Medically Assisted Reproduction and Hormone-Related Cancers

This cohort study used a retrospective emulated target trial design and was conducted as part of a broader body of work on cancer after MAR in Australia. 25 The methods are detailed fully in the eMethods in Supplement 1 . The study was approved by all relevant human research ethics committees, including the Australian Institute of Health and Welfare human research ethics committees. Data were accessed and used under a waiver of informed consent, and researchers were granted access only to linked anonymized data. Our reporting follows the Transparent Reporting of Observational Studies Emulating a Target trial ( TARGET ) guidelines for reporting of emulated target trials. 26 , 27 The ideal randomized trial for this study is described in eTable 1 in Supplement 1 . Briefly, we conducted a linkage of Australian state and national data to identify women who received MAR between January 1, 1991 and December 31, 2018. A description of the datasets used is given in eTable 2 in Supplement 1 . While the term women is used, this study includes all people assigned female at birth, including gender diverse and nonbinary people. Women who received MAR were identified by the Australian Institute of Health and Welfare as those who had a record of reimbursement for a MAR treatment on the Medicare Benefits Scheme (Australia’s national public health insurance scheme) or of reimbursement for a relevant MAR medication on the Pharmaceutical Benefits Scheme (Australia’s national medication subsidization program) (eTable 3 in Supplement 1 ). The Australian Institute of Health and Welfare then identified matched women who were never exposed to MAR (up to 4:1) and made the anonymized cohorts available to the research team. Women were matched on birth year, residential remoteness (metropolitan or rural), parity at matched exposed individual’s first MAR exposure date, and (if the matched exposed individual was parous at their first MAR) their age at their first birth (±365 days). Due to an error in data extraction, approximately 300 000 comparator women identified by the AIHW for study inclusion were not in the country at the beginning of analysis and had to be excluded, as we could not determine the presence or absence of cancer. This error created an imbalance in the number of nulliparous women in the MAR treatment and comparator groups, which was accounted for later through inverse probability of treatment weighting. eFigure 1 in Supplement 1 shows the steps taken to ascertain the cohort of eligible women for treatment strategy assignment. Three types of MAR treatment were identified by the presence of relevant procedural or prescription codes (eTable 4 in Supplement 1 ) in the first 28 days following first MAR exposure of any type: ART treatments (in vitro fertilization or intracytoplasmic sperm injection); IUI or OS with follicle stimulating hormone or ART cancelled before egg retrieval; and ovulation induction using clomiphene citrate (from 2002 only). This first 28-day period from first MAR treatment was defined as time zero (t0), with study-time partitioned into 28-day blocks from this point forward. Women could contribute to multiple treatment groups if they had the relevant treatment codes, although this was uncommon (<10% of included women). Outcomes included female cancers with sufficient evidence of an association with hormonal agents, specifically breast (invasive and in situ, separately), ovarian (serous and nonserous, separately and combined), and uterine. 28 Melanoma (invasive and in situ), colorectal cancer, and thyroid cancer were also included because hormones may be relevant to their development. 29 , 30 , 31 , 32 Cancers not considered hormone related (pancreatic, lung, and hematological) were included as negative controls. Only invasive tumors were considered ( International Classification of Diseases for Oncology, 3rd Edition [ ICD-O-3 ] behavior code 3), unless specified. All cancers were classified using ICD-O-3 topographies and morphologies in the Australian Cancer Database (eTable 5 in Supplement 1 ). Following calculation of descriptive statistics, we conducted an emulated target trial using flexible parametric survival modeling, setting α = .05 as statistically significant. We controlled for confounding using standardized inverse probability of treatment weighting. Confounding variables included age at t0, history of giving birth prior to t0, residential remoteness at t0 (metropolitan, inner regional, remote, or very remote), mean Index of Relative Socioeconomic Disadvantage percentile for area of residence, subsidized diabetes medication record prior to t0, and cancer registry notified cancer diagnosis prior to t0. Exposures were the 3 different MAR exposures, and outcomes were all hormone-related and negative control cancers. We tested for time-varying associations with treatment using likelihood ratio tests comparing 4 models: a baseline model with no time-varying association of treatment and 3 models with time-varying associations with differing numbers of splines in the hazard function (0, 1, or 3 splines). Women were followed up until either a relevant cancer outcome, December 31, 2019, or death (as recorded in the National Death Index), whichever occurred first. We report hazard ratios (HRs) and cumulative marginal survival difference per 100 000 population in expected cancer diagnoses. We estimated how much uncontrolled confounding would be needed to account for our observed associations using E-values. 24 E-values are the minimum association (on the risk ratio scale) an uncontrolled confounder would need to have between both the exposure and outcome to fully explain the observed association. For time-varying HRs, E-values are given at year 1, and then every 5 years starting at year 5. Data were analyzed using SAS version 9 (SAS Institute) and Stata version 18 (StataCorp) from April 2024 to July 2025.

We identified 1 748 927 women, including 396 661 women exposed to MAR ( Table 1 ). During follow-up for cancer, 0.4% to 1.3% of the cohorts were censored at death; 4.0% of women who received MAR identified by the Australian Institute of Health and Welfare were not eligible for inclusion due to missing data on state (or state recorded as Northern Territory), sociodemographic area of residence, or remoteness at t0. We assumed these women were missing at random and proceeded with complete-case analysis. eTable 6 and eTable 7 in Supplement 1 show the total number of incident cancers and the results of the model selection steps respectively. Abbreviations: ART, assisted reproductive technology; IUI, intrauterine insemination; NA, not applicable; OS, ovarian stimulation; t0, time zero. Percentages are by column, except for overall, which is given as percentage across treatment type. When approximating the number of treatments, only 1 treatment could occur in each 28-day analysis window. A summary of all modeling results is given in Table 2 . Eight emulated trials of hormone-related cancers showed evidence of a time-varying HR. The most common pattern was an association of MAR with higher risk of cancer immediately after first treatment that decreased over time, reducing to a small or no difference in risk after approximately 10 years ( Figure 1 and Figure 2 ). This was the case for any ovarian cancer, serous ovarian cancer (47% of ovarian cancers), uterine cancer following ART, invasive melanoma following IUI or OS, and invasive breast and uterine cancer following clomiphene citrate ( Figure 2 ). The observed association with an excess cancer risk was maintained across the follow-up period for only uterine cancer following clomiphene citrate ( Figure 2 ). ART was associated with increased risk for invasive melanoma, and IUI or OS was associated with increased risk of nonserous ovarian cancer up to approximately 5 years after treatment and then were associated with decreased risk, reaching no association with excess risk before 10 years after treatment. Abbreviations: ART, assisted reproductive technology; IUI, intrauterine insemination; MAR, medically assisted reproduction; OS, ovarian stimulation. Calculated from central estimate for cancers with a significant effect (α = 0.05) without a time-varying hazard ratio. Reported from 6 months. MAR indicates medically assisted reproduction. Reported from 6 months. MAR indicates medically assisted reproduction. In trials where the HRs were not observed to vary over time, all types of MAR treatment were associated with elevated risks of most cancers, with HR estimates ranging from 1.09 (95% CI, 1.03-1.16) (invasive breast cancer following IUI or OS) to 1.64 (95% CI, 1.31-2.05) (nonserous ovarian cancer following ART) ( Table 3 ). There was no elevation in risk for in situ breast, thyroid, or colorectal cancer following IUI or OS or in situ breast, serous ovarian, or colorectal cancer following clomiphene citrate use. Abbreviations: ART, assisted reproductive technology; HR, hazard ratio; IUI, intrauterine insemination; MAR, medically assisted reproduction; NA, not applicable; OS, ovarian stimulation. E values were not calculated for results where the HR’s 95% CI was entirely less than or included 1. Regarding the negative control cancers, pancreatic cancer showed no increased risk after IUI or OS or after clomiphene citrate and had an increase in risk following ART. This association with ART was time varying, observed in the 5 years after treatment, then decreasing over time, a pattern that mirrored most hormone-related cancers with time-varying HRs. A similar pattern was observed for hematological cancers following all 3 MAR treatments. ART and IUI or OS were associated with decreased risk of lung cancer, and there was no association of lung cancer with clomiphene citrate. Most treatments were estimated to be associated with fewer than 10 extra incident cases per 100 000 population per year, with only women treated with ART or IUI or OS approaching nearly 20 extra invasive breast cancers per 100 000 population per year ( Table 3 ; eTable 8 in Supplement 1 ). E-value analysis indicated the association between the treatment and cancer outcome could be explained by a moderately strong confounder (risk ratio, 2) at the lower-bound CI (taken at the effect after 5 years for trials with time-varying hazard ratios (eTable 9 in Supplement 1 ), with the exception of uterine cancer after both ART and clomiphene citrate use. We conducted a sensitivity analysis removing all women who had a reportable cancer (both invasive and in situ breast cancer and melanoma) prior to t0 (eTables 10-12 and eFigure 2 in Supplement 1 ). No major differences were observed that would substantially alter our conclusions.

In this cohort study applying an emulated target trial design to a cohort derived from national high-quality population-based datasets, we found MAR treatment was associated with an increase in risk of some hormone-related cancers. However, E-values and analysis of control non–hormone-related cancers suggested at least some of the increased risk was due to uncontrolled confounding and, for some cancers, bias due to differences in access to health care closer to the time of first treatment. The detection of such biases shows we did not emulate randomization sufficiently to emulate a target trial examining the impact of MAR on hormone-related cancers. However, we were able to leverage this framework to interrogate long-standing concerns with bias in studies on MAR and cancer. We also established that even with uncontrolled confounding, the absolute number of extra cancers expected following MAR treatment was relatively small. Understanding the pattern of cancer risk over time since MAR is important, as there has long been concern that the observed associations between MAR and cancer may be biased by medical surveillance either while undergoing treatment or while pregnant. 33 , 34 In our study, several emulated trials suggested a greater likelihood of incident cancer shortly after first treatment. This pattern suggests 2 possibilities: (1) MAR treatment (alone or in combination with pregnancy) promotes the growth of existing but undetected cancers or transformation of benign to invasive cancers or (2) there is increased surveillance during MAR treatment (and pregnancy), which leads to earlier detection of incident cancers (detection bias). Overall, although it is possible specific MAR treatments are associated with cancer progression, it is difficult to reconcile this explanation with our observed risk pattern for 2 of the 3 negative control cancers. Therefore, we believe detection bias is the more likely explanation for these results. This finding is the first empirical indication using a negative control that medical surveillance may be responsible for an increased risk of cancer following MAR. This conclusion echoes that of Vassard et al, 33 who observed a similar pattern in ovarian cancer risk after MAR using stratified Cox regression. In Australia, individuals are invited to attend breast and colorectal cancer screening, and there is opportunistic skin cancer screening for individuals at high risk. According to Andersen’s behavioral model for use of health services, 35 it is likely individuals who have engaged in fertility treatments are more likely to engage with these screening programs, and this disparity in seeking health services may also have affected our findings. Our E-value analysis suggested some of the associations we observed could be explained by uncontrolled confounding. One such factor is the underlying cause of infertility, which was not recorded in any of the available datasets. Recent meta-analyses suggested endometriosis is associated with both any ovarian cancer (risk ratio, 1.93—noting nonserous ovarian cancer after ART had a lower-bound E-value of 1.94) and thyroid cancer (risk ratio, 1.39), which may explain our observed results for these cancers following ART and IUI or OS. 36 , 37 The risk ratio for infertility for individuals with endometriosis compared with the general population has been estimated at 2.12, 38 indicating for these cancers both the confounder-exposure and confounder-outcome relationship may have sufficient strength to account for the associations we observed. Polycystic ovarian syndrome (PCOS) may also contribute to unmeasured confounding, particularly for uterine cancers and for clomiphene citrate users. Although it is uncommon to use clomiphene citrate for individuals with endometriosis-induced infertility (as endometriosis rarely causes anovulation), PCOS is partly defined by the presence of irregular ovulation, 39 and clomiphene citrate is commonly prescribed for individuals with PCOS-induced infertility. A 2023 meta-analysis found PCOS was associated with risk of uterine cancer (odds ratio, 4.07). 40 Furthermore, anovulation is a risk factor for uterine cancer 41 and more common in individuals with obesity and individuals with diabetes, 42 , 43 who are at greater risk of uterine cancer 44 and thyroid cancer. 45 , 46 Finally, a 2019 registry study found PCOS was significantly associated with ovarian cancer (HR, 2.16), although not thyroid cancer. 47 Overall, underlying causes of infertility, coupled with associated obesity and diabetes, may account for either all or a substantial proportion of the observed associations between MAR treatment and risk of ovarian, uterine, and thyroid cancers. Our findings for breast and colorectal cancer may be more robust to unmeasured confounding by the underlying cause of infertility. Recent studies found no associations between infertility history and overall risk of breast cancer 48 or colorectal cancer, 49 and in our study, both these cancers had the lowest observed increase in relative risk following MAR treatment. For IUI or OS, there was no association of the treatment with the risk of colorectal cancer, and for invasive breast cancer following clomiphene citrate, increased risk disappeared over time. Due to the high background rate of invasive breast cancer in the population, this association with a small increase in relative risk translated to an estimated mean of 400 extra cancers per 100 000 women treated over the course of the study period. This associated excess risk of breast cancer is similar to that following regular use of combined and progesterone-only hormonal contraceptives at ages 35 to 39 years. 50 Findings on melanoma and its associations with fertility and pregnancy hormones (like estrogen) are mixed, 30 so it is difficult to determine whether there is infertility-related confounding for this malignant neoplasm. MAR treatment was associated with increased risk of hematological cancer, despite no theorized underlying biological mechanisms. This suggests other sources of bias we could not account for. International estimates show White women are more likely to undergo fertility treatments. 51 As melanoma and hematological, uterine, breast, ovarian, and thyroid cancers are more common in White people, 52 confounding by race and ethnicity may be responsible for the associations we observed for these malignant neoplasms. Notably, we observed a negative association between MAR treatment and lung cancer. We posit this may reflect uncontrolled confounding by smoking behavior, with women who undergo MAR being substantially less likely to smoke before pregnancy 53 and potentially in the prenatal period. There is evidence smoking is a risk factor for ovarian, colorectal, and breast cancers. 28 By contrast, smoking may be protective for melanoma. 54 Therefore, failure to control for smoking history may have attenuated our observed associations between MAR and these cancers. Clinically, our findings indicate the associations of MAR with increased cancer risks may be partially or fully explained by underlying infertility conditions (ovarian, uterine, and thyroid cancer), heightened medical surveillance (breast and colorectal cancer), and demographic characteristics (melanoma), rather than MAR itself. However, given the numerous associations observed between MAR treatments and hormone-related cancers, there is a clear need for routine surveillance of patients after treatment. When advising patients considering undertaking MAR, health care professionals should emphasize cautious interpretation and communication of cancer risk. Our study highlights 2 areas for further research. MAR treatment may increase the risk of ovarian, uterine and thyroid cancers, albeit less than we observed. To improve our ability to answer this question, it is critical that we collect and make available for research information on key confounding variables at the population level. Such enhanced data collection efforts would significantly strengthen the capacity to identify and account for complex confounding pathways, ultimately improving the validity of causal inferences related to MAR and risk of cancer and other health outcomes. Second, future studies should aim to replicate the time-varying associations across cancers to test for robustness. This study has some limitations. We were limited to routinely collected national data, which did not include several important confounders, such as underlying cause of infertility, race and ethnicity, individual-level socioeconomic factors (eg, education and income, which may also be markers for engagement with preventive health services), family history of cancer, smoking, and oral contraceptive use. Although the optimal emulated trial would account for all potential confounding factors, this is likely an impossible task for causal questions regarding use of MAR. 12 Our research represents the most comprehensive attempt to address the effect of this missing confounder available, to our knowledge, through bias analysis. Due to concerns of unmeasured confounding from cause of infertility, as well as computational feasibility, we were unable to assess a mediating effect of pregnancy. Given pregnancy may be associated with a changed risk of several hormone-related cancers 55 it is likely there may be a different effect of MAR treatment on cancer dependent on whether a pregnancy is achieved. We were unable to examine associations with specific medications and dosages used in ART and IUI or OS treatments. We only included women who received government reimbursement for their MAR treatments, and those who paid out-of-pocket for fertility services (such as those in same-sex relationships without a diagnosis of infertility, which is not covered under Medicare) were not included. We underascertained clomiphene citrate use between January and June 2009 and January and April 2012 due to systematic undercapture in the Pharmaceutical Benefits Scheme Dataset, which may have led to misclassification of some women as not exposed to MAR. Additionally, our study was conducted within an Australian context, which subsidizes the use of most MAR treatments with no exceptions for age or prior children conceived, but historically did not subsidize those not deemed to be medically infertile. Thus, caution must be taken when applying the results presented here directly to other countries with different MAR eligibility criteria.

In this cohort study of MAR and cancer using target trial emulation, we examined associations of 3 types of MAR treatment, ART, IUI or OS, and clomiphene citrate, with hormone-related cancers in women. Although we observed increased relative risk for most hormone-related cancers following MAR, this corresponded to only small increases in estimated absolute excess risk. Bias analysis and the inclusion of negative control cancers showed the elevated risk was potentially fully attributable to uncontrolled confounding in some cases. Clinicians and patients should be aware that a small excess risk of cancer following MAR treatment may exist, but this excess risk may be partially or fully due to the health and sociodemographic profile of women who receive MAR and increased surveillance during treatment.

Medically assisted reproduction (MAR) refers to a collection of treatments that assist individuals to achieve a successful pregnancy. MAR treatments involving the extraction and in vitro handling of ova (eggs) are referred to as assisted reproductive technologies (ART). These treatments are generally conducted with medical stimulation of the ovaries to retrieve multiple ova in a single menstrual cycle (ie, ovarian stimulation [OS]). These eggs are fertilized through in vitro fertilization, either by mixing the sperm with the egg, or intracytoplasmic sperm injection. MAR treatments without in vitro handling of ova include ovulation induction through hormonal medications or by intrauterine insemination (IUI), in which prepared sperm is inserted into the uterus around the time of ovulation. MAR treatments have become increasingly common, with approximately 3 million ART cycles performed each year worldwide. 1 In most cases, MAR therapies involve the administration of medications that are either hormonal or hormone-modulating agents, including gonadotrophin-releasing hormone analogues, follicle stimulating hormone, human chorionic gonadotrophin, progesterone, estrogen, and clomiphene citrate or letrozole. There has long been concern that these medicines might be associated with the development of hormone-related cancers, including ovarian, uterine, breast, melanoma, thyroid, and possibly colorectal cancers. 2 , 3 , 4 , 5 In ART treatment, it is also possible the repeated puncture of follicles during ovum retrieval might cause breakdown and repair of the ovarian surface epithelium similar to that in natural ovulations, consistent with the incessant ovulation hypothesis for ovarian carcinogenesis. 6 , 7 Despite the biological plausibility of these hypotheses, 6 , 8 , 9 , 10 , 11 strong evidence for any carcinogenic effect of MAR medicines is limited. 12 , 13 Due to the rarity of some cancers, the long latency period for most solid cancers, and ethical concerns around withholding treatment for infertility, randomized clinical trials cannot be used to examine this question. 12 Recent meta-analyses on observational studies suggest few to no associations of ART with hormone-related cancer risk in female patients. 14 , 15 , 16 However, there is substantial heterogeneity in study designs and availability of confounding variables. This heterogeneity is compounded by changes in patterns of MAR use over time and between-country differences in eligibility for and access to MAR. In general, most studies conducted since 2010 adjusted for relevant confounders, like parity, maternal age, and socioeconomic status. 14 When meta-analyzing only these studies, no significant associations of MAR treatment with breast, ovarian, or uterine cancer were found. 14 Considering specific studies reported over the past 10 years, Vassard et al 17 reported a significant association between ART and invasive breast cancer, and Lundberg et al 18 reported a heightened risk of ovarian cancer following ART in individuals who gave birth following treatment. Kessous et al 19 reported high excess risks for both ovarian and uterine cancer but no increased risk for breast cancer (combining in situ and invasive). Reigstad et al 20 found increased risk following ART treatment for thyroid cancer, but not other types of hormone-related cancer (noting wide CIs). However, Reigstad et al 20 did find increased risks of uterine cancer following use of clomiphene citrate, invasive breast cancer following clomiphene citrate for parous individuals, and ovarian cancer for nulliparous individuals. Spaan et al 21 found evidence that borderline ovarian cancers, but not invasive ovarian cancers, were more common following ART. Overall, there is substantial variability in both confounder adjustment and analysis approach, with this variability contributing to heterogenous results. Understanding the risks after undertaking MAR treatments is critical for female patients and their health care practitioners. 14 , 15 , 16 , 22 This study aims to untangle the complex relationship between MAR and hormone-related cancers using the emulated target trial framework. 23 To our knowledge, no studies have leveraged the target trial framework to analyze the association of MAR with cancer. We also include bias assessment (through calculation of E-values), 24 negative controls, and assessment for time-varying treatment effects. In doing so, we build on previous work by directly assessing the likely impact of missing confounders, and the likelihood that detection bias might contribute to reported associations between MAR and cancer.