Risk of miscarriage after benzodiazepine use during pregnancy: updated systematic review and meta-analysis

Abstract

Background

Benzodiazepine use during pregnancy remains common despite guidelines discouraging it, except for short-term treatment of severe anxiety or agitation. Previous studies have suggested an increased risk of miscarriage after in utero exposure, raising concern at the European level. However, the most recent meta-analysis, published in 2020, included neither several large-scale observational studies nor an assessment by specific benzodiazepine agents. This study aimed to provide an updated synthesis of evidence on the association between early pregnancy exposure to benzodiazepines and miscarriages (excluding studies exclusively involving women with epilepsy), including agent-specific analyses.

Methods

A systematic review and meta-analysis were conducted according to Cochrane recommendations. Eligible studies evaluated the association between benzodiazepine use during pregnancy and miscarriage (pregnancy loss before 22 weeks of gestation); studies limited to women with epilepsy were excluded. Risk of bias was assessed using ROBINS-I. Pooled odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using a random-effects model. Publication bias, heterogeneity, and sensitivity analyses by study design, control group, psychiatric indication and risk of bias were performed. Analyses by specific benzodiazepine agents, dose-response relationships, and E-values for unmeasured confounding were also conducted.

Results

Of 1,142 records screened, ten studies including over 8,000 exposed pregnancies were retained. Benzodiazepine exposure during early pregnancy was associated with a significantly increased risk of miscarriage (pooled OR: 1.68; 95% CI: 1.48–1.90; I² = 60%). After adjusting for publication bias, the association remained (adjusted OR: 1.58; 95% CI: 1.39–1.80). The E-value (2.74) suggested moderate robustness to unmeasured confounding. Sensitivity analyses confirmed the main findings. Pooled ORs for the seven most frequently used agents (e.g., lorazepam, clonazepam, alprazolam) ranged from 1.42 to 1.82, supporting a class effect. All three studies investigating the dose–response relationship found a dose–response trend.

Conclusion

This updated meta-analysis indicates that early pregnancy exposure to benzodiazepines is associated with an increased risk of miscarriage. The consistency across analyses and evidence of dose-response strengthen confidence in this association. Clinicians should carefully weigh risks and benefits, consider non-pharmacological alternatives, and ensure close monitoring when prescribing benzodiazepines to women of childbearing potential.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12884-025-08574-0.

Introduction

Maternal use of benzodiazepines during pregnancy is frequent, with a worldwide prevalence of 1.9% (95% CI: 1.6%–2.2%) [1], and has increased over the past several decades [2]. However, the National Institute of Clinical Excellence (NICE) in the United Kingdom and the Haute Autorité de Santé (HAS) in France do not recommend benzodiazepines in pregnancy because of risks for the baby, except for the short time in case of severe anxiety and extreme agitation. Gradually stopping benzodiazepines is also recommended for women who are planning a pregnancy or are already pregnant [3, 4].

Benzodiazepine use during pregnancy has been associated with many adverse birth outcomes, such as preterm birth, low birth weight, and neonatal intensive care unit admission [5]. The most recent meta-analysis to date reported an 86% increased risk of miscarriages associated with prenatal benzodiazepine exposure (pooled odds ratio (OR): 1.86; 95% confidence interval (CI): 1.43–2.42) [5]. This finding is supported by the potential role of benzodiazepines in cell proliferation and differentiation processes that could disrupt fetal development, potentially resulting in miscarriage. However, this meta-analysis included studies published up to June 2018 and several large-scale and/or well-designed studies have been published since. In particular, in 2024, a study on the use of benzodiazepines during pregnancy published by Meng et al. raised concerns about the risk of miscarriage [6].

In this context, in 2024, as part of the upcoming benzodiazepines periodic safety update report single assessment (PSUSA), the Pharmacovigilance Risk Assessment Committee (PRAC) has requested real-world evidence on commonly used benzodiazepines during pregnancy, and additionally, on the background rates of pregnancy losses (miscarriages and stillbirths) to guide the assessment of treatment safety during pregnancy.

Given the increasing use of benzodiazepines during pregnancy, the ongoing concerns at the European regulatory level, and the previously reported association with miscarriage, an updated systematic review and meta-analysis is critically needed. While earlier meta-analyses have highlighted a potential risk of miscarriage following prenatal benzodiazepine exposure, they have not incorporated the most recent high-quality studies published since 2018, nor have they addressed the following key aspects.

Indeed, to our knowledge, no previous meta-analysis has evaluated the association between specific benzodiazepine agents and the risk of miscarriage. Yet, benzodiazepines differ in their pharmacokinetic properties (such as half-life), potency, and indications, which may result in varying levels of risk. Understanding these molecule-specific effects is essential for informing safer prescribing practices during pregnancy.

In addition, dose–response relationships have not been systematically examined in previous literature syntheses. Assessing whether higher doses are associated with greater risks could inform the need for dose adjustments. Moreover, identifying a dose–response relationship would strengthen the level of evidence between benzodiazepine exposure and the risk of miscarriage.

This updated meta-analysis was conducted to address existing evidence gaps by assessing the risk of miscarriage associated with individual benzodiazepine agents and examining dose–response relationships across included studies examining this association.

Moreover, as some benzodiazepines are indicated for the treatment of epilepsy as well as psychiatric and sleep disorders, dosage, treatment duration and co-exposure may vary depending on the underlying condition. Therefore, the present study focused on studies that does not include solely women with epilepsy, in order to ensure greater homogeneity in terms of potential confounding factors, particularly comorbidities and to minimize the risk of co-exposure to antiepileptic drugs, some of which have been associated with an increased risk of miscarriage [7].

The main objective of this study was to assess the risk of miscarriages associated with early pregnancy exposure to benzodiazepines, through a systematic review and meta-analytic approach, excluding studies that exclusively involve women with epilepsy. Moreover, the risk of miscarriages by specific benzodiazepine agents and dose-response relationships were also explored.

Materials and methods

A systematic review and meta-analysis were conducted to investigate the association between prenatal exposure to benzodiazepines and miscarriages.

The procedures for this review and meta-analysis adhered to established best practices and Cochrane standards [8] and are reported in accordance with the Meta-analyses Of Observational Studies in Epidemiology (MOOSE) guidelines [9] and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [10] (see additional Table 1). The study protocol was registered in PROSPERO (CRD4 2023468915). Data management and analyses were conducted using metaPreg (http://metapreg.org), a semi-automatized proprietary collaborative WEB-based meta-analysis platform. The master protocol was finalized prior to the literature search and data analysis and was previously published [11]. This publication focuses on miscarriages, but in accordance with the master protocol, all outcomes are available on metaPreg website (http://metapreg.org).

Search strategy and identification of studies

Relevant studies were identified through a search of 2 electronic databases (Pubmed/Medline and EMBASE), with good coverage (about 88.0%, 95% CI [86.2, 89.9]) [12, 13], from database inception date to February 2025. To maintain a comprehensive evidence base and reduce the potential for publication bias, no temporal cutoff, language or regional restrictions were applied. The detailed search strategy is presented in Additional Table 2. This search strategy was modified for the EMBASE database to account for differences in syntax and thesaurus headings. A snowballing approach was also used to identify potentially relevant studies using the reference lists of review articles (systematic review or meta-analysis).

Inclusion criteria

Eligible studies were all studies with a comparator group reporting specific data on miscarriages after in utero exposure to benzodiazepines. Studies exclusively including women with epilepsy were excluded in order to maintain relative homogeneity with respect to potential confounding factors, particularly comorbidities, as well as dosage and/or treatment duration, which may differ for this indication.

Types of studies

Prospective cohort studies, historical cohort studies, case–control studies and randomized clinical trials were included, regardless of publication status. Studies with inappropriate design (case reports, case series, disproportionate analysis, …), studies without original data (review, comments…) or animal studies were excluded. Observational studies not presenting quantitative results (e.g., OR, hazard ratio, relative risks, 95% confidence intervals, numbers of cases/population, observed and expected cases) or sufficient data to calculate treatment effect size were also excluded. In case of iterative studies using the same database, only the most recent publication was included. In case of overlapping data from studies using the same dataset, only the one with the largest sample size or with a methodology that provided a better consideration of the confounding factors was retained.

Type of outcome

The outcome of interest was miscarriages, most commonly referred to in the literature as “miscarriages” or “spontaneous abortions.” The definition used was the one provided by the included studies, as long as the event occurred before 22 completed gestational weeks (GW).

Type of exposure

The exposure of interest was exposure to benzodiazepines as a whole or at least one benzodiazepine (for main objective) and to each specific benzodiazepine agent (for the secondary objective). Studies that did not distinguish benzodiazepines and z-hypnotics were excluded. No restriction was applied regarding the timing of exposure during pregnancy, provided that it occurred prior to the miscarriage. Exposure was not limited to monotherapy, although a sensitivity analysis was conducted including only studies that restricted to monotherapy benzodiazepine exposure (see below).

Study selection

Study selection followed a two-stage process with a semi-automated platform described elsewhere [11]. First, the abstracts of all studies identified through the literature search were screened by one reviewer (CP). Subsequently, the full-text articles of potentially eligible studies were assessed by the same reviewer to determine whether they met the predefined inclusion criteria. In cases of uncertainty regarding study eligibility, the issue was resolved through discussion with the project scientific directors (JC and MC) until a consensus was reached.

Data extraction

For all included studies, data were extracted using a standardized electronic data collection form. Data extraction was performed by one reviewer (CP) with semi-automated tools. Following completion of the meta-analysis, a quality control check was performed by an independent second reviewer (YP), who verified all extracted data, including completeness and consistency. Any discrepancies were discussed and resolved by consensus among the data curators or during project team meetings. The feasibility and acceptability of this semi-automated and single-biocurator approach were evaluated. Overall, findings indicated that this process significantly reduced the time required to conduct the meta-analysis, without compromising expert confidence in its methodological and scientific rigor [14, 15].

The following information was recorded:

• Study characteristics: first author, year of publication, primary outcome, country of study, data source, study period, population description, definitions of exposure and non-exposure, type of control group, case and control descriptions, and sample size.

• Methodological details: study design, exposure assessment, outcome measurement, follow-up duration, and confounding factors considered in the analyses.

• Results: number of events, total number of participants in each study group, and reported effect sizes. When available, adjusted effect sizes were used; otherwise, crude effect sizes were extracted. If no effect measure was reported, effect sizes were calculated from the raw data provided in the published articles.

For the primary analysis, when a study reported results for multiple control groups, only one was selected in the following hierarchical order: (a) unexposed with the same underlying condition (“unexposed sick”), (b) unexposed population (not otherwise specified or general population), (c) unexposed disease-free, (d) those exposed to other treatments for the same condition and (e) sibling. Sensitivity analyses were conducted based on the type of comparison group to explore potential sources of heterogeneity.

To support the interpretation of the results, studies assessing the dose–response relationship were identified. From studies reporting effect sizes by dose categories, the following data were extracted: dose levels (preferably standardized as defined daily doses (DDD) or diazepam-equivalent doses, depending on availability), and adjusted odds ratios (ORs) with 95% confidence intervals (CIs) for each exposure level; and the author’s interpretation of the existence of dose–response relationship.

Risk of bias assessment of eligible studies

The risk of bias was evaluated using the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool [16], adapted to observational studies in pregnancy. This tool assesses six domains of potential bias: confounding, selection of participants, classification of exposures, measurement of outcomes, missing data, and selection of the reported results. The domain related to deviations from intended interventions was not assessed, as it is specific to randomized controlled trials and was deemed not applicable to observational studies investigating medication-related risks during pregnancy. For the domain of confounding, four levels of risk were considered: low, moderate, serious, and critical. For the remaining domains, the following levels were used: low, moderate, and critical. For each bias type, an unclear category was added, when the data reported was insufficient to allow assignment to the aforementioned categories. Risk of bias assessments were used to interpret the results and did not serve as a criterion for study exclusion. However, sensitivity analyses were conducted based on the level of risk of bias due to confounding and excluding studies with substantial risk of bias (i.e., with more than three domains rated as critical or unclear).

Statistical analyses

A random effects analysis model using inverse variance method was used to estimate the pooled ORs and 95% confidence intervals (CIs). Heterogeneity between the studies was assessed using the I2 statistic [17]. Publication bias was visually assessed with funnel plots. Their asymmetry was tested with the Egger’s test [18, 19] when at least 10 studies were included. Trim and fill method was used to determine the number of missing studies and to adjust for publication bias [20].

The E-value was presented to assess the robustness of the combined association and estimate the minimum strength of association that an unmeasured confounder would need to have with both the treatment and the outcome to fully explain the calculated effect size [21].

Sensitivity analyses were conducted to evaluate the potential impact of study design, type of comparator groups, co-exposure or not with antidepressants and/or psychotropic medications, restriction of psychiatric diseases as the indication and degree of bias in the risk bias assessment.

Lastly, data on each specific benzodiazepine agent were pooled to investigate their respective associations with miscarriage, with a focus on the most studied benzodiazepines (i.e. with at least 100 exposed pregnancies).

Results

Search results and study selection

The electronic searches and the screening of previously published systematic reviews retrieved 1,142 records. Among these, 195 were included for full text review (Fig. 1), and 10 studies were included in the meta-analysis [6, 22–30].

Fig. 1.

PRISMA flow diagram

Characteristics of included studies

The main characteristics and findings of the included studies are presented in Table 1. Among these ten studies, three were case-control studies (two nested case-control) and seven were cohort studies (three population-based retrospective cohorts, two prospective cohorts, one retrospective cohort, and one cohort with unspecified design).

Table 1.

Overview of the main characteristics of the included studies

Study	Country Study period Study design	Data source	Indications of BZDs	Exposition period and definition	Non-exposure definition	Miscarriage definition	Estimate [95%CI] x1/n1 x0/n0	Confounding bias - Potential confounders taken into account
Ishikawa - BZDs 2024 [22]	Japan 2005–2022 Nested case control	The administrative claims database from JMDC Inc. (Tokyo, Japan), which contains all inpatient (including those during hospitalization), outpatient, and pharmacy claims received from the insurers.	Not specified (data only available for antipsychotics)	Mothers who were prescribed benzodiazepines from the onset of pregnancy until the day before the index date.	Unexposed (general population or NOS) Mothers who were not prescribed benzodiazepines from the onset of pregnancy until the day before the index date.	Miscarriage (4–22 weeks of gestation)	OR = 1.43 [1.30; 1.57] 848/2,639 43,270/173,796	Serious ROB. Exclusion of women with a history of recurrent pregnancy loss, with antiphospholipid syndrome, or prescribing medications that potentially posed a risk of miscarriage. Adjusted for schizophrenia, other indications of antipsychotics (manic episodes, bipolar affective disorder, depressive disorder, or anxiety disorder), uterine diseases (endometriosis, congenital uterus and cervix malformations, …), other maternal comorbidities (polycystic ovarian syndrome, diabetes, obesity, or thyroid disorders), alcohol or tobacco dependence, number of medications/diagnoses prescribed 6 months before pregnancy onset.
Meng - BZDs 2024 [6]	Taiwan 2004–2018 Case-Time-Control (CTC) design	The Taiwan’s National Birth Certificate Application database and the National Health Insurance database.	Not specified (psychiatric medical conditions in all cases and controls were: - Anxiety (0.7%) - Insomnia (0.9%) - Depression (0.6%) - Schizophrenia (0.06%) - Epilepsy (0.1%) - Bipolar disorder (0.08%)	Women receiving at least 1 prescription of benzodiazepine during the risk period only (1–28 days before miscarriage).	Unexposed, sick Women receiving at least 1 prescription of benzodiazepine during the reference period only (181–208 days before the last menstrual period).	Miscarriage (between 8th-19th weeks of gestation; ICD-9 codes 631, 632, 634, 637, 640; ICD-10 codes O02.1 and O03)	OR = 1.69 [1.52; 1.87] Not specified	Low ROB. Illness. Controls matched for the birth year and a disease risk score (based on age, psychiatric medical conditions, lifestyle factors (obesity, tobacco, alcohol, drug misuse), chronic comorbidities (diabetes, hypertension, hyperlipidemia, …), medication use, and health care use). Adjusted for the comedication use (antidepressants, opioid analgesics, anticonvulsant, Z-hypnotics, and other anxiolytics). Negative control analysis.
Zanisi - BZDs 2022 [23]	Italy 2018–2020 Cohort	The Bergamo Teratology Information Service, Bergamo, Italy.	Not specified.	Pregnant women exposed to benzodiazepines alone.	Unexposed (general population or NOS) Pregnant women exposed to acetaminophen/amoxicillin.	Spontaneous abortion (not otherwise specified)	OR = 5.78 [1.17; 28.42] -/99 -/158	Critical ROB. No adjustment.
Kitchin - Long acting BZDs 2022 [28]	Spain 2002–2015 Case Control	The Spanish database BIFAP (Database for Pharmacoepidemiological Research in Primary Care).	Not specified.	At least one prescription of long acting benzodiazepine during prepregnancy (within the 90 days prior to LMP date) and the first trimester of pregnancy.	Unexposed (general population or NOS) No prescription of long acting benzodiazepine during prepregnancy (within the 90 days prior to LMP date) and the first trimester of pregnancy.	Miscarriage (ICD-9634–639 Pregnancy losses between 4–22 weeks)	OR = 1.30 [1.04; 1.61] 122/18,070 252/54,209	Serious ROB. Controls individually matched to cases (maternal age, gestational age, and year of Last Menstrual Period date). Adjusted by number of General practitioner visits, obesity, smoking, HTA, diabetes.
Sheehy - BZDs, 2019 [24]	Canada 1998–2015 Nested case control	The Quebec Pregnancy Cohort, Montreal, Quebec, Canada.	Not specified (Proxy: ≥1 Diagnosis within 12 months before LMP until index date in all pregnancies: - Mood and anxiety (8.2%) - Insomnia (0.04%)	Pregnancies who had filled at least 1 prescription for any type of benzodiazepine from the last menstrual period until the index date.	Unexposed, sick Pregnancies who had not filled a prescription for any type of benzodiazepine from the last menstrual period until the index date.	Spontaneous abortion (6th to 19th completed week of gestation) - ICD-9 diagnosis code 634 or ICD-10 diagnosis code O03	OR = 2.85 [1.72; 4.72] 198/570 3,221/15,382	Low ROB. Illness. Adjusted for (1) maternal sociodemographic variables (2) maternal chronic conditions (hypertension, diabetes, depression/anxiety, alcohol …), (3) health care resources utilization, (4) pregnancy-associated variables, (5) concomitant exposure to antidepressants and/or antipsychotics. Matched by gestational age and calendar year. Exclusion of women with epilepsy and exposed to known teratogens. Large E-value.
Bech - Clonazepam 2014 [25]	Denmark 1997–2008 Population based cohort retrospective	The Danish medical birth register, the Danish national hospital discharge register and the Danish Register of Medicinal Product Statistics.	Not specified but use of data for women never having a diagnosis of epilepsy	Pregnancies that never have a diagnosis of epilepsy and with a prescription of clonazepam redeemed during pregnancy (1st and/or 2nd trimester).	Unexposed (general population or NOS) Pregnancies that did not redeem any antiepileptic drug prescription in the exposure window.	Spontaneous abortions (ICD codes O02.0-O03.9)	RR = 1.91 [1.54; 2.37] 60/219 108,273/812,862	Critical ROB. Adjusted for maternal age, cohabitation, income, education, history of severe mental disorder, and history of drug misuse.
Ban - BZDs 2012 [26]	The United Kingdom (UK). 1990–2009 Population based cohort retrospective	The Health Improvement Network (THIN), a nationally representative database of computerised primary care medical records collected at 446 general practices (primary health care units).	Not specified	Pregnant women with prescriptions for any benzodiazepines (alone - i.e. no other psychotropic medication of interest) during the first trimester.	Unexposed, sick Pregnant women with un-medicated depression or anxiety, i.e. with current depression or anxiety but no prescriptions during the first trimester.	Miscarriage (NOS)	RRR = 1.60 [1.30;1.90] 520/3,392 -/-	Serious ROB. Illness. Adjusted for maternal age at the end of pregnancy, household socioeconomic status, maternal smoking status before delivery and body mass index before pregnancy.
Ornoy - BZDs 1998 [27]	Israel 1988–1996 Prospective cohort	The Israeli Teratogen Information Service.	Not specified	Pregnant women exposed to benzodiazepines at least during the 1st trimester.	Unexposed, disease free Healthy pregnant women with nonteratogenic exposures.	Spontaneous abortions (NOS)	OR = 2.00 [1.10;3.40] 40/460 22/424	Critical ROB. No adjustment.
Milkovich - Chlordiazepoxide 1974 [30]	USA 1959–1966 Retrospective cohort	The Child Health and Development Studies based on pregnant women included in the East San Francisco Bay Area facilities of the Kaiser-Permanent Medical Care Program.	The indication for meprobamate and chlordiazepoxide noted in the medical records was chiefly anxiety, tension or mild depression (no other details).	Pregnant women who received prescriptions of Chlordiazepoxide, for anxiety, during pregnancy.	Unexposed, sick Pregnant women with no prescription of Chlordiazepoxide, for anxiety, during the first 42 days of pregnancy (and very few were given between the 43rd and the 84th days).	Fetal deaths before 20 weeks of gestation	OR not provided 20/186 31/532	Critical ROB. No adjustment.

BZDs Benzodiazepines, CI Confidence interval, ICD International Classification of Diseases, LMP Last Menstrual Period, NOS Not Otherwise Specified, OR Odd ratio, ROB Risk Of Bias, RRR Relative Risk Ratio. Bold statistically significant association

Data sources included national population registers for three studies [6, 25, 26], claims databases linked to hospital records for four studies [22, 24, 28, 30], and teratology information services for the remaining three [23, 27, 29].

Miscarriages were clearly defined in six studies using International Classification of Diseases (ICD) codes and/or gestational age thresholds [6, 22, 24, 25, 28, 30], whereas the others referred to only spontaneous abortions or miscarriages without further specification.

Comparator groups consisted of: (i) pregnancies unexposed to the treatment but with the same underlying condition (“unexposed sick”) in four studies [6, 24, 26, 30]; (ii) unexposed pregnancies from the general or unspecified population in five studies [22, 23, 25, 28, 29]; and (iii) unexposed, disease-free pregnancies in one study [27].

Regarding risk of bias due to confounding, five of the ten studies were rated as having a critical risk [23, 25, 27, 29, 30], two as serious [26, 28], one as moderate [22] and the remaining two as low [6, 24].

Quantitative synthesis

Meta-analysis of benzodiazepines as whole (without distinction between specific benzodiazepine agents)

In total, the ten included studies encompassed data from over 8,035 pregnancies exposed to benzodiazepines. All individual studies reported a statistically significant increased risk of miscarriage following in utero exposure to benzodiazepines.

In the primary meta-analysis, benzodiazepine exposure during pregnancy was associated with a significantly 68% increased risk of miscarriage (pooled OR 1.68; 95% CI: 1.48–1.90; p < 0.0001; I² = 60%; Fig. 2).

Fig. 2.

Forest plot of the meta-analysis on the association between prenatal exposure to benzodiazepines and miscarriages. BZDs: Benzodiazepines; TE: Treatment effect; CI: Confidence Interval; I2, Higgins I2 statistic of heterogeneity; ROB: Risk of Bias

Visual inspection of the funnel plot and results of Egger’s test suggested the presence of publication bias, as smaller sample sized studies tended not to report both positive and negative associations equally (Fig. 3a). To further address potential publication bias, the Trim and Fill method was applied. The adjusted analysis yielded a similar result (4 added studies; adjusted pooled OR = 1.58, 95% CI 1.39–1.80; p < 0.0001; Fig. 3b).

Fig. 3.

Bias of publication for miscarriage. (a) Funnel plot (asymmetry test p-value [by Egger’s regression] = 0.0531). (b) Funnel plot with Trim and fill (pooled OR with Trim and fill: 1.58, 95% CI 1.39–1.80; p < 0.0001)

The E-value was 2.74 (95% CI: 2.33; …), indicating that an unmeasured confounder would need to be associated with both exposure and outcome by a risk ratio of nearly 3 to fully account for the observed effect size.

All sensitivity analyses, stratified by study design, type of control group, co-exposure or not to antidepressants and/or other psychotropic medications, restriction to psychiatric populations, level of confounding bias and excluding studies with substantial risk of bias were consistent with the main analysis (Table 2; Figs. 4 and 5).

Table 2.

Sensitivity analyses

		Pooled OR 95% CI	Hetero- geneity (I2)	Number of studies
Type of studies	Cohort studies	1.73 [1.59; 1.87]	0%	7
	Case-control studies	1.56 [1.20; 2.02]	75%	3
Type of controls	Unexposed (general population, disease free or unspecified)	1.63 [1.34; 1.98]	61%	6
	Unexposed, sick	1.74 [1.50; 2.02]	35%	4
Co-exposure with antidepressants and/or psychotropic medications	Yes	2.00 [1.14; 3.52]	/	1
	No	2.36 [0.74; 7.49]	59%	2
	Not specified	1.67 [1.44; 1.93]	68%	7
Level of risk of confounding bias	Low	2.07 [1.26; 3.40]	75%	2
	Moderate	1.43 [1.30; 1.57]	/	1
	Serious	1.45 [1.19; 1.78]	49%	2
	Critical	1.98 [1.65; 2.38]	0%	5
Excluding studies with substantial risk of bias (i.e. with more than three domains rated as critical or unclear)		1.65 [1.46; 1.87]	64%	8
Indication restricted to psychiatric disorders (depression, anxiety, …)		2.43 [1.65; 3.56]	0%	2

I2 Higgins I2 statistic of heterogeneity, CI Confidence Interval. Bold statistically significant association

Fig. 4.

Sensitivity analyses according to the type of studies, the type of control groups, the co-exposure with antidepressants and/or psychotropic medications (yes/no/not specified), the restriction to psychiatric diseases. ATD: Antidepressants; BZDs: Benzodiazepines

Fig. 5.

Sensitivity analysis - Forest plot of meta-analysis on the association between prenatal exposure to benzodiazepines and miscarriages restricting to studies with low risk of confounding bias. BZDs: Benzodiazepines; TE: Treatment effect; CI: Confidence Interval; I2, Higgins I2 statistic of heterogeneity; ROB: Risk of Bias

Meta-analysis of specific benzodiazepine agents

Among the 40 specific benzodiazepine agents identified, fourteen different benzodiazepines had at least one study with data on miscarriage risk. In total, six studies contributed to the specific benzodiazepine agent analyses. The number of included studies in each meta-analysis by specific benzodiazepine agents varied between one to four. The results of each specific benzodiazepine agent were presented in Table 3.

Table 3.

Meta-analysis results for benzodiazepines, categorized by type (short- or long-acting) and ranked by decreasing number of exposed pregnancies

Substance	OR 95%CI	Hetero-geneity (I2)	Number of studies	Included studies	Number of exposed pregnancies	E-value
Long acting (half-life > 24 h)
Clonazepam	1.82 [1.54, 2.16]	0%	3	Sheehy 2019 [24] Kitchin 2022 [28] Bech 2014 [25]	468	3.05 [2.44; .]
Chlordiazepoxide	1.42 [0.38, 5.36]	29%	2	Sheehy 2019 [24] Milkovich 1974 [30]	193	/
Diazepam	1.70 [1.21, 2.39]	45%	3	Meng 2024 [6], Kitchin 2022 [28] Sheehy 2019 [24]	> 132	2.79 [1.71; .]
Flurazepam	2.25 [0.47, 10.84]	58%	2	Kitchin 2022 [28] Sheehy 2019 [24]	28	/
Nitrazepam	1.65 [0.07, 40.48]	NA	1	Sheehy 2019 [24]	1	/
Cloxazolam	1.54 [1.27, 1.86]	NA	1	Meng 2024 [6]	Not specified	2.45 [1.86; .]
Short acting (half-life ≤ 24 h)
Lorazepam	1.48 [1.23, 1.79]	54%	3	Meng 2024 [6] Kitchin 2022 [28], Sheehy 2019 [24]	> 896	2.33 [1.75; .]
Alprazolam	1.48 [1.12, 2.38]	65%	4	Meng 2024 [6] Lee 2022 [29] Kitchin 2022 [28] Sheehy 2019 [24]	> 455	2.33 [1.48; .]
Bromazepam	1.55 [0.85, 2.84]	58%	2	Kitchin 2022 [28] Sheehy 2019 [24]	197	/
Oxazepam	1.48 [1.02, 2.14]	NA	1	Sheehy 2019 [24]	160	2.32 [1.17; .]
Lormetazepam	1.33 [0.84, 2.10]	NA	1	Kitchin 2022 [28]	84	/
Temazepam	2.74 [1.29, 5.83]	NA	1	Sheehy 2019 [24]	29	4.92 [1.90; .]
Triazolam	1.14 [0.11, 11.82]	NA	1	Sheehy 2019 [24]	5	/
Fludiazepam	2.52 [1.89, 3.36]	NA	1	Meng 2024 [6]	Not specified	4.48 [3.19; .]

No available data for: Adinazolam, Bentazepam, Brotizolam, Camazepam, Cinolazepam, Clobazam, Clorazepate, Clotiazepam, Doxefazepam, Estazolam, Etizolam, Flunitrazepam, Halazepam, Ketazolam, Loflazepate, Loprazolam, Medazepam, Mexazolam, Midazolam, Nimetazepam, Nordazepam, Pinazepam, Prazepam, Quazepam, Remimazolam, Tofisopam

I2 Higgins I2 statistic of heterogeneity, CI Confidence Interval, NA Not Applicable. Bold statistically significant association

Given that results were more robust for substances supported by multiple studies and larger numbers of exposed pregnancies, only the seven benzodiazepines with at least 100 exposed pregnancies were described below. These include lorazepam (> 896 exposed pregnancies), clonazepam (n = 468), alprazolam (> 455), bromazepam (n = 197), chlordiazepoxide (n = 193), oxazepam (n = 160), and diazepam (> 132).

Among long-acting benzodiazepines (half-life > 24 h), the pooled ORs were in the same order of magnitude, ranging from 1.42 to 1.8. Clonazepam showed a statistically significant increased risk (pooled OR = 1.82; 95% CI: 1.54–2.16) with no heterogeneity (I²=0%), based on 3 studies and 468 exposed pregnancies. The E-value of 3.05 suggested robustness to unmeasured confounding. Chlordiazepoxide had wide confidence intervals (pooled OR = 1.42; 95% CI: 0.38–5.13) with low heterogeneity (I²=29%), based on 2 studies and 193 exposed pregnancies. A significant association (pooled OR = 1.70; 95% CI: 1.21–2.39) was obtained for diazepam, with moderate heterogeneity (I²=45%) based on 3 studies with over 132 exposed pregnancies. The E-value is 2.79, indicating moderate robustness.

Among short-acting benzodiazepines (half-life ≤ 24 h), the pooled ORs were similar, ranging from 1.48 to 1.55. Lorazepam and alprazolam had statistically significant increased risks (pooled OR = 1.48 for both), with moderate to high heterogeneity (54% and 65% respectively), supported by multiple studies and a large sample size (> 896 and > 455 pregnancies respectively). Bromazepam had wide confidence intervals (pooled OR = 1.55; 95% CI: 0.85–2.84) with moderate heterogeneity (I²=58%), based on 2 studies and 197 exposed pregnancies. Oxazepam also showed a significant association (pooled OR = 1.48; 95% CI: 1.02–2.14) but based on a single study and 160 exposed pregnancies.

Dose-response relationship

Among the ten studies included in this meta-analysis, three specifically investigated dose–response relationships between benzodiazepine exposure during early pregnancy and miscarriage risk [6, 24, 25] (Table 4). Due to heterogeneity in exposure definitions and dose categorization, the findings were only synthesized descriptively.

Table 4.

Overview of the dose-response relationship between benzodiazepines in early pregnancy and miscarriages

Study Drug	Dose categories	Adjusted OR (95% CI)	Author’s interpretation
Bech 2014 [25] Clonazepam	Low (≤ 50% DDD)	1.84 (1.41–2.39)	Women might have an increased risk of spontaneous abortion, especially when using high doses of clonazepam.
	High (> 50% DDD, > 4 mg/day)	4.50 (2.93–6.93)
Meng 2024 [6] Benzodiazepines	Low (< 1.0 DDD)	1.61 (1.43–1.82)	A dose-response association was observed between benzodiazepine exposure and miscarriage.
	High (≥ 1.0 DDD)	1.86 (1.53–2.25)
Sheehy 2019 [24] Benzodiazepines (as Diazepam equivalents)	Mean daily dose: ≤5 mg/day	1.73 (1.44–2.08)	The risk increased with increasing daily dose of benzodiazepines, which may suggest a dose-response effect.
	Mean daily dose: 6–20 mg/day	1.96 (1.59–2.43)
	Mean daily dose: >20 mg/day	2.55 (1.08–6.01)

BZD Benzodiazepine, CI Confidence Interval, DDD Defined Daily Dose, OR Odd ratio. Bold: statistically significant association

In the study conducted by Bech et al. [25], the adjusted OR increased from 1.84 (95% CI: 1.41–2.39) for low-dose clonazepam exposure (≤ 50% DDD) to 4.50 (95% CI: 2.93–6.93) for high-dose exposure (> 50% DDD, > 4 mg/day). Similarly, Meng et al. [6] reported higher odds of miscarriage among women exposed to benzodiazepines at ≥ 1.0 DDD (OR 1.86, 95% CI: 1.53–2.25) compared with lower doses (< 1.0 DDD; OR 1.61, 95% CI: 1.43–1.82). In the study by Sheehy et al. [24], the risk increased progressively from an OR of 1.73 (95% CI: 1.44–2.08) for ≤ 5 mg/day to 2.55 (95% CI: 1.08–6.01) for > 20 mg/day (diazepam-equivalent).

Discussion

The meta-analysis, based on ten observational studies and over 8,000 benzodiazepine-exposed pregnancies, supports the presence of an association between benzodiazepine use during early pregnancy and an increased risk of miscarriage, with a pooled OR of 1.68 (95% CI: 1.48–1.90). This result remained robust across multiple sensitivity analyses and after correction for potential publication bias using the Trim and Fill method (adjusted OR: 1.58; 95% CI: 1.39–1.80), thus reinforcing the reliability of the observed association.

Three studies investigated the dose–response relationship between early prenatal benzodiazepine exposure and miscarriage [6, 24, 25]. All three studies reported a consistent trend suggesting a dose–response relationship. Indeed, in Bech et al. [25], the adjusted OR increased from 1.84 (95% CI: 1.41–2.39) for low-dose clonazepam (≤ 50% DDD) to 4.50 (95% CI: 2.93–6.93) for high-dose exposure (>50% DDD, >4 mg/day). Meng et al. [6] reported ORs of 1.61 (95% CI: 1.43–1.82) for < 1.0 DDD and 1.86 (95% CI: 1.53–2.25) for ≥ 1.0 DDD. In Sheehy et al. [24], the OR increased progressively from 1.73 (95% CI: 1.44–2.08) for ≤ 5 mg/day to 2.55 (95% CI: 1.08–6.01) for >20 mg/day (diazepam-equivalent). While a quantitative synthesis of these three dose–response studies could have strengthened this assessment, substantial heterogeneity in study design and dose categorization precluded a formal meta-analysis. This limitation underscores the need for future studies with standardized methodologies to provide more robust quantitative evidence on dose–response relationships.

Across these three studies, a consistent pattern of increasing miscarriage risk was observed with higher benzodiazepine doses, suggesting a dose–response relationship. The observation of similar patterns across independent datasets, conducted in different populations and using varying exposure definitions, strengthens the consistency of these findings. Nevertheless, each category of dose was associated with an increased risk of miscarriage compared to the comparator. Therefore, no threshold dose could be identified below which benzodiazepine exposure did not confer increased risk. Further research using standardized dose definitions and categorizations is warranted to confirm these results and to better define potential dose thresholds.

Meta-analyses conducted by specific benzodiazepine agents revealed that many substances have limited or no available data, highlighting gaps in research and the need for further investigation into substance-specific safety profiles. Of the 40 benzodiazepines identified, only 14 had at least one study reporting data on miscarriages. Both long- and short-acting benzodiazepines were associated with an increased risk of miscarriage when used in early pregnancy, with no clear difference in the magnitude of risk between the most studied agents. This is consistent with a class effect of benzodiazepines. Indeed, among the seven substances with more than 100 exposed pregnancies (lorazepam, clonazepam, alprazolam, bromazepam, chlordiazepoxide, oxazepam, and diazepam), pooled ORs ranged from 1.42 to 1.82. Some of these associations reached statistical significance, depending on the number of included studies, sample size, and the level of heterogeneity across studies. Although direct comparisons between substances could not be conducted, the results suggest no substantial differences in risk magnitude among the most studied benzodiazepines.

These findings strengthened and refined the conclusions of the last published meta-analysis, which reported a significant association between benzodiazepine exposure and miscarriage (pooled OR: 1.86, 95% CI: 1.43–2.42) [5]. Indeed, this update (i) incorporates six additional studies published after 2018, including large population-based cohorts that yielded narrower confidence intervals, increasing precision; (ii) benefits from a more rigorous control for confounding in recent studies (e.g [6, 24]).

Limitations included risk of bias due to confounding, inherent to observational studies. Indeed, five of the ten included studies were rated as having a critical risk of bias due to confounding. Nevertheless, the results were consistent across all studies and sensitivity analyses to explore confounding bias and degree of bias. Indeed, to limit confounding by indication, we restricted the meta-analysis to the four studies including an unexposed group with the same underlying condition and similar results were obtained. Additionally, restricting the analysis to the two studies with low risk of confounding bias produced results fully in line with the overall estimate, with a slightly higher effect size (pooled OR around 2), further supporting the robustness of the association. Results were unchanged after exclusion of the 2 studies with substantial risk of bias (i.e., with more than three domains rated as critical or unclear). The E-value of 2.74 suggests that an unmeasured confounder would need a risk ratio of nearly 2.7 to fully explain the observed effect size, indicating that residual confounding is unlikely to account for the effect entirely. Overall, while residual confounding cannot be completely excluded, its likely impact on the pooled OR is limited, and the association appears robust and consistent.

Another methodological challenge identified across the included studies is the heterogeneity in the definition of miscarriage. Definitions varied in both gestational age cutoffs and the diagnostic codes used to identify events. For instance, some studies defined miscarriage as pregnancy loss between 4 and 22 weeks of gestation [22, 28], while others applied narrower windows, such as 8 to 19 weeks [6], or 6 to 19 weeks [24]. Similarly, the ICD codes used to capture miscarriage events ranged from specific codes (e.g., ICD-10 O03) to broader categories (ICD-10 O02.xx, O03.xx, O20.xx), and some studies did not report the codes used. This variability may introduce heterogeneity and classification bias, as differences in gestational age thresholds can affect which miscarriages are captured, particularly at the earliest stages of gestation, where losses may go unrecognized or undocumented.

Another limitation is the potential under-ascertainment (censoring) of early miscarriages in healthcare databases. Many spontaneous abortions, especially those occurring before 6–8 weeks of gestation, are not clinically recognized or are managed outside formal healthcare systems, and thus not recorded in registries or claims databases. As a result, studies relying on such data may underestimate the true incidence of miscarriage, particularly in the early first trimester. If underreporting is non-differential (occurs equally in exposed and unexposed groups), effect sizes would probably be biased toward the null, meaning true associations may be stronger than observed. If underreporting is differential (e.g., medicated patients have earlier medical contact and more losses are recorded), the bias could go in either direction, making effect sizes unpredictable. Nevertheless, given that the results from all different data sources (e.g., healthcare databases vs. self-reported vs. registry) were consistent, the impact on the conclusion would be limited.

Despite these limitations, the consistency of the association in meta-analysis results across the sensitivity analyses, combined with the trend for a dose-response relationship and an E-value of 2.74, support the association between benzodiazepine exposure during early pregnancy and increased miscarriage risk. It draws attention to the importance of risk-benefit assessment before considering the prescription of these medications to women who are pregnant or may become pregnant. Shared decision-making, close monitoring, and exploration of non-pharmacological alternatives should be considered in clinical practice.

Lastly, benzodiazepines readily cross the human placenta and may accumulate in fetal tissues at concentrations exceeding those observed in maternal serum. They bind to γ-aminobutyric acid (GABA) receptors in both peripheral tissues and the brain, where they play roles in cell proliferation and differentiation, potentially influencing embryonic and fetal development. Evidence from animal studies suggests that benzodiazepines may disrupt fetal development, particularly affecting neurodevelopment and immune system maturation [31–35].

Conclusion

This updated systematic review and meta-analysis (which excluded studies exclusively involving women with epilepsy), strengthens previously raised concerns about an increased risk of miscarriage associated with benzodiazepine exposure during pregnancy. While the inherent limitations of observational study designs must be acknowledged, the consistency of findings across all sensitivity analyses, along with the dose–response relationship reported in three included studies, supports the robustness of this association. Although direct comparisons between substances could not be conducted, the results suggest a class effect of benzodiazepine with no substantial differences in risk magnitude among the most studied benzodiazepines (lorazepam, clonazepam, alprazolam, bromazepam, chlordiazepoxide, oxazepam and diazepam).

Abbreviations

ATDs

Antidepressants

BZDs

Benzodiazepines

CI

Confidence Interval

DDD

Defined Daily Doses

GW

Gestational Weeks

HAS

Haute Autorité de Santé

ICD

International Classification of Diseases

LMP

Last Menstrual Period

MOOSE

Meta-analyses Of Observational Studies in Epidemiology

NA

Not applicable

NICE

National Institute of Clinical Excellence

NOS

Not Otherwise Specified

OR

Odd Ratio

PSUSA

Periodic Safety Update Report Single Assessment

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

ROBINS-I

Risk of Bias in Non-Randomized Studies of Interventions

TE

Treatment effect

Authors’ contributions

MC, JC, and CP for the study protocol; CP for screening eligibility, data extraction and risk-of-bias assessment; YP for data reviewing; JC and CP for drafting the manuscript; AB and JP for critical revision of the manuscript for important intellectual content. All authors were involved in the analysis and interpretation of the data and read and approved the final manuscript.

Funding

This research received a grant from the French National Agency for the Safety of Medicines and Health Products. The funding body did not influence the design of the study and collection, analysis and interpretation of data and the writing of the manuscript.

Data availability

The datasets generated and/or analysed during the current study are available in the metaPreg repository at metapreg.org, and are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. Patient consent for publication is not required because this work, involving meta-analysis of previously published studies, is exempt from direct ethical review board approval.

Consent for publication

Not applicable.

Competing interests

Dr Bérard has a Canada Research Chair Tier 1 on Medications and Pregnancy. No other disclosures were reported.