Friday, February 11, 2011

Doctor’s Orders

Last month marked the 38th anniversary of the United States Supreme Court decision in the case of Roe v. Wade, which established abortion as a woman’s fundamental right under the U.S. constitution. President Obama spoke on his support of the decision, asking Americans to, “recommit ourselves more broadly to ensuring that our daughters have the same rights, the same freedoms, and the same opportunities as our sons to fulfill their dreams.” Normally, I couldn’t have cared less about relaying such information here. As its name asserts, this blog deals with science. I try to avoid politics. It’s messy and uncouth. But sometimes politics worms its way into science and I am thus coaxed into acknowledging it.

Due to recent shifts in the political balance of power in this country, the abortion debate has been more front and center than usual. It has figured prominently in challenges to the health care act and is a popular topic in conservative rhetoric preparing for the next presidential election. But, as I said, I try to stick to science here. So I’m not going to write about how, in a country where 35% of women live in counties without a single abortion provider, state laws requiring in-person counseling followed by a 24-hour waiting period create an undue burden for lower-income women. Nor will I explain how excluding abortion coverage from federally funded health programs similarly punishes the poorest women, for whom the average $450 cost of a non-hospital abortion can be prohibitively high. These are matters of politics.

Instead, I will focus exclusively on the science behind certain states’ abortion policies. While federal law overrides any state laws banning abortion, states have the power to create some restrictions to abortion access. Often these take the form of waiting periods and parental consent for minors, but several states also have laws mandating that health care providers “inform” women seeking abortions of one or both of the following; A) that abortion increases their risk of developing breast cancer and B) that abortion can cause depression and other mental health problems. These are frightening things to be told and one would hope that lawmakers would not create these policies without good reason. But is there any evidence behind such claims? Well, it sort of depends on what you’re willing to accept as evidence.

The trouble with many of the studies on both of these alleged links is that they are correlation studies. Randomized controlled trials are the approach used in much of medical testing (efficacy of new drugs for instance), but it is considered unethical to randomly assign abortion/non-abortion outcomes to pregnancies in human subjects. The only way to examine something in which subjects rather than researchers makes such choices is by looking for correlations between two variables. For instance, are individuals diagnosed with lung cancer more likely to be life-long, heavy smokers than those without the disease? Easy enough. However, you can’t actually infer causality from correlation. Such inferences can miss a third variable that is responsible for the other two. A favorite teaching example of this error is the conclusion that ice cream consumption causes drowning, because rates of both rise and fall in the same pattern. Of course, it is a third variable – hot weather – that increases both swimming (and therefore chance of drowning) and the interest in frozen desserts. Interpretation of correlation data can also mistake a symptom for a cause - for instance, the assumption that marriage counseling causes divorce. What is useful about correlation studies is that they can uncover relationships (or lack thereof) that suggest causality. But the only way to verify cause and effect is with actual experimentation.

Case-control vs. Cohort
In addition to the pitfalls of interpretation mentioned above, correlation studies also vary in format. In case-control studies, a group of people with a particular diagnosis (case subjects, such as women with breast cancer) are matched with another group without the diagnosis (control subjects, such as women without breast cancer) and both groups are reviewed for a variable that is a suspected link (such as reproductive history). A cohort study instead looks at a large population and records a variety of life-history information, and then examines the correlation between variables of interest. A cohort can be a group of individuals living in a particular country, or working in a particular profession, and so on. Often, multiple studies can be created from data gathered on one cohort. Cohort studies are considered to be more reliable than case-control studies.

Retrospective vs. Prospective
One of the reasons case-control studies can be inferior to cohort studies is that the former is generally retrospective, while the latter can be either retrospective or prospective. A retrospective study begins with the outcome (cancer) and looks backward in time to find the potentially correlating variable (reproductive history). Often this is done through interviews or questionnaires, which can be biased. Conversely, prospective studies start with the variable of interest and then follow a cohort to see if a particular outcome occurs. While they can be lengthy and expensive to conduct, prospective cohort studies are about as reliable as it gets in the world of correlation.

While sporadic research into a possible connection between breast cancer risk and reproductive factors goes back to the early 20th century, interest in the effect of abortion really picked up during the 1980’s. Doing a literature search on the subject, one encounters a sea of contradictory conclusions, with some finding increased risk, others finding no correlation and some actually reporting decreased risk. As you may have guessed, much of this frenzied, first wave of research took the form of retrospective case-control studies. More recent prospective cohort studies have greater consistency in their results - usually finding no correlation between abortion and breast cancer.

Research interest in this subject was probably fueled in part by controversy over abortion, but it is not unreasonable to consider whether there might be a connection. There has been some indication that full-term pregnancies early in life confer protection against breast cancer. Conversely, pregnancy later in life (and, it turns out, post-menopausal hormone replacement therapy) can increase cancer risk. Since the hormones associated with pregnancy can have positive and negative repercussions, it is worth examining what effect an incomplete dose of these hormones would have on women’s health. But recall that not all incomplete pregnancies are the result of induced abortions. Spontaneous abortions (aka miscarriages) are another possible outcome of pregnancy. Interestingly, studies looking at spontaneous abortions generally find no correlation with breast cancer.

Why should induced abortion have greater correlation with breast cancer than does miscarriage? Well, now we come back to the problematic biases of the dreaded retrospective study. As you may have noticed, induced abortion carries with it more social stigma than miscarriage. Women who obtained abortions are thus more likely to omit this information than women who suffered miscarriages. Additionally, it has been suggested that the tendency to inaccurately report induced abortion history may differ in women who have and have not been diagnosed with breast cancer. Like many people with life-threatening diseases, women with breast cancer, hoping to provide accurate information that might contribute to disease cure or treatment, are more inclined to admit to having had an abortion. Healthy women lack this motivation for honesty. Thus, in retrospective studies, women with breast cancer may report a greater number of abortions not because their rate is actually higher than healthy women, but due to their greater willingness to disclose the information.

One way of getting around the problem of biased self-reported data is to rely exclusively on medical records. Several interesting cohort studies have been conducted using the handy national registries of Denmark. Unlike the U.S., Denmark provides its citizens with universal health care, and all such medical transactions (cancer diagnosis, mental health care, births, abortions, etc.) are logged into various registries. Information from the registries can be anonymously matched using numbers assigned by the Civil Registration System. A 1997 study published in The New England Journal of Medicine used the Danish registries to follow one and half million women. They found no correlation between induced abortion and breast cancer.

While the volume of correlations studies of this subject continues to increase, there is still a conspicuous absence of a demonstrated mechanism for how induced abortion might impact cancer development. Several pregnancy hormones have been associated with the cancer-protective benefits of pregnancy. One of these is human chorionic gonadotrophin (hCG), which actually peaks during the first trimester of pregnancy rather than lingering throughout like estrogen and progesterone. Since pregnancy hormones have been implicated in both increase and decrease of cancer risk, establishing a mechanism would be essential to demonstrating a link between abortion and breast cancer, particularly given the absence of any good correlation evidence pointing in that direction.

Psychology can be a more nebulous field than cancer research, and so data gathering for depression presents additional challenges. However, there is one advantage. Unlike cancer, whose risk rises with age, many mental illnesses emerge during youth and persist throughout the course of patients’ lives. When examining a proposed causative link between abortion and depression, researchers at least have the luxury of looking at what their subjects’ state of mental health was like prior to an abortion or pregnancy. Stunningly, numerous studies opt not to gather such data. A 2008 study published in The Journal of Psychiatric Research reported a connection between abortion and incidence of various mood, anxiety and substance abuse disorders.§ However, the authors only looked at mental health status after abortion. Despite having no data on subjects’ psychiatric condition prior to abortion, and thus no reason to even suspect (much less proclaim) causality, they felt comfortable writing that, “…abortion is responsible for more than 10% of the population incidence of alcohol dependence, alcohol abuse, drug dependence, panic disorder, agoraphobia, and bipolar disorder…”.

The New England Journal of Medicine recently published a more thorough study on the alleged link between mental disorders and abortion. Once again, Demark’s national registries supplied data that was free from the possible biases of self-reported information. The cohort study looked at “psychiatric contact” – a visit to mental health professionals – in 2 groups of women; those undergoing abortions and those giving birth. Data were taken both before and after these reproductive events. The population of women who had abortions showed no significant change in rate of psychiatric contact afterward, while group of women who gave birth showed an increase in psychiatric visits. The latter should come as no surprise, since postpartum depression and its rarer cousin postpartum psychosis are well documented phenomena. While the study found no link between psychiatric contact increase and abortion, it did show a greater overall incidence of psychiatric contact (ie. both before and after) in the women who had abortions than in women who gave birth. Had the authors only gathered data only from after abortions and births, they too could have reported a link between abortion and psychiatric disorders. But seen in full context, the data suggest entirely different possibilities; such as that depressed women may be more likely to seek abortions. This study should cast doubt on correlation studies claiming to find a link between abortion and mental disorders but failing to include before and after data on mental health.

The New England Journal of Medicine study has its problems too. It does not, for instance, include data on women who neither gave birth nor had abortions. Additionally, taking data from the national registry rather than through interviews, it has no way of even attempting to separate wanted pregnancies from unwanted ones. It has been suggested that what is potentially stressful to emotional health is not abortion but the experience of an unwanted pregnancy. Some studies focusing on three categories of women – abortion, birth and no pregnancy – have reported that while women who obtained abortions had a higher incidence of mental health problems, it was those who never got pregnant in the first place that were the most psychologically healthy. The fact that women who gave birth fell somewhere in between the troubled abortion group and the emotionally stable non-pregnant group, might suggest that some of these births were the result of stressful unwanted pregnancies. Or alternately, it could indicate that more emotionally stable women are less likely to become pregnant unintentionally. Ah, the vagueness of correlation studies.

Science journalists often report on interesting new findings that are far from established facts, but it’s hard to think of any other subject for which isolated and contested data is converted into policy. Doctors advise us not to smoke and to reduce our consumption of saturated fat, but they shy away from recommending a daily dark chocolate supplement or herbal remedies with no demonstrated efficacy in clinical trials, and the government has yet to instruct them to do otherwise. There is no demonstrated link between abortion and either breast cancer or depression, and no reputable medical organization claims that such a links exist. And yet multiples states have opted to jeopardize the doctor-patient relationship by forcing health care providers to be deliverers of political ideology disguised as the protection of informed consent.

At no point in my reading did I encounter any single perfect study that addressed all variables and decisively established or refuted either of the anti-choice movement’s purported health risks of abortion. But one wouldn’t expect to find such a thing. That’s not how medical research works. However, one might expect to find numerous, well-designed and reasonably consistent correlation studies pointing the way to further lab-based research on possible mechanisms. And one might expect that, during this period, the research would not be presented to patients as proven and indisputable truths. And one would certainly expect that doctors would not be required by law to lie to their patients about these un-established risks, and that patients would not be required to lie to their doctors that they accepted and understood the confusing misinformation given to them. In short, one would expect less politics and more science.

* If you already know everything there is to know about research methodology, I encourage you to skip this section. And if you know even less than I’m giving you credit for, check out this useful glossary compliments of The British Medical Journal.

In case you’re wondering, the answer is yes.

hCG is what pregnancy tests are checking for and is also the prime suspect in the cause of “morning sickness” during pregnancy.

§ The data were taken from National Comorbidity Survey (NCS) interviews conducted in the early 1990s.

Who told you this?

Guttmacher Institute website: (U.S. abortion statistics)

National Cancer Institute website:

Henderson, K. et al. 2008. “Incomplete pregnancy is not associated with breast cancer risk: the California Teachers Study.” Contraception 77: 391-396.

Michels, K. B. et al. 2007. “Induced and Spontaneous Abortion and Incidence of Breast Cancer Among Young Women. A Prospective Cohort Study.” Arch Intern Med 167: 814-820.

Reeves, G. K. et al. 2006. “Breast cancer risk in relation to abortion: Results from the EPIC study.” Int. J. Cancer 119: 1741-1745.

Melby, M. et al. 1997. “Induced Abortion and the Risk of Breast Cancer.” The New England Journal of Medicine 336: 81-85.

Jacobson, H.I. et al. 2010. “A Proposed Unified Mechanism for the Reduction of Human Breast Cancer Risk by the Hormones of Pregnancy.” Cancer Prevention Research 3: 212-220.

Munk-Olsen, T. et al. 2011. “Induced First-Trimester Abortion and Risk of Mental Disorder.” The New England Journal of Medicine 364: 332-339.

Coleman, P. K. et al. 2009. “Induced abortion and anxiety, mood, and substance abuse disorders: Isolating the effects of abortion in the national comorbidity survey.” Journal of Psychiatric Research 43: 770-776.

Dwyer, J. M. and Jackson, T. 2008. “Unwanted pregnancy, mental health and abortion: untangling the evidence.” Australia and New Zealand Health Policy 5: 1-6.

Lazzarini,, Z. 2008. “South Dakota’s Abortion Script – Threatening the Physician-Patient Relationship.” The New England Journal of Medicine 359: 2189-2191.

Jasen, P.2005. “Breast Cancer and the Politics of Abortion in the United States.” Medical History 49: 423-444.


  1. Bravo, Alex. Well done.

  2. Bravo, Alex, indeed. Excellent job, demonstrating what sorts of things various kinds of scientific studies can (or cannot) assert. But then, of course, politics enter the scene and all these fine points go out the window. “Everyone is entitled to their own opinions, but they are not entitled to their own facts” - thus spoke Daniel Patrick Moynihan in the olden days. I’m sure he meant well. But the opinions that do not follow from facts are many and their consequences are forced on us. Our recently voted out chairman of Texas Board of Education believed the world was created 6000 years ago and wanted to educate kids accordingly. The craziness of forcing doctors to broadcast non-sequiturs to their patients is not alone. It is a part of a large and seemingly growing craziness.