Risks of ovarian, breast, and corpus uteri cancer in women treated with assisted reproductive technology in Great Britain, 1991-2010: data linkage study including 2.2 million person years of observation

Abstract Objective To investigate the risks of ovarian, breast, and corpus uteri cancer in women who have had assisted reproduction. Design Large, population based, data linkage cohort study. Setting and participants All women who had assisted reproduction in Great Britain, 1991-2010, as recorded by the Human Fertilisation and Embryology Authority (HFEA). Interventions HFEA fertility records for cohort members were linked to national cancer registrations. Main outcome measures Observed first diagnosis of ovarian, breast, and corpus uteri cancer in cohort members were compared with age, sex, and period specific expectation. Standardised incidence ratios (SIRs) were calculated by use of age, sex, and period specific national incidence rates. Results 255 786 women contributed 2 257 789 person years’ follow-up. No significant increased risk of corpus uteri cancer (164 cancers observed v 146.9 cancers expected; SIR 1.12, 95% confidence interval 0.95 to 1.30) was found during an average of 8.8 years’ follow-up. This study found no significantly increased risks of breast cancer overall (2578 v 2641.2; SIR 0.98, 0.94 to 1.01) or invasive breast cancer (2272 v 2371.4; SIR 0.96, 0.92 to 1.00). An increased risk of in situ breast cancer (291 v 253.5; SIR 1.15, 1.02 to 1.29; absolute excess risk (AER) 1.7 cases per 100 000 person years, 95% confidence interval 0.2 to 3.2) was detected, associated with an increasing number of treatment cycles (P=0.03). There was an increased risk of ovarian cancer (405 v 291.82; SIR 1.39, 1.26 to 1.53; AER 5.0 cases per 100 000 person years, 3.3 to 6.9), both invasive (264 v 188.1; SIR 1.40, 1.24 to 1.58; AER 3.4 cases per 100 000 person years, 2.0 to 4.9) and borderline (141 v 103.7; SIR 1.36, 1.15 to 1.60; AER 1.7 cases per 100 000 person years, 0.7 to 2.8). Increased risks of ovarian tumours were limited to women with endometriosis, low parity, or both. This study found no increased risk of any ovarian tumour in women treated because of only male factor or unexplained infertility. Conclusions No increased risk of corpus uteri or invasive breast cancer was detected in women who had had assisted reproduction, but increased risks of in situ breast cancer and invasive and borderline ovarian tumours were found in this study. Our results suggest that ovarian tumour risks could be due to patient characteristics, rather than assisted reproduction itself, although both surveillance bias and the effect of treatment are also possibilities. Ongoing monitoring of this population is essential.


Introduction
Assisted reproduction cycles usually involve exposure to supraphysiological levels of oestradiol, exogenous gonadotropins, and multiple ovarian punctures, all potentially carcinogenic. 1 2 Most concern surrounds the risks of breast, endometrial, and ovarian cancers after such exposures. [3][4][5][6][7][8][9][10][11][12][13][14][15][16] Studies investigating breast cancer risks in women who underwent assisted reproduction are inconsistent. [3][4][5][6][7][8][9][10][11][12] Although some studies have shown an increased risk, 17 most studies do not show an overall increase of breast cancer in exposed women. 3-8 10 However, some suggest a possible increased risk within subgroups, 8 9 including women treated at younger ages 9 and with multiple cycles. 8 Most studies investigating endometrial cancer risk in exposed populations have not found a significant increased risk. 3 4 6 7 18 However, most studies have provided very imprecise estimates doi: 10.1136/bmj.k2644 | BMJ 2018;362:k2644 | the bmj due to small sample size and few events. 3 4 6 18 One study suggested an increased risk of endometrial cancer associated with exposure to gonadotrophins, commonly used as part of assisted reproductive technology. 19 Some early studies investigating fertility drugs used alone, such as single agent oral clomifene, suggested increased risks of ovarian cancer. 20 Others found no association between fertility drugs and ovarian cancer risk. 21 Recent investigations into their use as part of assisted reproduction have generally been more reassuring, but remain inconsistent and at risk of bias. 4 5 11 Some 13 14 but not all studies 6 have found an increase in borderline tumours.
Given previous inconsistent results, small study size, and lack of information on potential confounders, we undertook a population based linkage study in Britain to provide risk estimates for ovarian, breast, and corpus uteri cancer, in a cohort of over 266 000 women undergoing assisted reproduction, with information on potential confounders such as parity and infertility diagnosis.

study population
We defined assisted reproduction as "treatments or procedures that include in vitro handling of both human oocytes and sperm or embryos, for the purpose of reproduction." 22 Records for all women undergoing assisted reproduction from January 1991 to September 2009, and those undergoing the same from October 2009 to December 2010 who gave their prospective consent, in England, Wales, and Scotland were obtained from the Human Fertilisation and Embryology Authority (HFEA).
UK law mandates reporting of all assisted reproduction cycles to the HFEA. For cycles performed before October 2009, research use of these data was permitted, but consent could be withdrawn retrospectively. Fewer than 300 women had done so before this study began (based on the level of reporting detail provided by the HFEA). The study cohort, January 1991 to September 2009, therefore represents about 99.7% of the at-risk population. For cycles performed October 2009 onwards, prospective consent was required. Overall consent was not provided for an estimated 7% of women undergoing assisted reproduction in 1991-2010 (about 20 000 women, based on reports from the HFEA), who were therefore not included in this study, representing a loss of less than 1% of person years' follow-up (figure S1, supplementary appendix).
Outcome data HFEA records were linked to the National Health Service Central Registers of England, Wales, and Scotland (from which emigrations, deaths, and cancer registrations are reported to authorised medical researchers) in a one-off linkage. Completeness and accuracy of these registers have been described. [23][24][25] Overall, records of 266 787 (95.1%) eligible women were linked (box S1 and figure S1, supplementary appendix). Cancer diagnosis date, topography code (ICD-9/ICD-10 (international classification of diseases, 9th and 10th revisions)), morphology (ICD-O-2/  ICD-O-3 (international classification of diseases for  oncology, second and third revisions)), and behaviour  (ICD-O-2/ICD-O-3) were available where an incident cancer was diagnosed. Women with cancer diagnoses (including non-melanoma skin cancer) recorded before the first treatment year were excluded from analyses. We obtained data relating to potential confounding factors such as infertility diagnosis, parity (as recorded at last treatment cycle completion), and treatment details (including number of stimulated cycles and age at first treatment) for each cohort member from the HFEA database. These data are a combination of patient reported and clinic reported information (table S1). Information regarding infertility diagnoses are reported to the HFEA by assisted reproduction clinics, based on investigations undertaken by that clinic; by the referring clinician; or occasionally by patient self report.

statistical analyses
Follow-up was calculated from date of first treatment (estimated as the mid-point of the first treatment year) until the date of any cancer diagnosis, death, emigration, or study end (March 2011), whichever came first. For analyses involving number of cycles, infertility duration, and live and multiple births, person years at risk were calculated from date of last treatment (estimated as mid-point of the last treatment year), because the HFEA did not record intermediate dates required for time dependent analysis. To calculate expected cancers, we multiplied the person years at risk by corresponding national incidence rates (by 5 year age band and individual calendar year) for the general female population of England and Wales.
Standardised incidence ratios were calculated by the comparison of observed values with expected values. We calculated 95% confidence intervals, two sided P values, and trends assuming a Poisson distribution. 26 Sensitivity analyses excluded the first 12 months of follow-up, to investigate potential surveillance bias in the period immediately following assisted reproductive treatment (which could arise as a result of treatment and or after-care; supplementary appendix). Absolute excess risks represent an estimate of the increased risk in the study group as compared with the general population and gives a direct measure of excess risk. They are presented per 100 000 person years, with corresponding 95% confidence intervals, based on exact confidence intervals for Poisson counts. Analyses were performed using Stata, version 12. 27 patient involvement and study approval Representatives from patient support groups were consulted on the original research question, design, and planning of this study. Approval of the study and waiver of the requirement for individual consent were obtained from the UK Health Research Authority Confidentiality Advisory Group and London Research Ethics Committee (references 5.04(b)/10 and 10/ H0720/18, respectively). Given the anonymous nature of the final dataset, it is not possible to disseminate results to individual study participants; instead results will be shared with fertility practitioners and clinics through the Human Fertility and Embryology Authority networks.

characteristics of study participants
In total, 255 786 women contributed 2 257 789 person years' follow-up. Average follow-up was 8.8 years (range 1-19 years), with 105 436 (41%) followed for at least 10 years. Average age at first treatment was 34.5 years. Infertility cause involved at least one female factor in 111 658 women (44%; including endometriosis, ovulatory disorders (predominantly polycystic ovary disease), and tubal disease). Infertility was unexplained in 47 757 (19%) women, and was due only to male factors in 84 871 (33%). Average infertility duration was 4.9 years. Women had 1.8 stimulated cycles on average, with only 20% (n=50 485) having more than two stimulated cycles. About half the study population had at least one live birth after treatment completion (table 1).
We found significant risk reductions with increasing duration since treatment completion (P=0.01; table 3), and in women with any female factor or only male factor infertility (table 3), but no difference between risks at premenopausal and postmenopausal ages separately (age <50 years, standardised incidence ratio 0.98 (95% confidence interval 0.94 to 1.02); ≥50 years, 0.97 (0.89 to 1.06); data not shown). After exclusion of the first 12 months of follow-up, breast cancer risk was significantly reduced compared with age standardised expectation (standardised incidence ratio 0.95 (0.92 to 0.99), P=0.02; supplementary appendix ). There was no increased risk of invasive breast cancer (standardised incidence ratio 0.96 (0.92 to 1.00); absolute excess risk −4.4 cases per 100 000 person years (95% confidence interval −8.5 to −0.2); carcinoma of the corpus uteri Risk of corpus uteri cancer was not significantly raised (standardised incidence ratio 1.12 (95% confidence interval 0.95 to 1.30); absolute excess risk 0.8 cases per 100 000 person years (95% confidence interval −0.3 to 2.0); table 2). Over 92% (n=152) of corpus uteri tumours were epithelial, 70% (n=107) of which were endometrioid; 8% were non-epithelial or unspecified (n=12). We found a significantly increased risk of corpus uteri cancer in women with an ovulatory disorder (standardised incidence ratio 1.59 (1.13 to 2.17); table 3). There was a highly significant trend of increasing risk with decreased parity (P<0.001), and a significantly decreased risk with women having a multiple birth (standardised incidence ratio 0.42 (0.14 to 0.99);

Ovarian cancer
An overall increased risk of ovarian cancer was observed in our study population (standardised incidence ratio 1.39 (95% confidence interval 1.26 to 1.53); absolute excess risk 5.0 cases per 100 000 person years (95% confidence interval 3.3 to 6.9); table 2). Increased risks were seen across most age groups at first treatment, but there was a highly significant trend of increasing risk with decreasing age at first treatment (P<0.001; table 3). Significantly increased risks were found in women who had any diagnosis of female factor infertility (standardised incidence ratio 1

discussion
Assisted reproduction is practiced worldwide, and more than five million children have been born as a result. 28 It is important to establish related disease risks for affected individuals, public health systems, and for counselling of potential patients. In this large population based cohort, we found no overall increased risk of breast cancer associated with assisted reproduction, consistent with most 3-10 but not all 12 published studies. We found no significant association between the risk of breast cancer and age at first treatment, in contrast to a small number of earlier studies. 8 9 29 Reasons for significant decreases in breast cancer risk seen in some subanalyses-such as women who had assisted reproduction for female factor infertility-are unclear, but could reflect beneficial levels of lifestyle related risk factors for breast cancer. 30 31 However, details of these risk factors and also age at first birth were not available. Menopausal status did not seem to account for the significant reduction in risk with increasing follow-up. Despite no increased risk of invasive breast tumours, there was a significant increase in in situ tumours which was significantly associated with increasing number of stimulated cycles. Interpretation of these findings is challenging: the significant association with increasing number of cycles suggests a causal association, yet there was no overall increased risk of breast cancer. Other potential explanations include surveillance bias, chance, and potential confounding by factors such as socioeconomic status, given that most cycles within our cohort were privately funded. To our knowledge, this study is the first to analyse risks of in situ and invasive breast cancers after assisted reproduction separately, so there are no previous data with which to compare.
Risk of corpus uteri cancer overall was not raised in our study. Women with the known risk factor of nulliparity 32 and those with a history of ovulatory problems (mainly the known risk factor polycystic ovary disease 33 ) were found to have an increased risk of corpus uteri cancer. Most similar studies contained few events. 3 5 6 The largest studies included 15 4 and 49 cases 7 of endometrial cancer in women after assisted reproduction, and neither suggested an increased risk.
We found an excess of ovarian cancer compared with age standardised expectation. Significant increases were observed for both invasive and borderline tumours, but were not seen in women without the known risk factors of endometriosis 34 35 and nulliparity. 35 Ovarian cancer risks were not associated with number of treatment cycles, time since treatment completion, or male factor or unexplained infertility, which argues against a causal role for assisted reproduction procedures. However, we did find a significant association between age at first treatment and risk of all, invasive, and borderline ovarian cancers. Previous studies investigating invasive ovarian tumour risk after assisted reproduction 3-711 13 15 16 have generally found increased risks in comparison with the general population when potential confounding effects of infertility have not been considered, 16 but not when such factors were taken into account. 3 4 11 16 While our study compared cancer incidence with that in the general population (standardised for age and calendar year), it had sufficient size to stratify by potential confounding factors and thereby to investigate characteristics of associations. We found an increased risk of borderline ovarian tumour in women having assisted reproduction compared with the general population. As with invasive ovarian tumours, this increased risk was not seen in parous women without endometriosis. Few studies have investigated the risk of borderline ovarian cancer in women after assisted reproduction, 6 13 14 but increased risks have been found in studies in the Netherlands 13 and Australia. 14 Although the increased risk in borderline ovarian cancer in women with assisted reproduction could be genuine, it could also be due to surveillance bias. The frequency of borderline tumour diagnosis is increased in ovarian cancer screening studies using ultrasound, 36 and women who have undergone assisted reproduction might have more frequent ultrasound scans after treatment than the general population. This potential bias is supported by the reduction in overall risk after we excluded the first 12 months of followup. However, sensitivity analyses looking at time to diagnosis, age at diagnosis, diagnosis in women of high socioeconomic status, and clinical presentation in other studies suggested surveillance bias an unlikely cause of increased risks. 13 14 We are not able to further differentiate surveillance bias from a genuine increase in borderline tumours. Women with unrecorded cause of infertility had significantly increased rates of breast, ovarian, and corpus uteri cancers. Reasons are unclear but might include reverse causality (box S2, supplementary appendix).

strengths and limitations of the study
Most studies investigating risks of cancer in women after assisted reproduction have been small, 6 8 with few events and short follow-up. [4][5][6][7] Two of the largest studies published so far include 67 608 4 and 113 226 7 women treated with assisted reproduction. Systematic reviews have included at most 70 753 treated women for analyses of breast cancer risk, 10 79 143 for ovarian cancer risk, 16 and 118 320 for analysis of all gynaecological cancer risk. 37 Our study comprised over 250 000 treated women, including almost 65 000 person years of follow-up for at least 15 years beyond last treatment with an average follow-up of 8.8 years and a maximum follow-up of 19 years (table S2, supplementary appendix). However, we cannot exclude the possibility of different risk profiles for any studied cancer on longer follow-up, at ages when most reproductive related cancers occur. 35 Women treated with assisted reproduction are likely to differ from the general population in their parity, age at first birth, age at menopause, and the incidence of predisposing conditions such as endometriosis. More information on these and other factors (eg, socioeconomic status, oral contraceptive use, body mass index, and breastfeeding) would be useful. Comparison to women with untreated infertility problems might have been beneficial, although interpretational problems would remain because of potential selection factors for treatment. Although our study was not able to compare with such a group as some smaller studies have done, 4 13 14 large study size enabled us to stratify for some important potential confounders and draw inferences despite using general population rates as our comparator. While comparator rates do include cohort participants, less than 5% of the population of reproductive age women underwent assisted reproduction, and our standardised incidence ratios were generally lower than 2.0; therefore, resulting bias will have been minimal. 38 Infertility diagnoses were reported by treating fertility clinics to the HFEA. No data were available about how such diagnoses were made. Further details of specific treatments could have enabled detailed analysis of risk by treatment type. However, over our 19 year study period, ovarian stimulation regimens as part of assisted reproductive cycles have been relatively constant, with the majority of advances leading to better success rates having occurred in assisted reproduction laboratories. Gonadotrophin injections have been used for ovarian stimulation and human chorionic gonadotropin for triggering ovulation throughout the study period, and while new highly purified and recombinant versions have been used in more recent years, they are essentially equivalent. Clomifene citrate was used as additional ovarian stimulation in the pioneering years of assisted reproduction treatment, but this was uncommon by 1991. Downregulated cycles using GnRH (gonadotrophin releasing hormone) agonists were standard by 1991 and not replaced by GnRH antagonists as standard until after the study period. Progesterone support was used throughout the study period. The number of ovarian punctures per cycle and information about fertility treatment before assisted reproduction were not available.

conclusions and implications
In this large, national population based study of British women after assisted reproductive technology treatment, no increased risk of corpus uteri or invasive breast cancer was detected. There was an increased risk of in situ breast cancer associated with increasing number of treatment cycles. We also observed an excess of all types of ovarian cancer. However, our results suggest that this finding is more likely due to underlying patient characteristics, rather than assisted reproduction itself. We were not able to distinguish between a genuine increase in risk of borderline ovarian tumours and other explanations including surveillance bias. Further investigation of this and longer follow-up is warranted to continue monitoring these important outcomes in this ever growing population.
We thank the Human Fertilisation and Embryology Authority for access to their original dataset and advice on data governance, and the National Health Service Digital and National Records for Scotland for their roles in the data linkage process and access to cancer registration data.
Contributors: CLW jointly conceptualised and designed the study, devised the linkage protocol, supervised the linkage, carried out the analysis, interpreted data, drafted the initial manuscript, and approved the final manuscript as submitted. MEJ jointly conceptualised and designed the study, jointly supervised the analysis, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. AJS jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. BJB jointly conceptualised and designed the study, jointly supervised the analysis, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. MCD jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. IJ jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. KJB jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. MFGM jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and approved the final manuscript as submitted. AGS jointly conceptualised and designed the study, interpreted data, reviewed and revised the manuscript, and had the final decision over submission of the approved the final manuscript. All authors had access to the data and take responsibility for the integrity of the data and accuracy of the data analysis. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. AGS acts as guarantor for this study. Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: support for this study from Cancer Research UK and NIHR; MEJ additionally received funding from Breast Cancer Now, and KJB and MFGM received funding from the UK Department of Health and Children with Cancer UK during the study; IJ reports personal fees from Abcodia and Women's Health Specialists, and receives royalties as coinventor of the ROCA algorithm; MCD reports personal fees from the Centre for Reproductive and Genetic Health; the authors declare no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: Approval of the study and waiver of the requirement for individual consent were obtained from the UK Health Research Authority Confidentiality Advisory Group and London Research Ethics Committee (references 5.04(b)/10 and 10/ H0720/18, respectively).
Data sharing: No additional data are available, in compliance with ethical and governance regulations under which this research was undertaken.
AGS affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/.