JAMA Psychiatry Publishes More Questionable Science

Priscilla K. Coleman, PhD

Last week JAMA Psychiatry irresponsibly published a study led by Julia Steinberg titled “Examining the Association of Antidepressant Prescriptions with First Abortion and First Childbirth” 

Using Danish population registries, relative to women with no history of abortion or childbirth, women who had a single abortion were found to have a higher risk of first-time antidepressant use when age, calendar year, prior history of psychiatric contact, prior history of antianxiety and antipsychotic medication. maternal and paternal psychiatric history, and maternal and paternal education levels were controlled. More specifically, prior to an abortion, women with an abortion had a 46%, 54%, 24%, and 12% higher risk for securing an antidepressant prescription compared to women who had not aborted in the year before an abortion, the year after an abortion, 1 to 5 years post-abortion, and 5 years post-abortion respectively. The authors claim that abortion is not a significant risk factor for depression because there was not a statistically significant change in prescription rates before and after abortion. There are numerous problems associated with the methodology employed and the conclusions drawn.   

First, the sample is not representative of women who choose abortion for the following reasons: 

1) The authors’ results are based on a mere 4.4% of the population of women examined as they looked exclusively at those who had only one abortion and no children during the study period.  

2) They eliminated women under age 18 who are at a well-documented increased risk of post-abortion psychological problems compared to other women.

3) They did not include women prone to depression, because those with a prior prescription for an antidepressant drug were not included.

Second, the assessment of pre-abortion antidepressant use should have extended back well beyond a year when data for 40 years is available. What is the justification? Obviously, the year leading up to an abortion is likely to be one of relationship and other challenges, enhancing the probability of depression manifesting.    

Third, the outcome measure of first antidepressant medication prescription is not an appropriate assessment for the experience of post-abortion depression. Antidepressant medications are most typically prescribed for clinical depression, as opposed to reactive depression resulting from a traumatic event.   

In the conclusion of their study the authors stated: “the risk of first-time antidepressant use for all times relative to an abortion decreased in the fully adjusted model relative to the basic model. This finding indicates that preexisting mental health conditions and the other covariates are confounding the association between abortion and antidepressant use.” However, there is a fundamental problem with this conclusion related to the statistical issue of multicollinearity, well-known to students of statistics and statisticians, but ignored by the authors and the peer-reviewers of this article.

Multicollinearity occurs when there are significant correlations between two or more predictor variables, meaning the variables tend to change in unison. For example, if one has been prescribed an antianxiety drug, then one is also likely to have received professional help for a mental health problem. This creates redundant information, which can easily skew the results in regression models. The basic problem is multicollinearity results in unstable parameter estimates which make it very difficult to assess the effect of independent variables on dependent variables; in this case abortion and antidepressant use. In the Steinberg and colleagues paper several of the predictor variables are likely to be highly correlated with each other, violating the independence assumption of the statistical test employed. The authors convey no understanding of this potentially serious violation and provide no data on the correlations among the set of predictors employed. The differences between the abortion and no abortion groups may have been even more pronounced had multicollinearity not been a likely issue.

In conclusion, all that the results indicate is women who abort once, compared to those who do not, are more likely to secure a first prescription for an anti-depressant drug both in the year before and for years after the procedure. This is a small fraction of women who choose abortion and more research is needed to examine biological pre-dispositions, life-circumstances, relationship histories, beliefs and feelings about the abortion, etc. to understand the reasons women in this small segment sought relief for depression prior to and after an abortion. Different sets of variables could be operative during the distinct time periods and a lack of difference in antidepressant rates pre- and post-abortion is inadequate data to conclude abortion does not lead to depression, as the news headlines suggest.