magnitude of the associations we observed were small and require confirmation in additional studies. Nevertheless, given that heavy menstrual bleeding is common, even small relative risks can translate to a relatively large impact on heavy menstrual bleeding occurrence.47
Additional research is warranted to investigate the role of early life exposures on heavy menstrual bleeding in adulthood. The impact of heavy menstrual bleeding on daily life can be substantial due to the extraordinary strategies needed to contain blood loss and prevent leaking through pads and tampons and staining clothing, bedding and furniture;4, 5 some women with heavy menstrual bleeding report scheduling plans based on proximity and access to bathrooms, stopping daily activities, avoiding social activities, or missing days of work.3-5 Similar to that reported in other studies of reproductive-age women,2-5 our sample of young African American women reported a high frequency of heavy menstrual bleeding as well as the adverse impacts of menstrual bleeding on the quality of daily life.
Conclusion
In our study of young African American women, our data suggested that soy formula feeding during infancy is associated with an increased risk of heavy menstrual bleeding, but replication studies are needed. Given that heavy menstrual bleeding is common, even the small increase in risk we observed could translate to an impact on a large number of women. Our results add to the data on infancy as a susceptible developmental window for female reproductive function.
Supplementary Material
Acknowledgments
We thank Drs. Matthew Longnecker and Anne Marie Jukic for their comments on this manuscript.
Source of Funding: This research was supported by the Intramural Research Program of the National Institutes of Health, National Institute of Environmental Health Sciences (1Z1A ES049013-10) and in part by funds allocated for health research by the American Recovery and Reinvestment Act.
Footnotes
Conflicts of Interest: The authors declare no conflict of interest.
References
- 1.National Institute for Health and Clinical Excellence. Heavy Menstrual Bleeding: Clinical Guideline. London: RCOG Press; Jan, 2007. 2007. [Google Scholar]
- 2.Marsh EE, Brocks ME, Ghant MS, Recht HS, Simon M. Prevalence and knowledge of heavy menstrual bleeding among African American women. Int J Gynaecol Obstet. 2014;125:56–9. doi: 10.1016/j.ijgo.2013.09.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Santer M, Wyke S, Warner P. Women's management of menstrual symptoms: findings from a postal survey and qualitative interviews. Soc Sci Med. 2008;66:276–88. doi: 10.1016/j.socscimed.2007.08.018. [DOI] [PubMed] [Google Scholar]
- 4.Matteson KA, Clark MA. Questioning our questions: do frequently asked questions adequately cover the aspects of women's lives most affected by abnormal uterine bleeding? Opinions of women with abnormal uterine bleeding participating in focus group discussions. Women Health. 2010;50:195–211. doi: 10.1080/03630241003705037. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Karlsson TS, Marions LB, Edlund MG. Heavy menstrual bleeding significantly affects quality of life. Acta Obstet Gynecol Scand. 2014;93:52–7. doi: 10.1111/aogs.12292. [DOI] [PubMed] [Google Scholar]
- 6.Cote I, Jacobs P, Cumming D. Work loss associated with increased menstrual loss in the United States. Obstet Gynecol. 2002;100:683–7. doi: 10.1016/s0029-7844(02)02094-x. [DOI] [PubMed] [Google Scholar]
- 7.Whiteman MK, Kuklina E, Jamieson DJ, Hillis SD, Marchbanks PA. Inpatient hospitalization for gynecologic disorders in the United States. Am J Obstet Gynecol. 2010;202:541 e1–6. doi: 10.1016/j.ajog.2009.12.013. [DOI] [PubMed] [Google Scholar]
- 8.Critchley HO, Maybin JA. Molecular and cellular causes of abnormal uterine bleeding of endometrial origin. Seminars in reproductive medicine. 2011;29:400–9. doi: 10.1055/s-0031-1287664. [DOI] [PubMed] [Google Scholar]
- 9.Maybin J, Critchley H. Repair and regeneration of the human endometrium. Expert Review of Obstetrics & Gynecology. 2009;4:283–98. [Google Scholar]
- 10.Maybin JA, Critchley HO. Steroid regulation of menstrual bleeding and endometrial repair. Reviews in endocrine & metabolic disorders. 2012;13:253–63. doi: 10.1007/s11154-012-9228-2. [DOI] [PubMed] [Google Scholar]
- 11.Munro MG. Classification of menstrual bleeding disorders. Reviews in endocrine & metabolic disorders. 2012;13:225–34. doi: 10.1007/s11154-012-9220-x. [DOI] [PubMed] [Google Scholar]
- 12.Kuiri-Hanninen T, Haanpaa M, Turpeinen U, et al. Postnatal ovarian activation has effects in estrogen target tissues in infant girls. J Clin Endocrinol Metab. 2013;98:4709–16. doi: 10.1210/jc.2013-1677. [DOI] [PubMed] [Google Scholar]
- 13.Kuiri-Hanninen T, Kallio S, Seuri R, et al. Postnatal developmental changes in the pituitary-ovarian axis in preterm and term infant girls. J Clin Endocrinol Metab. 2011;96:3432–9. doi: 10.1210/jc.2011-1502. [DOI] [PubMed] [Google Scholar]
- 14.Valdes-Dapena MA. Chapter 3. The development of the uterus in late fetal life, infancy, and childhood. In: Norris HJ, Hertig AT, Abell MR, editors. The Uterus. International Academy of Pathology. Monographs in pathology, no. 14. Baltimore: Williams & Wilkins; 1973. pp. 40–67. [Google Scholar]
- 15.Winter JS, Faiman C, Hobson WC, Prasad AV, Reyes FI. Pituitary-gonadal relations in infancy. I. Patterns of serum gonadotropin concentrations from birth to four years of age in man and chimpanzee. J Clin Endocrinol Metab. 1975;40:545–51. doi: 10.1210/jcem-40-4-545. [DOI] [PubMed] [Google Scholar]
- 16.Winter JS, Hughes IA, Reyes FI, Faiman C. Pituitary-gonadal relations in infancy: 2. Patterns of serum gonadal steroid concentrations in man from birth to two years of age. J Clin Endocrinol Metab. 1976;42:679–86. doi: 10.1210/jcem-42-4-679. [DOI] [PubMed] [Google Scholar]
- 17.Woods HF. Committee on Toxicity of Chemicals in Food, Consumer products and the environment: Phytoestrogens and Health. London, UK: Food Standards Agency; 2003. [Google Scholar]
- 18.McCarver G, Bhatia J, Chambers C, et al. NTP-CERHR expert panel report on the developmental toxicity of soy infant formula. Birth defects research Part B, Developmental and reproductive toxicology. 2011;92:421–68. doi: 10.1002/bdrb.20314. [DOI] [PubMed] [Google Scholar]
- 19.Bateman HL, Patisaul HB. Disrupted female reproductive physiology following neonatal exposure to phytoestrogens or estrogen specific ligands is associated with decreased GnRH activation and kisspeptin fiber density in the hypothalamus. Neurotoxicology. 2008;29:988–97. doi: 10.1016/j.neuro.2008.06.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Faber KA, Hughes CL., Jr Dose-response characteristics of neonatal exposure to genistein on pituitary responsiveness to gonadotropin releasing hormone and volume of the sexually dimorphic nucleus of the preoptic area (SDN-POA) in postpubertal castrated female rats. Reprod Toxicol. 1993;7:35–9. doi: 10.1016/0890-6238(93)90007-t. [DOI] [PubMed] [Google Scholar]
- 21.Jefferson WN, Padilla-Banks E, Newbold RR. Adverse effects on female development and reproduction in CD-1 mice following neonatal exposure to the phytoestrogen genistein at environmentally relevant doses. Biol Reprod. 2005;73:798–806. doi: 10.1095/biolreprod.105.041277. [DOI] [PubMed] [Google Scholar]
- 22.Nagao T, Yoshimura S, Saito Y, Nakagomi M, Usumi K, Ono H. Reproductive effects in male and female rats of neonatal exposure to genistein. Reprod Toxicol. 2001;15:399–411. doi: 10.1016/s0890-6238(01)00141-1. [DOI] [PubMed] [Google Scholar]
- 23.Strom BL, Schinnar R, Ziegler EE, et al. Exposure to soy-based formula in infancy and endocrinological and reproductive outcomes in young adulthood. JAMA. 2001;286:807–14. doi: 10.1001/jama.286.7.807. [DOI] [PubMed] [Google Scholar]
- 24.Moore KR, Smith JS, Laughlin-Tommaso SK, Baird DD. Cervical neoplasia-related factors and decreased prevalence of uterine fibroids among a cohort of African American women. Fertil Steril. 2014;101:208–14. doi: 10.1016/j.fertnstert.2013.09.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Wyatt KM, Dimmock PW, Walker TJ, O'Brien PM. Determination of total menstrual blood loss. Fertil Steril. 2001;76:125–31. doi: 10.1016/s0015-0282(01)01847-7. [DOI] [PubMed] [Google Scholar]
- 26.Final Report National Institute of Health. Evidence-based Methodology Workshop on Polycystic Ovary Syndrome. 2012 Dec 3-5; Executive Summary. [Google Scholar]
- 27.Wegienka G, Baird DD, Hertz-Picciotto I, Harlow SD, Hartmann KE. Uterine leiomyomata (fibroids): are bleeding symptoms more likely to be reported after diagnosis? J Clin Epidemiol. 2004;57:318–20. doi: 10.1016/j.jclinepi.2003.09.004. [DOI] [PubMed] [Google Scholar]
- 28.Wegienka G, Baird DD, Hertz-Picciotto I, et al. Self-reported heavy bleeding associated with uterine leiomyomata. Obstet Gynecol. 2003;101:431–7. doi: 10.1016/s0029-7844(02)03121-6. [DOI] [PubMed] [Google Scholar]
- 29.Blizzard L, Hosmer DW. The log multinomial regression model for nominal outcomes with more than two attributes. Biometrical journal Biometrische Zeitschrift. 2007;49:889–902. doi: 10.1002/bimj.200610377. [DOI] [PubMed] [Google Scholar]
- 30.Janssen CA, Scholten PC, Heintz AP. Menorrhagia--a search for epidemiological risk markers. Maturitas. 1997;28:19–25. doi: 10.1016/s0378-5122(97)00065-0. [DOI] [PubMed] [Google Scholar]
- 31.Fox MP, Lash TL, Greenland S. A method to automate probabilistic sensitivity analyses of misclassified binary variables. Int J Epidemiol. 2005;34:1370–6. doi: 10.1093/ije/dyi184. [DOI] [PubMed] [Google Scholar]
- 32.Lash TL, Fox MP, Fink AK. In: Applying Quantitative Bias Analysis to Epidemiologic Data. Gail M, Krickeberg K, Samet J, Tsiatis A, Wong WS, editors. New York, NY: Springer Science+Business Media, LLC; 2009. [Google Scholar]
- 33.Orsini N, Bellocco R, Bottai M, Wolk A, Greenland S. A tool for deterministic and probabilistic sensitivity analysis of epidemiologic studies. The Stata Journal. 2008;8:29–48. [Google Scholar]
- 34.Jefferson WN, Patisaul HB, Williams CJ. Reproductive consequences of developmental phytoestrogen exposure. Reproduction. 2012;143:247–60. doi: 10.1530/REP-11-0369. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kuiper GG, Carlsson B, Grandien K, et al. Comparison of the ligand binding specificity and transcript tissue distribution of estrogen receptors alpha and beta. Endocrinology. 1997;138:863–70. doi: 10.1210/endo.138.3.4979. [DOI] [PubMed] [Google Scholar]
- 36.Kuiper GG, Lemmen JG, Carlsson B, et al. Interaction of estrogenic chemicals and phytoestrogens with estrogen receptor beta. Endocrinology. 1998;139:4252–63. doi: 10.1210/endo.139.10.6216. [DOI] [PubMed] [Google Scholar]
- 37.Mueller SO, Kling M, Arifin Firzani P, et al. Activation of estrogen receptor alpha and ERbeta by 4-methylbenzylidene-camphor in human and rat cells: comparison with phyto- and xenoestrogens. Toxicol Lett. 2003;142:89–101. doi: 10.1016/s0378-4274(03)00016-x. [DOI] [PubMed] [Google Scholar]
- 38.Irvine CH, Shand N, Fitzpatrick MG, Alexander SL. Daily intake and urinary excretion of genistein and daidzein by infants fed soy- or dairy-based infant formulas. Am J Clin Nutr. 1998;68:1462S–5S. doi: 10.1093/ajcn/68.6.1462S. [DOI] [PubMed] [Google Scholar]
- 39.Adgent MA, Umbach D, Zemel B, et al. Abstracts of the 2014 Conference of the International Society of Environmental Epidemiology (ISEE) Abstract number 2578. Research Triangle Park, NC: Environmental Health Perspectives; 2014. Soy-based infant feeding is associated with estrogenized urogenital epithelium in girls at 24 weeks of age. http://dx.doi.org/10.1289/ehp.isee2014. [Google Scholar]
- 40.Greathouse KL, Bredfeldt T, Everitt JI, et al. Environmental estrogens differentially engage the histone methyltransferase EZH2 to increase risk of uterine tumorigenesis. Molecular cancer research: MCR. 2012;10:546–57. doi: 10.1158/1541-7786.MCR-11-0605. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Tang WY, Newbold R, Mardilovich K, et al. Persistent hypomethylation in the promoter of nucleosomal binding protein 1 (Nsbp1) correlates with overexpression of Nsbp1 in mouse uteri neonatally exposed to diethylstilbestrol or genistein. Endocrinology. 2008;149:5922–31. doi: 10.1210/en.2008-0682. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Forsyth BW, McCarthy PL, Leventhal JM. Problems of early infancy, formula changes, and mothers' beliefs about their infants. J Pediatr. 1985;106:1012–7. doi: 10.1016/s0022-3476(85)80260-2. [DOI] [PubMed] [Google Scholar]
- 43.Polack FP, Khan N, Maisels MJ. Changing partners: the dance of infant formula changes. Clin Pediatr (Phila) 1999;38:703–8. doi: 10.1177/000992289903801202. [DOI] [PubMed] [Google Scholar]
- 44.Fomon SJ. Reflections on infant feeding in the 1970s and 1980s. Am J Clin Nutr. 1987;46:171–82. doi: 10.1093/ajcn/46.1.171. [DOI] [PubMed] [Google Scholar]
- 45.World Health Organization Task Force on Psychosocial Research in Family Planning SPoR, Development and Research Training in Human Reproduction. Women's Bleeding Patterns: Ability to Recall and Predict Menstrual Events. Stud Fam Plann. 1981;12:17–27. [PubMed] [Google Scholar]
- 46.American College of Obstetricians and Gynecologists. Clinical Management Guidelines for Obstretrician-Gynecologists. Practice bulletin no. 128: Diagnosis of abnormal uterine bleeding in reproductive-aged women. Obstet Gynecol. 2012;120:197–206. doi: 10.1097/AOG.0b013e318262e320. [DOI] [PubMed] [Google Scholar]
- 47.Koepsell TD, Weiss NS. Epidemiologic methods: studying the occurrence of illness. New York, NY: Oxford University Press; 2003. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.