Cancer Risk in Women Treated with Fertility Drugs According to Parity Status-A Registry-based Cohort Study.

Pregnancy is known to protect against ovarian ( 1 ), breast( 2 ) and endometrial( 3 ) cancers, and nulliparity is consequently an established risk factor for these cancers. Furthermore, some studies have suggested that older ages at first birth may relate to increased risk of cutaneous malignant melanoma (CMM)( 4 ) and increasing parity to an elevated risk of thyroid cancer ( 5 ). A continuing expansion of the use of assisted reproductive techniques (ART) means that growing numbers of women are exposed to a variety of fertility drugs ( 6 ), and monitoring the safety of these relatively new drugs is of importance. Some studies have found associations between fertility drug use and risks of ovarian ( 7 , 8 ) breast ( 9 – 11 ) and other cancers ( 12 , 13 ), whilst others have not ( 14 – 17 ), including two meta-analyses ( 18 , 19 ). With reproductive factors being modifiers of cancer risk at several sites, a question that remains is whether effects of fertility drugs are different among nulliparous and parous women. Only a limited number of studies have been able to perform analyses stratified by parity ( 8 , 20 , 21 ), and even fewer are able to look separately at risks in women who remain nulliparous after treatment ( 22 ). We previously examined cancer risk associated with ART in parous women in Norway and found elevated risks of breast ( 11 ) and central nervous system cancers ( 23 ). We attempted in the present study to expand on our previous studies by examining risks in both parous and nulliparous women, and by analyzing exposure to both ART and clomiphene citrate. The novelty of this study is that we are able to present results for nulliparous women alone, to assess whether these women harbor an especially high risk of cancer that parous women. The aim of the study was to compare cancer risk in nulliparous women exposed to fertility drugs to nulliparous women not treated with fertility drugs. By using data from four nationwide registries we were able to establish a nation-wide cohort of considerable size. For comparison, analyses on parous women were also included. The study assessed all-site cancer risk, and the risk of breast, cervical, endometrial, ovarian, colorectal (CRC), central nervous system (CNS) thyroid cancers and cutaneous malignant melanoma (CMM).

In the National Registry 1 470 476 women were registered as born between 1960 and 1996, of which 1 353 724 (92%) women were eligible for study ( figure 1 ). The total follow-up time was 12 354 392 person-years, median 11 years ( table 1 ). Apart from region of residence (485 , <0.1% missing), no other variables had any missing values. A total of 598 983 (44%) women were classified as nulliparous, of which 14 645 (2.4%) had received fertility treatment. The corresponding number of parous women with fertility treatment was 41 549 (5.5%, figure 1 ). Of those receiving fertility drugs, 33 431 received treatment with ART and 38 927 with CC. Median age at entry was 27 years for nulliparous women with fertility treatment, and 18 years for nulliparous women without ( table 1 ). Nulliparous women with cancer were younger at diagnosis (median 40 years and 37 years for those without and with fertility treatment respectively) compared to parous women (median 43 and 38 years). Of the total cohort, 20 128 women were registered with at least one cancer diagnosis, with 920 (4.6%) of these occurring in exposed women ( table 1 ). The risk of all-site cancer in ART exposed compared to unexposed women was 1.10 (95% CI 0.98, 1.23). For nulliparous women the HR was 1.00 (95% CI 0.81, 1.24), compared to 1.14 (95% CI 1.00, 1.29) among parous women ( table 2 ). ART was not associated with an elevated risk of breast cancer, either in nulliparous (1.11 (95% CI 0.75, 1.66)) nor parous women (0.96 (95% CI 0.76, 1.22)). ART-women (both parous and nulliparous) appeared to have a lower risk of cervical cancer although neither risk was statistically significant. Risk of endometrial cancer was slightly but not statistically significantly elevated in parous women exposed to ART (1.62 (95% CI 0.70, 3.85)). No elevation was observed among nulliparous women (0.39 (95% CI 0.15, 1.03)). Women exposed to ART did not have a significantly elevated risk of ovarian cancer, (1.29,95% CI 0.73, 2.28) compared to unexposed. For nulliparous women risk was slightly higher, although not statistically significant, (1.62 (95% CI 0.78, 3.35)). For parous women alone, no risk elevation was observed (0.87 (95% CI 0.33, 2.27)). A p-value of 0.05 indicates a borderline-significant difference in risk between nulliparous and parous women for ART and ovarian cancer. Risk of borderline ovarian tumors was elevated for all ART exposed women, (1.95 (95% CI 1.18, 3.23)). The stratified analyses on parity, showed that there was no significant difference in risk between nulliparous (1.69 (95% CI 0.75, 3.79)) and parous women (2.12 (95% CI 1.11, 4.04))(p= 0.9). No differences in risk were observed with increasing number of cycles of ART ( supplementary table 2 ). The risk of thyroid cancer was elevated for all women exposed to ART compared to non-ART women (1.53 (95% CI 1.01, 2.31)), with significant risks in nulliparous women (2.19 (95% CI 1.08, 4.44)) and non-significant risk in parous women (1.31 (95% CI 0.78, 2.19)). ART treatment was not associated with elevated risk of CRC, CNS tumors or CMM in exposed women, regardless of parity and dose ( supplementary table 2 ). CC exposure was associated with an elevated risk of all-site cancer, (1.14 (95% CI 1.03, 1.26)), and the risk estimates were similar for parous and nulliparous women ( table 3 ). CC exposure was associated with increased risk of breast cancer, in parous women (1.26 (95% CI 1.03, 1.54)) (p = 0.02), but no dose response relationship for CC and breast cancer ( supplementary table 3 ). For CC exposed women the risk of cervical cancer was the same as in unexposed women, although slightly but not statistically significantly lower in the parous group (0.83 (95% CI 0.62, 1.18)). The risk of endometrial cancer was elevated in women treated with CC, (2.91 (95% CI 1.87, 4.53) and risk was highest for nulliparous women (4.49 (95% CI 2.66, 7.60), (p = 0.04), table 3 ), and among parous women with more than 6 cycles, (4.68 (95% CI 1.74, 12.6)), p-value for trend analysis was 0.011. CC exposed nulliparous women had increased risk of ovarian cancer (2.49 (95% CI 1.30, 4.78)), while risk was not increased in parous women (1.37 (95% CI 0.64, 2.96) (p = 0.04), table 3 ). The magnitude of the HRs appeared to increase with increasing doses of CC, 1.76 (95% CI 0.68, 4.58) at the lowest dose, vs 3.46 (95% CI 1.19, 10.0) with the highest dose, although a test for trend revealed a p-value of 0.269. CC exposure was not associated with risk of borderline tumors, thyroid cancers, CRC, CNS tumors or CMM. When stratifying on different histologic subtypes of ovarian cancer, CC exposure was associated with risk of endometrioid ovarian cancers (4.75 (95 % CI 1.95, 11.6); data not shown). No differences were seen for risk of neither serous nor mucinous tumors (data not shown). When stratifying on age above and below 30 and 40 years at start of follow-up, no differences were observed for ovarian, breast or endometrial cancer. When looking at time since diagnosis, no differences could be seen due to few cases of cancers in the exposed group (data not shown). When removing ovarian and thyroid cancers, from analyses of ART, the estimate for all site cancer was unchanged (1.08 95 % CI 0.96, 1.21)). Neither did removing ovarian and endometrial cancers, change estimates for CC exposure (1.12 (95 % CI 1.01, 1.25)) (data not shown).

This population based registry study is one of the largest to date to assess risk of cancer in women receiving fertility treatments. We observed elevated risk of ovarian and endometrial cancer, and risk appeared to be highest among nulliparous women following exposure to CC. An enhanced risk of thyroid cancer was observed for nulliparous women exposed to ART. Further, a modest increase in risk of breast cancer was observed with CC treatment, of similar magnitude as in our previous study on breast cancer risk after ART among parous women ( 11 ). Results demonstrate elevated risk of ovarian cancer after CC exposure, and suggest that also ART exposure may be associated with elevated risk, albeit among nulliparous women. For CC the risk among nulliparous women increased with increasing drug dosage. Fathalla suggested in 1971 that repeated involvement of the ovarian surface epithelium during ovulation could be related to development of ovarian neoplasms, coining the term “incessant ovulation” ( 34 ). Subsequent research has suggested that ovulatory pauses such as oral contraceptives (OC), pregnancy and lactation could reduce ovarian cancer risk ( 35 ). Our results indicate that an additional risk pertains to nulliparous women treated with fertility drugs, due not only to the lack of ovulatory pause associated with pregnancy, but also possibly to exposure to COH. One of the first studies looking at fertility drugs and ovarian cancer, also found that women who remained childless had a higher risk than women conceiving after fertility treatment ( 36 ). Later, two US based studies reported higher risks associated with CC in women remaining nulliparous ( 37 , 38 ). An Australian study also found non-significantly elevated risk of ovarian cancer in nulliparous IVF women ( 21 ). Two further studies, one from Sweden ( 8 ), and another from the Netherlands, both detected elevated risk of ovarian cancers after treatment with ART ( 7 ), but none provided separate estimates for nulliparous women. In our previous study, we found higher risk of ovarian cancer in women with primary infertility and those conceiving only one child ( 23 ). On the other hand, several other studies observe no increase in risk of ovarian cancers ( 39 – 41 ), and/or no difference in risk between nulliparous and parous women with fertility treatment ( 20 , 41 ). Our results suggest highest risk of ovarian cancer among those with the highest doses of CC. Although one earlier study also found an association between ovarian cancer and 12 or more cycles of CC ( 42 ), most other investigators observe no dose-response relationships with CC ( 16 ) nor ART ( 7 , 20 ) and ovarian cancer. Although our findings are noteworthy, it is important to keep in mind that women receiving multiple doses of fertility drugs may be a selected group of women. It may well be that women exposed to many treatment rounds suffer resistant infertility, for example, for women with polycystic ovarian syndrome (PCOS), only 20% become pregnant with each cycle of CC on average ( 43 , 44 ). These women require many cycles of CC, but may indeed have elevated risks of ovarian cancer due to their ovulation disorders, and not the treatment itself. Finally, it is important to mention that the dose response analyses are based on few cases, possibly representing a chance finding. We found elevated risk of borderline ovarian tumors in women treated with ART, in line with two previous studies ( 7 , 45 ), but in contrast to a further three ( 41 , 46 , 47 ). In our study, we could not observe difference in risk among nulliparous and parous women. This was in line with findings from an Australian study ( 45 ), but contrary to a Dutch study that found highest risk among nulliparous women ( 7 ). In Norway these non-malignant tumors are systematically registered in the Cancer Registry of Norway. Elevated risks of borderline tumors in women treated with fertility hormones have been suggested to reflect surveillance bias, and not a biological explanation, which may explain the absence of risk with increasing number of cycles of ART. In the current study we did not have sufficient data to evaluate potential surveillance bias in terms of time since diagnosis. We found elevated risk of endometrial cancers in women exposed to CC, highest among nulliparous women, and for those with more than 6 treatment cycles. In contrast to this, a recent meta-analysis consisting of six studies found no elevation in risk connected to neither fertility drugs nor ART ( 48 ). Notably, one of the studies in the meta-analysis ( 49 ) was unable to replicate their earlier findings ( 50 ) where they had demonstrated increased risk of endometrial cancer associated with CC for six or more cycles. It is worth mentioning that BMI and anovulatory infertility (PCOS) have been shown associated with endometrial cancer ( 51 ), and may cause confounding of our results, as this information is unavailable. Our study suggests thyroid cancer to be associated with ART treatment, with risks highest in the nulliparous group. Two other studies made similar findings, one detecting elevated risks among nulliparous women with use of CC ( 22 , 52 ), whereas another discovered increased risk among parous women ( 12 ). Thyroid tumors are more frequent in women than in men ( 53 ), giving reason to believe that female sex hormones may be involved. Ovarian stimulation has been shown to cause elevated levels of TSH ( 54 ) which promotes cellular proliferation in the gland. Both the normal thyroid gland and thyroid tumors exhibit estrogen receptors ( 55 ), although the exact mechanisms by which tumor growth is promoted are unclear ( 56 ). Thyroid cancer incidence has increased in recent years, possibly due to incidental findings of tumors with increased use of ultrasound, CT and MRI, however, when correcting for this in our analyses, risk was still elevated among ART women. We found an increased risk of breast cancer associated with CC treatment, but not with ART. The risk increase associated with CC use was restricted to parous women, and the magnitude of the estimate similar to our previous study of parous women ( 11 ). Although Brinton and colleagues reported an elevated risk of breast cancer in women exposed to multiple cycles (>12) of CC ( 15 ), we did not observe any relation of risk according to number of CC cycles. A recent meta-analysis concluded that the association between infertility treatment (any hormonal treatment) and risk of breast cancer was weak, but underlined that extensive use of CC should be limited due to concerning findings relating to breast cancer ( 19 ). No association was found between fertility treatment and risks of colorectal, CNS cancer nor CMM. Reassuringly, the elevated risk of CNS tumors found in our previous study on parous women ( 11 ) could not be replicated presently, possibly reflecting the shorter follow-up in the present study, with data from the Norwegian Prescription Databaseonly available from 2004. Two recent papers support our null-findings on fertility drugs and CMM ( 13 , 57 ). Two other studies conclude no association between use of CC and CRC ( 22 , 58 ). We observed a non-significant decrease in risk of cervical cancer among fertility treated women, in line with others studies ( 16 , 49 , 59 ), possibly due to infertile women’s regular gynecological examinations, including cervical screening tests leading to reduced risk of invasive cervical cancer ( 60 ). The main strength of this study is its size. By using population-based registry data all the way back to 1960, we are able to obtain a nationwide study cohort that may be the largest to date adressing fertility drugs and cancer risk. In this study, collection of information on drug exposure from the Norwegian Prescription Database minimized the risk of recall bias ( 61 , 62 ). Further, the Prescription Database only records prescribed drugs which are dispensed and collected by patients, reducing the risk of primary non-compliance ( 24 , 63 ) and subsequently risk of misclassification. Moreover, the registry-based collection of data on cancer from the Cancer Registry of Norway and childbirths from the Medical Birth Registry of Norway is advantageous as the mandatory reporting to both registries ensures high external validity and completeness ( 27 ). Another strength of the study is that we are able to look at both ART and CC as separate exposures. In contrast to several other studies data from Medical Birth Registry enabled us to separate nulliparous and parous women, accounting for the effects of childbearing on cancer risk. The study is at risk of some misclassification of exposure, since the Norwegian Prescription Databaseonly includes data from 2004. Thus, some women may be misclassified as unexposed if they only received fertility drugs before 2004. However, stratifying on age at inclusion did not reveal any differences in risk between women who were older compared to those that were younger in 2004. Misclassification of exposure may also occur if women receive fertility treatment abroad not recorded in the Norwegian Prescription Database ( 64 ). Using Prescription Database information does not provide the opportunity to assess the degree of adherence; although a drug is dispensed, the patient might not actually have taken it. Although, infertile patients seeking to conceive are likely to have high drug adherence ( 65 ). In the present study, comorbidity data are unavailable. Firstly, this is important as nulliparous women may be more likely to suffer from chronic diseases including cancer, which prevent them from having children. However, then risk elevations would likely be observed for several cancer sites, not just the specific ones we observed. Secondly, with respect to comorbidity, information on fertility diagnoses are unavailable to us in the present study. This is an issue, for example has endometriosis been suggested associated with elevated risk of ovarian cancer ( 21 , 66 – 68 ). In our study we are unable to disentangle the effects of the fertility treatment from underlying causes of infertility themselves. It may be that some women harbor pathological changes in the ovary leaving them prone to both infertility and oarian neoplasms. Thus, confounding by indication may be driving some of the observed associations between fertility drugs and ovarian cancer. This may also be the case for thyroid cancer as, it has been demonstrated that thyroxin substitution treatment is used more frequently by ART pregnant women, than by those pregnant after natural conception ( 69 ). It may therefore be that the preexisting thyroid disease may be a common cause of both infertility and thyroid neoplasms. Some factors associated with cancer and infertility were unavailable: BRCA mutations, socioeconomic factors, smoking and body mass index (BMI). BMI is an potential confounder associated with both infertility ( 70 ) and breast ( 71 ) and endometrial cancer ( 72 ), and may be the reason at least in part why we observe elevations in risk after CC exposure. Oral contraceptives (OC) are known to reduce risk of ovarian and endometrial cancers. If OC use is less in infertile than in fertile women, this factor may be mediating some of the observed effects of fertility drugs on ovarian cancer. Women treated with fertility drugs may be subject to some degree of surveillance bias, which could be a possible explanation for our findings of elevated risk of thyroid and borderline tumors. This could have been clarified by examining stage and mortality of cancers in exposed women in future studies. It is also important to note that this study includes women of relatively young ages, and as a consequence the follow-up time is short. Thus, the median ages at cancer diagnosis were below the ages where cancer is prevalent, (37 years for exposed nulliparous women and 38 years for exposed parous women), and for some site specific analyses, number of cases in the comparison groups are low. Another limitation is that correction for multiple analyses has not been performed. However, the need to do to may be subject for discussion, as there may be little correlation between breast cancer with for example malignant melanoma, both with respect to tumor biology but also differences in etiology. In this nationwide registry-based study, we used data on parity, drug exposure and cancer outcomes, to compare cancer risks associated with fertility drug exposures among parous and nulliparous women. Findings were reassuring for most cancers, although fertility treatment, particularly with CC, appeared to increase the risk of ovarian and endometrial cancer especially in nulliparous women. For some sites, including breast, endometrium and thyroid, there were some elevations in risk, although less consistently according to treatment type and parity. Although some risk elevations are observed, it must be kept in mind that the study population is young, and its absolute cancer risk low. Future research should continue to monitor women treated for infertility as they grow older. Assessing ovarian cancer risk in women remaining childless after treatment, and risks associated with cumulative doses of fertility drugs is of importance.

This population-based study includes all women born in Norway between 1960 and 1996 registered in the National Registry. Additional nation-wide registries provided data on dispensed fertility drugs [the Norwegian Prescription Database], pregnancies and births [the Medical Birth Registry of Norway], and cancer diagnoses [the Cancer Registry of Norway]. Since 2004, the Norwegian Prescription Database has included data on all prescribed drugs dispensed to individuals in ambulatory care ( 24 ). Drugs are classified according to the World Health Organization Anatomical Therapeutic Chemical (ATC) classification ( 25 ). The Medical Birth Registry of Norway was established in 1967 and contains information on all births in Norway, and is based on compulsory notification of every birth, from 16 completed weeks of gestation onwards ( 26 ). The Cancer Registry of Norway was established in 1952, and contains information on all persons diagnosed with cancer since 1953 ( 27 ). A unique personal identity number (PID) is assigned to each resident at birth or immigration, enabling data linkage across the registries. Reporting to the registries is mandatory and regulated by national laws. Exposure to ART: Controlled ovarian hyperstimulation (COH) is the process of using drugs to obtain several mature oocytes in a single menstrual cycle for use in in vitro fertilization (IVF). Hormone protocols used for COH in ART vary widely, but mostly the standard protocols include the following three medications: Gonadotropin releasing hormone (GnRH) analogues (agonists or antagonists), gonadotropins (Follicle stimulating hormone or human menopausal gonadotropin) and finally human chorionic gonadotropin (hCG) ( 28 ). Study subjects were considered as exposed to an ART treatment either when they had been prescribed either a combination of all three medications, or only the first two (GnRH analogues and other gonadotropins) within a two-month time period. Number of cycles of ART were categorized: 1, 2 and 3 or more ART cycles. Clomiphene citrate is a nonsteroidal ovarian stimulant used for ovulation induction since the 1960s. It binds to hypothalamic estrogen receptors and by negative feedback induces pulsatile GnRH secretion, increased gonadotrophin secretion and increased ovarian follicular activity ( 29 ). All women with at least one prescription of CC were considered as exposed to CC, all others were denoted non-CC women . Each treatment cycle consists of 50 mg for 5 consecutive days, and dose was categorized as ≤ 3 cycles, 3–6 cycles or < 6 cycles. The drugs and ATC codes included in the analyses are displayed in Supplementary Table 1 . Cancers diagnosed were categorized according to the International Classification of Diseases version 10 (ICD-10), (C00–96, up to 12 per individual). Analyses of all-site cancer risk considered the first cancer (at any site) and in site-specific analyses, the first case of the cancer of interest was used. Women diagnosed with a cancer before 2004 were excluded from the analyses. Analyses were conducted for the same sites as in our previous studies on parous women, namely breast (C50), cervical (invasive cancers only) (C53), endometrial (C54–55), ovarian (C56), colorectal (C18–20) CNS (C70–72) thyroid cancers (C73) and CMM (C43). Separate analyses were made for ovarian cancer subgroups, and for borderline ovarian tumors (see supplementary file for histology codes). We adjusted for region of residence because there may be regional differences in both use of fertility treatment and cancer incidence. We adjusted for birth year to account for potential cohort effects on cancer incidence. In the analyses of all women, adjustment was made for parity by splitting the data at the date of each woman’s first birth. When assessing ART exposure, adjustments were made for CC exposure (ever/never), and vice versa. Follow-up started on January 1 2004 for all women born between 1960 and 1985. Women who were born in 1986 or later, started follow-up on turning 18 years because receiving fertility treatment before this age was deemed unlikely. Women born before 1960 were not included as it was considered likely that they would be too old for fertility treatment during in the observational period 2004 to 2014. Follow-up ended upon diagnosis of the first cancer of interest, death, emigration or December 31 2014 (the latest update of the Cancer Registry of Norway). We used Cox regression models to compute hazard ratios (HR) and 95% confidence intervals (CI) for cancer risk in exposed versus unexposed women. Separate analyses were made for ART and CC exposure, and for parous and nulliparous women. Age was used as the timescale ( 30 ). Parity was treated as a time-dependent variable, with women switching classification from nulliparous to parous at the time of the first birth. Analyses were stratified by doses of ART and CC, with dose as a time-varying covariate.( 31 ) Stratified analyses were also made on age at follow-up (below and above 30 years), and age at inclusion (below and above 40 years). Sensitivity analyses were made excluding those receiving other hormones (such as progesterone, or monotherapy with GnRH analogues). Risk of all-site cancer was made after omitting any cancer sites with elevated risks. Testing for heterogeneity was done using a likelihood-ratio test to test any observed differences between groups (p-values below 0.05 were considered statistically significant). We attempted to analyze risk according to time since diagnosis to assess potential surveillance bias. The proportional hazards assumption was tested using the Schoenfeld residuals ( 32 ). Analyses were made using the STATA software package, version 14.0, and the STROBE guidelines for reporting observational studies were adhered to ( 33 ). The Regional Ethics Committee of the South Eastern Health Region and the Norwegian Data Inspectorate approved the study.