Incidence and recurrence of common mental disorders after abortion: Results from a prospective cohort study

Abstract

Research in the field of mental health consequences of abortion is characterized by methodological limitations. We used exact matching on carefully selected confounders in a prospective cohort study of 325 women who had an abortion of an unwanted pregnancy and compared them 1-to-1 to controls who did not have this experience. Outcome measures were incidence and recurrence of common DSM-IV mental disorders (mood, anxiety, substance use disorders, and the aggregate measure ‘any mental disorder’) as measured with the Composite International Diagnostic Interview (CIDI) version 3.0, in the 2.5–3 years after the abortion. Although non-matched data suggested otherwise, women in the abortion group did not show significantly higher odds for incidence of ‘any mental disorder’, or mood, anxiety and substance use disorders, compared to matched controls who were similar in background variables but did not have an this experience. Having an abortion did not increase the odds for recurrence of the three disorder categories, but for any mental disorder the higher odds in the abortion group remained significant after matching. It is unlikely that termination of an unwanted pregnancy increases the risk on incidence of common mental disorders in women without a psychiatric history. However, it might increase the risk of recurrence among women with a history of mental disorders.

Introduction

Over the last decade, a number of reviews on mental health consequences of abortion have been conducted (APA, 2008, Charles et al., 2008, Robinson et al., 2009, Steinberg and Russo, 2009, National Collaborating Centre for Mental Health (NCCMH), 2011, Coleman, 2011). These reviews showed that the field is severely hampered by methodological problems. With the exception of one review (Coleman, 2011), which has been critically refuted (e.g., Abel et al., 2012, Polis et al., 2012, Steinberg et al., 2012), all concluded that the highest quality empirical research so far has found no evidence that abortion increases the risk on mental disorders. As experiments would be unethical in this field, all claims are based on observational research. Even though some of these studies make use of carefully chosen reference groups and rigorously adjust for covariates, the methodological literature shows that other strategies, like matching, could offer additional insight (Kessler and Schatzberg, 2012), especially in situations in which standard regression methods perform poorly due to insufficient overlap in covariate distributions (Stuart, 2010, Rosenbaum and Rubin, 1983, Cook et al., 2008, Iacus et al., 2012). In the current study, we chose to use a prospective cohort design with 1-to-1 matching, to address a number of the most commonly mentioned methodological problems in this area of research.

One methodological problem in this field is that pre-existing mental health problems are inadequately measured, even though it is likely that these are predictive of post-abortion mental disorders (Gilchrist et al., 1995, Major et al., 2000). Former studies have shown that lifetime psychiatric disorders are highly prevalent among women having abortions (Van Ditzhuijzen et al., 2013, Steinberg et al., 2014), which could lead to incorrect attribution of these disorders to the abortion. Several studies have adjusted for pre-existing mental disorders, but in these studies abortion history was assessed through retrospective self-report (Mota et al., 2010, Fergusson et al., 2009, Steinberg and Finer, 2011, Steinberg et al., 2014), so that the order of abortion and mental disorder cannot be ascertained accurately (Charles et al., 2008, APA, 2008, Major et al., 2009, National Collaborating Centre for Mental Health (NCCMH), 2011). Other studies measured abortion history prospectively and controlled for pre-existing mental disorders, but these measured only a few disorders (Major et al., 2000, Biggs et al., 2015), or previous mental disorders were assessed by family doctors (Gilchrist et al., 1995) or based on psychiatric contact (Munk-Olsen et al., 2011), which could lead to serious underreporting of mental disorders. These findings should therefore be complemented with research using a strong structured diagnostic instrument to assess a wide variety of clinical-level mental disorders, hereby taking into account the onset of these disorders in order to determine whether disorders were already present before the pregnancy was terminated. The current study fills this gap.

A second serious threat to validity in this type of research is confounding by covariates (APA, 2008; NCCMH, 2011). Since it is impossible to randomly assign women to an abortion or control condition, and there is no rigorous way to guarantee comparability between groups (Kessler and Schatzberg, 2012), alternative methods to regression adjustment could offer additional insight. One of the techniques that is strongly recommended in the methodological literature is matching, which allow one to analyze a non-experimental study so that it mimics characteristics of an experiment, because the distribution of baseline covariates is similar between groups (Austin, 2011). In the current study, we matched participants 1-to-1 on confounding covariates. This type of matching induces balance in the distributions of the measured covariates between the two cohorts, and simulation studies have also shown that it can reduce imbalance in unmeasured covariates considerably (Cook et al., 2008, Stuart, 2010, Stürmer et al., 2010).

A third methodological issue which has repeatedly been mentioned by review studies (e.g., APA, 2008, National Collaborating Centre for Mental Health (NCCMH), 2011), is the use of inappropriate reference groups, such as women who never had been pregnant, miscarried, or delivered a (presumably intended) pregnancy. The choice of reference group is dependent on the type of causal question (Charles et al., 2008). In this area of research, two different types of causal questions are often confused (Fergusson et al., 2009, Kessler and Schatzberg, 2012). The first question is whether abortion of an unwanted pregnancy has more, or less, adverse effects on mental health than its realistic alternatives (such as carrying the unwanted pregnancy to term). This question has recently been investigated by the Turnaway Study (e.g., Foster et al., 2015, Biggs et al., 2015, Rocca et al., 2013), using a reference group of women whose abortion requests were declined. This is an adequate design to investigate effects of abortion versus being denied an abortion, when the unwanted pregnancy is a given. The second question is whether the life event of termination of an unwanted pregnancy increases the risk on mental disorders. It is this question that is being addressed in the current study. In this study design, measures based on before-after intra-person comparisons are compared between groups which are matched on confounding covariates. This design provides information about the consequences of the whole life event, which is lost when focusing on the specific effect of abortion versus denial of abortion.

The main goal of this study is to investigate the incidence and recurrence of mental disorders after termination of an unwanted pregnancy. The research question is whether having an abortion of an unwanted pregnancy is associated with an increased risk for the development of (1) incident and (2) recurrent mental disorders at follow-up (2.5–3 years later). We matched women who had an abortion of an unwanted pregnancy 1-to-1 on confounding covariates to women from the general population who did not have an abortion.