Choosing an SSRI in pregnancy: Clinical context versus statistical significance

A meta-analysis of selective serotonin re-uptake inhibitor (SSRI) medications and risks of congenital malformations published in this Journal by Myles et al. (2013) concluded that fluoxetine and paroxetine should be avoided in the first trimester of pregnancy because they were associated with small but statistically significant increases in birth defects and specifically cardiac defects following paroxetine exposure. Conversely and reassuringly, sertraline and citalopram were not found to be associated with such risks.

Unfortunately, a simplistic statement of avoiding an antidepressant medication in the first trimester seems inadequate to cover the many complexities of clinical practice, including whether or not a pregnancy is planned and weighing up the risks to both mother and baby of untreated mood disorder against the potential risks of medications. Odds ratios of 1.14 (fluoxetine) and 1.29 (paroxetine) are not necessarily deal-breakers in terms of dictating best practice and it is difficult to be convinced about the clinical significance of these findings when one considers the data on which they are based.

Because of obvious ethical concerns, drug companies are unable to include pregnant women in randomised controlled studies looking at reproductive outcomes following medication exposures. Therefore, data regarding pregnancy outcomes must be obtained from other sources, including prospective observational studies, retrospective case-controlled studies, case reports and series, population prescription and birth defects registries, spontaneous drug company reports as well as drug company registries. All of these methods have their pros and cons but, needless to say, none is a true ‘gold standard’. Although meta-analysis has some advantages, there are still some inherent problems because of the nature of the data being analysed.

Although they were not designed to assess drug safety, databases which link prescriptions and birth defects are being increasingly used worldwide to determine pregnancy outcomes following exposures. Despite their statistical power in analysing large population datasets, these studies have significant methodological problems, including, first and foremost, the assumption that just because a prescription was filled the medication was actually taken (and in the critical first trimester) during the pregnancy. It is well recognised that adherence or compliance to prescribed medication is sub-optimal during pregnancy (Sawicki et al., 2011) and may be even lower for antidepressants. At least one study has shown that despite filling a prescription, many women chose not to actually take their prescribed antidepressant medication (Kallen et al., 2011). However, these women would have been included as ‘exposed’ on a prescription database. This suggests that exposures are likely to be over-estimated and thus makes it impossible to determine whether infants with defects were truly exposed. Given the paucity of details about the malformations reported in most of these studies, it is also very difficult to assess either their severity or clinical significance and thus contextualise the risks in a meaningful way for both patients and their health-care providers.

Some factors that can affect estimates of rates of birth defects are whether or not prenatally diagnosed anomalies (which may result in termination of pregnancy) are included, the length of follow-up from birth, specific surveillance of infants, as well as recall and ascertainment bias. Studies have shown that the rate of birth defects is around 3% when reporting is up to 12 months of age but can be as high as 6% if the period of surveillance is extended until the age of 5 years. For example, if the cut-off for reporting of birth defects is 1 year then a child with unilateral ureteric duplication would not be reported if asymptomatic until the age of 2 years. However, if reporting is until age 5 and the same child develops a urinary tract infection which requires long-term antibiotic treatment and urological intervention at age 3, then this would be reported as a major malformation.

Likewise, post-natal echocardiography may be routinely ordered because of maternal exposure to an SSRI in a healthy neonate with no cardiac murmur and may detect a small asymptomatic ventricular septal defect (VSD) which would go undetected in a baby not exposed to an SSRI where there were no specific concerns about cardiac defects and therefore no post-natal cardiac ultrasound performed.

Given that in retrospective case–control studies much of the data are obtained by interview/questionnaire months or even years after the birth of an affected child, there may be significant recall bias in terms of the nature and timing of exposure. It is well recognised that mothers of affected infants may have inaccurate recollection of what they did or did not take during the critical period of embryonic development.

In 2005, concerns were raised by the American Food and Drug Administration (FDA) following reports of an increase in the rates of cardiovascular malformations, predominantly ventricular and atrial septal defects (VSD and ASD) following exposure to paroxetine during pregnancy. As a result, paroxetine was re-classified by the ADEC (Australian Drug Evaluation Committee) and the FDA from category C to D in 2005. Subsequently, there have been numerous studies that have shown both positive and negative associations between the use of paroxetine (and other SSRIs) and congenital cardiac defects. A meta-analysis by Wurst et al. (2010) of the most published studies on pregnancy outcomes following paroxetine exposure had very similar findings to those reported in this study, with a prevalence odds ratio for cardiac defects of 1.46 (95% confidence interval 1.17–1.82) and for all defects of 1.24 (95% confidence interval 1.08–1.43), although, somewhat inexplicably, a prospective study of over 1100 paroxetine-exposed pregnancies which did not show an increased risk of cardiac or other malformations was not included in that meta-analysis (Einarson et al., 2008).

The authors suggest re-categorising fluoxetine to category D as was done for paroxetine in 2005, based on concerns about its teratogenicity. However, there are some real concerns about the A–X categorisation of drugs and this should not be used as a bible to dictate prescribing but rather as a guide (Kennedy, 2011). Countless women are counselled erroneously (and directively) about the safety or otherwise of a particular medication based on misinterpretation of the categories.

When counselling my patients about the risks of medication use in pregnancy I always start with the statement of background risk (i.e. that no pregnancy comes with a guarantee and that every couple has a 3% chance of having a live-born baby with a major birth defect and that this is as high as 5% if other problems including neurodevelopmental sequelae are included) .

Of all categories of birth defects, those involving the cardiovascular system are the most common, affecting between 1/100 and 1/200 live-born babies (i.e. 0.5–1%). If one assumes the worst-case scenario, as suggested in this paper, which is that taking SSRIs such as paroxetine and fluoxetine increases that risk (and for the sake of convenience let’s call it a doubling of that risk although the risk appears less than that), then this translates to a relative risk of 2. What this means for a couple is that their absolute risk of having a baby with a major birth defect is now 3.5–4% – compared to the background risk of 3%. Clearly it sounds far more worrying to the patient (and her doctor) to be told her risks of having a baby with a heart defect are twice that of her friend who doesn’t take that particular SSRI, but putting it as an absolute risk makes it a far more realistic and less daunting prospect for all concerned.

Thus, based on this meta-analysis, the ‘high-risk’ SSRIs paroxetine and fluoxetine, with hardly earth-shatteringly elevated odds ratios of 1.14 and 1.29, respectively, are not sufficiently of concern (at least in my opinion) to necessarily dictate a change in prescribing practice.

Agreed that the data suggest that if prescribing a woman of reproductive age an antidepressant for the first time, then paroxetine or fluoxetine would probably not be the SSRI antidepressants of first choice, but again there may well be other more compelling therapeutic reasons to influence such decisions, including side-effect profile and efficacy.

MotherSafe, the largest Teratogen Information Service in Australia, analysed all calls regarding psychotropic drugs since its establishment in 2000 and noted a statistically significant reduction in the number of calls regarding both paroxetine and fluoxetine, with a concomitant increase in calls regarding the newer SSRI antidepressants citalopram and escitalopram in recent years as compared with the earlier years of the service (Eamus et al., 2010).

This change is consistent with findings from the USA (Mitchell et al., 2011) as well as recently published Australian psychotropic prescribing data which also noted a decrease in paroxetine prescribing over this period and attributed this, at least in part, to concerns about teratogenicity (Stephenson et al., 2013).

Despite this there will be a number of women who are best treated by either fluoxetine or paroxetine and I do not believe that this meta-analysis (or in fact any publication to date) has provided us with sufficiently robust data to justify the recommendation that women should be advised not to fall pregnant while taking these medications or should switch to another antidepressant once pregnant for perceived ‘fetal safety reasons’. It is also worth noting that there are more long-term neurodevelopmental follow-up data available about children exposed in utero to fluoxetine than many of the newer antidepressants and for some women this reassuring information may outweigh the slightly higher potential risks of cardiac defects.

In summary, it is important that decisions weighing up the relative risks and benefits of the optimal treatment of anxiety and depression during pregnancy are made by women and their health-care providers based on clinical context and that they are aware of the quality and limitations of the data on which their so-called evidence-based decisions are made.