## Abstract

Background Fertility drugs stimulate hyperovulation, which may have implications for breast cancer. We examined the association between use of fertility drugs (clomiphene citrate [CC] and follicle-stimulating hormone [FSH]) and subsequent risk of young-onset (<50 years at diagnosis) breast cancer.

Methods We conducted the Two Sister Study, a sister-matched case–control study, by enrolling 1422 women between September 2008 and December 2010, who were younger than age 50 years at diagnosis with breast cancer and were enrolled within 4 years of diagnosis, and 1669 breast cancer–free control sisters from the Sister Study. Participants reported their use of fertility drugs (CC and FSH) and ever-users reported whether a pregnancy had resulted that lasted 10 or more (10+) weeks. Conditional logistic regression was used to estimate confounder-adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for fertility drug use with or without conception of a 10+ week pregnancy.

Results A total of 288 participants reported having used ovulation-stimulating drugs (193 CC only, 29 FSH only, and 66 both). Overall, women who had used fertility drugs showed a non-statistically significantly decreased  risk of breast cancer, compared with nonusers (OR = 0.82, 95% CI = 0.63 to 1.08). Women who had used fertility drugs but had not conceived a 10+ week pregnancy under treatment showed a statistically significantly  decreased risk of breast cancer compared with nonusers (OR = 0.62, 95% CI = 0.43 to 0.89). Women who had used fertility drugs and conceived a 10+ week pregnancy under treatment showed a statistically significantly increased risk of breast cancer compared with unsuccessfully treated women (OR = 1.82, 95% CI = 1.10 to 3.00), although their risk was not increased compared with women who had not used fertility drugs (OR = 1.13, 95% CI = 0.78  to 1.64).

Conclusions In the absence of a 10+ week pregnancy under treatment, exposure to ovulation-stimulating fertility drugs was associated with reduced risk of young-onset breast cancer. This apparent association was absent in women who conceived a 10+ week pregnancy under treatment, for whom risk was higher than that of unsuccessfully treated women, but similar to that of untreated women.

Treatment with ovulation-stimulating fertility drugs causes recruitment and maturation of multiple ovarian follicles, temporarily elevating estrogen; in women treated with clomiphene citrate (CC), the peak serum levels of estrogen are two- to threefold higher than normal (average: 200 pg/mL at peak) (3,4), and peak levels are even higher in women who undergo in vitro fertilization (IVF) treatment, which typically involves the use of follicle- stimulating hormone (FSH)-containing drugs (5,6). If pregnancy results, the supernumerary ovarian corpora lutea produce both estrogen and progesterone and raise the levels far above normal in the first 9 weeks (7,8). It is estimated that 6% of babies born in the United States in 2005 were conceived with ovulation-stimulating drugs (9).

The widespread use of ovulation-stimulating fertility drugs has raised concern about possible implications for breast cancer. Some, but not all, studies report increased risk following infertility treatment (10–31). A recent meta-analysis was inconclusive, citing methodological limitations of previous studies, which included infrequent use of fertility drugs, unspecified treatments, incomplete control for confounding, and small numbers of case subjects (32).

Breast cancers in women younger than age 50 years are rare but can be aggressive and carry a worse prognosis compared with those in older women (33), and distinct risk factors are associated with such young-onset disease. We conducted a sister-controlled study and examined whether use of ovulation-stimulating fertility drugs is associated with risk of breast cancer in women younger than age 50 years. To examine the possible consequences of the high hormonal exposures experienced in early pregnancy by women who conceive with ovarian hyperstimulation, we separated fertility-drug exposures according  to whether or not that treatment had resulted in a pregnancy lasting at least 10 weeks.

## Methods

### Study Design

The Two Sister Study is a family-based retrospective study of environmental and genetic factors related to young-onset breast cancer, which was developed from the Sister Study (http://www.sisterstudy.org/2Sisters_English/2Sisters.htm). The Sister Study is an ongoing prospective cohort study of more than 50 000 women aged 35–74 years, enrolled between 2004 and 2009, who had a sister diagnosed with breast cancer but had not been diagnosed themselves at the time of enrollment (http://www.sisterstudy.org). This study was undertaken to study environmental and genetic risk factors for breast cancer, using a cohort of women known to be at increased risk. We identified Sister Study participants whose sister had been diagnosed within 4 years and had been younger than age 50 years at diagnosis (case sisters), and asked them to forward to their affected sister our invitation to enroll in the Two Sister Study. Enrollment was accomplished between September 2008 and December 2010.

To allow for treatment time, case sisters were not contacted until at least a year after diagnosis. Each enrolled case sister completed the same two-part computer-assisted telephone interview (CATI) on demographics, lifestyle, reproductive factors, health conditions, and medications as their control sister(s) had when they enrolled in the Sister Study. In addition, case sisters completed a CATI on breast cancer diagnosis, tumor characteristics, and treatment,  providing contact information for their medical providers, and authorizing release of cancer-related medical records.

Of the 3283 families identified as potentially eligible (ie, the Sister Study participant had reported she had a sister diagnosed with breast cancer at age younger than 50 years and had been diagnosed within 4 years), 403 Sister Study participants did not respond. Of the remaining 2880 participants, 91 case sisters were deceased, and 77 were determined to be ineligible (mostly because the control sister had misreported the case sister’s age at diagnosis or the 4-year limit had been exceeded by the time we made contact). Of the remaining 2712 families, 2492 Sister Study participants agreed to send their sister our letter of invitation between 2008 and 2010. Of those, 411 young-onset case sisters did not respond, 63 declined to participate, and the remaining 2018 expressed interest. However, 264 of those were then found to be ineligible (mostly because the time since diagnosis was >4 years). Of the remainder, 1422 of 1754 (81.1%) case sisters who completed all telephone interviews were included along with their 1669 control sisters, after excluding 20 control sisters who had been more than 7 years younger than their case sister’s age at diagnosis when they completed their interviews, and for whom there was at least one alternate control sister available.

Medical records of cancer diagnosis and treatment were obtained for 1245 of 1422 (88%) case sisters. Agreement between CATI self-report and medical records was excellent. The positive predictive value of self-report was 99.5% for invasive cancer, 98.8% for estrogen receptor–positive (ER+) cancer, 98.9% for progesterone receptor–positive (PR+) cancer, and 98.8% for ductal carcinoma. Therefore, for subset analyses restricted to families with specific tumor characteristics, self-report was substituted when medical records were unavailable.

Both the Sister Study and the Two Sister Study were approved by the institutional review boards of the National Institute of Environmental Health Sciences, National Institutes of Health, and the Copernicus Group. Written informed consent was provided by all but 119 case sisters who completed just the telephone interview and provided verbal consent.

### Exposure Assessment

Before the interview, participants were mailed memory aids including visual cue cards, lists of relevant medications, and a chronological life calendar to record landmark events, such as births or major surgeries. Women were asked in the CATI whether they had ever sought medical help to become pregnant. Women who reported ever taking fertility medications provided the medication names, when they first took them, the number of menstrual cycles of use, and whether any  treatment had resulted in a pregnancy lasting 10 or more (10+) weeks.

Medications were coded using the Slone Drug Dictionary (34), with linking to active ingredients and American Hospital Formulary Service (AHFS) drug classes using Pharmacologic–Therapeutic Classification codes from AHFS. We considered two major types of ovulation-inducing fertility drugs: CC and gonadotropins (eg, FSH), identified by class codes 68:16:12:00 and 68:18:02:00, respectively (Supplementary Table 1, available online). A multicomponent fertility drug, Pergonal, which lacks a class code and is often used in IVF treatment protocols, was included as “FSH” because it contains FSH.

### Statistical Analysis

In this sister-matched case–control study, case sisters were younger than age 50 years at diagnosis, whereas the matched control sisters may have been older when they enrolled in the Sister Study. To ensure comparable opportunity for exposure in within-sibship comparisons, we assessed time-varying variables with reference to an “index age,” which was defined for each set of sisters as the smallest of the numbers reported for the age of the case sister at diagnosis and the age(s) of her control sister(s) at their completion of the CATIs. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using conditional logistic regression to account for sibling matching.

Fertility-drug use was categorized as nonuser, CC only, FSH only, or both CC and FSH. Because biological consequences  of exposure may differ for users of ovulation-stimulating drugs who do vs who do not conceive a 10+ week pregnancy with treatment, we conceptualized seven exposure categories (nonuser, CC only [conceived vs did not conceive with treatment], FSH only [conceived vs did not conceive with treatment], and both CC and FSH [conceived vs did not conceive with treatment]) based on full cross-classification. We compared the model that included all seven unordered categories (using six dummy variables) with a reduced model that included the four categories for ovulation-stimulating drugs together with a dummy variable for occurrence of a resulting stimulated pregnancy (three dummy variables for drugs and one for stimulated pregnancy). The latter model fit as well and was adopted as a joint exposure model (Model I). We further collapsed the four drug-exposure categories to aggregate into just two (users vs nonusers) and this aggregation caused very little loss of fit. The final parsimonious model thus included just users vs nonusers of ovulation-stimulating drugs (Model II).

The exposure parsed in this way has some unusual features: a stimulated pregnancy by definition can only happen in women with exposure to ovulation-stimulating drugs; consequently, the relative risk for women with a stimulated pregnancy compared with women without fertility-drug exposure is estimated as the exponentiated sum of the coefficient for drug use plus the coefficient for stimulated pregnancy (ie, as the product of the two odds ratios, one for her drug exposure and one for her pregnancy, given drug exposure). In addition, the exponentiated coefficient for stimulated pregnancy is interpretable as estimating the odds ratio for the  pregnancy exposure relative to women with exposure to fertility drugs who did not conceive.

In further exploratory analyses, subcategories were considered based on age at first use (<35 vs ≥35 years), and whether the first treatment came before vs after the first birth. Stimulated pregnancies were also classified according to whether a stimulated pregnancy was their first birth.

We used directed acyclic graphs (35) to identify potential confounders. The following were considered: relative birth order among participating sisters, education, household income per person, age at menarche (<12, 12–13, ≥14 years), age at first birth (<25, 25–29, 30–34, ≥35 years, or nulliparous), parity (0, 1, 2, ≥3 children), infertility (had ever been at risk of pregnancy for 12 months without conceiving or had ever sought help to become pregnant), menopausal status (premenopause, postmenopause, and premenopausal hysterectomy with retained ovarian tissue) at index age, duration of breast-feeding (treated as a continuous variable), body mass index (BMI) at ages 30–39 years (<18.5, 18.5–24.9, 25.0–29.9, ≥30.0 kg/m2), hormonal birth control history (nonuser, user for <10 years, user for ≥10 years, and user with unknown duration), smoking (nonsmoker, <1, 1–9.9, ≥10 pack-years) and average alcohol consumption during the previous 10 years (nondrinker, <13, 13–<48, 48–<180, ≥180 drinks/year). Age at first birth (<25, 25–<30, 30–<35, ≥35 years), parity (0, 1, 2, ≥3 children), and duration of breastfeeding were cumulated to the exact index age (in days). Other time-varying variables were cumulated up through index age minus one (in years). Age at first birth and menopausal status at index age were included in all models. We adjusted for relative birth order among participating sisters to account for design-induced differences between case sisters and control sisters. Other potential confounders (including age at menarche, self-reported BMI at ages 30–39 years, and infertility) had negligible impact on the estimates and were not included in the final models. Subset analyses were also performed according to tumor characteristics (invasive, ER+, and ductal carcinoma; there were more cancers with missing PR status than missing ER status [67 vs 26], and the negative predictive value for self-reported PR status was lower than that for ER, so PR status was not used in the analyses).

All statistical analyses were carried out using the SAS software, PHREG procedure, version 9.2 (SAS Institute Inc, Cary, NC). Maximum likelihood methods were used to estimate the odds ratios and P values are based on likelihood ratio tests. All statistical tests were two-sided and P values less than .05 were considered statistically significant.

## Results

The mean age of control sisters was 47.7 years (standard deviation [SD] = 6.2 years) at enrollment, and the mean age of case sisters was 44.7 years (SD = 4.0 years) at diagnosis. The youngest age at diagnosis was 28 years. Of the 1422 case sisters, 621 (43.7%) were diagnosed before age 45 years, and 203 (14.3%) were diagnosed before age 40 years. A total of 1185 (84.0%) case sisters had invasive tumors (n = 12 missing data), 1095 (78.4%) had ER+ tumors (n = 26 missing data), 962 (71.0%) had PR+ tumors (n = 67 missing data), and 1229 (87.8%) had ductal tumors (n = 23 missing data).

Most (90%) of the families were non-Hispanic white and most participants (90%) had completed some college (Table 1). Compared with control sisters, case sisters reported a statistically significantly (P < .05) younger age at menarche, older age at first birth, were more likely to report having taken hormonal birth control pills for more than 10 years, were less likely to have had low BMI at age 30–39 years, and less likely to have experienced menopause before the index age. There were no substantial differences in education, parity, breastfeeding, smoking, or alcohol drinking.

Table 1.

Characteristics of participant sisters at index age in the  Two Sister Study*

Characteristic Control sisters  (n = 1669) Case sisters  (n = 1422)
Race, No. (%)
Non-Hispanic white 1487 (89.1) 1253 (88.1)
Black 75 (4.5) 71 (5.0)
Hispanic 63 (3.8) 57 (4.0)
Other 44 (2.6) 41 (2.9)
Relative birth order among  participating sisters, No. (%)†
First (oldest) 915 (54.8) 530 (37.3)
Second 610 (36.5) 792 (55.7)
Third or younger 144 (8.6) 100 (7.0)
Education, No. (%)
High school or less 217 (13.0) 177 (12.4)
Some college but no degree 280 (16.8) 210 (14.8)
Associate or technical degree 253 (15.2) 200 (14.1)
Bachelor degree 525 (31.5) 480 (33.8)
Master or doctoral degree 394 (23.6) 355 (25.0)
Household income per person  in US dollars‡, mean (SD)  $40 179 (25 998)$38 780 (24 883)
Age at menarche‡, No. (%)
<12 y 276 (16.5) 268 (18.8)
12–<14 y 956 (57.3) 833 (58.6)
≥14 y 436 (26.1) 321 (22.6)
Parity, No. (%)
0 child 358 (21.5) 304 (21.4)
1 child 259 (15.5) 227 (16.0)
2 children 618 (37.1) 554 (39.0)
≥3 children 433 (26.0) 337 (23.7)
Age at first birth among parous  women§, No. (%)
<25 y 501(30.0) 346 (24.3)
25–<30 y 465 (27.9) 423 (29.7)
30–<35 y 242 (14.5) 255 (17.9)
≥35 y 102 (6.1) 94 (6.6)
Total duration of breastfeeding  in weeks, mean (SD) 46.3 (70.8) 44.5 (64.4)
Body mass index at ages 30–39 y‡,  No. (%)
<18.5 kg/m2 53 (3.2) 33 (2.3)
18.5–24.9 kg/m2 1157 (69.5) 1002 (70.8)
25.0–29.9 kg/m2 312 (18.7) 283 (20.0)
≥30.0 kg/m2 143 (8.6) 98 (6.9)
Use of hormonal birth control‡,  No. (%)
Nonuser 163 (9.8) 135 (9.5)
<10 y 889 (53.4) 700 (49.3)
≥10 y 590 (35.4) 568 (40.0)
Unknown duration 24 (1.4) 17 (1.2)
Cigarette smoking‡, No. (%)
Never-smoker 1044 (62.6) 859 (60.4)
<1 pack-year 148 (8.9) 157 (11.0)
1–<10 pack-years 235 (14.1) 217 (15.3)
≥10 pack-years 240 (14.4) 189 (13.3)
Alcohol drinking in the 10 years  preceding index age‡, No. (%)
Nondrinker 143 (8.7) 151 (10.7)
<13 drinks/y 388 (23.7) 313 (22.1)
13–<48 drinks/y 306 (18.7) 260 (18.4)
48–<180 drinks/y 425 (25.9) 373 (26.4)
≥180 drinks/y 378 (23.0) 318 (22.5)
Menopausal status‡, No. (%)†
Premenopausal 1396 (83.7) 1269 (89.2)
Postmenopausal 147 (8.8) 76 (5.3)
Premenopausal hysterectomy,  with retained ovarian tissue 124 (7.4) 77 (5.4)
Characteristic Control sisters  (n = 1669) Case sisters  (n = 1422)
Race, No. (%)
Non-Hispanic white 1487 (89.1) 1253 (88.1)
Black 75 (4.5) 71 (5.0)
Hispanic 63 (3.8) 57 (4.0)
Other 44 (2.6) 41 (2.9)
Relative birth order among  participating sisters, No. (%)†
First (oldest) 915 (54.8) 530 (37.3)
Second 610 (36.5) 792 (55.7)
Third or younger 144 (8.6) 100 (7.0)
Education, No. (%)
High school or less 217 (13.0) 177 (12.4)
Some college but no degree 280 (16.8) 210 (14.8)
Associate or technical degree 253 (15.2) 200 (14.1)
Bachelor degree 525 (31.5) 480 (33.8)
Master or doctoral degree 394 (23.6) 355 (25.0)
Household income per person  in US dollars‡, mean (SD)  $40 179 (25 998)$38 780 (24 883)
Age at menarche‡, No. (%)
<12 y 276 (16.5) 268 (18.8)
12–<14 y 956 (57.3) 833 (58.6)
≥14 y 436 (26.1) 321 (22.6)
Parity, No. (%)
0 child 358 (21.5) 304 (21.4)
1 child 259 (15.5) 227 (16.0)
2 children 618 (37.1) 554 (39.0)
≥3 children 433 (26.0) 337 (23.7)
Age at first birth among parous  women§, No. (%)
<25 y 501(30.0) 346 (24.3)
25–<30 y 465 (27.9) 423 (29.7)
30–<35 y 242 (14.5) 255 (17.9)
≥35 y 102 (6.1) 94 (6.6)
Total duration of breastfeeding  in weeks, mean (SD) 46.3 (70.8) 44.5 (64.4)
Body mass index at ages 30–39 y‡,  No. (%)
<18.5 kg/m2 53 (3.2) 33 (2.3)
18.5–24.9 kg/m2 1157 (69.5) 1002 (70.8)
25.0–29.9 kg/m2 312 (18.7) 283 (20.0)
≥30.0 kg/m2 143 (8.6) 98 (6.9)
Use of hormonal birth control‡,  No. (%)
Nonuser 163 (9.8) 135 (9.5)
<10 y 889 (53.4) 700 (49.3)
≥10 y 590 (35.4) 568 (40.0)
Unknown duration 24 (1.4) 17 (1.2)
Cigarette smoking‡, No. (%)
Never-smoker 1044 (62.6) 859 (60.4)
<1 pack-year 148 (8.9) 157 (11.0)
1–<10 pack-years 235 (14.1) 217 (15.3)
≥10 pack-years 240 (14.4) 189 (13.3)
Alcohol drinking in the 10 years  preceding index age‡, No. (%)
Nondrinker 143 (8.7) 151 (10.7)
<13 drinks/y 388 (23.7) 313 (22.1)
13–<48 drinks/y 306 (18.7) 260 (18.4)
48–<180 drinks/y 425 (25.9) 373 (26.4)
≥180 drinks/y 378 (23.0) 318 (22.5)
Menopausal status‡, No. (%)†
Premenopausal 1396 (83.7) 1269 (89.2)
Postmenopausal 147 (8.8) 76 (5.3)
Premenopausal hysterectomy,  with retained ovarian tissue 124 (7.4) 77 (5.4)

* Index age was defined as the smallest of the numbers reported for the age at diagnosis for the case sister and the age(s) at interview of her control sister(s), to ensure equal opportunity for exposures. All time-varying variables (parity, age at first birth among parous women, total duration of breastfeeding, use of hormonal birth control, cigarette smoking, alcohol drinking, and menopausal status) were defined with reference to the index age, except household income, which was assessed at the interview.

P < .01; calculated using two-sided likelihood ratio tests based on conditional logistic regression.

‡ Missing data: household income per person (46 control sisters and 35 case sisters), age at menarche (one control sister), parity (one control sister), duration of breastfeeding (three control sisters), body mass index at age 30 (four control sisters and six case sisters), hormonal birth control (three control sisters and two case sisters), smoking (two control sisters), alcohol drinking (29 control sisters and seven case sisters), and menopausal status (two control sisters).

§ P < .05; calculated using two-sided likelihood ratio tests based on conditional logistic regression.

Women who reported use of CC alone (n = 193 women) or both CC and FSH (n = 66 women) showed a non-statistically significantly reduced odds of young-onset breast cancer compared with nonusers of these drugs (Table 2). However, after adjusting for exposure to a 10+ week stimulated pregnancy, compared with nonusers, women who had taken CC only or both CC and FSH, but without success (ie, without a pregnancy that lasted ≥10 weeks), were less likely to be diagnosed with young-onset breast cancer (CC only, OR = 0.61, 95% CI = 0.41 to 0.90; CC and FSH, OR = 0.53; 95% CI = 0.29 to 0.96; Table 2). Use of FSH only (n = 29 women: 17 case sisters and 12 control sisters) without success was not associated with risk (OR = 1.03, 95% CI = 0.44 to 2.40), but the number of exclusive users of FSH was small. When the fit of the model (Model I) that treated the fertility-drug-use histories as four categories (nonusers, users of CC only, FSH only, or both CC and FSH) was compared with that of a simpler model where the ovulation-stimulating drugs were aggregated to form just two categories (exposed or not), as in Model II (Table 2), the loss of goodness of fit was not statistically significant (P > .30), indicating that the effects of CC and FSH were not distinguishable in our data. Women with a history of use of fertility drugs (aggregated) showed a non-statistically significantly decreased risk compared with women who were nonusers (OR = 0.82, 95% CI = 0.63 to 1.08). However, as shown in Table 2, when the occurrence of pregnancy was included in the model, women who used fertility drugs, but had not conceived a 10+ week pregnancy under treatment, showed a statistically significantly decreased risk of breast cancer compared with nonusers (OR = 0.62, 95% CI = 0.43 to 0.89). Women who used fertility drugs and had conceived a 10+ week pregnancy under treatment showed a statistically significantly increased risk of breast cancer compared with other users (unsuccessfully treated; OR = 1.82, 95% CI = 1.10 to 3.00; Table 2). Note that in this model, the comparison for those treated but without a consequent 10+ week pregnancy is with women without a history of use of ovulation-stimulating drugs, whereas the comparison for treated women with a stimulated  10+ week pregnancy is with other (unsuccessfully treated) women who also used ovulation-stimulating drugs. Overall, women who received treatment and conceived under treatment did not  have a statistically significantly increased risk of breast cancer compared with untreated women (OR = 0.62 × 1.82 = 1.13, 95% CI = 0.78 to 1.64).

Table 2.

Associations of young-onset breast cancer with fertility drugs and stimulated pregnancies*

Variable Control sisters  (n = 1669) Case sisters (n = 1422) Adjusted  OR (95%  CI)† Adjusted OR (95% CI)‡
No. (%) No. (%)
Model I
Nonusers of fertility drugs 1511 (90.5) 1292 (90.9) 1.00 (referent) —
CC only 107 (6.4) 86 (6.0) 0.80 (0.58 to 1.09) 0.61 (0.41 to 0.90)
FSH only 12 (0.7) 17 (1.2) 1.40 (0.63 to 3.12) 1.03 (0.44 to 2.40)
CC and FSH 39 (2.3) 27 (1.9) 0.73 (0.43 to 1.24) 0.53 (0.29 to 0.96)
Stimulated pregnancy 69 (4.1) 72 (5.1) — 1.82 (1.10 to 3.02)
Model II
Nonusers of fertility drugs 1511 (90.5) 1292 (90.9) 1.00 (referent) —
Use of fertility drug(s) 158 (9.5) 130 (9.1) 0.82 (0.63 to 1.08) 0.62 (0.43 to 0.89)
Stimulated pregnancy 69 (4.1) 72 (5.1) — 1.82 (1.10 to 3.00)
Variable Control sisters  (n = 1669) Case sisters (n = 1422) Adjusted  OR (95%  CI)† Adjusted OR (95% CI)‡
No. (%) No. (%)
Model I
Nonusers of fertility drugs 1511 (90.5) 1292 (90.9) 1.00 (referent) —
CC only 107 (6.4) 86 (6.0) 0.80 (0.58 to 1.09) 0.61 (0.41 to 0.90)
FSH only 12 (0.7) 17 (1.2) 1.40 (0.63 to 3.12) 1.03 (0.44 to 2.40)
CC and FSH 39 (2.3) 27 (1.9) 0.73 (0.43 to 1.24) 0.53 (0.29 to 0.96)
Stimulated pregnancy 69 (4.1) 72 (5.1) — 1.82 (1.10 to 3.02)
Model II
Nonusers of fertility drugs 1511 (90.5) 1292 (90.9) 1.00 (referent) —
Use of fertility drug(s) 158 (9.5) 130 (9.1) 0.82 (0.63 to 1.08) 0.62 (0.43 to 0.89)
Stimulated pregnancy 69 (4.1) 72 (5.1) — 1.82 (1.10 to 3.00)

* ”Young-onset” refers to case sisters diagnosed with breast cancer at age less than 50 years. Model II is like Model I, except that the drug categories (use of CC only, use of FSH only, and use of both CC and FSH) have been aggregated. Model II fits as well based on a non-statistically significant loss of goodness of fit (P > .30). OR = odds ratio; CI = confidence interval; CC = clomiphene citrate; FSH = follicle-stimulating hormone; — = not applicable.

† Adjusted for relative birth order among included sisters, age at first birth (with nulliparous status treated as a separate category), and menopausal status at index age.

‡ Adjusted for relative birth order among included sisters, age at first birth (with nulliparous status treated as a separate category), and menopausal status at index age; stimulated pregnancy was simultaneously included in the model. Note that we do not show a referent here because the nonusers are the referent category for effects of drug use with no resulting 10+ week pregnancy, whereas the referent for effects of a 10+ week stimulated pregnancy is users of ovulation-stimulating drugs who did not have a 10+ week pregnancy.

We conducted further analyses to explore the factors (menopausal status at index age, whether or not the first use preceded the first birth, whether or not their first pregnancy was a stimulated pregnancy, age at first use, ER status of the cancer, and whether or not it was invasive) that might modify the associations of fertility-drug use and stimulated pregnancy with risk of breast cancer, and to explore evidence for a dose–response relationship with the number of treated cycles or the number of stimulated 10+ week pregnancies. There were no statistically significant findings (data not shown), but there was a suggestion of a stronger association with stimulated pregnancy if it had been their first birth (OR = 2.04, 95% CI = 1.16 to 3.58),  especially if the cancer was invasive and ER+ (OR = 2.69, 95%  CI = 1.32 to 5.49). The class of fertility drug did not appear to modify the association between stimulated pregnancy and risk (P = .86).  In exploratory dose–response analyses, the number of stimulated pregnancies was not statistically significantly related to risk (P = .13), nor was the number of treated menstrual cycles (P = .88).

When analyses were restricted to ER+ invasive breast cancer  (n = 907 case sisters), the results were similar (Table 3). Stratification by combined ER and PR status or restriction to ductal carcinoma did  not substantially modify the associations between fertility drugs and  stimulated pregnancies with risk of breast cancer (data not shown).

Table 3.

Associations of invasive estrogen receptor–positive  young-onset breast cancer with fertility drugs and stimulated pregnancies*

Variable Control sisters  (n = 1067) Case  sisters  (n = 907) Adjusted  OR (95% CI)† Adjusted OR (95% CI)‡
No. (%) No. (%)
Model I
No fertility-drug use 968 (90.7) 816 (90.0) 1.00 (referent) —
CC only 67 (6.3) 58 (6.4) 0.88 (0.60 to 1.31) 0.61 (0.37 to 1.01)
FSH only 7 (0.7) 11 (1.2) 1.71 (0.60 to 4.89) 1.20 (0.40 to 3.62)
CC and FSH 25 (2.3) 22 (2.4) 0.99 (0.54 to 1.81) 0.65 (0.32 to 1.32)
Stimulated pregnancy 39 (3.7) 51 (5.6) — 2.23 (1.19, to 4.16)
Model II
No fertility-drug use 968 (90.7) 816 (90.0) 1.00 (referent) —
Use of fertility drug 99 (9.3) 91 (10.0) 0.96 (0.68 to 1.34) 0.66 (0.42 to 1.04)
Stimulated pregnancy 39 (3.7) 51 (5.6) — 2.23 (1.20 to 4.14)
Variable Control sisters  (n = 1067) Case  sisters  (n = 907) Adjusted  OR (95% CI)† Adjusted OR (95% CI)‡
No. (%) No. (%)
Model I
No fertility-drug use 968 (90.7) 816 (90.0) 1.00 (referent) —
CC only 67 (6.3) 58 (6.4) 0.88 (0.60 to 1.31) 0.61 (0.37 to 1.01)
FSH only 7 (0.7) 11 (1.2) 1.71 (0.60 to 4.89) 1.20 (0.40 to 3.62)
CC and FSH 25 (2.3) 22 (2.4) 0.99 (0.54 to 1.81) 0.65 (0.32 to 1.32)
Stimulated pregnancy 39 (3.7) 51 (5.6) — 2.23 (1.19, to 4.16)
Model II
No fertility-drug use 968 (90.7) 816 (90.0) 1.00 (referent) —
Use of fertility drug 99 (9.3) 91 (10.0) 0.96 (0.68 to 1.34) 0.66 (0.42 to 1.04)
Stimulated pregnancy 39 (3.7) 51 (5.6) — 2.23 (1.20 to 4.14)

* “Young-onset” refers to case sisters diagnosed with breast cancer at age less than 50 years. Model II is like Model I, except that the drug categories (use of CC only, use of FSH only, and use of both CC and FSH) have been aggregated. Model II fits as well based on a non-statistically significant loss of goodness of fit (P > .30). OR = odds ratio; CI = confidence interval; CC = clomiphene citrate; FSH = follicle-stimulating hormone; — = not applicable.

† Adjusted for relative birth order among included sisters, age at first birth (with nulliparous status treated as a separate category), and menopausal status at index age.

‡ Adjusted for relative birth order among included sisters, age at first birth (with nulliparous status treated as a separate category), and menopausal status at index age; stimulated pregnancy was simultaneously included in the model. Note that we do not show a referent here because the unexposed group is the referent category for effects of drug use with no resulting 10+ week pregnancy, whereas the referent for effects of a 10+ week stimulated pregnancy is the group with exposure to ovulation-stimulating drugs who did not have a 10+ week pregnancy.

Age at first use showed no statistically significant modification of the associations across categories of fertility drugs. Although there was some evidence for elevated risk in women who had used both CC and FSH, with first use after age 35 years (Supplementary Table 2, available online), the numbers in this category were too small for reliable inference. The patterns of association were similar for women whose first use of these drugs preceded their first birth compared with those whose first use came later (data not shown; Pinteraction = .49).

## Discussion

In this sister-matched case–control study, the Two Sister Study, we analytically differentiated between users of ovulation-stimulating drugs who conceived vs those who did not conceive under treatment, to account for the fact that there are known hormonal effects of ovarian stimulation in the first trimester of pregnancy. We found statistically significantly reduced risk of young-onset breast cancer in women with a history of unsuccessful use of ovulation-stimulating fertility drugs compared with nonusers. Women who had used fertility drugs and had conceived a 10+ week pregnancy under treatment were at statistically significantly increased risk of young-onset breast cancer compared with unsuccessfully treated women, but had risk similar to that of nonusers.

The most widely used fertility drug, CC, is classified as a selective estrogen-receptor modulator (SERM) and acts as an estrogen antagonist in the hypothalamus by binding to the ERs and blocking the effects of estrogen (36). Given its mode of action and its short duration of use, we doubt that unsuccessful cycles of CC use (usually fewer than six) would plausibly pharmacologically protect against breast cancer. However, a history of CC use could serve as a marker for historically lower average endogenous estrogen levels in women with categories of infertility for which CC is prescribed. CC is often the preferred first treatment for patients with polycystic ovarian syndrome or suspected ovulatory dysfunction.

Others have reported apparent protection (ie, association with reduced risk) in women who took CC. In the Nurse’s Health Study II, women who had undergone CC treatment for ovulatory infertility had reduced risk of breast cancer (15). Some studies have found increased risk in specified subgroups, such as CC-treated infertile women with nonovulatory disorders (19) and CC users more than 20 years after treatment (31).

We were unable to distinguish the associations of CC with risk of breast cancer from those of FSH, which is the primary ovulation stimulator used in IVF protocols, but we had a few participants who had only been exposed to FSH and not CC. However, the point estimate for unsuccessful use of FSH alone was 1.03, suggesting that conception under treatment with FSH could confer increased risk compared with untreated women (OR = 1.03 × 1.82 = 1.87). Consistent with our data, several studies have reported increased risk of breast cancer in FSH users (26,30). IVF (using FSH) is attempted for a diverse array of diagnoses, including tubal occlusion and male factor infertility. However, for women with either ovulatory dysfunction or infertility of uncertain cause but with patent oviducts, FSH is probably prescribed only after CC has failed. Thus, the small category of women with history of FSH use but no CC use would likely comprise women with fertility problems for which CC was not tried because it could not succeed, such as tubal occlusion, and specific indications for nonuse of CC might explain a lack of protective association in these women.

No previous study simultaneously accounted for stimulated pregnancy and use of fertility drugs, instead effectively aggregating two very different exposure histories. Our data suggest that exposure to a stimulated pregnancy is enough to undo the reduction in risk associated with a history of exposure to ovulation-stimulating drugs. In young women, increased breast cell differentiation during pregnancy can lead to a long-term protective effect. Ovulation-stimulating treatment causes abnormally high exposure to ovarian hormones during the first 9 weeks of pregnancy (7,8) and could potentially raise risk by modifying breast tissue remodeling (37).

Fourteen of 23 earlier studies followed up infertile women or women who sought infertility treatment in clinics (13,14,16–22,25,26,28,29,31), of which one study (comparing women with IVF-based deliveries to other women delivering during the same interval in Sweden) found a statistically significant reduced risk (25) and others found no association or increased risk. Two studies (11,23) recruited participants who had given birth. The Jerusalem Perinatal Study reported a borderline statistically significantly increased risk of breast cancer among CC users compared with women who conceived spontaneously (11). A study of Swedish women with live births found no statistically significant association with IVF, but the follow-up period was short (23). In both designs, treated women with stimulated pregnancy would be overrepresented.

Our sister-based design provided several advantages and  opportunities. Sisters tend to be well matched for social factors, health care–seeking behaviors, and other unmeasured potential confounders. We were able to identify and enroll a large sample of young-onset case sisters by approaching their already-enrolled unaffected sister. This strategy produced a large sample of highly motivated participants.

This study has a few limitations. We relied on self-reported fertility-drug use, and recall bias may be present. We enhanced recall by providing lists of medication names in the interview materials. In addition, infertility treatment is a major life event, and likely to be remembered. A second limitation is that we do not have data on the specific diagnosis for infertility. However, reliable information on diagnosis is hard to obtain, because infertility workups are complex, time-consuming, often inconclusive, and frequently cut short by conception. Patients may be managed empirically (without a definitive diagnosis) with CC, which is widely regarded as inexpensive, safe, noninvasive, and effective. Another limitation of our design is that case sisters were on average younger than control sisters. By considering exposures up to the same index age for both case sisters and control sisters, we ensured that sisters being compared had similar opportunities for exposure. Nevertheless, by adopting this strategy, we had to neglect exposures that may have occurred between the index age and the age at diagnosis (for case sisters) or the age at interview (for control sisters). However, only one control sister (and no case sister) was first exposed to fertility drugs during that interval. Finally, we delayed contact for at least a year after diagnosis to allow for treatment and recovery, and some case sisters with more aggressive cancers may have died before we could contact them. Although only 91 case sisters were excluded because the control sister reported they had died, some of the other nonresponders may have been deceased or too sick, and consequently some of our findings may not apply to particularly aggressive cancers. The use of sister controls, however, confers internal validity through case–control comparisons.

In conclusion, in this sister-based case–control study, a history of unsuccessful use of ovulation-stimulating fertility drugs was associated with a statistically significantly reduced risk of young-onset breast cancer, but the occurrence of a stimulated 10+ week pregnancy appeared to offset the protective association. Use of fertility drugs, particularly CC, may have pharmacological effects (38) and could (more plausibly) serve as a marker for a different hormonal milieu in women with categories of infertility for which CC is used. For those who achieve conception through treatment, the elevated production of ovarian hormones early in a stimulated pregnancy might increase the risk of breast cancer by modifying pregnancy-related remodeling of breast tissue (37).

## Funding

Intramural Research Program of the National Institutes of Health; National Institute of Environmental Health Sciences (Z01-ES044005 [CRW] and Z01-ES102245 [DPS]). Additional funding: Susan G. Komen for the Cure (FAS0703856 to CRW).

## Notes

We thank Dr Donna Baird and Dr Allen Wilcox of the National Institute of Environmental Health Sciences for helpful comments on the paper.

Aside from receiving yearly progress reports and their original decision to approve the project, Susan G. Komen for the Cure had no role, and the authors are responsible for the design and conduct of this study; collection, analysis, and interpretation of the data; and preparation of this article.

Affiliations of authors: Biostatistics Branch (CF, CRW) and Epidemiology Branch (LAD, DPS), National Institute of Environmental Health Sciences, Research Triangle Park, NC.

## References

1.
Rossouw
JE
Anderson
GL
Prentice
RL
et al
Risks and benefits of  estrogen plus progestin in healthy postmenopausal women: principal results from the women’s health initiative randomized controlled trial.
JAMA.

2002
;
288
3
):
321
333
2.
Kelsey
JL
Gammon
MD
John
EM
Reproductive factors and breast  cancer
Epidemiol Rev.

1993
;
15
1
):
36
47
3.
Wu
HH
Wang
NM
Cheng
ML
Hsieh
JN
A randomized comparison  of ovulation induction and hormone profile between the aromatase  inhibitor anastrozole and clomiphene citrate in women with infertility.
Gynecol Endocrinol.

2007
;
23
2
):
76
81
4.
A
Elnashar
A
Totongy
M
Clomiphene citrate or aromatase inhibitors for superovulation in women with unexplained infertility undergoing intrauterine insemination: a prospective randomized trial.
Fertil Steril.

2009
;
92
4
):
1355
1359
5.
Barrenetxea
G
Agirregoikoa
JA
Jimenez
MR
de Larruzea
AL
Ganzabal
T
Carbonero
K
Ovarian response and pregnancy outcome in poor-responder women: a randomized controlled trial on the effect of luteinizing hormone supplementation on in vitro fertilization cycles.
Fertil Steril.

2008
;
89
3
):
546
553
6.
Lass
A
Monitoring of in vitro fertilization-embryo transfer cycles by ultrasound versus by ultrasound and hormonal levels: a prospective, multicenter, randomized study.
Fertil Steril.

2003
;
80
1
):
80
85
7.
Dickey
RP
Hower
JF
Effect of ovulation induction on uterine blood flow and oestradiol and progesterone concentrations in early pregnancy.
Hum Reprod.

1995
;
10
11
):
2875
2879
8.
Hassiakos
D
Mantzavinos
T
Kalomiris
K
Zourlas
PA
Comparison of maternal serum estradiol and progesterone levels in pregnancies after induced and spontaneous ovulation.
Arch Gynecol Obstet.

1991
;
248
3
):
145
150
9.
Schieve
LA
Devine
O
Boyle
CA
Petrini
JR
Warner
L
Estimation of the contribution of non-assisted reproductive technology ovulation stimulation fertility treatments to US singleton and multiple births.
Am J Epidemiol.

2009
;
170
11
):
1396
1407
10.
Ricci
E
Parazzini
F
Negri
E
Marsico
S
La Vecchia
C
Fertility drugs and the risk of breast cancer.
Hum Reprod.

1999
;
14
6
):
1653
1655
11.
Calderon-Margalit
R
Friedlander
Y
Yanetz
R
et al
Cancer risk after exposure to treatments for ovulation induction.
Am J Epidemiol.

2009
;
169
3
):
365
375
12.
Braga
C
Negri
E
La Vecchia
C
Parazzini
F
Dal Maso
L
Franceschi
S
Fertility treatment and risk of breast cancer.
Hum Reprod.

1996
;
11
2
):
300
303
13.
Venn
A
Watson
L
Bruinsma
F
Giles
G
Healy
D
Risk of cancer after use of fertility drugs with in-vitro fertilisation.
Lancet.

1999
;
354
9190
):
1586
1590
14.
Venn
A
Watson
L
Lumley
J
Giles
G
King
C
Healy
D
Breast and ovarian cancer incidence after infertility and in vitro fertilisation.
Lancet.

1995
;
346
8981
):
995
1000
15.
Terry
KL
Willett
WC
Rich-Edwards
JW
Michels
KB
A prospective study of infertility due to ovulatory disorders, ovulation induction, and incidence of breast cancer.
Arch Intern Med.

2006
;
166
22
):
2484
2489
16.
Rossing
MA
Daling
JR
Weiss
NS
Moore
DE
Self
SG
Risk of breast cancer in a cohort of infertile women.
Gynecol Oncol.

1996
;
60
1
):
3
7
17.
Potashnik
G
Lerner-Geva
L
Genkin
L
Chetrit
A
Lunenfeld
E
Porath
A
Fertility drugs and the risk of breast and ovarian cancers: results of a long-term follow-up study.
Fertil Steril.

1999
;
71
5
):
853
859
18.
Pappo
I
Lerner-Geva
L
Halevy
A
et al
The possible association between IVF and breast cancer incidence.
Ann Surg Oncol.

2008
;
15
4
):
1048
1055
19.
Orgeas
CC
Sanner
K
Hall
P
et al
Breast cancer incidence after hormonal infertility treatment in Sweden: a cohort study.
Am J Obstet Gynecol.

2009
;
200
1
):
72.e1
72.e7
20.
Modan
B
Ron
E
Lerner-Geva
L
et al
Cancer incidence in a cohort of infertile women.
Am J Epidemiol.

1998
;
147
11
):
1038
1042
21.
Lerner-Geva
L
Keinan-Boker
L
Blumstein
T
et al
Infertility, ovulation induction treatments and the incidence of breast cancer—a historical prospective cohort of Israeli women.
Breast Cancer Res Treat.

2006
;
100
2
):
201
212
22.
Lerner-Geva
L
Geva
E
Lessing
JB
Chetrit
A
Modan
B
Amit
A
The possible association between in vitro fertilization treatments and cancer development.
Int J Gynecol Cancer.

2003
;
13
1
):
23
27
23.
Kristiansson
P
Bjor
O
Wramsby
H
Tumour incidence in Swedish women who gave birth following IVF treatment.
Hum Reprod.

2007
;
22
2
):
421
426
24.
Kotsopoulos
J
Librach
CL
Lubinski
J
et al
Infertility, treatment of infertility, and the risk of breast cancer among women with BRCA1 and BRCA2 mutations: a case-control study.
Cancer Causes Control.

2008
;
19
10
):
1111
1119
25.
Kallen
B
Finnstrom
O
Lindam
A
Nilsson
E
Nygren
KG
Olausson
PO
Malignancies among women who gave birth after in vitro fertilization.
Hum Reprod.

2011
;
26
1
):
253
258
26.
Jensen
A
Sharif
H
Svare
EI
Frederiksen
K
Kjaer
SK
Risk of breast cancer after exposure to fertility drugs: results from a large Danish cohort study.
Cancer Epidemiol Biomarkers Prev.

2007
;
16
7
):
1400
1407
27.
Gauthier
E
Paoletti
X
Clavel-Chapelon
F
E3N Group
Breast cancer risk associated with being treated for infertility: results from the French E3N cohort study.
Hum Reprod.

2004
;
19
10
):
2216
2221
28.
Doyle
P
Maconochie
N
Beral
V
Swerdlow
AJ
Tan
SL
Cancer incidence following treatment for infertility at a clinic in the UK.
Hum Reprod.

2002
;
17
8
):
2209
2213
29.
Dor
J
Lerner-Geva
L
Rabinovici
J
et al
Cancer incidence in a cohort of infertile women who underwent in vitro fertilization.
Fertil Steril.

2002
;
77
2
):
324
327
30.
Burkman
RT
Tang
MT
Malone
KE
et al
Infertility drugs and the risk of breast cancer: findings from the National Institute of Child Health and Human Development Women’s Contraceptive and Reproductive Experiences Study.
Fertil Steril.

2003
;
79
4
):
844
851
31.
Brinton
LA
Scoccia
B
Moghissi
KS
et al
Breast cancer risk associated with ovulation-stimulating drugs.
Hum Reprod.

2004
;
19
9
):
2005
2013
32.
Zreik
TG
Mazloom
A
Chen
Y
et al
Fertility drugs and the risk of breast cancer: a meta-analysis and review.
Breast Cancer Res Treat.

2010
;
124
1
):
13
26
33.
Axelrod
D
Smith
J
Kornreich
D
et al
Breast cancer in young women.
J Am Coll Surg.

2008
;
206
3
):
1193
1203
34.
Kelley
KE
Kelley
TP
Kaufman
DW
Mitchell
AA
The Slone Drug Dictionary: a research driven pharmacoepidemiology tool.
Pharmacoepidemiol Drug Saf.

2003
;
12
suppl 1
):
S168
S169
35.
Glymour
M
Greenland
S
Causal diagrams
. In:
Rothman
K
Greenland
S
Lash
T
, eds.
Modern Epidemiology.

3rd ed
:
Lippincott Williams & Wilkins
;
2008
:
183
209
36.
Shelly
W
Draper
MW
Krishnan
V
Wong
M
Jaffe
RB
Selective estrogen receptor modulators: an update on recent clinical findings.
Obstet Gynecol Surv.

2008
;
63
3
):
163
181
37.
Polyak
K
Pregnancy and breast cancer: the other side of the coin.
Cancer Cell.

2006
;
9
3
):
151
153
38.
Homburg
R
Clomiphene citrate—end of an era? A mini-review.
Hum Reprod.

2005
;
20
8
):
2043
2051