On Antipsychotics in Pregnancy and the Problem of Confounding

Second-generation antipsychotics, also known as atypical antipsychotics, are frequently used in psychiatry, prescribed not just for psychotic disorders, but also for treatment of bipolar disorder, and for augmentation of depression or OCD treatment. Studies have shown that there is a very high risk of relapse of psychiatric disorders in pregnancy--for example, women with a history of bipolar disorder who discontinue their medications while pregnant have an 85% chance of recurrence of a mood episode while pregnant, compared to a 37% chance of recurrence when continuing to take their medications. Thus, it's important for women who are taking antipsychotics prior to pregnancy to know about the risks and benefits of continuing these medications while pregnant. Here we will focus on a recent article which looked at how taking antipsychotics in pregnancy may affect maternal and perinatal outcomes.

A large population-based cohort study looked at the question of whether or not antipsychotics affected certain maternal and perinatal health outcomes. It looked at all women who had delivered between April 2003 and December 2012 in Ontario, Canada. Data were collected from multiple linked healthcare administrative databases, which included information about medications women were taking, inpatient and outpatient claims, and demographic information. Antipsychotic exposure was defined as at least 2 consecutive filled prescriptions for an antipsychotic between conception and delivery. There were a total of 41,523 women in the study. Of these women, 1,209 were prescribed an antipsychotic, and 40,314 were not. The women were compared in 2 ways--there was an unmatched cohort, and a matched cohort. In the unmatched cohort, maternal and perinatal outcomes between women who were prescribed an antipsychotic and those who were not were compared directly. In the matched cohort, 1,021 women who were prescribed antipsychotics were matched with 1,021 women who were not, using something called a high dimensional propensity score (HDPS) algorithm.

What is a high dimensional propensity score (HDPS), you ask? It's a method commonly used in pharmaco-epidemiological studies to create less biased estimates of treatment effects. In other words, these epidemiological studies are subject to many confounders--things like age, ethnicity, multiple gestations, multi-parity, pre-existing medical problems like hypertension, diabetes, renal disease, smoking, drug use, other medications, and number of healthcare visits. Many of these confounders are observed, or known (e.g., it's easy to determine age from the databases). Researchers sometimes create a propensity score, which is a summary confounding score. The HDPS algorithm attempts to further minimize residual confounding by also incorporating proxy variables (healthcare diagnoses, procedures, and drug claims), which, when combined, can account for important confounders that are unobserved. This is extremely important, because without controlling for the unobserved confounders, we cannot know if a certain maternal or perinatal outcome is due to medication exposure, or to something else. In fact, women who take medications during pregnancy are often very different from women who do not; they tend to have more severe illness with higher risk of relapse...without controlling for this, we won't know if an outcome is related to a woman's underlying illness (severity), or to the medication she took. And this is how the HDPS is very useful.

Back to the study: the researchers looked at the following maternal medical outcomes: gestational diabetes, hypertensive disorders of pregnancy (gestational hypertension, pre-eclampsia, eclampsia), and venous thromboembolism (VTE), and the following perinatal outcomes: preterm birth (<37 weeks gestation), and extremes of newborn weight (a birth weight <3rd percentile or >97th percentile for the same sex and gestational age). What the researchers found illustrates not only the possible effects of antipsychotics on pregnancy outcomes, but also the importance of controlling for confounding. In terms of maternal outcomes in the matched cohort (where HDPS was used), there were no differences in terms of gestational diabetes, hypertensive disorders of pregnancy, and VTE between women who took antipsychotics in pregnancy and those who did not. However, in the unmatched cohort, there was an increased risk of gestational diabetes and hypertensive disorders in the antipsychotic group. In terms of perinatal outcomes, there were similar differences between the matched and unmatched cohorts. In the matched cohort, the 2 groups had no differences in preterm birth, and extremes of newborn weight. However, in the unmatched cohort without HDPS, the infants exposed to antipsychotics had an increased risk of preterm birth and weight >97th percentile. The study thus shows that when these antipsychotic users were matched with non-users who had similar pre-pregnancy morbidity (e.g., weight,  other medical conditions, diagnoses, and other variables), then the risk was about the same for both groups. Here is an illustration of the potential magnitude of bias that is created by not controlling for confounding variables. However, it also showed that the absolute rates of adverse maternal and perinatal outcomes in antipsychotic users were higher than those in the general population. Thus, for women who require antipsychotics during pregnancy, it's imperative to closely monitor their health before and during pregnancy, paying close attention to diabetes, hypertension, preterm birth and fetal growth.