Maternal, Fetal, and Child Outcomes of Mental Health Treatments in Women: A Meta-Analysis of Pharmacotherapy.

Objective: The authors systematically reviewed evidence on pharmacotherapy for perinatal mental health disorders.
Methods: The authors searched for studies of pregnant, postpartum, or reproductive-age women with mental health disorders treated with pharmacotherapy in MEDLINE, EMBASE, PsycINFO, the Cochrane Library, and trial registries from database inception through June 5, 2020 and surveilled literature through March 2, 2021. Outcomes included symptoms; functional capacity; quality of life; suicidal events; death; and maternal, fetal, infant, or child adverse events.
Results: 164 studies were included. Regarding benefits, brexanolone for third-trimester or postpartum depression onset may be associated with improved depressive symptoms at 30 days when compared with placebo. Sertraline for postpartum depression may be associated with improved response, remission, and depressive symptoms when compared with placebo. Discontinuing mood stabilizers during pregnancy may be associated with increased recurrence of mood episodes for bipolar disorder. Regarding adverse events, most studies were observational and unable to fully account for confounding. Evidence on congenital and cardiac anomalies for treatment compared with no treatment was inconclusive. Brexanolone for depression onset in the third trimester or the postpartum period may be associated with risk of sedation or somnolence, leading to dose interruption or reduction when compared with placebo. Conclusions: Evidence from few studies supports the use of pharmacotherapy for perinatal mental health disorders. Although many studies report on adverse events, they could not rule out underlying disease severity as the cause of the association between exposures and adverse events. Patients and clinicians need to make informed, collaborative decisions on treatment choices.

Untreated mental health disorders in perinatal (pregnant and postpartum, including breastfeeding) women can have devasting sequelae. Pregnancy‐associated suicide kills more women than either hemorrhage or preeclampsia (1). Depressive symptoms are associated with reduced child safety, increased harsh punishment, impaired development of infant emotional regulation and attachment, and greater risk of child psychiatric disease (2, 3, 4, 5, 6). Some psychotropic medications that can improve outcomes may be accompanied by potential adverse events specific to pregnancy, lactation, and males and females with reproductive potential, as laid out in FDA prescription labels (7). For women who are pregnant or planning pregnancy, a critical question for women and their clinical providers is whether the potential maternal and fetal benefits of treating psychiatric illness with pharmacologic interventions outweigh the potential harms. This challenging benefit‐harm tradeoff is further complicated by the potential for underlying mental health disorders to serve as confounders. Underlying mental health disorders result in the use of psychotropic medications. Underlying mental health disorders may also result in harms regardless of exposure to medications. Uncertainty in the management of maternal mental health disorders (8, 9, 10) points to the need for a systematic review to help clarify the balance of benefits and harms from psychotropic drugs for these disorders.

METHODS

This review followed an a priori protocol (11) and adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) (12) statement. None of the authors reported financial conflicts of interest.

Scope of the Review

Figure 1 shows the key questions (KQs) guiding the review. Detailed methods (searches, list of excluded studies, approach to risk of bias, grading, and synthesis), evidence tables, results (including detailed results for all outcomes rated as insufficient), and meta‐analyses are available in the full evidence report at https://doi.org/10.23970/AHRQEPCCER236.

FIGURE 1

Key questions: Maternal, fetal, and child outcomes of mental health treatments in women

Data Sources and Searches

PubMed, the Cochrane Library, PsycInfo, and EMBASE were searched for English language articles published from database inception through June 5, 2020. ClinicalTrials.gov, Cochrane Register of Controlled Trials, and the World Health Organization International Clinical Trials Registry Platform were also searched. Additionally, reference lists of pertinent articles and studies suggested by reviewers were searched. Ongoing surveillance of article alerts and targeted journal searches through March 2, 2021 identified no major studies published since the last search.

Inclusion and Exclusion Criteria

For benefits (KQ 1 and KQ 2), pregnant or postpartum women were included; for harms (KQ 3 and KQ 4), pregnant, postpartum, or reproductive‐age women were included. Because benefits are disorder‐specific but harms may not be, the review's population inclusion criteria were more inclusive for harms questions than benefits questions. An anxiety, depression, bipolar disorder, or schizophrenia diagnosis was required for benefits questions but not for harms questions. All U.S. Food and Drug Administration (FDA)‐approved drugs for mental health disorders and off‐label drugs used for mental health disorders were eligible. Outcomes included maternal benefits (symptoms, functional capacity, quality of life, delivery mode, breastfeeding, weight change, change in suicidal events); maternal harms (miscarriage, abruption, preterm labor/preterm birth, preeclampsia, gestational hypertensive disorders, gestational diabetes mellitus); and fetal, infant, or child harms (preterm birth, small or large for gestational age, congenital anomalies, Apgar score, withdrawal, respiratory distress, neonatal intensive care unit time, persistent pulmonary hypertension, delayed development, child mental health disorders, and death). Given limited trial evidence, all designs with a comparison arm were eligible (randomized controlled trials [RCTs], nonrandomized clinical trials, cohorts with comparisons, and case‐controls). However, because of the lack of controls for potential confounding in observational studies, the review cataloged all evidence but focused synthesis on more robust evidence. Specifically, for benefits/harms (KQ 1 and KQ 3, comparing treatment with no treatment), the synthesis was limited to studies controlling for confounding. For comparative benefits/harms (KQ 2 and KQ 4), the synthesis included studies that did not attempt to control for confounding, but these studies were downgraded for risk of bias.

Study Selection

Two investigators independently reviewed titles, abstracts, and full‐text articles using prespecified inclusion criteria (Table S1 in the online supplement); conflicts were resolved by discussion or by a third reviewer. The Supplement also includes a complete list of all excluded studies along with reasons for exclusion.

Data Extraction and Quality Assessment

For each included study, one reviewer abstracted relevant study characteristics and outcomes into a structured form. A second reviewer checked all data for completeness and accuracy. Two senior reviewers independently assessed each study's methodological quality using predefined criteria using the ROBINS‐I (13) tool for observational studies and the Cochrane ROB 2.0 (14) tool for RCTs; a third senior reviewer helped resolve conflicts.

Data Synthesis and Analysis

Data were synthesized in tabular and narrative formats. When at least three similar studies were available, a quantitative synthesis was performed using random effects models with the DerSimonian and Laird inverse‐variance weighted method in Comprehensive Meta‐Analysis (Version 3.3) software (15) to generate pooled estimates of the absolute risk difference (ARD) and the RR (16).

The strength of evidence (SOE) was assessed through the Grading of Recommendations Assessment, Development and Evaluation working group guidance (17), and guidance established for the Evidence‐based Practice Center Program (18). Based on the assessment of study limitations, directness, consistency, precision, and reporting bias for each intervention comparison and major outcome of interest, evidence grades were assigned as high, moderate, low, or insufficient. Two senior reviewers independently developed initial SOE assessments for each relevant outcome and comparison across the KQs and resolved disagreements through discussion.

RESULTS

A total of 164 studies (168 articles) met eligibility criteria. Figure S1 in the online supplement depicts the article flowchart. The results below focus on primarily on evidence offering at least a low level of certainty. However, all results, including those with insufficient certainty to arrive at conclusions are described in Tables S2–S109 in the online supplement and in the full report (https://doi.org/10.23970/AHRQEPCCER236). Key results are presented below in order of certainty.

Effectiveness of Perinatal Pharmacotherapy

Evidence on benefits of pharmacotherapy in pregnant and postpartum women is sparse (9 RCTs and 10 observational studies).

When evidence was available, the benefit of pharmacotherapy for depression and bipolar disorder was graded as having low to moderate strength. Table 1 maps all the evidence on benefits for perinatal pharmacotherapy.

TABLE 1

Summary of evidence for maternal benefit for treatment versus placebo, no treatment, or active comparators for mental health disorders in pregnancy and postpartum

Note: I: Insufficient evidence for conclusions on the outcome, that is, an insufficient rating indicates that the evidence does not permit estimation of an effect because multiple domain ratings indicate weakness in the evidence base (i.e., the evidence base may comprise studies with limitations such as uncontrolled or poorly controlled confounding or high and differential attrition; be inconsistent, indirect, or imprecise; or be biased in reporting); M: Moderate evidence of benefit for at least one measure for the outcome domain, that is, a moderate rating implies moderate confidence that the evidence reflects the true effect. Further research may change our confidence in the estimate of the effect and may change the estimate; L: Low evidence of benefit for at least one measure for the outcome domain, that is, a low rating implies low confidence that the evidence reflects the true effect. Further research is likely to change our confidence in the estimate of the effect and is likely to change the estimate.; ‐: No eligible evidence.

Abbreviations: SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.

Sedative hypnotics may be prescribed for sleep disturbances that occur during any mental health disorder as well as in the presence of no diagnosable mental health disorder; they may also be used, at times, off label as an anti‐anxiety alternative.

No prespecified outcomes were evaluated but one study reported on psychiatric admissions; the evidence was insufficient to judge the effectiveness of lamotrigine versus lithium.

Specifically, for depression, moderate SOE from three RCTs suggests that postpartum brexanolone for depression onset in the third trimester or postpartum is associated with improved depressive symptoms shortly after infusion (60 h) (least square [LS] mean difference in the Hamilton Rating Scale for Depression [HAM‐D], −4.1, standard error [SE] 0.9, p < 0.001) and at 30 days after treatment (LS mean difference, SE −2.6, 1.1, p = 0.02) (30, 31). Brexanolone results were mixed for other depression measures.

Two RCTs provide low SOE suggesting that compared with placebo, sertraline for postpartum depression may improve response (Table 2) (RR, 4.00; 95% CI, 0.98 to 16.31; N = 36), remission (RR, 4.22; 95% CI, 0.98 to 18.12; N = 36), and depressive and anxiety symptoms (regression coefficients range from 0.91 to 1.18, p‐values range from 0.01 to 0.05) (24, 25).

TABLE 2

Detailed results for key outcomes rated low strength of evidence

Population	Outcome a	Exposure	Comparator	Incidence	N	Results	ARD per 1000 women b (95% CI)	NNTB/NNTH c (95% CI)	Key considerations
Women with postpartum depression	Response	Sertraline	Placebo	10/17 (59%) versus 5/19 (26%)	36	RR in low risk‐of‐bias study: 2.24 (95% CI, 0.95 to 5.24) (24) no difference between two arms in high risk‐of‐bias study, p = 0.054 (25)	326 (11 to 1116)	NNTB:4 (1 to 91)	RCT evidence, small sample, imprecise results
Women with postpartum depression	Remission	Sertraline	Placebo	9/17 (53%) versus 4/19 (21%)	36	RR: 2.52 (95% CI, 0.94 to 6.70); (24) no difference between two arms in high risk‐of‐bias study, p = 0.372 (25)	320 (4 to 619)	NNTB:4 (2 to 250)	RCT evidence, small sample, imprecise results
Women with bipolar disorder	Recurrence from discontinuation	Discontinuedmood stabilizers	Continued mood stabilizers	53/62 (85.5%) versus.10/27 (37%)	89	AHR: 2.2 (95% CI, 1.2 to 4.2) (32)	268 (56 to 486)	NNTH: 4 (3 to 18)	Observational evidence, small sample, imprecise results
Women with bipolar disorder	Time‐to‐25%‐recurrence from discontinuation	Discontinued mood stabilizers	Continued lamotrigine	16/16 (100%) versus 3/10 (30%)	26	AHR: 12.1 (95% CI, 1.6 to 91.7) (33)	687 (135 to 700)	NNTH: 2 (2 to 8)	Observational evidence, study has limitations, small sample, imprecise results
Women with at least one anxiety diagnosis in the year before conception	Ectopic pregnancy	Benzodiazepine exposure 90 days before conception	No benzodiazepine exposure before conception	249/9188 (2.71%) versus 1730/81,291 (2.13%)	90,479	ARR: 1.33 (95% CI, 1.17 to 1.51) (37)	7 (4 to 11)	NNTH: 143 (91 to 250)	Observational evidence, potential for residual confounding
Women with mood or anxiety disorder	Postpartum hemorrhage	Exposed to SSRIs during delivery	Unexposed to SSRIs during delivery	503/12,710 (3.96%) versus 1896/69,044 (2.75%)	81,754	ARR, 1.47 (95% CI, 1.33 to 1.62) (38)	13 (9 to 17)	NNTH: 77 (59 to 112)	Observational evidence, potential for residual confounding
Women with mood or anxiety disorder	Postpartum hemorrhage	Exposed to SNRIs during delivery	Unexposed to SNRIs during delivery	35/702 (5.0%) versus 1896/69,044 (2.75%)	69,746	ARR, 1.90 (1.37 to 2.63) (38)	25 (10 to 45)	NNTH: 40 (23 to 100)	Observational evidence, potential for residual confounding
SNRI exposure or depression diagnosis, through second trimester	Preeclampsia	SNRIs exposure through second trimester	Unexposed depressed	107/1216 (9%) versus 3215/59,219 (5%) (39); 23/408 (5.6%) versus 1569/65,392 (2.4%) (40)	65,800	ARR, 1.52 (95% CI, 1.25 to 1.83) (39); ARR, 1.59 (95% CI, 1.26 to 3.03) (40)	14 (6 to 49)	NNTH: 72 (21 to 167)	Observational evidence, potential for residual confounding
Women with depression	Preeclampsia	Exposed to TCAs in pregnancy	Unexposed to TCAs in pregnancy	47/441 (10.7%) versus, 3215//59,219 (5.4%) (39); 14/146 (9.59%) versus 1569/65,392 (2.40%) (40)	65,538	ARR, 1.62 (95% CI, 1.23 to 2.12) (39); ARR, 3.23 (95% CI, 1.87 to 5.59) (40)	54 (21 to 110)	NNTH: 19 (10 to 48)	Observational evidence, potential for residual confounding
Women prescribed second‐generation antipsychotic	Gestational diabetes	Quetiapine continued in pregnancy	Quetiapine discontinued in pregnancy	110/1543 (7.1%) versus 122/2990 (4.1%)	4533	ARR, 1.28 (95% CI, 1.01 to 1.62) (41)	11 (<1 to 25)	NNTH: 88 (40 to 2439)	Observational evidence, potential for residual confounding
Women prescribed second‐generation antipsychotic	Gestational diabetes	Olanzapine continued in pregnancy	Olanzapine discontinued in pregnancy	46/384 (12.0%) versus 49/1041 (4.7%)	1425	ARR, 1.61 (95% CI, 1.13 to 2.29) (41)	29 (6 to 61)	NNTH: 35 (17 to 167)	Observational evidence, potential for residual confounding
Pregnant women with depression or anxiety	Spontaneous abortion	Benzodiazepine exposure in first trimester (42) or within the first 19 weeks (43) versus untreated or a history of mood disorders or anxiety during pregnancy	Unmedicated mental illness	386/2384 (16%) versus 442/3647 (12%) (42)	6031	ARR, 1.6 (95% CI, 1.3 to 1.9) (42)	73 (36 to 109)	NNTH: 14 (10 to 28)	Observational evidence, potential for residual confounding
				198 cases/570 controls versus 3221 cases/15 382 controls (43)		AOR: 2.85 (95% CI, 1.72 to 4.72) (43)
Women with SNRI exposure or depression diagnosis in past 4 years	Spontaneous abortion	SNRI exposure in 1st trimester	Unexposed with depression diagnosis in past 4 years	20/90 (22%) versus 720/7034 (10%); results corrected for induced abortions: 20/137 (15%) versus 720/8877 (8.1%)	9014	ARR, 2.1 (95% CI, 1.4 to 3.0); corrected for induced abortions ARR, 1.7 (95% CI, 1.2 to 2.6) (44)	62 (16 to 130)	NNTH: 17 (8 to 63)	Observational evidence, potential for residual confounding
Women with a mental health disorder	NICU admission	Benzodiazepine exposure during pregnancy	Unexposed to benzodiazepine during pregnancy	32/144 (22.2%) versus 125/649 (19.3%)	793	AOR, 2.02 (95% CI, 1.11 to 3.66) (45)	133 (17 to 274)	NNTH: 8 (4 to 59)	Observational evidence, potential for residual confounding, benzodiazepine exposure may result in NICU admission by policy
Women exposed to SSRIs or unexposed with a psychiatric diagnosis	Apgar score <7 at 5 min	Exposed to SSRIs during pregnancy	Exposed to SSRIs before pregnancy or unexposed with a psychiatric diagnosis	28/2664 (1.1%) versus 31/5141 (0.6%) (46)	25,381	Adjusted prevalence ratio: 1.69 (95% CI, 1.02 to 2.79) (46)	8 (4 to 13)	NNTH: 125 (77 to 250)	Observational evidence, potential for residual confounding, transient outcome
				376/15,729 (2.4%) versus 113/9652 (1.2%) (47)		AOR, 1.68 (95% CI, 1.34 to 2.12) (47)
Women with depression	Persistent pulmonary hypertension of the newborn	Exposed to SSRIs during pregnancy	Unexposed to SSRIs during pregnancy	94/54,281 (0.2%) versus 669/567,118 (0.1%) (48)	621,399	Adjusted OR, 1.28 (95% CI, 1.01 to 1.70) (48)	1 (<1 to 1)	NNTH: 3031 (1220 to 11,112)	Observational evidence, potential for residual confounding
						AOR, when not restricted to full‐term or by outcome (persistent pulmonary hypertension rather than primary persistent pulmonary hypertension): 1.08 (95% CI, 0.92 to 1.27) (48)
Women with a mental health disorder	Childhood depression	Exposed to SSRIs during pregnancy	Unexposed to SSRIs during pregnancy	60/15,729 (0.4%) versus 30/9651 (0.3%)	25,380	AHR, 1.78 (95% CI, 1.12 to 2.82) (49)	2 (0 to 6)	NNTH: 414 (178 to 2703)	Observational evidence, potential for residual confounding
Women with a mental health disorder	Autism spectrum disorder without intellectual disabilities	Exposed to citalopram during pregnancy	Unexposed to any antidepressants during pregnancy	46/1064 (4.3%) versus 291/12,325 (2.4%)	13,389	AOR, 1.75 (95% CI, 1.25 to 2.45) (50)	17 (6 to 32)	NNTH: 59 (32 to 167)	Observational evidence, potential for residual confounding

Abbreviations: AOR, adjusted odds ratio; ARD, absolute risk difference; ARR, adjusted relative risk; CI, confidence interval; NICU, neonatal intensive care unit; NNTB, number needed to treat for an additional beneficial outcome; NNTH, number needed to treat for an additional harmful outcome; SNRI, serotonin‐norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.

Key outcomes were selected from evidence bases rated as at least low strength of evidence.

The ARD was calculated based on adjusted effect sizes when available. When multiple studies were available on harms, cohort studies with the highest effect size to present the outer bound for harms were selected.

When multiple studies were available on harms, the NNTB and NNTH were based on the results reported for the ARD. NNTB and NNTH and their upper and lower confidence intervals are rounded up to the next whole number.

For bipolar disorders, two cohort studies provide low SOE that discontinuing mood stabilizers during pregnancy may be associated with increased recurrence when compared with continuing mood stabilizers (AHR, 2.2; 95% CI, 1.2 to 4.2; N = 89) (32) and time‐to‐25%‐recurrence of mood episodes when compared with continuing lamotrigine (2 vs. 28 weeks, AHR, 12.1; 95% CI, 1.6 to 91; N = 26) (33).

No evidence was available for schizophrenia. Evidence did not permit conclusions for any drug for anxiety disorder, nor for fluoxetine or paroxetine for depression.

Comparative Effectiveness of Perinatal Pharmacotherapy

One RCT and 10 observational studies of exposure provided insufficient evidence to judge the comparative effectiveness of a limited number of outcomes and interventions for depression and bipolar disorder (Table 1). For anxiety and schizophrenia, no eligible evidence on comparative effectiveness was found.

Harms of Perinatal Pharmacotherapy

Five RCTs and 70 observational studies were included. The authors judged the certainty of evidence to draw conclusions to be insufficient or low in all instances (Table 3), primarily because of lack of control for confounding. Table 2 indicates small absolute risk differences for all adverse events. Key outcomes—specifically, evidence from RCTs, for an association observed for more than one drug, or for serious adverse events—graded as low SOE are described below, as is evidence for congenital and cardiac anomalies (graded insufficient).

TABLE 3

Summary of evidence for harms from pharmacotherapy versus no treatment for mental health disorders in pregnancy or postpartum

Note: I: Insufficient evidence for conclusions on the outcome, that is, an insufficient rating indicates that the evidence does not permit estimation of an effect because multiple domain ratings indicate weakness in the evidence base (i.e., the evidence base may comprise studies with limitations such as uncontrolled or poorly controlled confounding or high and differential attrition; be inconsistent, indirect, or imprecise; or be biased in reporting); L: Low evidence of benefit for at least one measure for the outcome domain, that is, a low rating implies low confidence that the evidence reflects the true effect. Further research is likely to change our confidence in the estimate of the effect and is likely to change the estimate.; ‐: No eligible evidence.

Abbreviations: ADHD, attention‐deficit/hyperactivity disorder; SNRI, serotonin‐norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.

Included studies also reported a higher risk of ectopic pregnancy for benzodiazepine use when compared with no benzodiazepine exposure (graded low for harms).

Used as a sedative‐hypnotic to address sleep difficulty, not targeted at underlying mental health disorder.

Although no evidence of harms for brexanolone was found for a priori outcomes, the evidence on dose interruption or reduction due to somnolence or sedation was graded as low.

Data obtained by personal communication with authors after release of their publication (77).

Underlying mental health indication and exposure to medications are generally not specified; study populations are based on cohorts with prescriptions.

Maternal Harm

Postpartum Hemorrhage

Regarding postpartum hemorrhage, results from one propensity score–adjusted study suggested an association between exposure to several antidepressants and postpartum hemorrhage that may vary by timing (38). Current exposure (at the time of delivery) is associated with postpartum hemorrhage for selective serotonin reuptake inhibitors (SSRIs) as a class, citalopram, escitalopram, sertraline, serotonin‐norepinephrine reuptake inhibitors (SNRIs) as a class, and venlafaxine. Recent exposure (up to 1 month before delivery) is also associated with an increased risk of postpartum hemorrhage for SSRIs as a class, paroxetine, and sertraline (low SOE of harms). For bupropion, however, the association is with past exposure (supply of drug 1 to 5 months before delivery). Previous studies suggested that SSRIs in particular can reduce platelet function and result in bleeding because the drugs inhibit serotonin reuptake into platelets (85). In a sample with a prevalence of 2.75% in the unexposed arm, the ARD for SSRIs is 13 more cases per 1000 (95% CI, 9 more to 17 more); for SNRIs, the risk is 25 more per 1000 (95% CI, 10 more to 45 more). The mechanism of action for other antidepressants is not clear, and associations could potentially be attributable to chance (38). The risk of residual confounding also remains: the study was not able to control for confounding factors of inadequate diet, tobacco use, and disorder severity. Additionally, the study was not powered to examine the association between antidepressants and severe postpartum hemorrhage leading to blood transfusion, further morbidity, or mortality.

Preeclampsia

Low SOE from two studies (39, 40) suggests increased risk of preeclampsia for women exposed to non‐SSRI drugs, specifically SNRIs and TCAs; ARDs range from 14 more cases per 1000 to 54 cases per 1000. Risk factors for preeclampsia include maternal antiphospholipid antibody syndrome, family history, nulliparity, donor egg, diabetes, obesity, and preexisting hypertension. These factors may cause placental hypoxia and ischemia. However, antidepressants may increase serotonin and norepinephrine levels and therefore contribute to preeclampsia (40). Because both serotonin and norepinephrine are vasoconstrictors, SNRIs could affect preeclampsia risk through placental ischemia (86). Although both studies controlled for age, primiparity, multiple gestation, diabetes, proxies of depression severity (such as number of visits or claims), and general markers of comorbidity (such as healthcare and prescription use other than for depression), results are not consistent across exposures: these differences may be attributable to residual confounding because of inadequately controlled depression severity or another comorbid condition.

Gestational Diabetes

For gestational diabetes mellitus (GDM), low SOE from one cohort study (41) suggested that continuing use (as reflected in two or more dispensed prescriptions) during the first half of pregnancy of quetiapine (ARD, 11 per 1000; 95% CI, 0 to 25) or olanzapine (ARD, 29 per 1000; 95% CI, 6 to 61) may be associated with an increased risk of developing GDM compared with women discontinuing these medications before the start of pregnancy. Antipsychotics may result in changes in appetite and diet because of interactions with serotonergic, histaminergic, and dopaminergic neurotransmitter systems (87). FDA notes the risk of metabolic side effects from second‐generation antipsychotics in prescription labels for these drugs (88, 89). The other second‐generation antipsychotics analyzed that showed no clear difference in risk were aripiprazole, ziprasidone, and risperidone. Women who continued antipsychotic treatment during pregnancy generally had higher comorbidity and longer baseline antipsychotic use. Of note, these studies used a generalized linear model and propensity score stratification to obtain risks of developing GDM with adjustment for confounders, such as demographic data, psychiatric diagnoses, comorbidity, other medication use, history of gestational diabetes, and the duration of antipsychotic treatment received during the 3 months before the last menstrual period.

Early Pregnancy Loss

Evidence from one study (37) suggested an increased risk of ectopic pregnancy with benzodiazepine exposure 90 days before conception when compared with no exposure before conception, among pregnant women with at least one anxiety disorder diagnosis in the year before conception (graded low SOE of harms). Risk factors for ectopic pregnancy include previous ectopic pregnancy, history of pelvic infection, infertility, cigarette smoking, and age older than 35 years (90). Confounding by indication may explain these results; however, one suggested mechanism of action could be through the central relaxation of smooth muscle and the direct effect on gamma‐aminobutyric acid receptors in the fallopian tube, potentially resulting in a higher incidence of ectopic pregnancy (37). The ARD is 7 per 1000 (95% CI, 4 to 11).

One study suggested that SNRI exposure in the first trimester may be associated with a higher rate of spontaneous abortion when compared with no exposure (low SOE) (44). Analyses controlling for induced abortion showed a slightly attenuated but still statistically significant difference (ARD, 62 per 1000; 95% CI, 16 to 130). The comparison group includes women with a depression diagnosis in the 4 years preceding pregnancy. Although the authors adjusted for use of teratogenic medication in the first trimester, number of prescription medications in 3 months before pregnancy, and number of mental health visits in 3 months before pregnancy, the two comparison groups likely had different baseline severity, which is unaccounted for in this analysis.

Evidence from two studies (42, 43) suggested an increased risk of spontaneous abortion with benzodiazepine exposure 90 days before conception when compared with untreated women or women with a history of mood disorders or anxiety (graded low SOE of harms). To the extent that women experiencing greater psychological stress would be more likely to be treated pharmacologically, confounding by indication by underlying may explain the observed associations. Although residual confounding may explain these results, as with the results for ectopic pregnancy, the authors note that benzodiazepines cross the placental barrier easily and may accumulate in fetal issues (37). The ARD is 73 per 1000 (95% CI, 36 to 109).

Excessive Sedation or Loss of Consciousness

FDA includes a boxed warning on the prescribing information of excessive sedation or loss of consciousness in the active arm for brexanolone leading to dose interruption or reduction (5% vs. 0% for the placebo arm). In the prescribing instructions, the manufacturer reports higher rates of dizziness, loss of consciousness, and somnolence with brexanolone compared with placebo. The pooled RR for somnolence was 2.00 (95% CI, 0.78 to 5.16; Figure S2 in the online supplement) and the ARD was 0.05 (95% CI, −0.01 to 0.12). Results could not be pooled for loss of consciousness from the individual studies; they did not appear to increase with dose intensity: 5% of women randomized to brexanolone with a maximum dose of 60 μg/kg per hour versus 3% for brexanolone with a maximum dose of 90 μg/kg per hour experienced loss of consciousness.

Fetal, Infant, or Child Harms

Persistent Pulmonary Hypertension of the Newborn

For the study reporting an association between persistent pulmonary hypertension of the newborn and SSRIs, like with other observational cohorts, residual confounding and the potential for misclassification may exist. Risk factors like smoking, obesity, and Caesarean section are all more prevalent in populations of psychiatric patients (91). However, adjustments for potential sources of confounding (restricting the sample to full‐term births (91) and restricting the outcome to those without cardiac anomalies or hypoplasia) resulted in higher odds (odds ratio [OR], 1.28; 95% CI, 1.01 to 1.64) than results without these adjustments (OR, 1.10; 95% CI, 0.94 to 1.29) when compared with no exposure (48). Notably, the baseline risk (0.1% in the unexposed arm) and the ARD (33 more cases per 100,000 persons, 95% CI, 1 to 83 more cases) are very low; in the analyses adjusting for potential sources of confounding, the ARD is lower still at 9 more cases per 100​,000 (95% CI ranges from 9 fewer cases to 32 more cases). These results suggest that although the exposure may be associated with a higher risk of a potentially serious complication, the absolute risk of harm is very low (low SOE).

Congenital Anomalies

FDA prescription labeling suggests potential concerns for congenital anomalies for several psychotropic drugs, including paroxetine, temazepam, triazolam, alprazolam, diazepam, valproate, carbamazepine, and topiramate. The evidence was insufficient to judge the risks of congenital anomalies (Figure 2) and cardiac defects from studies included in our review. Figure 2 displays results for all studies reporting on major congenital malformations as an outcome in a forest plot (the results are not pooled because of potential overlap of participants across studies for some interventions). Figure 2 does not include studies reporting on subsets of major congenital malformations, such as cardiac malformations: these results are described after the synthesis on all major congenital malformations.

FIGURE 2

Exposure to psychotropic medications and major congenital anomalies. CI, confidence interval; OR, odds ratio; SNRI, serotonin‐norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant

Regarding major congenital anomalies overall, eight studies evaluated associations between various exposures (diazepam, temazepam, zopiclone, SSRIs as a class, citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, SNRIs as a class, TCAs as a class, quetiapine, risperidone, second‐generation antipsychotics, and first‐generation antipsychotics) and major congenital anomalies (51, 52, 53, 54, 55, 56, 57, 81, 84). Except for one study each on citalopram (53) and second‐generation antipsychotics (82), ORs span from 0.69 to 1.36, with wide CIs spanning the null. For both citalopram and second‐generation antipsychotics, the evidence base comprised two studies with conflicting results. Inconsistency was one factor in downgrading the evidence base overall; another was the potential study limitations of the evidence base. No eligible evidence was available on congenital anomalies for triazolam, alprazolam, valproate, carbamazepine, clonazepam, and topiramate, although evidence is available from studies of other populations ineligible for this review.

Regarding cardiac anomalies specifically, eight studies evaluated associations (53, 54, 55, 58, 59, 92, 93, 94) and used different criteria to identify affected infants. Odds risks from meta‐analyses for SSRIs, citalopram, fluoxetine, paroxetine, sertraline, and TCAs ranged from 0.86 to 1.26, with wide CIs spanning the null. Studies used varying definitions that did not consistently exclude cardiac anomalies associated with prematurity. Beyond differences in outcome measurement, differences between studies in design and controls for confounding could potentially explain variations in results.

Comparative Harms of Perinatal Pharmacotherapy

One RCT and 55 observational studies were identified, limiting causal inference regarding exposures and resultant harms. Evidence from one study suggested that the association between first trimester exposure to lithium and overall congenital anomalies (adjusted RR, 1.85; 95% CI, 1.23 to 2.78) and cardiac anomalies (adjusted RR, 2.25; 95% CI, 1.17 to 4.34) may be greater than the association between first trimester exposure to lamotrigine and the same outcomes (low SOE). This evidence could support a decision to transition from lithium to lamotrigine when clinically appropriate. The evidence is insufficient for all other comparisons and outcomes.

DISCUSSION

Despite the high reported prevalence for any mental health disorder among pregnant women (25.3% in the United States in 2001 to 2002 (95)), the evidence is very sparse on the benefits of pharmacotherapy, mostly reflecting how little controlled, high‐quality research has been undertaken in this population. Available evidence of benefit is limited in design (e.g., 2 cohort studies for bipolar disorder (32, 33)), number and size of studies (e.g., 2 RCTs for sertraline response and remission, N = 145 (24, 25)), and duration for outcome reporting (e.g., 30 days for brexanolone (30, 31)). Regarding benefits of treatment, the absence of evidence on the efficacy of many psychotropic drugs in pregnancy or in the postpartum period should not be interpreted as an absence of their benefit. Substantial evidence exists on the efficacy of psychotropic medications across a broad spectrum of persons with mental health disorders. Systematic reviews in the general population have found evidence of benefit for several pharmacological agents for anxiety, depression, bipolar disorder, and schizophrenia (96, 97, 98, 99, 100), may be interpreted as applicable to pregnant and postpartum women. However, doses of medication may need to be adjusted to maintain comparable blood levels of medication given the physiological changes in blood volume and other pharmacokinetic considerations during pregnancy (101, 102); specialist training may help attune clinical providers to these issues.

Regarding treatment harms, the review was restricted to studies comparing women receiving psychotropic drugs with women with a mental health disorder who were not receiving psychotropic drugs. Despite this requirement, no observational study could completely control for psychiatric illness severity that could have predicted both exposure and outcome. Fewer than one quarter of studies across the review considered dose of exposure, further complicating the extent to which inferences can be drawn between the stated exposure and the outcome. Furthermore, some harms are self‐limiting (as in the case of Apgar scores, which assess immediate need for resuscitation and do not predict individual neonatal mortality or neurologic outcome (103)) or very rare (as in the case of persistent pulmonary hypertension). The consideration of the balance of benefits and harms should include a discussion of the absolute risks of these harms. Additionally, the risks of proceeding without treatment should be compared with the specific pharmacologic treatment being considered. Given the well‐documented maternal and fetal risks of untreated maternal mental health disorders, shared and collaborative decision‐making tailored to a patient's specific mental health diagnosis is essential for clinical decision making.

Finally, the findings on comparative benefits and harms were very sparse. The paucity of definitive evidence on this topic offers many new opportunities for research. Clinical trials offer the greatest rigor, but feasibility and ethics have constrained their use. Routine exclusion of pregnant and lactating women from clinical trials forces patients and providers to make clinical decisions in the absence of evidence. In 2018, the Task Force on Research Specific to Pregnant Women and Lactating Women issued recommendations to include pregnant and lactating women in scientific studies and remove regulatory barriers to participation in research (104). Pragmatic trial designs and collaborative care models that allow ongoing data collection may permit greater rigor while addressing confounding.

A large proportion of studies included in our review are observational and often draw from registries and prescription databases. The limitations posed by these data sources, particularly with regard to data collection on symptoms, can skew the overall evidence base to focus on harms rather than benefits. New efforts, such as the Outcome Measures Framework, supported by the Agency for Healthcare Research and Quality, can help categorize outcomes and harmonize data collection. A recent publication suggested a minimum set of outcome measures for depression in patient registries and clinical practice and provides guidance for implementation and data collection (105). Claims data will also continue to be important in identifying harms but will need to offer better evidence of severity of disorders, dosing, and duration of exposure to adequately control for confounding. Linked databases of maternal and child outcomes can also help to control for selection bias and confounding.

Limitations

Few pharmacotherapy RCTs for mental health disorders during pregnancy or lactation were identified; observational studies comprised the bulk of this review. A significant constraint to interpreting the evidence is the widespread risk of confounding. Common data sources such as registry studies did not have data on severity of psychiatric illness and as a result were unable to control for confounding adequately. In some instances, controls for confounding reduced the effect size and reversed the direction of effect.

The restriction of the evidence to women with mental health disorders served as a means of reducing the potential for confounding in the evidence base. However, this criterion excluded studies of well‐conducted negative controls that might bolster the evidence on the association between exposure and outcome. Also, this criterion excluded studies reporting on relevant outcomes for pharmacotherapy exposures for other clinical conditions (e.g., epilepsy). Studies of multiple drug exposures presented results for each exposure but did not always present results separately for women with multiple drug exposures. In studies with overlapping arms, the association of the specific drug and the outcome could not be ascertained. As a result, these studies were excluded. The exclusion of studies with overlapping arms also restricted the comprehensiveness of the review. These limitations of the evidence and review criteria mean that the signals of harms identified may be partially or wholly attributable to residual confounding. Eligible studies were further limited by restriction to English language studies.

CONCLUSION

Evidence from few studies supports the use of pharmacotherapy for perinatal mental health disorders. Although many studies report on increased adverse events, they could not rule out underlying disease severity as the cause of the association between exposures and adverse events. Patients and clinicians need to make informed, collaborative decisions on treatment choices.

Author and Article Information

This work was supported by funding from the Agency for Healthcare Research and Quality; Contract No. HHSA‐290‐2015‐00011‐I, Task Order No. 13. The authors acknowledge the following individuals for their contributions to this project: current or prior AHRQ staff, Christine Chang, MD, Jill Huppert, MD, MPH, Aysegul Gozu, MD, MPH; EPC Associate Editor, Ian Saldanha, MBBS, MPH, PhD; current or prior RTI International–University of North Carolina at Chapel Hill EPC staff: Rania Ali, MPH, Sharon Barrell, MA, Rachel Clark, BA, Kayla Giger, BS, Valerie Forman Hoffman, PhD, MPH, Linda Lux, MPA, Kathleen Lohr, PhD, Loraine Monroe, Kara Suvada, BS, Rachel Palmieri Weber, PhD, and Carol Woodell, BSPH; and representatives from the APA (Jennifer Medicus, MBA, Laura Fochtmann, MD, MBI) and ACOG (Jessica Butler, MPH).

All authors report no financial relationships with commercial interests.

Investigators worked with AHRQ staff to develop the scope, analytic framework, and key questions for this review. AHRQ had no role in study selection, quality assessment, or synthesis. AHRQ staff provided project oversight, reviewed the report to ensure that the analysis met methodological standards, and distributed the draft for peer review. Otherwise, AHRQ had no role in the conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript findings. The opinions expressed in this document are those of the authors and do not reflect the official position of AHRQ or the U.S. Department of Health and Human Services. Additional Information: A draft version of the full evidence report underwent external peer review from seven content experts, review from partners (APA, ACOG), and public posting (June 15, 2020, to July 17, 2020). Comments from reviewers were considered in preparing the final evidence review.

Supplementary Material

Click here for additional data file.