Cohorts

The cohort consisted of registered live and stillbirths from 2004 to 2016 of women with a valid personal identification number aged 18 and above. Exclusion criteria for the malformation analyses were the following: mothers migrating between 365 days prior to last menstrual period (LMP) until 365 days postdelivery, stillbirths, invalid personal identification number of offspring, births with a chromosomal abnormality diagnosis (ICD-10 codes Q87.1, Q87.4, Q9X), and mothers with a redeemed prescription for a teratogenic drug in the period from LMP to 90 days post LMP (warfarin [ATC: B01AA03], antineoplastic agents [ATC: L01], isotretinoin [ATC: D10AD04, D10BA01, D10AD54], misoprostol [ATC: A02BB01, G02AD06, M01AE56], lithium [ATC: N05AN01], and thalidomide [ATC: L04AX02]). Exclusion criteria for the stillbirth analyses were the following: mothers migrated between 90 days prior to LMP until delivery and gestational age shorter than 22 weeks or longer than 45 weeks.

Exposure, comparison groups, and outcome

Maternal exposure to medication was defined as a redeemed prescription at a pharmacy. The exposure time window for malformations was from LMP to 90 days after LMP, corresponding to the first trimester of the pregnancy. The exposure time window for stillbirth was from LMP to end of pregnancy. With maternal exposure, fetal exposure is assumed.

Duloxetine exposure was defined as at least one redeemed prescription of duloxetine (ATC N06AX21) in the exposure time window. Four comparison groups were used, all with no redeemed prescription of duloxetine in the relevant exposure time window: (1) duloxetine nonexposed: no redeemed prescription of duloxetine in the exposure time window; (2) SSRI exposed: at least one redeemed prescription of an SSRI (ATC N06AB) in the exposure time window; (3) venlafaxine exposed: at least one redeemed prescription of venlafaxine (ATC N06AX16. Venlafaxine is an SNRI like duloxetine) in the exposure time window; and (4) duloxetine discontinuers: at least one redeemed prescription of duloxetine between 365 days prior to LMP to LMP and not during pregnancy. SSRI-exposed and venlafaxine-exposed women and duloxetine discontinuers were used as comparators to take confounding by indication and severity, and maybe even unmeasured confounding, into account, as they are expected to be similar to duloxetine-exposed women with regard to, e.g., the underlying psychological disease, comorbidity, and health behavior. The comparison groups were not mutually exclusive, and comparisons were analyzed separately with each comparison group. For both the malformation and the stillbirth analyses, an additional exclusion criterion was applied when comparing duloxetine-exposed women with duloxetine nonexposed, SSRI exposed, and venlafaxine exposed: women with duloxetine exposure from 90 days prior to LMP but no exposure from LMP to 90 days after LMP were excluded. This washout period was applied to avoid misclassification.

Major and minor malformations were classified according to the EUROCAT classification of congenital malformations version 1.4 [47]. Diagnoses of the offspring were gathered from the national patient registers as either a primary or secondary diagnosis registered within 365 days after birth. Major malformations were defined as the following ICD-10 codes: Q-chapter, D215, D821, D1810, P350, P351, P371, except for the ICD-10 codes used to define minor malformations. Minor malformations were defined as the ICD-10 codes in Table A in S1 Tables. Also, analyses of major malformation subtypes were performed: abdominal wall; cardiac; digestive system; ear, face, and neck; eye; genital; limb; nervous system; orofacial clefts; respiratory system; urinary system; and other anomalies (Table B in S1 Tables). Information on stillbirth was gathered from the medical birth registers and was defined as no signs of life at birth after week 22 of pregnancy and from the patient register if abortions were registered after week 22 [48].

Statistical methods

For each of the 4 comparator groups, an unadjusted, an adjusted multiple regression, and a propensity score (PS)-matched analysis were performed.

The PSs were estimated with logistic regression and greedy matching using the SAS macro OneToManyMTCH [49] with an extension to secure only women with a difference of maximum of 0.2 logit of the PS were used to match. When matching duloxetine exposed with duloxetine nonexposed, a 1:4 ratio was used. Because of limited data in the other comparison groups, duloxetine and SSRI exposed were matched with a 1:2 ratio and venlafaxine exposed and duloxetine discontinuers with a 1:1 ratio. Duloxetine-exposed individuals with no match were excluded from the PS-matched analyses. After PS matching, a conditional logistic regression, including the matched group id as a strata variable, was fitted to assess risk of minor and major malformations and stillbirths, respectively. To assess the balance of possible confounders after PS matching, standardized differences were calculated using the SAS macro stddiff macro [50].

For the adjusted multiple regression and the PS, prespecified covariates were used based on previous publications [21,23] and available data. When fitting each model, covariates were removed, if the model could not be estimated. This is the case if, e.g., no patients in one exposure group had any severe stress reaction. Then, the variable severe stress reaction cannot be part of the model and the covariate needs to be removed.

Covariates used for the malformation and stillbirth analyses are the following: data source (Sweden/Denmark), birth year of the offspring (3 categories: 2004 to 2008, 2009 to 2012, and 2013 to 2016), maternal age (4 categories: 18 to 24, 25 to 29, 30 to 34, and >34 years), previous spontaneous abortions (0/1/≥2), previous stillbirths (yes/no), smoking during pregnancy (yes/no), psychiatric hospitalizations (1 year prior to LMP: yes/no), psychiatric outpatient visits (1 year prior to LMP: yes/no), household income (year of LMP, grouped in quartiles), and highest completed education (year of LMP, 3 categories: <11, 11 to 15, and >15 years). Comorbidities (identified up to maximum 5 years prior to LMP; see Table C in S1 Tables for ICD-10 codes and ATC codes): affective disorder, anxiety or phobia, depression, diabetes during pregnancy, diabetes, diabetic peripheral neuropathy, hyperthyroidism and hypothyroidism, hypertension, obesity, renal failure, severe stress reaction, and stress urinary incontinence. Comedication (at least one redeemed prescription between 90 days prior to LMP to end of the relevant exposure time window; see Table D in S1 Tables for ATC codes): antiepileptics, antihypertensives, antipsychotics, anxiolytics, danazol, estradiol, fluconazole, glucose-lowering, NSAID, opioids, progesterone, steroid hormone, thyroid hormone, and triptans. SSRI and venlafaxine comedication were not part of the confounder selection, since these will be highly associated with the comparison group rather than the duloxetine exposed. For a complete list of covariates used for the individual analysis, see Table E in S1 Tables.

In general, there were very few missing values. If income was missing at year of LMP, first income was imputed from 1 year prior to LMP, and, if still missing, income was imputed from 1 year after LMP, where possible. If education was missing at year of LMP, it was imputed from 1 year after LMP, where possible. Data were analyzed assuming missing at random and persons with missing values were deleted from the analysis.

For the analyses of stillbirth and malformation subtypes with less than 30 outcome events in the exposed group, only unadjusted and PS-matched analyses were performed.

Four prespecified sensitivity analyses were conducted. To assess the exposure definition of minimum one redeemed prescription and evaluate potential misclassification of exposure, sensitivity analyses were conducted with (1) exposure redefined to minimum 2 redeemed prescriptions; and (2) exposure redefinition to overlap between the exposure time window and days’ supply of redeemed prescriptions. Days’ supply was based on the number and strength of redeemed pills compared with the WHO’s daily defined dose [51]. A woman’s first pregnancy may influence later pregnancies; therefore, sensitivity analyses (3) restricting the cohort to the first pregnancy within the study period were performed. The medical birth registers hold information on maternal BMI, but this is expected to be missing for a substantial number of women. The main analyses did not include BMI as a covariate, but (4) sensitivity analyses including BMI as covariate were conducted. The sensitivity analyses handled BMI as missing at random, although this may not be fully accurate.

SAS Enterprise Guide 7.15 was used, and a significance level of 5% was applied. Validation of the programming was performed; smaller programs (3 to 20 lines of coding) were reviewed, and longer programs were double coded by an independent statistical programmer. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (see S1 Checklist). The protocol was developed prior to data access and followed, with a few exceptions: (1) Incomes should have been standardized to 2015-year level, but when grouped in quartiles, stratified on calendar year, standardization was not needed. (2) ATC codes were included to identify comorbidity, although not described in the protocol. Utilizing both ICD-10 and ATC codes more women with the given comorbidities will be identified. (3) Comedication should have been identified 1 year prior to LMP but was changed to comedication during the relevant exposure time window, since it is more likely that comedication during pregnancy will affect the outcomes of malformation and stillbirth and therefore act as confounders.

Interpretation

The results of the present study are in line with previous studies [31,32] and case reports [26,52,53] finding no increased risk for malformations associated with duloxetine exposure. A review from 2015 concluded that there was insufficient data on duloxetine to draw definitive conclusions about its safety in pregnancy [54]. A review from 2016 concluded that women exposed to duloxetine during the first trimester (n = 668) had no increased risk of congenital malformations (OR 0.80, 95% CI 0.46 to 1.29) [33]. The present study corroborates these findings based on a considerably larger group of duloxetine-exposed women (n = 1,512) while addressing more potential confounding factors and including PS-matched analyses and sensitivity analyses. Huybrechts and colleagues [32] did a recent study in the US and found no increased risk for malformation overall, but a small increased risk of cardiac malformation. Although no increased risk of cardiac malformations was found in the present study, it must be noted that there was an increased risk of “other anomalies and syndromes” (See Table B in S1 Tables for the full list of other anomalies and syndromes, but it includes, e.g., craniosynostosis, situs inversus, and fetal alcohol syndrome). This must be interpreted with caution based on the low number of cases (n = 14) and the wide CIs. Since we find no clear pharmacological mechanism explaining the association, the result is interpreted to be a chance finding.

A small Swedish study from 2007 investigating SNRI/NRIs (not including duloxetine) found no increased risk for stillbirth when compared to the background population or to women exposed to SSRI during pregnancy [55]. The present study, with a bigger population and more recent data, supports this finding.

Strengths and limitations

Due to the nationwide coverage, high validity, and completeness of the included registers, the risk of selection (sampling, allocation, and lost to follow-up) bias was minimal. There was no risk of recall bias. The quality of malformation diagnoses was validated in Danish data and was found to have a predictive value of 88%, with a completeness of 90% [56]. Diagnoses of heart defects were found to have a positive predictive value of 98.4% [57]. For the present study, differential misclassification depending on exposure status was not suspected. Information on malformation among abortions and stillbirths was not available. This can lead to an underestimation of the risk if duloxetine exposure is associated with malformations leading to these outcomes. Information on LMP was precise as it was based on the mother’s self-report and 2 subsequent ultrasounds in the first and second trimester.

The primary exposure definition was the redemption of a single prescription. Although the medication has been prescribed, dispensed, redeemed, and paid for, there is a probability that the patient did not ingest the drug. Sensitivity analyses were performed with a stricter definition of exposure (>1 redeemed prescription) under the assumption that redeeming multiple prescriptions increases the likelihood that the medication was taken. These sensitivity analyses did not change the overall results. Information on drug exposure during hospitalization was unavailable. This might have led to misclassification of exposure as exposure is based on redeemed prescriptions from community pharmacies. However, we assume that women hospitalized because of depression also redeem prescriptions of antidepression medication. Also, it is expected that most women with the indication for antidepressants are treated outside the hospital since only 1.6% of the total duloxetine use in the time period was administrated in hospitals in Denmark [58]. Important unavailable potential confounders were alcohol, illicit drug use, and poor adherence to folic acid supplementation during pregnancy [59]. Women with depressive disorder are more likely to smoke, use alcohol or other substances, and, in general, not to adhere to recommended health behavior during pregnancy. This may confound the association between duloxetine and pregnancy outcomes and increased risk of malformations may erroneously be attributed to duloxetine [60,61]. Socioeconomic status (education and income) and smoking were included as covariates. Due to the concern of unmeasured confoundings (e.g., alcohol and illicit drug use) and confounding by indication, SSRI-exposed and venlafaxine-exposed women were used as comparators as they are expected to have similar health behavior as duloxetine-exposed women.

Indication for duloxetine was not available as a recorded covariate. In addition to depression, duloxetine is indicated for the treatment of neuropathy, anxiety, severe stress reaction, and stress urinary incontinence. In the analyzed cohort of pregnant women, the prevalence of these indications is, however, expected to be very low. Depression severity, which might also lead to confounding, and a direct measurement for depression severity was unavailable. This potential confounding was addressed by including diagnoses of neuropathy, anxiety, severe stress reaction, and stress urinary incontinence as covariates, as well as covariates that describe depression severity (depression diagnosis recorded at hospital contact, psychiatric hospitalization, and outpatient visits).

In the analyses of major malformation subtypes and stillbirth, adjusted analyses could not be consistently performed due to the low number of outcome events despite the cohort including more than 2 million births. As opposed to adjusted analyses, the PS-matched analyses were possible despite few outcome events.

The study included all registered pregnancies resulting in a birth (live or stillbirth). We believe that the results have a high external validity especially applicable to other western European countries with free and universal healthcare, where treatment regimens and population characteristics are comparable, as well as to the US, where indications and treatment guidelines are similar to the studied population. Despite the large cohort, the number of women exposed to duloxetine during pregnancy was limited, and future studies should focus on analyzing larger cohorts and additional safety outcomes (e.g., preterm birth, abortions, and SGA). Improved measurements of exposure, outcome, and covariates could also yield more precise estimates. In addition, information on duloxetine indication and depression severity could add relevant information.

Funding

This publication is based on a post-authorization safety study (PASS) on duloxetine requested by the Food and Drug Administration (FDA). Eli Lilly and Company funded the PASS study. The funder had the opportunity to comment study design, data collection and analysis, where to publish and preparation of the manuscript, but MZA and EJS had the final decisions.