Association of antidepressant continuation in pregnancy and infant birthweight

Abstract

Purpose:

To evaluate the association of antidepressant continuation in pregnancy with infant birthweight among women using antidepressants before pregnancy.

Methods:

This retrospective cohort study used electronic health data linked with state birth records. We identified singleton live births (2001–2014) to enrolled women with ≥1 antidepressant prescription filled ≤6 months before pregnancy, including “continuers” (≥1 antidepressant fill during pregnancy, n=1,775) and “discontinuers” (no fill during pregnancy, n=1,249). We compared birthweight, small or large for gestational age (SGA or LGA), low birthweight (LBW; <2500 g), and macrosomia (>4500 g) between the two groups, using inverse probability of treatment weighting to account for pre-pregnancy characteristics, including mental health conditions.

Results:

After weighting, infants born to antidepressant continuers weighed 71.9 g less than discontinuers’ infants (95% confidence interval [CI]: −115.5,−28.3 g), with a larger difference for female infants (−106.4 g, 95% CI: −164.6, −48.1) than males (−48.5 g, 95% CI: −107.2, 10.3). For female infants, SGA risk was greater in continuers than discontinuers (relative risk [RR]: 1.54, 95% CI: 1.02, 2.32). LBW risk was greater in continuers with ≥50% of days covered (RR: 1.69, 95% CI: 1.11, 2.58) and exposure in the second trimester (RR: 1.53, 95% CI: 1.02, 2.29), as compared with discontinuers.

Conclusions:

Depending on infant sex, as well as duration and timing of use, continuation of antidepressant use during pregnancy may be associated with lower infant birthweight, with corresponding increases in LBW and SGA.

INTRODUCTION

Approximately one of 10 pregnant women experiences depression and/or anxiety.^1–3 Although antidepressant medications are a primary treatment modality for depression and anxiety, many women choose to discontinue antidepressants in pregnancy due to concerns about fetal harm.

One concern is infant birthweight, a predictor of short-term and lifelong health. Low birthweight (LBW, defined as birthweight <2500 g), which affects 8% of births in the US,⁴ is associated with greater risk of infection, respiratory distress, and mortality in the neonatal period,^5,6 as well as developmental delays, cardiovascular disease and diabetes later in life.^7,8 Small for gestational age (SGA) is associated with similar adverse outcomes.^9–11 It is biologically plausible that prenatal antidepressant exposure could lower birthweight because studies of human cells in vitro indicate that the presence of antidepressants can impede placental development,¹² which could lead to intrauterine growth restriction. Additionally, animal models indicate that antidepressants may cause uterine vasoconstriction.¹³

On the other hand, macrosomia (defined here as birthweight >4500 g), which affects just over 1% of births in the US,¹⁴ is associated with greater risk of cesarean section and birth injuries,¹⁵ as well as obesity, diabetes, and cardiovascular diseases later in life.^16,17 Large for gestational age (LGA) is also associated with greater risks of these poor health outcomes.^18,19 Some antidepressants are associated with weight gain in non-pregnant populations,^20–22 potentially due to their action on monoamines, which are thought to affect appetite, satiety, and metabolism.^20,23,24 We hypothesize that a similar effect could be present in pregnancy, with prenatal antidepressant use increasing maternal gestational weight gain, which is associated with subsequent increased birthweight.²⁵

Some studies, but not all, have reported that prenatal antidepressant use was associated with lower mean birthweight and greater risk of LBW and SGA.^26–28 However, most did not sufficiently control for maternal depression, anxiety, or substance use. These factors are associated with lower birthweight,²⁹ so their findings may be due to confounding. A recent systematic review examined associations of mean birthweight and risk of LBW and SGA with prenatal antidepressant use. The authors of the review restricted to studies comparing infants of women using antidepressants in pregnancy with infants of women with depression who did not use antidepressants in pregnancy. They concluded, “the low number of studies meeting criteria highlights the need for further research to aid in risk-benefit analysis for women considering antidepressant use in pregnancy.”³⁰ The literature on risk of macrosomia or LGA associated with antidepressant use in pregnancy is even more limited.

To address this gap, we used electronic health data and linked state birth records to assess the association between continuing, versus discontinuing, antidepressant use during pregnancy and outcomes related to infant birthweight, including mean birthweight, SGA, LBW, LGA, and macrosomia, among women using antidepressants before pregnancy. We utilized inverse probability of treatment weighting (IPTW) to account for maternal characteristics that may bias associations of prenatal antidepressant use and perinatal outcomes. We also calculated associations by specific antidepressant medications, trimesters of exposure, and proportion of days covered, and separately by infant sex, given evidence that male and female infants can be differentially affected by in utero exposures.^31–33

MATERIALS AND METHODS

Overview

We conducted a retrospective cohort study within Kaiser Permanente Washington (KPWA), an integrated healthcare delivery system in Washington State. We required all women to have enrollment ≥6 months prior to pregnancy onset and an antidepressant prescription fill during that period. We chose this population for two key reasons. First, it allows us to address the clinical decision that these women and their providers face of whether to continue antidepressants in pregnancy. Second, it allows us to limit confounding by indication by creating a comparator group that is more similar to the exposed group than the general population of pregnant women in terms of mental health, substance use, and other characteristics that could impact birthweight. Electronic health data from KPWA were linked to Washington State birth records.³⁴ The study was approved by the KPWA Institutional Review Board and the Washington State Department of Health Institutional Review Board (both with waivers of consent).

Study Population and Design

Our cohort was drawn from live births to KPWA enrollees from 2001–2014, due to data availability. KPWA maintains extensive data on members including patient enrollment, demographics, encounters, diagnoses, procedures, and prescription fills. Nearly two-thirds of members receive comprehensive care from KPWA healthcare providers through the Integrated Group Practice, and for these women we ascertained additional data on vital signs and mental health questionnaires. We required KPWA births to be linked to a Washington State birth record to obtain information on gestational age, necessary to determine timing of antidepressant fills relative to pregnancy onset, and birthweight. Because 7% of women contributed more than one singleton birth to our cohort, our unit of analysis was technically “births” rather than “women”, but we have used the terms interchangeably here.

An exhaustive list of exclusions and associated counts is detailed in Figure 1; we describe key exclusions here. We aimed to limit the study population to women taking antidepressants for depression or anxiety, as 80–90% of antidepressant use among women is for these indications.^3,35 Because depression and anxiety are not always documented, we did not require such a code (Supplemental Table 1), but we excluded women who lacked a pre-pregnancy depression or anxiety code if they had a code for another antidepressant indication (Figure 1, Supplemental Table 2).³⁶

Figure 1.

Identification of the eligible cohort.

AD: antidepressant; Antidepressants of interest include paroxetine, sertraline, escitalopram, citalopram, fluoxetine, fluvoxamine, desvenlafaxine, venlafaxine, bupropion, mirtazapine.

LBW: low birthweight, LGA: large for gestational age, SGA: small for gestational age

^aWe excluded twin and higher order births because their birthweight is typically very different from singletons.

^bFibromyalgia, migraine, chronic pain, bulimia, premenstrual syndrome, attention deficit disorder, tobacco use disorder (only for bupropion, the only antidepressant indicated to treat tobacco use disorder).

^cThe exposure window spans the start of pregnancy through 4 weeks prior to delivery, when birthweight is ascertained.

The relevant window of antidepressant exposure was defined as pregnancy onset through four weeks before delivery, to allow time for an exposure effect (Figure 2). Women with their first antidepressant fill after this period were excluded (Figure 1).

Figure 2.

Study design.

Exposure

We utilized pharmacy dispensing data, including antidepressant name, fill date, and number of days supplied. Women with any antidepressant fill during the exposure window were “continuers”; women without such a fill were “discontinuers” (Figure 2).

Exposure included selective serotonin reuptake inhibitors (SSRIs; citalopram, escitalopram, fluvoxamine, fluoxetine, paroxetine, sertraline), serotonin-norepinephrine reuptake inhibitors (SNRIs; desvenlafaxine, venlafaxine), and other antidepressants (bupropion, mirtazapine). Continuers were not restricted to women who filled the same antidepressant during the six months pre-pregnancy and the exposure window. Women who had an antidepressant fill overlapping into pregnancy but did not have a fill during pregnancy were considered discontinuers, an assumption addressed by sensitivity analysis.

Outcomes

We assessed magnitude of birthweight (in grams [g]), SGA and LGA (<10^th percentile and >90^th percentile based on the Washington State sex-specific birthweight growth curve³⁷), LBW (birthweight <2500 g), and macrosomia (birthweight >4500 g).

Covariates

Birth records provided the following covariates: parity, maternal race/ethnicity, maternal education, and pre-pregnancy weight and body mass index (BMI). Primary analyses used pre-pregnancy weight instead of BMI because BMI was not available until 2003.

KPWA electronic health databases provided birth year, maternal age, and infant sex. From pre-pregnancy data, we identified the following variables: membership in the Integrated Group Practice, Medicaid insurance, chronic disease diagnoses, smoking and substance use disorder diagnoses, mental health condition diagnoses, psychotropic prescription medication fills, and utilization of psychotherapy, psychiatry, and inpatient psychiatric hospitalization (Supplemental Tables 1, 3, and 4). We allowed collection of covariate data to go back in time as long as women were enrolled.³⁸ We used baseline characteristics as covariates because adjusting for information from after the start of pregnancy could induce bias.³⁹

We assessed depression severity using scores from the Patient Health Questionnaire-9 (PHQ-9), a 9-item, self-reported depression screening questionnaire.⁴⁰ These data were available for women in the Integrated Group Practice starting in about 2008. During this time, KPWA did not practice universal depression screening, but instead used the PHQ-9 when patients initially presented with symptoms of depression and periodically during follow-up to assess treatment response. A score of 0–4 indicates no depression, 5–9 indicates mild depression, 10–14 indicates moderate depression, 15–19 indicates moderately severe depression, and 20–27 indicates severe depression.

Statistical Analysis

We summarized characteristics of eligible women stratified by exposure status. We used marginal models with generalized estimating equations and an exchangeable correlation matrix for all regression analyses, to account for the correlation among multiple births to the same woman.⁴¹ We calculated mean differences and 95% confidence intervals (CIs) of the associations of prenatal antidepressant use with birthweight using an identity link function. For the analyses of SGA, LBW, LGA, and macrosomia with antidepressant continuation in pregnancy, we used a logarithmic link function to calculate relative risks (RRs) and 95% CIs.

We employed IPTW estimation using propensity scores to weight observations by their likelihood of exposure to improve balance in baseline covariates between exposed and unexposed. This allowed the exposure model to include more potential confounders than possible in a multivariable adjusted model.⁴² More details about IPTW are included in the Supplemental Materials Methods Appendix.⁴³

Covariates were chosen a priori based on clinical knowledge and previous literature, after considering sample size limitations. For the overall analyses for all our outcomes, we weighted the IPTW model by all variables in Variable Set 4 (Supplemental Table 5).

We ran sub-analyses comparing women who continued specific antidepressants or continued in specific trimesters with discontinuers (the group of women who discontinued any antidepressant, as in the overall analysis). The smaller Variable Set 3 was used for these sub-analyses due to smaller numbers of exposed women (Supplemental Table 5). For analyses of venlafaxine, with the smallest number of continuers, we needed a parsimonious model. We began by identifying characteristics with standardized mean differences (SMDs) of ≥0.10 comparing continuers and discontinuers (SMDs in Supplemental Table 6), and then selected those associated with a >10% change in the mean difference in birthweight after their addition to the crude model (Variable Set 1 in Supplemental Table 5). We conducted the same process for paroxetine (Variable Set 2 in Supplemental Table 5). For consistency, we used the models determined from this process for all analyses of these two antidepressants.

We ran analyses separately by infant sex, to assess possible differential effects of antidepressant use. We ran analyses separately for continuers with <50% of days covered with antidepressant fills and continuers with ≥50% of days covered, to address the issue of duration-response by proportion of days exposed. For analyses stratified by proportion of days covered, we included only women with a fill for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants.

We restricted analyses to subjects with known information for variables included in each model. For prenatal weight, race/ethnicity, parity, or education, <1% of observations had missing values; other characteristics included in primary models did not have missing values.

Sensitivity Analyses

We conducted sensitivity analyses using IPTW models, weighting for characteristics described in Variable Set 3 (Supplemental Table 5). We redefined exposure as receiving ≥2 antidepressant fills to address potential exposure misclassification, which could arise if women who only filled once in pregnancy did not actually take the medication. We adjusted for pre-pregnancy BMI instead of pre-pregnancy weight to address residual confounding, restricting to years when BMI was available. To address possible lingering effects of medication and the assumption that women whose last fill overlapped pregnancy stopped before pregnancy onset, we conducted a sensitivity analysis that re-categorized these women as continuers. We restricted to women with an anxiety or depression diagnostic code in the two years before pregnancy. We ran an analysis excluding women with a bipolar disorder diagnosis in the 2 years before pregnancy onset or a fill for a mood stabilizer or an antipsychotic medication at any point before pregnancy onset, as these women may have had a more severe or substantially different type of mental health condition than women with unipolar depression. We ran a multivariable-adjusted model, adjusted for characteristics in Variable Set 3 (Supplemental Table 3), to assess whether associations were similar to IPTW analyses.

We conducted sensitivity analyses including gestational age in the model for magnitude of birthweight, LBW, and macrosomia, because studies that accounted for underlying mental health status have found an association of prenatal antidepressant use with shorter gestational age.^30,44 We did not adjust for gestational age in our main analyses of these outcomes, as it is in the causal pathway, and adjustment could lead to collider-stratification bias.³⁹ SGA and LGA inherently account for gestational age, so we did not conduct this sensitivity analysis for those outcomes.

All statistical analyses were conducted in R version 3.4.2 (R Core Team [https://www.r-project.org]).

RESULTS

There were 57,743 births identified in linked KPWA-Washington State birth record data during the study period. After exclusions, eligible women had 3,024 births (Figure 1), including 1,775 to women who continued antidepressants in pregnancy and 1,249 to women who discontinued.

Characteristics of exposed and unexposed women

Women who continued antidepressant medication during pregnancy were similar to discontinuers with a few exceptions: antidepressant continuers were slightly older, less likely to be nulliparous, more likely to be non-Hispanic white, more educated, and more likely to have a psychiatry visit or filled an antidepressant prescription more than one year before pregnancy (Table 1, Supplemental Table 6). The average length of enrollment prior to pregnancy, during which we collected covariate data, was 4.2 years (standard deviation [SD]: 4.1 years) for continuers and 4.5 years (SD: 4.3 years) for discontinuers. After weighting, SMDs indicated that baseline characteristics were not meaningfully different between continuers and discontinuers (Supplemental Figure 1).

Table 1.

Selected baseline characteristics of women with births.

Covariates	No antidepressant fill in pregnancy (n=1,249)	Antidepressant fill in pregnancy (n=1,775)
Maternal age at delivery, mean (SD)	29.7 (5.8)	31.4 (5.4)
Parity, n (%)
Zero	556 (44.9)	657 (37.2)
One	426 (34.4)	619 (35.1)
Two or more	172 (13.9)	331 (18.8)
Maternal race/ethnicity, n (%)
Hispanic	87 (7.0)	107 (6.1)
Non-Hispanic Asian	35 (2.8)	27 (1.5)
Non-Hispanic Black	50 (4.0)	38 (2.2)
Non-Hispanic Native American	15 (1.2)	26 (1.5)
Non-Hispanic Native Hawaiian or Other Pacific Islander	21 (1.7)	23 (1.3)
Non-Hispanic white	1033 (83.2)	1546 (87.5)
Maternal education level, n (%)
High school diploma or less	273 (22.0)	326 (18.5)
Some college	492 (39.7)	653 (37.1)
Bachelor’s degree or more	474 (38.8)	783 (44.4)
Medicaid, n (%)	48 (3.8)	61 (3.4)
Pre-pregnancy weight in lbs, mean (SD)	167.2 (46.2)	170.1 (45.4)
During the two years prior to pregnancy, n (%)
Alcohol use disorder	53 (4.2)	76 (4.3)
Tobacco use disorder	165 (13.2)	251 (14.1)
Drug use disorder	49 (3.9)	63 (3.5)
Depression disorder	906 (72.5)	1302 (73.4)
Anxiety disorder	482 (38.6)	745 (42.0)
Obsessive compulsive disorder	22 (1.8)	57 (3.2)
Post-traumatic stress disorder	41 (3.3)	57 (3.2)
Bipolar disorder	58 (4.6)	78 (4.4)
During the year prior to pregnancy, n (%)
Antipsychotic fill	23 (1.8)	40 (2.3)
Benzodiazepine fill	279 (22.3)	426 (24.0)
Mood stabilizer fill	59 (4.7)	95 (5.4)
Medication associated with weight gain fill	178 (14.3)	271 (15.3)
Psychotherapy visit	231 (18.5)	302 (17.0)
Psychiatry visit	177 (14.2)	281 (15.8)
Inpatient psychiatric hospitalization	9 (0.0)	17 (0.0)

SD: Standard deviation

Less than 1% of observations were missing prenatal weight, race/ethnicity, parity, or education. All other characteristics included in this table did not have missing values.

This table includes a subset of all covariates that were used to create the inverse probability of treatment weights and included in the overall regression models (Variable Set 4 in Supplementary Table 5). The full version of this table (with all included characteristics) is Supplementary Table 6.

Diagnostic codes for conditions are listed in Supplementary Table 1. Specific medications in medication categories are listed in Supplementary Table 3. Definitions of mental health care utilization are described in Supplementary Table 4.

Among continuers, we had a pre-pregnancy PHQ-9 score for 363 (~61% of women who could have a record, based on data availability) compared with 285 records among discontinuers (~67% of women who could have a record). The mean pre-pregnancy score was 9.5 for continuers and 10.4 for discontinuers (SDs: 6.6 and 6.7, respectively).

Counts of continuers filling specific antidepressants and extent of exposure in pregnancy are included in Supplemental Tables 7 and 8.

On average, continuers delivered babies about 2 days earlier than discontinuers (mean gestational age: 38.7 weeks [SD: 1.8] versus 39.0 weeks [SD: 1.8], respectively).

Birthweight outcomes

Antidepressant continuers had babies with slightly smaller unadjusted mean birthweights, compared with discontinuers (3405 g, SD: 566 g versus 3461 g, SD: 546 g, respectively, Table 2), and this difference remained after accounting for confounders (mean difference: −71.9 g, 95% CI: −115.5, −28.3). For female infants, we observed stronger associations than in the overall analysis, with lower mean birthweight (−106.4 g, 95% CI: −164.6, −48.1 g) comparing continuers versus discontinuers, whereas associations in male infants were attenuated (mean difference in birthweight: −48.5 g, 95% CI: −107.2, 10.3 g). Continuers of fluoxetine and those with a fill in the second trimester had the greatest deficits (mean difference: −101.4 g, 95% CI: −164.1, −38.7 g and mean difference: −115.3 g, 95% CI: −164.0, −66.6 g, respectively). Comparing continuers with ≥50% of days covered with discontinuers, the association strengthened for mean birthweight (mean difference: −125.4 g, 95% CI: −179.5, −71.2 g).

Table 2.

Antidepressant continuation in pregnancy and mean infant birthweight.

	No.	Birthweight in g Mean (SD)	Crude mean difference in g (95% CI)		Inverse probability of treatment weighted mean difference in g (95% CI)
No antidepressant fill in pregnancy	1249	3461 (546)	ref		ref
Any antidepressant fill in pregnancy	1775	3405 (566)	−57.6	(−97.7, −17.6)	−71.9	(−115.5, −28.3)a
SSRIs	1533	3402 (562)	−61.8	(−103.1, −20.4)	−83.1	(−126.1, −40.0)b
Citalopram	381	3396 (513)	−70.5	(−131.2, −9.9)	−62.0	(−129.2, 5.2)b
Fluoxetine	509	3399 (556)	−64.2	(−120.9, −7.5)	−101.4	(−164.1, −38.7)b
Paroxetine	160	3453 (497)	−12.7	(−96.7, 71.3)	−89.1	(−197.1, 19.0)c
Sertraline	534	3385 (602)	−73.2	(−133.5, −12.9)	−80.2	(−142.5, −17.8)b
Bupropion	240	3405 (583)	−49.1	(−129.1, 30.8)	−37.9	(−137.8, 62.0)b
Venlafaxine	110	3364 (525)	−84.8	(−186.3, 16.6)	−85.8	(−196.3, 24.7)d
Any first trimester fill	1534	3408 (565)	−55.2	(−96.6, −13.8)	−70.5	(−113.5, −27.6)b
Any second trimester fill	1045	3363 (562)	−99.6	(−145.5, −53.7)	−115.3	(−164.0, −66.6)b
Any third trimester fill	935	3405 (516)	−54.2	(−99.2, −9.1)	−83.3	(−131.4, −35.2)b
Fill in all trimesters	666	3396 (496)	−64.1	(−112.5, −15.6)	−87.9	(−142.4, −33.5)b
<50% of days covered with fillse	696	3464 (580)	−1.7	(−54.3, 51.0)	−20.8	(−75.6, 34.0)b
≥50% of days covered with fillse	797	3353 (573)	−109.1	(−159.3, −58.9)	−125.4	(−179.5, −71.2)b
Female infantf	859	3321 (571)	−86.4	(−143.6, −29.3)	−106.4	(−164.6, −48.1)b
Male infantf	916	3484 (549)	−32.1	(−87.2, 23.0)	−48.5	(−107.2, 10.3)b

g: grams; SD: standard deviation; CI: confidence interval; SSRIs: selective serotonin reuptake inhibitors

Numbers for specific medications may add to >100% due to some women using multiple antidepressants in pregnancy (89% of deliveries were to women who used one medication, 10% used 2, and 1% used three or four), and categories are overlapping.

SSRI antidepressants also include escitalopram and fluvoxamine, and other types of antidepressants include mirtazapine and desvenlafaxine, but there were too few births exposed to these medications to analyze them separately.

^aWeighted by characteristics included in Variable Set 4 in Supplementary Table 5.

^bWeighted by characteristics included in Variable Set 3 in Supplementary Table 5.

^cWeighted by characteristics included in Variable Set 2 in Supplementary Table 5.

^dWeighted by characteristics included in Variable Set 1 in Supplementary Table 5.

^eFor analyses of days covered, we only included continuers with fills for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants. Because of how days covered was calculated, we also excluded women who had a pre-pregnancy fill of an antidepressant that overlapped into pregnancy if it was a different antidepressant than the one she filled during pregnancy. Therefore, only 1,493 (84% of total continuers) are included in these analyses, instead of the 89% noted in the footnotes of Supplementary Table 7.

^fFemale infants born to continuers were compared with female infants born to discontinuers. Male infants born to continuers were compared with male infants born to discontinuers. There were 621 female infants born to discontinuers, with a mean weight of 3405 g (SD: 545 g). There were 628 male infants born to discontinuers, with a mean weight of 3516 g (SD: 541 g).

Among both continuers and discontinuers, the crude prevalence of SGA was 7% (Table 3). Using IPTW, overall analyses found that continuers did not have greater risk of SGA than discontinuers (RR: 1.14, 95% CI: 0.86, 1.51). For female infants, we observed stronger associations with antidepressant continuation than in the overall analysis, with increased risk of SGA (RR: 1.54, 95% CI: 1.02, 2.32), whereas no significant association was seen in male infants (RR: 0.86, 95% CI: 0.58, 1.28). Results were generally unchanged in analyses of specific antidepressant medications or trimesters of exposure.

Table 3.

Association of antidepressant continuation in pregnancy with small for gestational age.

	No.	No. (%) with small for gestational age	Crude RR (95% CI)	Inverse probability of treatment weighted RR (95% CI)
No antidepressant fill in pregnancy	1249	82 (7%)	ref	ref
Any antidepressant fill in pregnancy	1775	131 (7%)	1.13 (0.87, 1.48)	1.14 (0.86, 1.51)a
SSRIs	1533	108 (7%)	1.08 (0.82, 1.42)	1.09 (0.81, 1.46)b
Citalopram	381	27 (7%)	1.07 (0.70, 1.63)	0.91 (0.55, 1.50)b
Fluoxetine	509	33 (6%)	0.99 (0.67, 1.46)	1.04 (0.69, 1.58)b
Paroxetine	160	10 (6%)	0.94 (0.50, 1.78)	1.61 (0.48, 5.36)c
Sertraline	534	43 (8%)	1.24 (0.87, 1.78)	1.25 (0.84, 1.86)b
Bupropion	240	22 (9%)	1.35 (0.85, 2.16)	1.32 (0.79, 2.20)b
Venlafaxine	110	9 (8%)	1.20 (0.59, 2.43)	0.77 (0.35, 1.67)d
Any first trimester fill	1534	110 (7%)	1.10 (0.84, 1.45)	1.10 (0.82, 1.47)b
Any second trimester fill	1045	76 (7%)	1.11 (0.82, 1.51)	1.15 (0.83, 1.59)b
Any third trimester fill	935	67 (7%)	1.09 (0.80, 1.49)	1.13 (0.81, 1.57)b
Fill in all trimesters	666	46 (7%)	1.04 (0.73, 1.48)	1.03 (0.70, 1.51)b
<50% of days covered with fillse	696	53 (7%)	1.16 (0.83, 1.61)	1.14 (0.80, 1.61)b
≥50% of days covered with fillse	797	55 (7%)	1.05 (0.75, 1.46)	1.15 (0.81, 1.64)b
Female infantf	859	70 (8%)	1.49 (1.00, 2.21)	1.54 (1.02, 2.32)b
Male infantf	916	61 (7%)	0.87 (0.61, 1.25)	0.86 (0.58, 1.28)b

RR: relative risk; CI: confidence interval; SSRIs: selective serotonin reuptake inhibitors

Numbers for specific medications may add to >100% due to some women using multiple antidepressants in pregnancy (89% of deliveries were to women who used one medication, 10% used two, and 1% used three or four), and categories are overlapping.

SSRI antidepressants also include escitalopram and fluvoxamine, and other types of antidepressants include mirtazapine and desvenlafaxine, but there were too few births exposed to these medications to present separately here.

^aWeighted by characteristics included in Variable Set 4 in Supplementary Table 5.

^bWeighted by characteristics included in Variable Set 3 in Supplementary Table 5.

^cWeighted by characteristics included in Variable Set 2 in Supplementary Table 5.

^dWeighted by characteristics included in Variable Set 1 in Supplementary Table 5.

^eFor analyses of days covered, we only included continuers with fills for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants. Because of how days covered was calculated, we also excluded women who had a pre-pregnancy fill of an antidepressant that overlapped into pregnancy if it was a different antidepressant than the one filled during pregnancy. Therefore, only 1,493 (84% of total continuers) are included in these analyses, instead of the 89% noted in the footnotes of Supplementary Table 7.

^fFemale infants born to continuers were compared with female infants born to discontinuers. Male infants born to continuers were compared with male infants born to discontinuers. There were 621 female infants born to discontinuers, with 34 (5%) with SGA. There were 628 male infants born to discontinuers, with 48 (8%) with SGA.

Five percent of babies born to continuers had LBW, compared with 4% among discontinuers (Table 4). In IPTW overall analyses, there was not substantial evidence of differing risk of LBW for antidepressant continuers versus discontinuers (RR: 1.31, 95% CI 0.90, 1.90). Sertraline and bupropion continuers had suggestions of greater risk of LBW compared with discontinuers (RR: 1.56, 95% CI: 0.95, 2.55 and RR: 1.71, 95% CI: 0.83, 3.53, respectively), but these estimates were statistically unstable. Continuers with a fill in the second trimester had a greater risk of LBW than discontinuers (RR: 1.53, 95% CI: 1.02, 2.29). Comparing continuers with ≥50% of days covered with discontinuers, associations strengthened for risk of LBW (RR: 1.69, 95% CI: 1.11, 2.58). Associations were attenuated for LBW when restricting continuers to women with <50% of days covered.

Table 4.

Association of antidepressant continuation in pregnancy with low birthweight (<2500 g).

	No.	No. (%) with low birthweight	Crude RR (95% CI)	Inverse probability of treatment weighted RR (95% CI)
No antidepressant fill in pregnancy	1249	50 (4%)	ref	ref
Any antidepressant fill in pregnancy	1775	87 (5%)	1.22 (0.87, 1.72)	1.31 (0.90, 1.90)a
SSRIs	1533	74 (5%)	1.21 (0.85, 1.71)	1.29 (0.89, 1.89)b
Citalopram	381	17 (4%)	1.12 (0.65, 1.92)	1.16 (0.60, 2.23)b
Fluoxetine	509	21 (4%)	1.03 (0.63, 1.70)	1.21 (0.71, 2.06)b
Sertraline	534	32 (6%)	1.50 (0.97, 2.31)	1.56 (0.95, 2.55)b
Bupropion	240	15 (6%)	1.53 (0.86, 2.72)	1.71 (0.83, 3.53)b
Venlafaxine	110	6 (5%)	1.19 (0.48, 2.92)	1.21 (0.44, 3.38)c
Any first trimester fill	1534	76 (5%)	1.24 (0.87, 1.75)	1.32 (0.90, 1.91)b
Any second trimester fill	1045	58 (6%)	1.38 (0.95, 2.00)	1.53 (1.02, 2.29)b
Any third trimester fill	935	37 (4%)	0.98 (0.65, 1.49)	1.10 (0.69, 1.75)b
Fill in all trimesters	666	26 (4%)	0.96 (0.60, 1.54)	1.10 (0.63, 1.94)b
<50% of days covered with fillsd	696	31 (4%)	1.11 (0.72, 1.73)	1.19 (0.74, 1.89)b
≥50% of days covered with fillsd	797	48 (6%)	1.50 (1.02, 2.21)	1.69 (1.11, 2.58)b
Female infante	859	50 (6%)	1.30 (0.83, 2.03)	1.40 (0.87, 2.27)b
Male infante	916	37 (4%)	1.16 (0.69, 1.94)	1.20 (0.67, 2.14)b

RR: relative risk; CI: confidence interval; SSRIs: selective serotonin reuptake inhibitors

Numbers for specific medications may add to >100% due to some women using multiple antidepressants in pregnancy (89% of deliveries were to women who used one medication, 10% used 2, and 1% used three or four), and categories are overlapping.

SSRI antidepressants also include escitalopram, fluvoxamine, and paroxetine, and other types of antidepressants include mirtazapine and desvenlafaxine, but there were too few births exposed to these medications to present separately here.

^aWeighted by characteristics included in Variable Set 4 in Supplementary Table 5.

^bWeighted by characteristics included in Variable Set 3 in Supplementary Table 5.

^cWeighted by characteristics included in Variable Set 1 in Supplementary Table 5.

^dFor analyses of days covered, we only included continuers with fills for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants. Because of how days covered was calculated, we also excluded women who had a pre-pregnancy fill of an antidepressant that overlapped into pregnancy if it was a different antidepressant than the one she filled during pregnancy. Therefore, only 1,493 (84% of total continuers) are included in these analyses, instead of the 89% noted in the footnotes of Supplementary Table 7.

^eFemale infants born to continuers were compared with female infants born to discontinuers. Male infants born to continuers were compared with male infants born to discontinuers. There were 621 female infants born to discontinuers, with 28 (5%) with LBW. There were 628 male infants born to discontinuers, with 22 (4%) with LBW.

The crude prevalence of LGA was 10% among infants born to continuers and 11% among discontinuers (Table 5). After accounting for confounders, we found no evidence in overall analyses that antidepressant continuation was associated with having an LGA infant (RR: 0.84, 95% CI: 0.67, 1.05). There was evidence of lower risk of LGA associated with antidepressant continuation in female infants, and for specific antidepressants and trimesters of exposure, which are presented in Table 5.

Table 5.

Association of antidepressant continuation in pregnancy with large for gestational age.

	No.	No. (%) with large for gestational age	Crude RR (95% CI)	Inverse probability of treatment weighted RR (95% CI)
No antidepressant fill in pregnancy	1249	139 (11%)	ref	ref
Any antidepressant fill in pregnancy	1775	180 (10%)	0.91 (0.74, 1.12)	0.84 (0.67, 1.05)a
SSRIs	1533	150 (10%)	0.88 (0.71, 1.09)	0.78 (0.62, 0.98)b
Citalopram	381	27 (7%)	0.64 (0.43, 0.95)	0.59 (0.38, 0.92)b
Fluoxetine	509	51 (10%)	0.89 (0.66, 1.21)	0.78 (0.56, 1.08)b
Paroxetine	160	16 (10%)	0.88 (0.53, 1.46)	0.66 (0.22, 1.92)c
Sertraline	534	56 (10%)	0.95 (0.71, 1.27)	0.89 (0.65, 1.21)b
Bupropion	240	27 (11%)	0.98 (0.66, 1.47)	1.20 (0.73, 1.94)b
Venlafaxine	110	10 (9%)	0.81 (0.44, 1.48)	0.76 (0.38, 1.51)d
Any first trimester fill	1534	157 (10%)	0.91 (0.74, 1.13)	0.82 (0.66, 1.03)b
Any second trimester fill	1045	92 (9%)	0.79 (0.61, 1.02)	0.72 (0.55, 0.93)b
Any third trimester fill	935	86 (9%)	0.82 (0.64, 1.07)	0.73 (0.56, 0.96)b
Fill in all trimesters	666	56 (8%)	0.75 (0.56, 1.01)	0.67 (0.49, 0.92)b
<50% of days covered with fillse	696	79 (11%)	1.00 (0.77, 1.30)	0.93 (0.70, 1.23)b
≥50% of days covered with fillse	797	75 (9%)	0.85 (0.65, 1.11)	0.79 (0.60, 1.05)b
Female infantf	859	88 (10%)	0.83 (0.62, 1.10)	0.72 (0.54, 0.98)b
Male infantf	916	92 (10%)	1.02 (0.75, 1.38)	0.93 (0.67, 1.28)b

RR: relative risk; CI: confidence interval; SSRIs: selective serotonin reuptake inhibitors

Numbers for specific medications may add to >100% due to some women using multiple antidepressants in pregnancy (89% of deliveries were to women who used one medication, 10% used 2, and 1% used three or four), and categories are overlapping.

SSRI antidepressants also include escitalopram and fluvoxamine, and other types of antidepressants include mirtazapine and desvenlafaxine, but there were too few births exposed to these medications to present separately here.

^aWeighted by characteristics included in Variable Set 4 in Supplementary Table 5.

^bWeighted by characteristics included in Variable Set 3 in Supplementary Table 5.

^cWeighted by characteristics included in Variable Set 2 in Supplementary Table 5.

^dWeighted by characteristics included in Variable Set 1 in Supplementary Table 5.

^eFor analyses of days covered, we only included continuers with fills for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants. Because of how days covered was calculated, we also excluded women who had a pre-pregnancy fill of an antidepressant that overlapped into pregnancy if it was a different antidepressant than the one she filled during pregnancy. Therefore, only 1,493 (84% of total continuers) are included in these analyses, instead of the 89% noted in the footnotes of Supplementary Table 7.

^fFemale infants born to continuers were compared with female infants born to discontinuers. Male infants born to continuers were compared with male infants born to discontinuers. There were 621 female infants born to discontinuers, with 77 (12%) with LGA. There were 628 male infants born to discontinuers, with 62 (10%) with LGA.

Macrosomia was rare, with a crude prevalence of 2% in both continuers and discontinuers (Table 6). We observed no differing risk of macrosomia comparing continuers of any antidepressant with discontinuers in overall analyses (RR: 0.77, 95% CI: 0.45, 1.31), but we observed markedly lower risk after restricting continuers to women with any fill in the second trimester (RR: 0.46, 95% CI: 0.23, 0.91). Non-significant risk estimates suggesting lower LGA risk for women with any fill in the third trimester and fills in all trimesters are presented in Table 6.

Table 6.

Association of antidepressant continuation in pregnancy with macrosomia (>4500 g).

	No.	No. (%) with macrosomia	Crude RR (95% CI)	Inverse probability of treatment weighted RR (95% CI)
No antidepressant fill in pregnancy	1249	28 (2%)	ref	ref
Any antidepressant fill in pregnancy	1775	38 (2%)	0.95 (0.59, 1.55)	0.77 (0.45, 1.31)a
SSRIs	1533	32 (2%)	0.93 (0.56, 1.54)	0.79 (0.47, 1.32)b
Fluoxetine	509	13 (3%)	1.14 (0.60, 2.18)	1.20 (0.61, 2.36)b
Sertraline	534	12 (2%)	1.00 (0.51, 1.96)	0.86 (0.43, 1.71)b
Bupropion	240	5 (2%)	0.93 (0.36, 2.39)	1.05 (0.40, 2.76)b
Any first trimester fill	1534	34 (2%)	0.99 (0.60, 1.62)	0.87 (0.52, 1.46)b
Any second trimester fill	1045	12 (1%)	0.51 (0.26, 1.00)	0.46 (0.23, 0.91)b
Any third trimester fill	935	15 (2%)	0.72 (0.38, 1.33)	0.57 (0.30, 1.10)b
Fill in all trimesters	666	8 (1%)	0.55 (0.25, 1.19)	0.49 (0.22, 1.08)b
<50% of days covered with fillsc	696	24 (3%)	1.54 (0.90, 2.63)	1.46 (0.83, 2.55)b
≥50% of days covered with fillsc	797	10 (1%)	0.56 (0.27, 1.15)	0.53 (0.25, 1.12)b
Female infantd	859	13 (2%)	0.78 (0.36, 1.70)	0.69 (0.30, 1.57)b
Male infantd	916	25 (3%)	1.07 (0.58, 1.99)	0.89 (0.47, 1.69)b

RR: relative risk; CI: confidence interval; SSRIs: selective serotonin reuptake inhibitors

Numbers for specific medications may add to >100% due to some women using multiple antidepressants in pregnancy (89% of deliveries were to women who used one medication, 10% used 2, and 1% used three or four), and categories are overlapping.

SSRI antidepressants also include escitalopram, fluvoxamine, paroxetine, and citalopram, and other types of antidepressants include mirtazapine and desvenlafaxine, but there were too few births exposed to these medications to present separately here.

^aWeighted by characteristics included in Variable Set 4 in Supplementary Table 5.

^bWeighted by characteristics included in Variable Set 3 in Supplementary Table 5.

^cFor analyses of days covered, we only included women with fills for one specific antidepressant during the exposure window because it was difficult to distinguish sequential from simultaneous use for women who filled multiple antidepressants. Because of how days covered was calculated, we also excluded women who had a pre-pregnancy fill of an antidepressant that overlapped into pregnancy if it was a different antidepressant than the one she filled during pregnancy. Therefore, only 1,493 (84% of total continuers) are included in these analyses, instead of the 89% noted in the footnotes of Supplementary Table 7.

^dFemale infants born to continuers were compared with female infants born to discontinuers. Male infants born to continuers were compared with male infants born to discontinuers. There were 621 female infants born to discontinuers, with 12 (2%) with macrosomia. There were 628 male infants born to discontinuers, with 16 (3%) with macrosomia.

Sensitivity analyses

In the analysis defining exposure as receiving ≥2 antidepressant fills, we observed a stronger association than in the overall analysis for mean birthweight (mean difference: −106.1 g, 95% CI: −151.8, −60.4, Supplemental Table 9). Compared with results for the primary analyses of mean birthweight and LBW, adjustment for gestational age attenuated the associations (Supplemental Table 9 and 11).

Results were generally similar to overall analyses after restricting to women with anxiety or depression diagnoses, restricting to women without a bipolar disorder diagnosis or associated medications, re-categorizing women with prescriptions that overlapped pregnancy as exposed, adjusting for BMI, or using multivariable-adjusted models (Supplemental Tables 9–13).

DISCUSSION

In overall analyses, infants born to antidepressant continuers weighed slightly less than discontinuers’ infants, and this association was stronger in female infants. Overall, risks of SGA, LBW, LGA, and macrosomia did not differ by antidepressant continuation status. However, SGA risk was greater in continuers than discontinuers specifically in female infants, and LBW risk was greater in continuers with ≥50% of days covered than discontinuers. We observed corresponding lowered risk of LGA in certain sub-analyses. Continuers with antidepressant fills later in pregnancy were at lower risk of macrosomia compared with discontinuers. We are less certain our findings apply to antidepressants with few users in our study (e.g. bupropion, venlafaxine, escitalopram). Given the large number of associations examined, our findings warrant confirmation.

Similar to our study, most previous studies that analyzed differences in mean birthweight associated with antidepressant use in pregnancy reported small differences.^28,45–50 The majority studies of SGA do not sufficiently account for confounding by mental health conditions and severity, or other key potential confounders such as substance use disorders.^51–53 Studies that did account for these did not report greater risk of SGA, including a study that compared discordantly-exposed siblings (RR: 1.01, 95% CI: 0.81, 1.25)⁴⁴ and two studies comparing women who used antidepressants in pregnancy with women who had used antidepressants prior to pregnancy (RR: 0.93, 95% CI: 0.73, 1.19⁵⁴ and RR: 1.00, 95% CI: 0.87, 1.15⁵⁵). These studies were consistent with our overall findings (RR: 1.14, 95% CI: 0.86,1.51), potentially because their designs accounted for potential residual confounding.

We found potential effect measure modification by infant sex. Among female infants, those born to continuers had lower mean birthweight and greater risk of SGA, as well as corresponding lower risk of LGA, than those born to discontinuers; these associations were not found in male infants. To our knowledge, this is the first study to assess risk separately by infant sex. There is precedent for sex-specific fetal effects, including evidence that antidepressant exposure in utero may have a stronger deleterious effect on females’ neurodevelopment than males’,³¹ as well as evidence that other prenatal exposures, such as prenatal Vitamin D intake and physical activity, have differential impacts on male and female birthweight.^32,33

Many studies have assessed antidepressant use in pregnancy and LBW, and increased risk was found in meta-analyses (RR: 1.44, 95% CI: 1.21, 1.70⁵⁶ and RR: 1.38, 95% CI: 1.13, 1.69⁵²). However, few studies restricted comparator groups to women with a history of depression or pre-pregnancy antidepressant use, and these studies found conflicting results. Two reported no association (RR: 1.00, 95% CI: 0.74, 1.35⁵⁵ and RR: 1.1, 95% CI: 0.9, 1.3⁵⁷) and two reported >2-fold higher risks (odds ratio [OR]: 2.26; 95% CI: 1.31, 3.91⁵⁸ and OR: 2.73, 95% CI: 0.92, 8.09⁵⁹). In comparison, our overall estimate suggests moderately increased risk, but was statistically imprecise (RR: 1.31, 95% CI: 0.90, 1.90). Our findings that the association strengthened when restricting continuers to those with longer duration of use (≥50% of days covered: RR: 1.69, 95% CI: 1.11, 2.58) or use in the 2^nd trimester (RR: 1.53, 95% CI: 1.02, 2.29), were not consistent with the previous studies.^55,57–59 The conflicting findings between studies could be due to the variation in study populations across a number of different countries, with different antidepressant prescribing and use patterns.

Few studies have focused on antidepressant use in pregnancy and macrosomia or LGA. Consistent with our findings, a large, Swedish cohort study reported that use later in pregnancy and use both early and later in pregnancy was associated with decreased odds of macrosomia (OR: 0.86, 95% CI: 0.74, 0.99 and OR: 0.86, 95% CI: 0.73, 1.03), whereas use early in pregnancy was not associated with macrosomia (OR: 1.04, 95% CI: 0.95, 1.13).⁶⁰ However, their findings for LGA were inconsistent with ours: they did not observe lower risk of LGA with use later in pregnancy (OR: 1.08, 95% CI: 0.97, 1.20), use both early and later in pregnancy (OR: 1.15, 95% CI: 1.01, 1.30), or lower risk associated with SSRI use later in pregnancy (OR: 1.06, 95% CI: 0.93, 1.19). Limitations of their study include that they compared antidepressant users to the general population of pregnant women instead of restricting to those with depression or anxiety, or pre-pregnancy antidepressant use as a proxy for said conditions, and did not adjust for indicators of mental health condition severity or other key potential confounders, such as substance use disorders.

Our study was strengthened by utilizing a large, well-defined study population, prescription fill data (recorded prospectively, preventing recall bias), and extensive information about baseline characteristics, including mental health care utilization. We also had information about PHQ-9 depression screening scores, which showed that pre-pregnancy depression scores were similar between continuers and discontinuers.

One study limitation was that we only had small numbers of women exposed to some antidepressants, decreasing power to detect modest effects. We also cannot discount the possibility of residual confounding. For example, we used diagnosis data to create substance use disorder covariates, but we recognize that these conditions may be underdiagnosed in routine care, and we did not have detailed information on which disorders were active during the perinatal period. Given that we used antidepressant medication fills to define exposure, there could be exposure misclassification if women filled but did not actually take the medication, potentially attenuating risk estimates. There is some evidence of differing impacts on birthweight for specific trajectories of antidepressant use or dose level,^61,62 which we did not assess. Also, our study population consisted of live births to women who were largely commercially insured and non-Hispanic white, which may limit generalizability.

In conclusion, in our study, antidepressant continuation was associated with slightly lower birthweight, and for those with later or longer duration of prenatal exposure, and particularly for female infants, this translated to lower risk of LGA or macrosomia and greater risk of SGA or LBW. These results may be signaling an underlying true biological effect, although they need to be replicated in future research. If confirmed, the small differences in birthweight we found to be associated with antidepressant use are not clinically actionable in isolation but could be if antidepressant use was combined with other risk factors. Many outcomes beyond birthweight should be considered when deciding whether to continue antidepressants in pregnancy, including neonatal syndromes, childhood neurodevelopmental outcomes, and the woman’s mental health and wellbeing during and after pregnancy.