Dr. Jensen is Leon Speroff Professor and Vice Chair for Research, Department of Obstetrics and Gynecology, Oregon Health & Science University, Portland.
Dr. Speroff is Professor Emeritus of Obstetrics and Gynecology, Oregon Health & Science University, Portland.
Dr. Jensen reports he is a consultant for and receives grant/research support from Bayer, Merck, ContraMed, and FHI360; receives grant/research support from Abbvie, HRA Pharma, Medicines 360, and CONRAD; and is a consultant for the Population Council. Dr. Speroff reports no financial relationships relevant to this field of study.
SYNOPSIS: Highly publicized results from the Danish database demonstrate an increase in the risk of breast cancer associated with current use of hormonal contraception. Consistent with prior research, the risk is small, confined to current users, and disappears following discontinuation.
SOURCE: Mørch LS, Skovlund CW, Hannaford PC, et al. Contemporary hormonal contraception and the risk of breast cancer. N Engl J Med 2017;377:2228-2239.
Prior research supports that hormonal contraception results in a small increase in the risk of breast cancer in current users, with no residual effect on risk following discontinuation. Since the doses of estrogen used in oral contraceptives have decreased, and a variety of new progestins have been introduced since the publication of the results of the Collaborative Group on Hormonal Factors in Breast Cancer in 1996,1 the group headed by Øjvind Lidegaard used the Danish Sex Hormone Register (DSHR) study, a nationwide cohort study that includes all women living in Denmark, to evaluate breast cancer risk with current lower-dose methods. They used the Danish Cancer Registry to identify primary invasive breast cancers and used other registries to obtain data on exposures and confounders.
The DSHR follows a cohort of Danish women between 15 and 79 years of age to assess the influence of hormone use on the risks of cardiovascular disease and cancer. The cohort includes all Danish women ≤ 49 years of age on Jan. 1, 1995, and ≥ 15 years of age as of Dec. 31, 2012 (a total of 1,837,297 women). Exclusions for prior cancer diagnosis, venous thrombosis, or treatment for infertility resulted in 1,797,932 women included in this study.
The National Prescription Register provided data on exposure. The authors categorized contraceptive use as current use or recent use (discontinuation within the previous six months) or previous use (discontinuation more than six months previously). Start of use was the date that the prescription was purchased, and end of use was 28 days after the last purchase. They assumed women used the levonorgestrel-releasing intrauterine system (LNG IUS) for four years unless the woman became pregnant or received a prescription for another hormonal contraceptive before the end of that time period. Both the etonogestrel and levonorgestrel contraceptive implants are available in Denmark. The authors did not differentiate between the two progestins, and lumped all implant use in a single category. The two implants have different duration of approved use, and the authors did not describe how duration of exposure was classified for users of this method. Women changing methods contributed to multiple categories according to their pattern of use as current, recent, or past users. Women who became pregnant beyond 22 weeks did not contribute data for months during the pregnancy and until six months after delivery. The authors did not evaluate breast cancer diagnosis during pregnancy.
To obtain information on potential confounders, the authors relied on other national registries. Notably, the National Health Register provided medical treatment codes, surgical codes, and discharge diagnoses. However, this database does not include baseline information on body mass, alcohol use, family history, smoking, or lifestyle. Similar problems exist with the other databases used to assess confounders. For example, information on family history was available only for women from the cancer registry (an undisclosed number), and data on body mass index (BMI) were available on fewer than half of the sample collected as part of the pregnancy registry.
The authors reported results based on a total of 11,517 cases of invasive breast cancer that occurred over 19.6 million women-years of follow-up during the interval of study. Compared to never users of hormonal contraception, current and recent users of hormonal contraception showed a small increase in risk (relative risk [RR], 1.20; 95% confidence interval [CI], 1.14-1.26). This risk was not increased with less than one year of use 1.09 (95% CI, 0.96-1.23), but became significant after one year, peaking at 1.38 (95% CI, 1.26-1.51) after 10 years of use. Overall, the increased risk stopped rapidly after discontinuation of a hormonal method, although one analysis suggested that women using hormonal contraception for five to 10 years maintained a small increased risk up to 10 years following discontinuation (RR, 1.3; 95% CI, 1.06-1.58).
Comparing the risk estimates of the various hormonal contraceptive methods and progestins yields interesting and confusing results. For example, while levonorgestrel combined pills were associated with a significant increase in risk, low androgen desogestrel pills showed a lower point estimate, and the risk associated with drospirenone pills was lower yet and not significant. Although this suggests an androgen effect could modify risk, one of the highest point estimates was seen with the antiandrogenic cyproterone pill. What about an estrogen effect? No risk discrimination was observed between 50 mcg and 20 to 40 mcg ethinyl estradiol pills, and the highest risk estimate of all combined pills was associated with the recently introduced estradiol valerate/dienogest pill. What about a progestin dose effect? Low-dose LNG progestin-only pills (30 mcg) showed a higher risk (RR, 1.93; 95% CI, 1.18-3.16) than combined (100 to 150 mcg) LNG pills (RR, 1.33; 95% CI, 1.20-1.48), with the LNG IUS (20 mcg/d release) only slightly lower (RR, 1.21; 955 CI, 1.11-1.33). Adding to the confusion, the authors reported no increased risk associated with use of contraceptive implants and depo medroxyprogesterone acetate. Keep in mind the point estimates for all methods are low (below 2.0), with overlapping confidence intervals.
The authors concluded that their results support that current low-dose hormonal contraceptives increase the risk of breast cancer, and they recommended that women consider nonhormonal methods to reduce this risk. To put this risk in perspective, they noted the overall absolute increase in breast cancers diagnosed among current and recent users of any hormonal contraceptive in the study was 13 (10 to 16) per 100,000 person-years. This translates to approximately one additional breast cancer for every 7,690 women using hormonal contraception for one year.
I first learned of this paper from a New York Times headline delivered to my smartphone. Who could blame the Times for taking the bait? A prestigious medical journal reports a link between hormonal contraception use and breast cancer, topics of interest to virtually all women and many men. This drives readers, revenue, citations, and impact factor. Unfortunately, this sensational coverage highlights risk and lacks critical review, frequently resulting in concern from patients leading to discontinuation of hormonal contraception.2
I am delighted that Leon Speroff, the former editor and founder of OB/GYN Clinical Alert agreed to share his thoughts and long experience with hormonal contraception and contribute to this commentary. Leon noted that the greatest strength of this study is the large number of participants. This yields high statistical precision. At the same time, the larger the number of women in a study, the greater the effect of confounding variables. Table 1 provides a summary of common risk factors for breast cancer from the American Cancer Society. As I have mentioned repeatedly in my criticism of results from the Danish Registry studies, the absence of complete information on important baseline confounders limits the conclusions possible from database studies. While the authors claim to adjust for baseline confounders, a careful review of the supplemental material (available online) demonstrates that the various Danish registries have incomplete information on family history, BMI, and smoking for most of the women, and completely lack information on lifestyle choices, such as alcohol use, making adequate adjustment impossible. Since the study design does not effectively control for baseline confounding, the weak associations reported should be interpreted with extreme caution.3
Overall, Mørch et al reported a small increased risk of breast cancer confined to current/recent but not past users of hormonal contraception. This supports the hypothesis that hormonal therapy may promote the growth of pre-existing breast cancers, leading to early detection. This result first was described in the Collaborative Reanalysis study from 1996.1 The rapid disappearance of excess risk after discontinuation of use among women who used hormonal contraception for short periods is what you would expect to see if early detection is occurring with short-term use. The same effect is seen in the results with levonorgestrel-releasing IUDs; the increase was significant with short-term use and failed to reach statistical significance with use greater than 10 years. This raises other questions regarding attribution of risk. The analysis assumed that a woman used the LNG IUS for four years unless she received a prescription for a different hormonal method, potentially increasing exposure.
Leon pointed out that the authors’ conclusion that the risk increased with longer durations of use is especially vulnerable to the unknown effect of non-assessed confounders. To reach this conclusion, they relied on a “bias analysis” that assumed confounders not assessed were highly prevalent in the population and strongly associated with breast cancer. This convenient assumption lacks rigor. The database design simply does not permit us to adequately assess the effect of unmeasured confounders strongly associated with breast cancer risk: the degree of positive family history (number of first-degree relatives), the presence of inherited mutations, early menarche, breast feeding, and alcohol consumption.4 As mentioned earlier, data on BMI were available only for parous women. Although the authors stated that risk increased with duration of use, these results only achieved statistical significance among older women (> 35 years, see supplemental tables). Attempting to parse out the effects of various treatments over many years without rigorous control for confounding renders this conclusion particularly weak. We also believe the authors missed an opportunity to report the risk of breast cancer diagnosis during and following pregnancy among nonusers of hormonal methods.
In our summary of the results above, we pointed out our confusion with the lack of dose response with estrogens and progestins, and the absence of any effect with contraceptive implants and low androgenic progestins, such as drospirenone and desogestrel. The results showing some of the highest risk estimates with LNG products deserves comment. We find it ironic that the Lidegaard group repeatedly has made the case for the cardiovascular safety of LNG pills using the same database. Could the negative press regarding thrombosis risk have resulted in a prescription bias driving women at higher risk for breast cancer to the use of progestin-only LNG products? A well-recognized phenomenon is preferential prescribing. Clinicians tend to prescribe newer products or perhaps progestin-only products to patients with known risk factors. The effect of this phenomenon was not and probably could not be assessed in this study, but it may explain the high point estimate seen with the estradiol valerate/dienogest pill.
The authors recognize that their numbers, if real, translate into a very low number of additional cases of breast cancers in users of hormonal contraception. When the statistical conclusions of analyses are in the risk range of 1 to 2, clinicians should be skeptical that there is real clinical meaning in the numbers. Add to this the uncertainty in this study regarding adjustment for the effect of all factors that influence breast cancer risk, we see no reason to change current practice or patient recommendations.
Clinicians and patients also should find reassurance in the long-term safety data published last year from the UK Royal College of General Practitioners’ Oral Contraception Study that began in the late 1960s and involves up to 44 years of follow-up.5 In this cohort, ever use of oral contraceptives was associated with reduced colorectal, endometrial, ovarian, and lymphatic/hematopoietic cancers. An increased risk of lung cancer was seen only among ever users who smoked at recruitment. An increased risk of breast and cervical cancer was seen only in current and recent users, and not in past users. These data strongly support that the overall risk of cancer is not increased with the use of hormonal contraception.
- Collaborative Group on Hormonal Factors in Breast Cancer. Breast cancer and hormonal contraceptives: Collaborative reanalysis of individual data on 53 297 women with breast cancer and 100 239 women without breast cancer from 54 epidemiological studies. Lancet 1996;347:1713-1727.
- Shelton J. Scientists and the media must give a balanced view. Nature 2011;479:7.
- Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet 2002;359:248-252.
- Kim HJ, Jung S, Eliassen AH, et al. Alcohol consumption and breast cancer risk in younger women according to family history of breast cancer and folate intake. Am J Epidemiol 2017;186:524-531.
- Iversen L, Sivasubramaniam S, Lee AJ, et al. Lifetime cancer risk and combined oral contraceptives: The Royal College of General Practitioners’ Oral Contraception Study. Am J Obstet Gynecol 2017;216:580.e1-e9.