First trimester exposure to hydroxychloroquine during pregnancy and major congenital malformations: a systematic review and meta-analysis

Abstract

Objective

To assess the association between hydroxychloroquine (HCQ) use during the first trimester of pregnancy and the risk of major congenital malformations (MCMs).

Methods

In this systematic review and meta-analysis, we searched MEDLINE, EMBASE, CENTRAL, clinical trials registers and specialised databases from inception to 9 July 2025. We included randomised controlled trials (RCTs) and comparative observational studies. Study selection and data extraction were performed independently by two reviewers. Risk of bias was assessed using Risk of Bias in Non-Randomised Studies of Interventions or Risk of Bias 2. Primary outcome was any MCM. Meta-analyses used fixed-effects and random-effects models (REM) based on adjusted or crude ORs.

Results

From 4311 records, we included 20 studies (18 cohorts, 1 case-control and 1 RCT) including 4260 HCQ-exposed and 361 802 unexposed-pregnancies. The risk of bias was serious (n=4), critical (n=15) and ‘some concern’ (n=1, RCT). A significant association was found between HCQ and any MCM (OR=1.27, 95% CI (1.11 to 1.45) for REM, I²=0.0%, Tau²=0, n=19 studies). The pooled association based on adjusted OR was also statistically significant (aOR 1.29, 95% CI (1.09 to 1.53) for REM, I²=0.0%, Tau²=0, n=8). The certainty of evidence was low.

Conclusion

A slight association between HCQ exposure and MCM was observed, with low certainty of evidence. Given the well-established maternal/fetal benefits of HCQ in systemic lupus erythematosus, these results do not support changes in current clinical practice regarding its use during pregnancy. For other indications where alternatives exist, its use during pregnancy should be considered on a case-by-case basis.

PROSPERO registration number

CRD420250609725.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Hydroxychloroquine (HCQ), widely used to treat immune-mediated inflammatory diseases, is the cornerstone therapy for systemic lupus erythematosus (SLE) and is commonly continued during pregnancy because of well-established maternal and fetal benefits.

• Evidence regarding its teratogenic risk remains conflicting.

• A 2016 systematic review and meta-analysis reported no significant association between first-trimester HCQ exposure and major congenital malformations (MCMs).

• In contrast, a large US cohort study published in 2021 found a significant association between first-trimester HCQ exposure and MCM, leading to regulatory updates and renewed debate.

WHAT THIS STUDY ADDS

• We included 20 studies (18 cohort studies, 1 case-control study and 1 randomised controlled trial), comprising 4260 HCQ-exposed and 361 802 unexposed-pregnancies.

• First-trimester HCQ exposure was associated with a modest increase in the odds of MCMs, with a low certainty of evidence.

• The association remained consistent when the analysis was restricted to adjusted estimates. Influence analyses yielded effect estimates of similar magnitude for both the main analysis and the analysis based on adjusted ORs, although the 95% CI included the null value for the latter.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This meta-analysis provides updated evidence on the teratogenic risk of HCQ.

• Considering the observational nature of the available evidence and in light of well-established maternal and fetal benefits of HCQ for women with SLE, these results do not support changes in current clinical practice regarding its use during pregnancy.

• For other indications, HCQ use during pregnancy should be considered within a case-by-case risk-benefit assessment.

Introduction

Hydroxychloroquine (HCQ) is an antimalarial drug widely used in the treatment of systemic lupus erythematosus (SLE), rheumatoid arthritis (RA) and other immune-mediated inflammatory diseases (IMIDs).^1–7

HCQ has demonstrated efficacy in reducing flare frequency in both SLE and RA, including during pregnancy, thereby reducing pregnancy and neonatal complications and in reducing maternal long-term damage.^8–12 However, HCQ crosses the placenta, with fetal exposure levels comparable to maternal ones, raising concerns about a potential impact on embryofetal development.^13–15 Because SLE and RA disproportionately affect women of reproductive age, treatment safety, especially during pregnancy, is an important clinical concern.^{3 4 14} Approved by the United States Food and Drug Administration in 1955, HCQ is the cornerstone treatment for SLE, for which available alternatives are either teratogenic (eg, methotrexate, cyclophosphamide, mycophenolate) or consist of biologics with limited safety data during pregnancy (eg, rituximab, belimumab). In contrast, non-teratogenic alternatives are available for RA and other IMIDs.

Numerous studies have explored the association between HCQ exposure during pregnancy and major congenital malformations (MCMs). A 2016 systematic review and meta-analysis including seven cohort studies and one small randomised controlled trial (RCT) reported no significant association between HCQ exposure during at least the first trimester of pregnancy (T1) and MCM (adjusted OR (aOR) of 1.13, 95% CI (0.56 to 2.17)).¹⁶ Subsequently, a large US cohort study published in 2021 found a significant association for exposure during T1, with an adjusted risk ratio (aRR) of 1.26, 95% CI (1.04 to 1.54). Specifically, oral clefts and urinary tract malformations were found to be significantly associated with a daily dose ≥400 mg.¹⁷ The Summary of Product Characteristics for HCQ was therefore updated in 2023 to reflect this recent evidence, omitting previous studies that had found no significant association between HCQ use and MCM.^{11 16 18–20} This raised controversy²¹ and put both patients and their physicians in an uncomfortable position where risks of flares and potential risk of MCM were to be balanced.

In this context, we aimed to assess the association between maternal use of HCQ during the first trimester of pregnancy and the risk of any MCM. The secondary aims were (1) to assess the association between HCQ use and different types of MCM (eg, oral clefts, cardiovascular or genitourinary malformations, etc), (2) to assess the association among women with SLE and (3) to explore the association between HCQ and MCM in different subgroups based on daily HCQ dose, methodological characteristics and risk of bias.

Methods

This Systematic Review and Meta-Analysis (SRMA) followed the recommendations of the Cochrane handbook²² and is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) and the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guidelines.^{23 24} The protocol was registered in PROSPERO Modifications made to the protocol are presented in online supplemental appendix 1.

Data sources and search strategy

We searched from inception to 9 July 2025 PubMed, EMBASE, the Cochrane library, the WHO International Clinical Trials Registry Platform, ClinicalTrials.gov and EudraCT. We also searched specialised teratology databases: Reprotox, Shepard’s Catalog of Teratogenic Agents and TERIS.²⁵ A dedicated search algorithm was developed for each database (online supplemental appendix 2). Abstracts from main international congresses in the field (n=11) over the past 5 years were also reviewed (online supplemental appendix 3). Authors of relevant abstracts were contacted to provide additional data.

Subsequently, two clinicians with recognised expertise in the field (NCC and FC) reviewed the list of included articles and suggested additional references if necessary.

Eligibility criteria

We included RCTs and comparative observational studies assessing the association between in utero exposure to HCQ during the first trimester of pregnancy (T1) and the risk of MCM. The population of interest included all embryos, fetuses, newborns or children who had been exposed to HCQ during T1 while comparators had not been exposed to HCQ at T1, but could have been exposed to other medications, no medication or a placebo.

We excluded non-comparative studies, preclinical studies, expert opinions, editorials, letters and SRMA. However, reference lists from these reviews were screened for potentially eligible studies. There were no restrictions on publication dates or language.

Outcomes

The primary outcome was the occurrence of any MCM, defined according to the WHO as structural changes with significant medical, social or cosmetic consequences for the affected individual, and typically requiring medical intervention.²⁶ Secondary outcomes included specific types of MCM.

Selection process

Records were managed using Zotero V.5.0 and Rayyan. After duplicate removal, EC screened titles and abstracts. Then two review authors (EC and AT) independently reviewed the full texts to assess eligibility. In case of missing or unclear data, study authors were contacted. Disagreements were resolved through discussion with the help of two experts in teratology and epidemiology (BM and BC), until a consensus was reached. Duplicate publications and overlapping study populations were identified by cross-checking study characteristics retaining the study with the most complete or recent dataset.

Data extraction

Two authors (EC and AT) independently extracted data from each included study using a standardised data extraction grid.

The following variables were extracted: country, recruitment period, study design, sample size, data source, population characteristics (maternal age, body mass index (BMI), medical and obstetric history, parental family history, tobacco/alcohol use, drug exposure, details of HCQ exposure (indication, timing and duration of exposure), comparator group (pregnant women with or without disease, exposed to any drug or to a placebo or unexposed). For each arm, we collected the number of participants, the number of events for each outcome per group. During data extraction, when authors provided detailed description of malformations, these were verified to ensure that they could indeed be considered as MCM. Association measures between exposure and outcomes (crude and adjusted estimates) with 95% CIs were also collected. We assessed whether the authors had adjusted for the following key confounders defined a priori: maternal age, socioeconomic aspects, personal and familial history of congenital malformation, BMI/obesity, diabetes, comorbidities, exposure to other medication, smoking, alcohol consumption, teratogens (medications, infections).

The same reviewers independently assessed the risk of bias, using the Risk of Bias in Non-Randomised Studies of Interventions (ROBINS-I) for non-randomised studies²⁷ and the Cochrane risk-of-bias tool for randomised trials (Risk of bias 2 (ROB 2)).²⁸

Discrepancies were resolved by consensus with the help of a third reviewer (BM or BC). In case of missing or unclear data in the manuscript, we attempted to contact the study authors via email.

Data analysis

Meta-analyses for the primary and secondary endpoints

A meta-analysis of OR was performed for endpoints reported in at least three studies. As recommended by the Cochrane collaboration, fixed-effect model and random-effects model (REM) were performed and presented.²⁹ Nevertheless, in the manuscript, we preferentially report results from the REM as the latter is more appropriate when clinical and methodological heterogeneity is expected, which is the case here. The meta-analysis was based on aORs when available; otherwise, crude ORs (cOR) reported in the publication or calculated from the frequencies reported were used. Because the rate of events was low, OR and RR were combined in the same meta-analysis.³⁰ Statistical heterogeneity was assessed visually and by using the Cochrane Q test, I² and Tau².²⁹

Subgroup analyses were conducted according to daily HCQ dose (<400 mg/day vs ≥400 mg/day), methodological characteristics (data collection, classification of congenital malformations used, data source…) and risk of bias (online supplemental appendix 1).

We assessed the robustness of the results in sensitivity analyses by restricting the meta-analysis (1) to aORs because crude estimates may be influenced by confounding, including indication bias and (2) to studies with the highest level of confounder control. We also used alternative control groups for studies that reported multiple comparator groups.

An influence analysis was performed to assess the influence of Huybrechts’ study on results.¹⁷

Small study effect was assessed using funnel plots and Egger’s test.³¹

Statistical analyses were performed using R Statistical Software (R Core Team, V.4.5.0) (online supplemental appendix 4).³²

Certainty of evidence

The level of evidence was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach.³³ Results were summarised with GRADEpro software.³⁴

Role of the funding source

EC received a scholarship from Sorbonne University to support the completion of their Master’s degree internship.

Patient and public involvement

Patients and the public were not involved in the preparation or conduct of this study, as it was a systematic review and direct involvement was not feasible. Results will be disseminated through lay summaries, including on the CRAT website (www.lecrat.fr) and the websites of relevant scientific societies.

Results

Search results

From the 4289 references identified by the electronic search, 3550 records were screened based on title and abstract after duplicate removal. Of these, 51 reports were sought for full-text retrieval, and 14 were included. From the 22 references identified from other sources, six studies were included. In total, 20 studies were included (online supplemental eFigure 1).

Study characteristics

The 20 studies included 18 cohort studies,^{11 17–20 35–47} one case-control study,⁴⁸ and one RCT⁴⁹ (table 1, online supplemental eTable 1), with a recruitment period ranging from 1976 to 2021. This represented 4260 pregnancies exposed to HCQ and 361 802 pregnancies not exposed to HCQ. The study with the largest number of exposed pregnancies was that by Huybrechts et al (n=2045 exposed vs 21 679 unexposed). Several conditions could be reported within the same study, the most frequent being SLE (n=15) and RA (n=7). In 13 studies, the control group consisted of pregnant women with similar underlying diseases and not exposed to HCQ^{11 17 19 20 37–42 47 49} and in one study the control group included women exposed to sulfasalazine⁴³ (maternal indication not specified). The remaining five studies included controls from the general population (n=3),^{36 44 45} women not exposed to HCQ referred to the same Teratology Information Service (n=1)¹⁸ or uncharacterised controls.⁴⁶ Data mostly originated from medical charts (n=10)^{20 35 37 38 40–42 45 47 48} and medico-administrative databases (n=6).^{17 19 36 39 43 44}

Table 1. Summary of study characteristics included in our systematic review and meta-analysis.

Characteristics	Number	References
Study design
Cohort study	18	11 17–20 35–47
Population-based case-control study	1	48
Randomised controlled trial (RCT)	1	49
Sample size of hydroxychloroquine exposed pregnancies
Overall	4260
Max	2045	17
Min	10	49
Geographical origin
Europe	10	19 20 37–40 42 43 46 47
North America	7	11 17 35 36 41 45 48
Brazil	1	49
Australia	1	35
Israel	1	18
Type of data
Medico-administrative databases	6	17 19 36 39 43 44
Teratology information service (TIS)	2	11 18
Medical charts	10	20 35 37 38 40–42 45 47 48
Not reported	2	46 49
Data collection
Prospectively	11	11 17–19 35 36 41 44–46 49
Retrospectively	9	20 37–40 42 43 47 48
Type of included pregnancies
All pregnancies	6	11 18 20 42 47 49
Live birth only	10	17 35–38 40 43–45 48
Live birth and stillbirth	3	19 39 41
Not reported	1	46
Conditions*
Systemic lupus erythematosus	15	11 17–20 35 37 38 40–42 45 47–49
Rheumatoid arthritis	7	11 17–19 39 41 45
Isolated discoid lupus erythematosus	4	18 37 48 49
Other connective tissue diseases	2	11 46
Antiphospholipid syndrome	1	42
Control group among cohorts and RCT (unexposed to hydroxychloroquine)
Similar type of underlying diseases	13	1 17 19 20 37–42 47 49
Sulfasalazine	1	43
General population	3	36 44 45
Women referred to the same TIS	1	18
Not reported	1	46
Hydroxychloroquine exposure time period
First trimester exposure only	9	11 17 19 35–37 39 44 45
Throughout pregnancy	9	18 20 38 40–42 47–49
Unreported	2	43 46
Definition of major congenital malformations
EUROCAT	3	19 20 36
MADCP	2	11 41
Appropriate narrative definition	1	18
Not reported	14	21 35 37–40 42–49

^*In some studies, more than one condition was included.

Exposure definition

All but two studies reported first-trimester exposure; for these two, the timing was not specified. T1 exposure was defined from the last menstrual period to 90 days of gestation (n=4)^{11 17 19 35} or simply described as ‘first trimester’ without further details (n=5).^{36 37 39 44 45} Other studies (n=9) reported exposure throughout pregnancy (including T1).^{18 20 38 40–42 47–49} Five studies also included preconception period.^{20 35 39 41 42}

All studies considered the association between MCM and HCQ exposure at any dose. In five studies, associations between HCQ and MCM were additionally reported based on categorised doses.^{11 17 19 20 47}

Outcomes

The definition and classification used for MCM varied. Six studies used appropriate definitions and classifications (EUROCAT,^{19 20 36} MACDP,^{11 41} narrative definition¹⁸) and 14 did not define MCM. Most studies focused on any MCM with some reporting subgroups or specific subtypes of MCM. Only half of the studies provided a detailed clinical description of the congenital malformations (online supplemental eTable 2). Estimates of the association between HCQ and MCM were available or could be calculated for 42 specific malformative outcomes. Of these, 29 outcomes were assessed in only one or two studies, while 13 were evaluated in three or more studies (online supplemental eTable 3).

Risk of bias

Overall 4 studies had a serious risk^{11 19 41 48} and 15 a critical risk^{17 18 20 35–38 40 42–48} of bias while the RCT was classified at ‘some concerns’ risk of bias⁴⁹ (online supplemental eFigure 2 and 3). As recommended by Cochrane,⁵⁰ 11 observational studies were directly classified as having a critical risk of bias since they did not adjust for any of the relevant confounding factors.^{18 20 35 37 40 42 44 46–49} Eight studies accounted to some extent for key confounding factors and reported aOR (online supplemental eTable 4).

Association between HCQ exposure during T1 and MCM

The pooled OR between any HCQ exposure at T1 and any MCM was 1.27 (95% CI 1.11 to 1.45) in the REM, based on 19 studies: 8 reporting aOR and 11 cOR. No statistical heterogeneity was detected (p_het=0.81, I²=0.0%, Tau²=0) (figure 1).

Figure 1. Pooled ORs for major congenital malformation associated with exposure to any dose of hydroxychloroquine during first trimester of pregnancy. *adjusted OR. ǂOnly live births were retained.

Sensitivity analyses regarding the association between T1 HCQ use and any MCM

The association remained statistically significant when restricting the analysis to aORs only (pooled aOR 1.29, 95% CI (1.09 to 1.53) for the REM on eight studies) with no evidence of statistical heterogeneity (p_het=0.65, I²=0.0%, Tau²=0) (figure 2). The results were similar when restricting the analysis to studies with the highest level of confounding control (pooled aOR 1.33, 95% CI (1.02 to 1.74), n=5 studies) (online supplemental eFigure 4).

Figure 2. Sensitivity analysis—pooled adjusted OR of major congenital malformations associated with exposure to any dose of hydroxychloroquine. *adjusted OR.

Influence analysis excluding Huybrechts, retrieved consistent results (OR 1.28, 95% CI (1.02 to 1.59), p_het=0.76, I²=0.0%, Tau²=0, n=18 studies) in the REM (online supplemental eFigure 5). When restricted to aOR only (n=7 studies), the pooled OR was similar (REM 1.36 (95% CI (0.99 to 1.88)), indicating a similar magnitude and direction of the effect estimate, although the 95% CI included the null value (online supplemental eFigure 6).

Other sensitivity analyses also found a significant association between HCQ exposure and increased MCM risk (online supplemental eFigures 7–9).

Association between HCQ exposure during T1 and subtypes of MCM

There was a significant association between exposure to HCQ and nervous system defects in the REM (pooled cOR of 2.47, 95% CI (1.35 to 4.51), n=6 studies, p_het=0.93, I²=0.0%, Tau²=0) (figure 3a). The association was not significant for spina bifida (figure 3b).

Figure 3. Pooled ORs for subtypes of major congenital malformation associated with exposure to any dose of hydroxychloroquine (a: nervous system; b: spina bifida; c: respiratory defects; d: oral clefts; e: urinary defects; f: congenital heart defects; g: ventricular septal defects). *adjusted OR.

There was a significant association between exposure to HCQ and respiratory defects in the REM: OR of 1.88, 95% CI (1.05 to 3.37) (n=4 studies p_het=0.81, I²=0.0%, Tau²=0) (figure 3c). Only one study reported an aOR (Huybrechts et al; aOR=1.85; 95% CI 0.94 to 3.64), with a weight of 87%.

There was no evidence of association between exposure to HCQ and oral clefts in the REM (figure 3d). Only one study reported an aOR (Huybrechts et al; aOR=3.70; 95% CI 1.55 to 8.83), with a weight of 50%.

No significant association was identified for urinary defects, congenital heart defects, ventricular septal defects, gastrointestinal defects, polydactyly, eye anomalies, limb anomalies, musculoskeletal defects, genital defects and hypospadias (figure 3e–g and figure 4a–g).

Figure 4. Pooled ORs for subtypes of major congenital malformation associated with exposure to any dose of hydroxychloroquine (a: gastrointestinal defects; b: polydactyly; c: eye anomalies; d: limb anomalies; e: musculoskeletal defects; f: genital anomalies; g: hypospadias). *adjusted OR.

Subgroup analyses by daily dose of HCQ

For a dose <400 mg/day (around 27% of the exposed women), the pooled OR was 1.03, 95% CI (0.75 to 1.42) (p_het=0,86, I²=0%, Tau²=0, n=4, REM). No significant association was observed for a daily dose ≥ 400 mg (around 73% of the exposed women) in the REM with pooled OR of 1.26, 95% CI (0.94 to 1.69), n=4, p_het=0,45, I²=0%, Tau²=0. Interaction was not statistically significant (figure 5).

Figure 5. Subgroup analysis of major congenital malformations associated with exposure to daily dose of hydroxychloroquine (<400 mg/day vs ≥400 mg/day). *adjusted OR.

An influence analysis excluding the Huybrechts study revealed no significant difference between dose categories and a reduction in the pooled OR for the ≥400 mg/day category (online supplemental eFigure 10).

Association between HCQ exposure during T1 and MCM among SLE women

When the analysis was restricted to women with SLE, the association was not statistically significant (OR=1.04, 95% CI (0.59 to 1.82) for the REM, phet=0,68, I²=0%, Tau²=0, n=8). This result was based on seven studies reporting cOR only and one study—Nguyen et al which had a weight of 64%—reporting an aOR of 1.29, 95% CI (0.65 to 2.56) (figure 6).

Figure 6. Pooled OR of major congenital malformations associated with exposure to any dose of hydroxychloroquine among women with systemic lupus erythematosus. *adjusted OR.

Other subgroup analyses

There was no significant difference between the subgroups based on methodological characteristics, including type MCM classification (p=0.91), type of data collection (p=0.92), type of data source (p=0.38), type of pregnancy outcome (p=0.38), type of control group (diseased women vs from the general population, p=0.82) or risk of bias (p=0.20) (online supplemental eFigure 11–16). A subgroup analysis excluding studies based on live births only was also performed and yielded results consistent with those of the main analysis, although the 95% CI included the null value (pooled OR 1.44, 95% CI 0.97 to 2.13, n=9 studies p_het=0.45, I²=0.0%, Tau²=0; REM, online supplemental eFigure 17).

Small study effect

The funnel plot for the association between T1 HCQ use and any MCM did not reveal visual asymmetry. Egger’s test was not statistically significant (p value=0.58) (online supplemental eFigure 18).

Certainty of evidence

The certainty of evidence was rated as low for the main outcome, mainly due to risk of bias inherent to observational studies, particularly selection, confounding and classification biases, and was evaluated as moderate for the sensitivity analysis restricted on aOR (online supplemental eTable 5).

Discussion

In this SRMA including 20 studies with 4260 pregnancies exposed to HCQ and 361 802 unexposed pregnancies, there was a significant association between exposure to HCQ during the first trimester of pregnancy and MCM (OR=1.27, 95% CI (1.11 to 1.45), REM), with a low certainty of evidence and a low heterogeneity across studies. This association was consistent when restricted to aOR (1.29, 95% CI (1.09 to 1.53), REM). Influence analyses yielded effect estimates of similar magnitude for both the main analysis and the analysis based on aORs, although the 95% CI included the null value for the latter. When considering specific MCM, significant associations were found for central nervous system and respiratory defects. There was no heterogeneity by dose of HCQ. No association was found when restricted to women with SLE.

This SRMA has several major strengths. First, it is based on a rigorous methodology, in accordance with the Cochrane Handbook.²² We conducted a comprehensive literature search from multiple databases and also searched additional sources such as congress. Selection of studies and data extraction was performed independently by two authors. The risk of bias was evaluated using ROB 2 and ROBINS-I tool. Particular attention was paid to identifying adjustment on key confounding factors and the type of control group. When necessary, authors of the included studies were contacted to clarify missing or ambiguous information. Second, the geographic diversity of the included studies (Europe, North America, Australia, Brazil and Israel) enhances the generalisability of the findings beyond any single healthcare system or population. Third, there was no evidence of small-study effects, suggesting an absence of detectable publication bias for the studied outcome.

Nevertheless, there are several limitations. First, our main outcome ‘any MCM’ is, by definition, a composite outcome and may lack specificity to identify specific teratogenic risks. Nevertheless, we were able to perform several meta-analyses focusing on some congenital malformation subtypes. Second, some pre-defined analyses could not be performed for certain congenital malformation subtypes because they were reported in fewer than three studies. Third, some potentially eligible studies (11 RCTs registered on ClinicalTrials.gov and 5 abstracts) were excluded due to lack of response from authors despite multiple attempts of contact. Finally, some subgroup and sensitivity analyses included only few studies, limiting statistical power and precision.

This SRMA included over six times more HCQ-exposed pregnancies than the last SRMA published in 2016 (4260 vs 678), significantly expanding the evidence base. Kaplan found no significant increase in overall MCM (pooled OR 1.13, 95% CI (0.59 to 2.17) based on six studies. The current SRMA included 13 additional studies, which enable more refined and outcome-specific classification analyses, as well as increased statistical power. It also made it possible to perform subgroup analyses based on daily HCQ dose and to conduct analyses restricted to women with SLE.

Among the recent studies, the study of Huybrechts et al stands out as the largest cohort assessing the risk of MCM with first-trimester exposure to HCQ. In this population-based cohort on two large US administrative databases (Medicaid and MarketScan), they compared 2045 pregnancies exposed to HCQ to more than 21 679 pregnancies not exposed to HCQ, focusing on women with autoimmune rheumatic disorders. The study controlled confounding factors, through restriction and propensity-score matching and found a 26% increased risk of MCM among HCQ-exposed pregnancies (aRR 1.26; 95% CI (1.04 to 1.54). However, as the analysis was restricted to live births, the risk of bias was assessed as critical. The authors concluded that although the absolute risk remained low, a modest teratogenic risk cannot be excluded and warranted further research. Our study also found a similar slight increase in the risk of MCM, even in the influence analysis of the main result excluding Huybrechts.

While Huybrechts et al reported no statistically significant association for doses <400 mg/day (aRR 0.95; 95% CI 0.60 to 1.50), the association was significant for doses ≥400 mg/day (aRR 1.33; 95% CI 1.08 to 1.65). In our subgroup analysis, we observed a higher pooled OR for daily HCQ doses ≥400 mg compared with doses <400 mg. However, the interaction test was not statistically significant, possibly due to limited statistical power. Therefore, results of this exploratory analysis should be interpreted with caution and do not allow conclusions to be drawn regarding the presence or absence of a dose-response relationship.

Several sources of methodological heterogeneity were identified across studies, notably differences in the definitions of MCM, type of pregnancy outcomes and exposure window, which should therefore be taken into account when interpreting the results, particularly in the context of observational exposure data.⁵¹ However, no statistical heterogeneity was detected (I²=0.0%), and subgroup analyses did not identify potential sources of statistical heterogeneity. Among the limitations of the existing literature, the issue related to potential selection bias deserves particular attention. Indeed, 10 studies out of 20 were restricted to live births (along with 6 including all pregnancies, 3 including both live births and stillbirths and 1 for which this information was not reported). This may introduce an important selection bias, a well-recognised limitation in perinatal pharmacoepidemiology.⁵² Incomplete capture of early pregnancy outcomes and conditioning on live births may lead to left truncation, whereby pregnancies ending in early loss or termination are not captured. For a teratogenicity question, this is highly consequential, because pregnancies affected by severe congenital malformations are more likely to result in termination or fetal loss, this may lead to an underestimation of the true prevalence of MCMs. This under-ascertainment may also vary across study designs, healthcare systems and exposure groups, thereby affecting comparability between studies. Ultimately, this bias may lead to an underestimation of the measure of association, potentially biasing results towards the null and reducing the likelihood of detecting a true difference between groups. In our study, the subgroup analysis based on the type of pregnancy outcome did not identify any differences between groups (live births, live births+stillbirths and all pregnancies, p=0.38). In addition, an analysis excluding studies based on live births only yielded estimates consistent with those of the main analysis, although the 95% CI included the null value.

Furthermore, we cannot formally rule out that the maternal disease indication and its severity may independently contribute to the risk of malformations. Although 14 studies included a control group of women with SLE, RA, or other autoimmune disorders, five studies relied on controls from the general population. Nevertheless, to date, no association has been reported between maternal indications for HCQ or disease activity and the occurrence of MCM. When the analysis was restricted to results based on women with SLE, the association was not significant. However, this result was based on seven studies reporting crude OR and only one study reporting aOR (Nguyen et al, aOR=1.29, 95% CI (0.65 to 2.56)), which was of similar magnitude to our main result.

Another limitation of the existing evidence lies in the fact that among the included studies, over two thirds did not report a definition of MCMs and only a third used appropriate definition and classification. This may have introduced outcome misclassification and complicated the interpretability of pooled estimates. Indeed, when a composite teratology endpoint is pooled across studies with heterogeneous or absent definitions, the apparent precision of the pooled estimate may be misleading. This concern becomes even more important for the subtype analyses, where several associations appear statistically fragile, model-dependent, based on a limited number of studies and sparse outcome counts and, in some instances, heavily influenced by a single large study. For example, for respiratory defects, our analysis included four studies among those only one study provided an adjusted estimate and accounted for approximately 87% of the statistical weight. As a result, subtype analyses should be considered hypothesis-generating rather than confirmatory.

Besides, in most studies, the clinical description of malformations was not provided: detailed information was available, or could be obtained from the authors, for only 9 out of 20 studies. This underlines a major lack of clinical teratology information (particularly in studies based on medico-administrative databases). Moreover, no specific pattern of malformations was identified, and no clear correlation could be established between the phenotype of the malformations described in the original reports and the results of our meta-analysis on malformation subtypes. This therefore warrants caution in the interpretation of our conclusions as it is not possible to identify a specific embryofetal tissue target.

Such limitations of the literature highlight the need for future research to better assess HCQ safety during pregnancy particularly through the use of large datasets with standardised exposure and outcome definitions, along with detailed clinical description of the malformations, carefully accounting for relevant confounding factors in particular adequately controlling for maternal indication and including all pregnancy outcomes. This aligns with the EULAR agenda, which emphasises the importance of strengthening the safety profile of HCQ through large-scale data and robust methodologies.

Cautious interpretation is warranted when evaluating emerging drug-associated adverse event signals.⁵³ Also, in observational exposure meta-analyses, careful consideration of confounding control and certainty of the evidence is also essential when interpreting pooled estimates.⁵⁴ Our primary analysis combined adjusted and crude estimates in accordance with our protocol allowing inclusion of the largest possible body of evidence. Restricting the primary analysis exclusively to adjusted estimates would have substantially reduced the number of eligible studies, the precision of the pooled estimates and the power of our analysis. Nevertheless, the certainty of evidence for our main outcome, according to GRADE, was rated as low. This was mainly driven by the risk of bias inherent to observational studies, particularly selection, confounding and classification. In particular, 15 studies were at critical risk of bias, including 11 observational studies classified as critical because they did not adjust for any relevant confounders. These issues led to an ‘extremely serious’ rating for the risk of bias domain in the GRADE assessment. In contrast, no serious concerns were identified for the other domains, including indirectness, imprecision and inconsistency.

The result based on aORs is likely to provide the principal clinically interpretable analysis (with a moderate level of evidence). This is particularly important in the context of HCQ exposure during pregnancy, where confounding by indication or related to maternal disease activity, co-medications, folate supplementation, diabetes, obesity for example may influence observed associations. Online supplemental eTable 4 reports, for each study, the original effect measure, covariates taken into account, statistical method used, comparator population and study weight for both the meta-analysis including all studies and the meta-analysis based on aORs. As shown in this table, adjustment strategies were heterogeneous across studies and several relevant confounders were not consistently accounted for. Therefore, residual confounding cannot be excluded even in analyses restricted to adjusted estimates, and these results should be interpreted accordingly.

The use of chronic immunomodulatory therapy necessitates a structured benefit-risk assessment.⁵⁵ The use of HCQ during pregnancy must be evaluated by carefully weighing its potential risks in light of its well-established maternal benefits.^{21 56} In RA, the availability of non-teratogenic therapeutic alternatives allows consideration of other treatments in the preconception period and during pregnancy. In SLE, HCQ remains a cornerstone therapy for maintaining maternal disease control and preventing flares,⁵⁷ thereby reducing the risk of maternal complications^{58 59} and adverse pregnancy outcomes, including miscarriage, preeclampsia, fetal growth restriction and preterm birth.^{3 9 18 21 35 40 58} If the association observed between HCQ and MCM is causal, and considering the 95% CI of the estimate from our main analysis (1.11–1.45) and a baseline risk of approximately 2.5% for MCMs in the general population according to EUROCAT, this would translate to between 3 and 11 additional MCM per 1000 exposed pregnancies. On the other hand, based on the literature for 1000 HCQ exposed pregnancies in SLE, one can estimate between 20 and 120 fewer very preterm births and 30 and 100 fewer cases of preeclampsia.^{40 60–62}

The evidence presented in the SLE subgroup is itself limited and not statistically significant, with only one adjusted study contributing materially to that subgroup analysis. This finding therefore appears too fragile to support recommendations for discontinuing HCQ during pregnancy. Moreover, many second-line treatments for SLE (steroids, immunosuppressive drugs—including mycophenolate, methotrexate, biologics) are not without maternal or fetal consequences, when they are not established teratogens. In other indications, HCQ use during pregnancy should be considered within a case-by-case risk-benefit assessment, taking into account disease severity, medication alternatives and the limited evidence available on fetal safety. In all instances, HCQ use during pregnancy should also rely on shared medical decision-making, involving a well-informed patient and appropriate prenatal monitoring.

Conclusion

A slight but significant and consistent association between MCM and HCQ exposure during the first trimester of pregnancy was found. Given the observational nature of the data and the low certainty of evidence provided here, these findings do not by themselves establish a causal teratogenic effect. HCQ may remain clinically necessary for disease control in many pregnant patients, and the modest association reported here must be interpreted cautiously and discussed within individualised maternal-fetal risk-benefit decision-making.

The funders had no role in considering the study design or in the collection, analysis, interpretation of data, writing of the report or decision to submit the article for publication.

Data availability statement

Data are available upon reasonable request.