Pregnancy outcomes

Clinically recognized abortions were identified in the Danish National Hospital Registry. This registry contains data on in- and outpatient contacts in Denmark coded according to a Danish version of the 10th revision of the International Classification of Diseases (ICD-10) since 1995 [7]. Abortions (ICD-10: O02.0–O06.9) were divided in spontaneous abortions (ICD-10: O02.0–O03.9), induced abortions (ICD-10: O04.0–O05.2, O05.5–O06.9), or induced abortions due to fetal disease (ICD-10: O05.3, O05.4). Molar or ectopic pregnancies (ICD-10: O00.0–O01.9) were excluded from the main analyses. Furthermore, diagnoses regarding failed induced abortions (ICD-10: O07) were disregarded as we assume that a failed induced abortion was followed by an abortion, a stillbirth or a live birth, which were all diagnosed separately. Live births and stillbirths were identified in the Danish Medical Birth Registry, which holds information on all births in Denmark [8].

The pregnancy period was defined from the estimated date of conception to the date of the outcome (induced or spontaneous abortion, stillbirth or live birth). For abortions (spontaneous or induced), the gestational age was based on the Danish National Hospital Registry, and for the live births and stillbirths it was based on the Medical Birth Registry. Gestational age before 12 weeks was normally based on last menstrual period and on ultrasound scans for later gestations when available. Missing gestational age for abortions was replaced by the median of non-missing values of the gestational age for abortions (N = 7,125; 2.5% of all abortions).

We performed a hierarchical coding of the pregnancies to take repeated contacts into consideration. Any stillbirth or live birth resulted in recoding of other prior endpoints in the index pregnancy period. Pregnancies with multiple codes for abortion were coded as a spontaneous abortion if the pattern of codes indicated such, even if the index period also included codes for induced abortion (e.g. in case of a spontaneous abortion in the time period between the initial contact and the planned surgical termination). Pregnancies including codes for induced abortions due to fetal disease (ICD-10: Q05.3,Q05.4), were coded as such, and the remaining pregnancies with codes for abortion were coded as induced abortion not due to fetal disease.

Antidepressant exposure

We obtained information on all redeemed prescriptions in Denmark from the Registry of Medicinal Product Statistics. We included information on all redeemed prescriptions from 1 January 1996 to 31 December 2008. Antidepressants were defined according to the Anatomical Therapeutic Chemical (ATC) code N06A, and we specifically assessed the prescriptions of selective serotonin reuptake inhibitors (SSRI), ATC code: N06AB, and tricyclic antidepressants (TCA), ATC code: N06AA, and ‘other’ (serotonin–norepinephrine reuptake inhibitors (SNRI), ATC codes: N06AX16 and N06AX21; noradrenergic and specific serotonergic antidepressants (NaSSA), ATC codes: N06AX03 and N06AX11; and norepinephrine-dopamine reuptake inhibitors, ATC code: N06AX12). Pregnancies were considered “exposed” to antidepressants if the mother had redeemed a prescription for antidepressant medication at any time from 30 days before conception up to 1 day before the end of pregnancy. Pregnancies were classified as “unexposed” if the mother had not redeemed any prescription for antidepressant medication from 6 months before conception up to 1 day before the end of the pregnancy. Thus, in an attempt to avoid misclassification, we excluded from the analyses pregnancies in which the women redeemed a prescription for an antidepressant anytime from 6 months to 30 days prior to conception but without redeeming any prescription for an antidepressant in the index pregnancy period from 30 days before conception up to 1 day before the end of pregnancy (N = 11,834 pregnancy outcomes). In sensitivity analyses, we included all these women as either exposed or unexposed, respectively. In further sensitivity analyses, we considered unexposed to be women who did not redeem a prescription during the 12 months before conception and up to 1 day before the end of pregnancy.

In a sensitivity analysis, we excluded women with exposure to antiepileptic medication, antipsychotic medication, and insulin (ATC codes: N03A, N05A, and A10A).

Maternal psychiatric illness

Information on psychiatric illness was obtained from the Danish Psychiatric Central Registry, which includes information on all admissions to psychiatric hospitals since 1970 as well as information on all psychiatric outpatient contacts since January 1995 [9]. From 1970 to 1993 psychiatric disorders were coded according to the ICD-8 and from January 1994 according to the ICD-10. Women were categorized as depressed during pregnancy if they had received a diagnosis of depression (ICD-10: F32–F33) any time in the interval from 6 months prior to conception up to 1 day before the end of the index pregnancy. History of severe mental disorder was defined as previous or ongoing diagnosis of bipolar disorder including mania or schizophrenia (ICD-8: 296.1–296.8, 298.1 and 295; ICD-10: F30–F31 and F20), and history of misuse was defined as previous or ongoing diagnosis of alcohol or drugs abuse (ICD-10: F10–F19).

The association between antidepressant use during pregnancy and spontaneous abortions was first valuated by comparing the risk of spontaneous abortion in pregnancies exposed to antidepressant medication with the risk in unexposed pregnancies regardless of a registry-based diagnosis of depressive disorder in the pregnant women, and thereafter by conducting the analyses stratified on registry-based diagnoses of depressive disorders during pregnancy.

We had no information on psychiatric diagnoses in women treated by a general practitioner or a private psychiatrist. As a consequence, for these women we had no comparison group of women with comparable depression but no use of antidepressants (discussed below).

Covariates

Information on the following potential confounders was obtained from Statistics Denmark and subsequently coded as shown in parenthesis: maternal age at conception (years), cohabitation at time of conception (yes/no), income at time of conception (quartiles), and education level at time of conception (<10, 10–12, >12 years). From the Registry of Medicinal Product Statistics we obtained information on use of other drugs: Antipsychotic drugs (yes/no), antiepileptic drugs (yes/no), insulin (yes/no), and any reimbursed medication (yes/no); and from the Danish Psychiatric Central Registry we obtained information on psychiatric morbidity: severe mental disorder (yes/no), and drug abuse (yes/no).

Statistics

Risk ratios (RR) for spontaneous abortion were estimated by using binomial regression with robust variance estimation to allow for correlation between pregnancy outcomes of each woman. RRs for spontaneous abortion were adjusted (aRR) for maternal age (age, age squared, and age to the power three, if the regression parameters could be estimated, and age in three categories otherwise), cohabitation, income (above or below the median), education, history of severe mental disorder, and history of drug abuse. The models were evaluated using directed acyclic graphs (not shown) [10]. To avoid unreliable results, RR analysis was only performed when at least 5 exposed events, i.e. spontaneous abortions or stillbirths, were observed. The RRs for spontaneous abortion were estimated not only for women exposed to any antidepressant but also for women exposed to individual or major subgroups of antidepressants (SSRI, TCA and ‘other’). No adjustments could be made in the latter analyses due to sparse data. In the analyses we excluded pregnancies resulting in induced abortions, i.e. results on spontaneous abortion are calculated from the numbers on spontaneous abortion, stillbirth and live birth. An induced abortion makes follow-up incomplete as the pregnancy could potentially have ended in a spontaneous abortion, but treating induced abortion as a competing risk in a time-to-event setup would result in the same conclusions as those obtained from the chosen strategy. However, we performed sensitivity analyses where the pregnancies were censored at the time of the induced abortion. Statistical analyses were performed using Stata 12 (StataCorp, Texas, USA).

Ethics

The study was approved by the Danish Data Protection Agency. All analyses were performed on encrypted and anonymized data in a protected environment under Statistics Denmark, and the researchers had no access to the personal identifier (the CPR-number). Further, care was taken not to present results that might be identifiable. Under these conditions, no consent is needed from the participants under Danish law.

Limitations

The study attempts to disentangle the effects of maternal depression and antidepressant exposure by using information from nationwide registries. We defined women as depressed if they in the Danish Psychiatric Central Registry had received an ICD-10 code of depression 6 months prior to conception or during pregnancy. Women coded as depressed in the Danish National Hospital Registry only, and not in the Danish Psychiatric Central Registry, were not classified as depressed. Conversely, women with a less severe degree of depression treated by a general practitioner or with no contact during the exposure period were not coded as depressed. Approximately 80% of the women who used antidepressants during pregnancy were treated in the primary care system only and we had consequently no information on the underlying psychiatric disorder.

We adjusted for available potential confounders and attempted to take the underlying disease (confounding by indication) into account in the design of the study. We recognize that residual and uncontrolled confounding may still be a problem in this study as in other studies without random allocation of exposure. Furthermore, we had no information on important potential confounders, including smoking, alcohol intake, and diet, and, importantly, the group of women with non-medicated depressive disorder is not a perfect comparison group, as they apparently differ in other aspects than antidepressant use.

Some pregnancies that end in spontaneous abortions, in particular very early in pregnancy, are not recorded in the hospital registry. The symptoms may be interpreted as a late menstrual period or treated outside the hospital system. In a system with free, universal health care the number of spontaneous abortions treated outside the hospital system will decrease dramatically with increased gestational age. If the propensity for hospital contact or admission is associated with the exposure, it will lead to information bias [13], although we found no sign of this particular type of bias when investigating the timing of spontaneous abortions. Women who used antidepressants more often terminated their pregnancy but taking this competing risk into consideration did not change the estimates much indicating that the outcome was not related to the risk of spontaneous abortion.

The study had very limited loss to follow-up owing to the completeness of the population-based registries and limited emigration. Selection bias may, however, still occur if a covariate in the models was associated with an underlying cause of the exposure and was itself a consequence of the outcome. Our directed acyclic graphs did not, however, suggest such collider bias even in the analyses stratified on registry-based diagnosis of maternal depression.

There is a risk of exposure misclassification in the use of redeemed prescriptions to define exposure. For drug use overall, the risk seems to depend on the type of medication, i.e. the correlation between the time of redemption and actual use will be low for migraine medication and high for antibiotics [14]. For the women defined as exposed to antidepressants, this misclassification is likely to bias results toward the null [14]. Conversely, for women defined as depressed with no antidepressant use, a potential drug exposure misclassification (e.g. if the women received treatment during an inpatient admission and redeemed no prescription to antidepressants after discharge) will lead to a bias away from the null in the main analyses comparing women with antidepressant use with the “non-medicated depression cohort”. In sensitivity analyses with “washout periods” up to 12 months we found no change suggesting that misclassification is not a likely cause of the main findings.