Ovarian Cancer and Fertility Medications: A Critical Appraisal

S. Kashyap, M.D., F.R.C.S.(C), O.K. Davis, M.D., F.A.C.O.G.


Semin Reprod Med. 2003;21(1) 

In This Article


The cohort consists of 29,700 women who were investigated and/or treated for infertility in any of 10 Australian IVF clinics before January 1994 (29.6% before 1986; 39.8% from 1986 to 1990; 30.5% after 1990). There was no significant difference between age at entry and age at end of follow-up for the exposed and unexposed cohorts. Exposed women included patients who had received at least one IVF treatment cycle whether or not the cycle proceeded to oocyte retrieval. Unexposed patients were patients who were registered for IVF but did not undergo treatment or patients who had natural cycle IVF. Selection bias may be introduced here. Patients who had natural cycle IVF may have been older, had failed previous ovarian stimulation elsewhere, or already been patients who had a specific predisposition to estrogen-dependent tumors. However, such a bias would be expected to skew away from the hypothesis that ovarian cancer might be related to fertility treatment.

Exclusion criteria included: women who had preexisting cancer prior to referral, permanent residence out-side Australia (n = 623); and age/date of birth unknown. If women registered with more than one IVF clinic, they were only included once but the total number of IVF cycles was combined. Canceled stimulation cycles (no oocyte retrieval) were included as exposure.

Causes of infertility were divided into the following categories: tubal factor, male infertility, endometriosis, and ovarian disorders. Ovarian disorders included a broad spectrum of processes: polycystic ovarian syndrome, premature ovarian failure, and bilateral salpingoophorectomy. It is not physiologically rational to combine polycystic ovarian syndrome with disorders such as premature ovarian failure, as the mechanisms of disordered ovulation are different. Unexplained infertility and other causes (cervical, uterine, and genetic) complete the categories.

Of note, there was a higher proportion of exposed patients (33.9%) with male factor infertility versus unexposed (19.3%). This might act as a confounder if an underlying ovarian dysfunction predisposes a patient to ovarian cancer and therefore exposed patients with male infertility could skew the results away from the hypothesis. Endometriosis was more common in exposed patients (17.5%) versus unexposed patients (13.2%). Some authors have suggested that the incidence of clear cell and/or endometrioid ovarian tumors is increased in patients with endometriosis.[8,9,10]

Important confounders such as parity and oral contraceptive (OCP) use were not addressed. Such confounders are difficult to address in a cohort study. If these questions are not asked at the beginning, it can be very difficult technically and financially, as well as methodologically controversial, to collect this information later. Also, the exposure may change over time. For example, patients who experience infertility may be nulliparous at the start of the study but may conceive as a result of treatment. Parity protects against ovarian cancer. It would be important, although difficult, to assess the impact of treatment on this confounding factor. Also, patients with disordered ovulation might use OCPs to regulate their cycles. As described previously, OCP use appears to confer significant protection against ovarian cancer.

The authors were unable to address treatments undertaken prior to the subjects' registration for IVF. It is likely that many patients had initial investigations and possibly treatment with their primary gynecologists. Prolonged clomiphene citrate therapy (greater than 12 cycles) is reportedly associated with an increased incidence of ovarian cancer.[11] It is not apparent whether there was an equal distribution of pre-IVF therapy between the exposed an unexposed cohorts. The authors describe three time periods (before 1986, 1986 to 1990, and after 1990) in an attempt to account for differences in medication protocols but they did not distinguish between time periods in their final analysis. The authors did try to account for different doses of medications by using cycle numbers as a proxy.

The unexposed cohort of patients included women who were registered for IVF but did not proceed. There were many reasons for not proceeding with treatment, including the occurrence of pregnancy. Patients who achieve pregnancy have a better prognosis regarding the occurrence of ovarian cancer than nulliparous patients. Infertile patients have a higher incidence of ovarian cancer than the general population. Unfortunately, the proportion of patients who did not proceed with therapy because they conceived spontaneously is not stated. It is also not clear whether patients who decided not to proceed with IVF because they conceived spontaneously truly had subfertility.

The study attempts to compare the observed incidence of ovarian cancer in the cohort to the expected incidence through the application of standardized incidence ratios. However, it is not clear whether the authors used the entire infertile cohort in this comparison or just the exposed infertile cohort. If the whole cohort was used, the unexposed proportion may dilute the effects if fertility drugs are responsible for an increased incidence of ovarian cancer. In either case, the comparison is not relevant for our purposes. We are concerned with whether or not infertile patients who are treated with ovulation induction agents have a higher incidence of ovarian cancer than untreated infertile patients. Infertility patients have a higher incidence of ovarian cancer than general population controls as a result of their infertility and nulliparity. Therefore a comparison between infertile treated patients and general population controls is not valid. Also, standardized incidence ratios are hypothetical numbers that assume the incidence of a disease would be the same in a population over time. It assumes that the only factor that may change over time is the size of the population.

Treatments prior to registration with one of the 10 IVF clinics were not addressed. The implications are as described above. Time periods were described as before 1986, 1986 to 1990, and after 1990. Given that the availability of medications differed over these times (e.g., gonadotropin-releasing hormone analogs), this differentiation is appropriate. However, time period is not addressed later in the analysis.

Exposed and unexposed periods are measured as person-years. Person-years from time of registration to treatments are included as unexposed. Only person-years after treatments are included as exposed. Again, this raises the issue of how many patients conceived spontaneously in the unexposed group. Also, a significant proportion of unexposed person-years were derived from the years between registration and first treatment. Therefore, a significant proportion of these years were early in the cohort cycle and therefore at less risk of ovarian cancer.

Outcomes were measured as invasive cancers of the ovaries and tubes. Borderline ovarian tumors were not included because of the difficulty with classification. Ascertainment of cancer cases was by record linkage to population based cancer registries. These registries are held by the state of by the National Cancer Statistics Clearing house and the National Death Index. Linkage was based on name and date of birth, with address as an additional identifier. Women who used a different family name than their partner were linked on both family names. Diagnosis was made by record linkage and not by confirmation of histopathological diagnosis. This approach gives rise to a serious potential for misclassification, although it should be nondifferential, that is, it should occur similarly in exposed and unexposed individuals.

Follow-up was from the time each woman entered the cohort until the first of date of cancer diagnosis, death, or December 31 of the year of complete cancer data for her state of residence (for 80% of the cohort this was 1996). Any diagnosis of cancer immediately before a women's registration with an IVF program was checked to confirm that the cancer preceded the referral to IVF. Four women with cancer were referred for IVF with the aim of producing embryos for freezing before cancer therapy commenced and they were excluded from the analysis.

No. The exposed population was followed for 7 years (median) and the unexposed population was followed for 10 years (median). The median age at entry for the exposed and unexposed was 31 and 30 years, respectively. The median age at the end of follow-up for the exposed and unexposed was 39 and 40 years, respectively. The peak age of incidence for ovarian cancer is 62.

Borderline tumors are not included in the analysis. Borderline ovarian tumors tend to occur at an earlier age (peak in the 40s) and have a more indolent course. It is possible that women exposed to fertility medications may have a higher incidence of borderline tumors. However, the authors recognized that it would be difficult to assure consistency in the reporting of borderline tumors. Because they were unable to confirm histopathological diagnosis, they excluded borderline tumors from the analysis.

According to the authors' criteria, none of the diagnoses of ovarian cancer preceded IVF treatment. Exposure was documented to have occurred before diagnosis. However, ovarian cancer has a long latency period and, unlike the breast cancer cases identified in this study, the timing of diagnosis of ovarian cancer relative to the treatments is not clear. For example, if the diagnosis occurred within 1 year of treatment, it is likely that the tumor existed before the treatment.

There is no clear evidence for a dose-response gradient. First, the authors compared the incidence of ovarian cancer from the cohort to general population controls by means of a standardized incidence ratio. Secondly, it would have been difficult to calculate a specific dose gradient for the various drugs. However, they attempted to use number of cycles as a proxy. The raw numbers are not available to recalculate whether the incidence increases in the exposed population in proportion to number of cycles. Also, it is not clear how they arrived at an expected number of cases in a "general population" of women who would not have been infertile and therefore would not have received treatment, because the general population does includes some infertile women.

A dechallenge-rechallenge procedure is not feasible with these event outcomes. Once exposed, the exposure cannot be removed nor can the outcome revert to baseline to observe whether the event is reinitiated with subsequent exposure.

These results are consistent with a recent meta-analysis of case control studies.[12] Also, other cohort studies have found similar results.[13,14]

A biologically plausible mechanism would support a cause-and-effect relationship. There are two predominant theories that explain the possible link between induction of ovulation and ovarian cancer. Although these theories are dated, they remain physiologically plausible in the context of infertility. In 1971, Fathalla proposed the "incessant ovulation theory."[15] In a letter to the editor, the author proposed that repeated ovulation induces microtrauma to the ovarian epithelium and, without protected anovulatory intervals, may lead to mitotic abnormalities, which may lead to cancer. The risk reduction afforded by pregnancy and OCP use supports this theory. The other theory is that extended periods of exposure to high gonadotropin levels, such as leutinizing hormone concentration in polycystic ovarian syndrome, induce tumorigenic cellular abnormalities. Some authors suggest that women with anovulatory infertility, such as polycystic ovarian syndrome, may have a particularly high risk of ovarian cancer secondary to excess leutinizing hormone. However, patients with polycystic ovarian syndrome have not been shown to have an increased incidence of ovarian cancer. The "gonadotropin theory" was suggested in animal studies by Biskind and Biskind in which rats, stimulated with gonadotropins, developed granulosa cell tumors.[16] Reports of the presence of follicle-stimulating hormone and leutinizing hormone receptors on ovarian epithelial tumors are inconsistent.[16,17] Confirmation of such receptors would lend more support to the theory of gonadotropin stimulation of epithelial ovarian tumors. However, one would expect that if a true association were present, the findings would be significant and persistent from study to study. So far there is no convincing evidence that fertility drugs are in fact associated with ovarian cancer in excess of the risk that infertility itself imparts.

The results from this study suggest that fertility treatment may actually protect against ovarian cancer. This would make biological sense if those patients who are treated actually do conceive. As mentioned in the introduction, parity affords protection against ovarian cancer.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.