SSRI Use During Pregnancy and Risk of Stillbirth and Neonatal Mortality

Since 1968, all Danish citizens have been given a unique 10-digit identification number at birth (10), which enables individualized information to be linked across databases. Using the Danish Fertility Database, we identified 974,805 births between 1995 and 2008. We excluded 1,562 records with coding errors, 20 duplicate records, and 52,603 records with missing data (date of birth and duration of pregnancy), leaving us with valid records of 920,620 births (94.4% of all births).

The Danish Fertility Database contains unique identification numbers for the mother, father, and child, along with information on the mother’s age and prior births as well as the child’s sex, gestational age, and, when applicable, time of death (11). Time of conception is based on ultrasound estimates and date of last menstrual period. We linked these data with information on redeemed prescriptions from the Register of Medicinal Product Statistics, which contains records from 1995 onward of date of redemption, type of drug, quantity dispensed, strength, and other data (12). The international Anatomical Therapeutic Chemical classification system was used to code all SSRIs. Information on smoking was gathered from the National Hospital Register, which contains information on all hospitalizations in the country, including admittance data and discharge diagnoses (13). Data on income and education were gathered from Statistics Denmark (http://www.dst.dk/en).

We identified exposure to five SSRIs: fluoxetine, citalopram, paroxetine, sertraline, and escitalopram. Other SSRIs were not included because the number of exposed women was too small (N<50).

We used the Register of Medicinal Product Statistics to identify all SSRI prescriptions filled during the study period. Exposure periods and dosages were estimated by using the date of prescription, the strength, and the number of tablets prescribed. This method for calculating drug treatment periods has been described previously (14). We defined first-trimester exposure as one or more days of exposure between conception and day 84 of pregnancy, first- and second-trimester exposure as exposure throughout the first trimester and between day 85 and day 196 of pregnancy (second trimester), and exposure in all trimesters as exposure throughout the first and second trimesters and between day 197 of pregnancy and birth (third trimester). Women who began treatment during the first (N=186), second (N=84), or third (N=59) trimester were not included in the study because of the low number of cases.

Stillbirth was defined as a child showing no signs of life at birth. Fetal deaths that occurred before 2004 were considered stillbirths if death occurred after 28 completed weeks of gestation. After 2004, fetal deaths were recorded as stillbirths if they occurred after 22 completed weeks of gestation. The method by which data on perinatal mortality are recorded has been described previously (15). Neonatal death was defined as death within 28 days of birth.

Data management and all statistical analyses were conducted using SAS, version 9.2 (SAS Institute, Cary, N.C.). Logistic regression models were used to identify a possible statistical association for dichotomous variables. Odds ratios are presented with 95% confidence intervals (CIs). Baseline characteristics are presented as frequencies and percentages. Chi-square tests were used to assess differences in baseline characteristics for categorical variables. The threshold for statistical significance was set at a p value of 0.05. All statistical tests were two-sided.

Mother’s age at birth was stratified into five groups: <20, 21–25, 26–30, 31–35, >35 years (0% missing values). Annual gross household income during the year of birth was divided into quartiles (<1% missing values). Education level was divided into three groups (low, medium, and high) according to the highest level of education attained by the end of the birth year; for missing values, we used information from the next calendar year, resulting in 3.6% missing values. Parity was defined as the number of births, including stillbirths, and divided into four classes: 1, 2, 3, and >3 births (<1% missing values). Multivariate analyses in model 1 were adjusted for these variables and for birth year, which was divided into five categories (1995–1997, 1998–2000, 2001–2003, 2004–2006, and 2007–2008). Analyses for stillbirth were additionally adjusted for prior stillbirths. The largest subgroup within each category was used as the reference.

Multivariate analyses in model 2 were adjusted for all variables in model 1 and for smoking. These analyses included all pregnancies between 1996 and 2007; 1995 and 2008 were excluded because information on smoking during pregnancy was not available for those two years. Smoking was divided into four classes according to the number of cigarettes smoked daily: 0, 1–10, 11–20, and >20 (9.3% missing values). We corrected for multiple testing using the Bonferroni method.

To ensure that no individuals could be identified, all personal information held in the registers was encrypted and analyzed on computers held by Statistics Denmark. The study was approved by the Danish Data Protection Agency. Retrospective register studies do not require ethical permission in Denmark.

There were 3,919 stillbirths (0.43% of all births) in our study population: 2,713 (0.41%) between 1995 and 2004 and 1,206 (0.47%) between 2005 and 2008. We identified 75 (0.60%) stillbirths among women exposed to an SSRI during pregnancy (Table 2); 21 (0.53%) with first-trimester exposure, 12 (0.58%) with first- and second-trimester exposure, and 42 (0.66%) with exposure in all trimesters (Table 2).

In unadjusted analyses, exposure to an SSRI in all trimesters (but not for first-trimester or first- and second-trimester exposure) was significantly associated with stillbirth (odds ratio=1.55, 95% CI=1.14–2.10) compared with unexposed pregnancies. Adjusting our model rendered this association nonsignificant (odds ratio_{model 1}=1.19, 95% CI=0.87–1.63; odds ratio_{model 2}=1.06, 95% CI=0.71–1.58). When stratifying for different SSRIs, we did not find an increased risk of stillbirth with exposure in any trimester (Table 2).

There were 3,138 neonatal deaths (0.34%) between 1995 and 2008 in our study population. We identified 47 (0.38%) neonatal deaths among women exposed to an SSRI during pregnancy (Table 3): eight (0.20%) with first-trimester exposure, nine (0.44%) with first- and second-trimester exposure, and 30 (0.47%) with exposure in all trimesters (Table 3).

We found no association between SSRI exposure and neonatal mortality (first-trimester exposure: odds ratio_{model 2}=0.56, 95% CI=0.25–1.24; first- and second-trimester exposure: odds ratio_{model 2}=0.90, 95% CI=0.37–2.17; exposure in all trimesters: odds ratio_{model 2}=1.27, 95% CI=0.82–1.99) (Table 3).

Stratifying exposure to different SSRIs revealed an association between three-trimester exposure to citalopram and neonatal mortality (unadjusted odds ratio=2.50; 95% CI=1.50–4.16; adjusted odds ratio_{model 1}=2.13, 95% CI=1.25–3.62; adjusted odds ratio_{model 2}=2.49, 95% CI=1.33–4.65). Estimates for the remaining SSRIs and trimesters were not statistically significant (Table 3).

We identified 1,629 pregnancies with exposure to two different SSRIs simultaneously. Among these pregnancies were seven that resulted in stillbirths and five that resulted in neonatal deaths. The most frequently used combinations during pregnancy were fluoxetine and citalopram (N=537) and fluoxetine and sertraline (N=439). Treatment with two SSRIs during pregnancy was not associated with stillbirth (adjusted odds ratio_{model 1}=0.81, 95% CI=0.38–1.70; odds ratio_{model 2}=0.47, 95% CI=0.15–1.48), nor was it associated with neonatal mortality (adjusted odds ratio_{model 1}=0.86, 95% CI=0.36–2.08; odds ratio_{model 2}=1.21, 95% CI=0.50–2.93).