Safety of macrolide antibiotics in pregnancy: a review of the epidemiological evidence - GOV.UK

1. Plain language summary

Key message

The Commission on Human Medicines (CHM) has reviewed the available safety data for the use of the macrolide antibiotics erythromycin, clarithromycin and azithromycin during pregnancy.

The available evidence is insufficient to confirm with certainty whether there is a small increased risk of malformations (birth defects) or miscarriage when macrolides are taken in early pregnancy.

There remains a need for high-quality research into the effect of erythromycin, clarithromycin or azithromycin in pregnancy. Further data are needed to draw firm conclusions.

Certain infections in pregnancy can cause serious harm to both to the mother and baby if not treated. In such cases, pregnant women should receive treatment with an appropriate antibiotic. Decisions by the prescriber of which antibiotic to use should be based on the benefits and risk to mother and baby. If a prescriber views that the potential benefits of treatment will outweigh the risks and that no suitable and safe alternative is available, for example in true penicillin allergy, a macrolide can be used during pregnancy.

Introduction

The Medicines and Healthcare products Regulatory Agency (MHRA) is the government agency responsible for regulating medicines and medical devices in the UK and ensuring their safety, quality and effectiveness. We continually review the safety of all medicines in the UK and inform healthcare professionals and the public of the latest updates.

The CHM advises government ministers and the MHRA on the safety, efficacy and quality of medicines. The MHRA safety public assessment reports aim to discuss evidence-based reviews of safety issues linked with a particular medicine or group of medicines.

This report presents our review of the safety of macrolide antibiotics erythromycin, clarithromycin and azithromycin during early pregnancy. The review evaluated the quality of the safety evidence in relation to three outcomes: major malformations of the baby in the womb, heart or blood vessel malformations, and miscarriage.

About macrolides

The macrolide drug group are antibiotics used to treat acute and chronic infections. These medicines are effective treatments for a range of infections similar to penicillin, another antibiotic, and so macrolides are often used in patients allergic to penicillin.

The three main macrolide antibiotics used in the UK are azithromycin, clarithromycin and erythromycin:

• Azithromycin is used in infections of the respiratory tract, ear, skin and soft tissue infections, infections of the urethra and sexually transmitted infections including chlamydia and gonorrhoea.
• Clarithromycin is used to treat infections of the respiratory tract, ear, skin and soft tissue infections and also treating stomach ulcers caused by the bacteria Helicobacter pylori
• Erythromycin is used in infections of the respiratory tract, ear, eyes or mouth, skin and soft tissue infections, infections of the stomach and intestines, infections of the urethra and sexually transmitted infections including syphilis, chlamydia and gonorrhoea.

A previous review of the respective Summaries of Product Characteristics (SmPCs) for macrolides marketed in the UK indicated varying levels of information and advice regarding use in pregnancy. There is therefore inconsistent and conflicting information in the various SmPCs regarding the safety and risk-benefit balance of macrolides in pregnancy.

Reason for the review

This review was initiated following the publication of a large cohort study in the UK (Fan and others, 2020), which reported a small increased risk of the baby being born with major malformation associated with use of macrolide antibiotics during pregnancy. Specifically, the study reported increased risks of malformations relating to the heart or blood vessels in babies born to mothers who were prescribed macrolide antibiotics in the first trimester of pregnancy.

The CHM recommended that the MHRA should undertake a review on whether the available data raise any new safety concerns or change current understanding about the safety of macrolide antibiotics during pregnancy.

Conclusions of the review

A range of studies were evaluated in this systematic review of the evidence. The overall quality of observational data on the safety of erythromycin, clarithromycin and azithromycin by mothers was low. Most studies included in the review were judged to have a serious risk of bias, according to a bias evaluating tool (ROBINS-I). There is also a lack of a known biological mechanism for malformations in the womb associated with macrolide antibiotics.

The overarching findings of the review are that the available evidence is insufficient to confirm with certainty the presence or absence of a small increased risk of malformations or miscarriage when macrolides are taken in early pregnancy.

The data were insufficient to confirm:

• small increased risk of major congenital malformations or cardiac malformations following exposure to erythromycin
• increased risk of miscarriage following exposure to clarithromycin or azithromycin

The data were also insufficient to establish:

• the absence of small increased risk of major congenital malformations or cardiovascular malformations following azithromycin or clarithromycin exposure
• the absence of small increased risks of miscarriage following exposure to erythromycin

There remains a need for high-quality research into the effect of erythromycin, clarithromycin or azithromycin prescription in pregnancy.

2. Introduction

The Medicines and Healthcare products Regulatory Agency (MHRA) is the government agency responsible for regulating medicines and medical devices in the UK. We continually review the safety of all medicines in the UK and inform healthcare professionals and the public of the latest updates. The aim of MHRA safety public assessment reports is to discuss evidence-based assessments of safety issues associated with a particular medicine or group of medicines.

A glossary is provided for an explanation of the terms used in this report.

The following report discusses our review of the relating to the safety of medicines containing macrolide antibiotics during pregnancy.

The information and analyses contained in this report reflect evidence that was available at the time of the review in 2020. They are not intended to provide clinical advice. The MHRA will continue to monitor the safety of all medicines. The information in this report will not be actively updated with new data or studies unless major new safety information is available that results in critical changes.

Reason for this review

In 2020, a large UK cohort study reported a significant association between maternal use of macrolide antibiotics in early pregnancy and major congenital malformations, specifically cardiovascular malformations (Fan and others, 2020).

Following publication of this study, the Commission on Human Medicines (CHM) recommended that the MHRA review all available epidemiological evidence to understand reasons why the findings conflicted with that of other studies and to determine whether an updated meta-analysis of evidence may be a feasible option.

Macrolide antibiotics

The 3 main macrolide antibiotics authorised and used in the UK are azithromycin, clarithromycin and erythromycin.

These macrolides have a similar antibacterial spectrum to penicillin and are frequently used as an alternative to penicillin, for example in patients allergic to penicillin:

• Azithromycin is indicated for respiratory tract infections (RTIs), otitis media, skin and soft tissue infections, chlamydia, gonorrhoea and urethritis
• Clarithromycin is indicated for RTIs, otitis media, skin and soft tissue infections and Helicobacter pylori eradication
• Erythromycin is indicated for RTIs, ear, eye and oral infections, skin and soft tissue infections, gastrointestinal infections and various other infections such as chlamydia, gonorrhoea and urethritis

Clinical guidelines on use of macrolides in pregnancy

The NHS website states that erythromycin is the only macrolide which can be taken during pregnancy.

National Institute for Health and Care Excellence (NICE) antimicrobial prescribing guidelines for managing common infections includes macrolides within their recommended antibiotic choices. Examples include, treatment of acute sore throat, acute sinusitis, acute otitis media, community-acquired pneumonia, cellulitis, erysipelas, diabetic foot infection and leg ulcer infection. Erythromycin is recommended as the preferred choice for pregnant women where a macrolide is deemed suitable, if, for example, true penicillin allergy is present.

NICE guidelines for use of these medicines in pregnancy are informed by recommendations from the UK Teratology Information Service (UKTIS). The UKTIS monograph on the use of macrolides in pregnancy was updated in April 2020 to reflect the findings of Fan and others (2020) and to highlight that macrolides, should only be used in pregnancy when clinically necessary and if the benefit of treatment is expected to outweigh any small increased risks which may exist.

Pregnancy warnings in macrolide product information

A previous review of the respective Summaries of Product Characteristics (SmPCs) for macrolides marketed in the UK indicated varying levels of information and advice regarding use in pregnancy.

All the azithromycin and clarithromycin SmPCs include a risk-benefit statement regarding use during pregnancy, stating that the prescriber should consider whether the benefits to use outweigh the risks and there is a clinical need. However, information on use in pregnancy appears to be more variable in the SmPCs for erythromycin products, with some stating that there is 'no evidence of hazard from erythromycin in human pregnancy' and some mentioning the increased risk of cardiovascular malformations, which is likely to have been prompted by findings of Källén and others (2005, 2014)

There is therefore inconsistent and conflicting information in the various SmPCs regarding the safety and risk-benefit balance of macrolides in pregnancy.

3. Epidemiological studies on macrolide safety in pregnancy

Recent studies presenting conflicting findings on macrolide safety in pregnancy

Findings by Fan and others: a population-based cohort study

In February 2020, a large cohort study conducted in the UK Clinical Practice Research Datalink (CPRD) and published in the BMJ (Fan and others, 2020) reported an increased risk of major congenital malformations in infants whose mothers received a prescription for a macrolide antibiotic during early pregnancy compared to those who received a prescription for penicillin (adjusted risk ratio (aRR) 1.55, 95% confidence interval (CI) 1.19 to 2.03).

Increased risks were also reported for cardiovascular defects with macrolide prescribing in the first trimester (aRR 1.62, 1.05 to 2.51), and for major congenital malformations with erythromycin prescribing in the first trimester (aRR 1.50, 1.13 to 1.99).

This contrasted with several studies performed in other countries that did not find a statistically significant increased risk of these outcomes with macrolides as a class or for erythromycin specifically (Damkier and others, 2019; Muanda and others, 2017b; Lin and others, 2013; Bahat Dinur and others, 2013; Romoren and others, 2012; Bar-Oz and others, 2012; Crider and others, 2009; Cooper and others, 2009; Bar Oz and others, 2008, and Czeizel and others, 1999).

The study by Fan included a retrospective cohort of women who received one prescription for either a macrolide or a penicillin antibiotic during pregnancy in routine clinical practice between 1990 and 2016.

Fan's cohort study recorded a range of adverse outcomes in the children such as major malformations, cerebral palsy, epilepsy, attention deficit hyperactivity disorder (ADHD), and autism spectrum disorder. The selection of outcomes of interest was based on the authors' theory that the proarrhythmic effects of macrolides induce fetal hypoxia, which may in turn lead to these events.

Findings by Damkier and others: a population-based study (2019)

When considering the findings of the Fan study (2020), the findings of another recently published cohort study were noted (Damkier and others, 2019).

In contrast to the UK study, this large Danish cohort study investigated associations between in-utero exposure to various antibiotics and the risk of congenital malformations.

Damkier did not report increased risks for major congenital malformations or cardiovascular malformations in infants exposed to erythromycin or azithromycin in the first trimester when compared to infants exposed to penicillin (adjusted odds ratio (aOR) for major congenital malformations 0.94, 95% CI 0.80 to 1.11 (erythromycin) and 1.09, 0.93 to 1.27 (azithromycin); aOR for cardiovascular malformations: 0.94, 0.69 to 1.28; and 1.14, 0.86 to 1.51, respectively).

Further analysis of study findings

The MHRA conducted further analysis of these studies to try and understand the reasons for the different findings (see table at Annex 1).

In summary, both studies presented an appropriate approach to a collection of population-based data from national databases and their analysis. However, with the difference in the selection of outcomes and grouping of exposures (as in Fan and others, 2020), and potentially due to fundamental differences in the distribution of exposures between the 2 populations (as discussed below), the comparison of findings of the 2 studies is difficult. Neither study presented rates of outcomes of interest in the general population, which would provide a context and a reference.

Both studies were considered to have used an appropriate statistical method. The analytical approach taken by Fan is appropriate, and Cox regression is generally the stronger modelling approach, since it utilises more information, while logistic regression (used in Damkier's study) estimates are prone to bias and reduced precision with increasing follow-up time.

The use of propensity scores (used by Fan's study) to match exposed and unexposed individuals can help to control heterogeneity of characteristics between patients, but the effect of variables on the outcome of interest that were matched on cannot be estimated.

Propensity score matching was used by Fan to adjust for covariates by applying weighting of the sample of units in each treatment group to match covariate distribution of a target population. The propensity scores derived from the macrolide-exposed group were used to weight children from the penicillin-exposed group.

Although this non-parametric balancing strategy is a recognised approach, application of this method commonly excludes unmatched units or units with extremely large weights. The remaining sample may only represent a subpopulation of the original targeted population, which can vary from study to study. This is referred to as 'trimming'.

Fan's study does not report on the number of observations that were trimmed during the matching and discrepancies between treatment groups were not discussed. However, given the large sample size, it might be assumed that this might not have had a significant effect on the estimates.

Fan's study evaluated macrolides as a class and presented analyses restricted to erythromycin, whereas Damkier studied only the effects of individual macrolides azithromycin and erythromycin within their study. Damkier also only presented limited data on the prevalence of outcomes by exposures in descriptive statistics.

Logistic regression analysis (used by Damkier) is an appropriate method; the estimates from this analysis approximate those from Cox regression for rare outcomes. However, as previously highlighted the Cox model would usually be preferred.

Data on outcomes were collected by Damkier within 1 year after birth, while Fan had 14 years of follow-up, which allowed inclusion of developmental disorders that can manifest or be diagnosed in later life. Damkier considered a smaller number of potential confounders compared to the study by Fan. No adjustment for previous or chronic maternal infection status was considered by Damkier. As such, confounding by indication cannot be ruled out, although this might be expected to bias the analyses away from the null and produce increased risk estimates for the outcomes, which was not the case in Damkier's study

The difference in proportions of penicillin prescriptions between the studies cannot be overlooked. The difference in prescribing practice regarding antibiotic choice may influence the respective findings. In Damkier's study, 37% of all prescriptions were penicillin, while in Fan's study prescriptions of penicillin constituted over 64% of antibiotic prescriptions in pregnant study participants requiring antibiotics.

If the choice of macrolide is associated with the severity or type of infection, then this could show systematic differences between groups of pregnant women prescribed macrolides or penicillin.

Although Fan tried to adjust their analysis for infections (for example sexually transmitted or genito-urinary), these variables might not have captured the whole variation in types of infection. Additionally, while Fan attempted to adjust for severity of infection by excluding pregnancies exposed to more than one course of macrolides, this would not fully eliminate confounding related to infection severity.

If the rationale for the choice of macrolide is non-differential across different indications, then confounding by severity or type of infection may not be present. Thus, the absence of difference in the effect of macrolides to those of penicillin could be true for the study population.

However, if penicillin is prescribed for specific types or severity of infection associated with adverse fetal outcomes and analysis failed to adjust for the effect of infection, this would increase the rate in penicillin group and bias any effect of macrolides toward the null.

Given the apparent higher use of macrolides in the Damkier study, it is not clear whether they were used predominantly in women with a reported penicillin allergy, as seems to be mainly the case in the Fan study.

Furthermore, the rate of any malformations in infants exposed to erythromycin (283 in 5563) was 50.87 per 1000 (calculated from data in supplementary tables) in the study by Damkier. This was much higher than the rate reported by Fan of 27.6 per 1000, albeit this rate was calculated using data on prescriptions in the first trimester only, which might have under-represented the rate by not capturing the whole duration of pregnancy.

However, the rates of malformation in control groups (exposed to penicillin and non-exposed) might have been relatively high to render non-significant results when compared to the rate of malformations in macrolide-exposed groups in the study by Damkier (the data were not presented to enable the calculation). This raises questions about the level of outcome misclassification.

While external estimates of the rate of congenital malformations vary, and these estimates are within the normal range, limitations to the data in both studies (which rely on the secondary use of data) may have led to different levels in the capture of relevant outcomes.

Finally, there were key differences in the classification of first trimester exposure, with Fan restricting to 4 to 13 weeks' gestation, thereby eliminating the first few weeks of pregnancy, prior to organogenesis.

Like several other studies that did not report associations between macrolide exposure during pregnancy and fetal harm, Damkier and others classed pregnancies as exposed if a macrolide prescription was recorded between 0 and 14 weeks' gestation, which may have resulted in a dilution of any risk. This is further discussed in the overall discussion section below.

Meta-analyses on the safety of macrolides during pregnancy

Two meta-analyses have been conducted to evaluate the body of evidence on macrolides and adverse child or pregnancy outcomes (Fan and others, 2019; Mallah and others, 2019). However, the authors considered different outcomes and applied slightly different study inclusion criteria. A comparison of the studies can be found in Annex 1.

Based on the hypothesis that short-term fetal hypoxia induced by fetal arrhythmia could possibly be the underlying mechanism of observed adverse effects of macrolides, Fan's 2019 meta-analysis included outcomes that could potentially result from short-term fetal hypoxia.

These outcomes included fetal and neonatal death, congenital malformations, and conditions resulting from central nervous system damage such as epilepsy, cerebral palsy, ADHD and autism.

Mallah and others commented that Fan's 2019 meta-analysis only assessed 'general adverse child outcomes' and only included a limited number of original studies, therefore their review included any congenital malformation as an outcome.

Both meta-analyses demonstrated a systematic approach to search, selection, evaluation and analysis of studies. The objectives of both analyses were to consolidate evidence on the effects of the use of macrolides in pregnancy on paediatric and fetal outcomes.

Although both were methodologically appropriate, each of the studies demonstrated their individual approach to the conduct of systematic review and meta-analysis. This was primarily due to subjectivity inherent in the selection of search terms, databases, selection and application of inclusions, and inclusion criteria to primary research studies.

The meta-analyses were composed of 2 separate sets of studies. There was partial overlap between the studies in the two metanalyses (6 of the 16 total studies). Despite this, there was mild agreement that there was either no association, or a very weak association between first-trimester exposures to macrolides and congenital malformations.

Fan's 2019 meta-analysis reported a pooled OR for major malformations of 1.13 (95% CI 0.99 to 1.29) compared to alternative antibiotics, based on 3 studies: Einarson and others, 1998; Romoren and others, 2012; and Muanda and others 2017b (pooled OR 1.03, 95% CI 0.86 to 1.22 based on 4 studies for all malformations).

Mallah and others reported an OR for all malformations of 1.06 (1.00 to 1.12) compared to the group exposed to non-macrolide anti-bacterials and other non-teratogenic drugs, based on 9 unspecified studies.

Use of mixed study designs (randomised control studies and observational studies) might not be appropriate but they were handled separately in the review by Fan and others. In contrast, Mallah and others only included observational studies.

Outcomes used by Fan were more inclusive and included central nervous system damage and miscarriage, while Mallah only considered outcomes in live births by different anatomic locations.

Ultimately, given the heterogeneity of conduct and findings of the 2 meta-analyses it is difficult to compare the findings presented in the 2 publications.

A systematic appraisal of their corresponding primary research studies was considered be a more appropriate approach for consolidation of available evidence on macrolide safety in pregnancy.

In order to facilitate this approach, a previously developed tool for assessing the quality of evidence was used, the ROBINS-I tool.

4. Systematic review of the evidence on the safety of macrolide use in early pregnancy

The objective of this systematic review by the MHRA was to evaluate the available epidemiological evidence on the effects of exposure to macrolides in early pregnancy.

The quality of studies was assessed using the ROBINS-I tool for non-randomised studies of interventions. The review was conducted to ascertain the effects of individual macrolides (erythromycin, azithromycin, and clarithromycin) on miscarriage, major congenital malformations, and cardiovascular malformations.

Full details of the systematic quality review are available upon request.

Materials and methods

Studies previously included in 2 meta-analyses on the effects of macrolide antibiotics (Mallah and others, 2020; Fan and others, 2019), along with the 2 most recent primary research publications (Damkier and others, 2019; Fan and others, 2020) were included. This resulted in a total of 38 studies for initial review to determine whether they qualified according to the agreed criteria in terms of the review questions.

This review evaluated the evidence under 3 separate objectives for each of the 3 reviewed macrolide antibiotics, developed using the Population Intervention Control Outcome criteria (PICO) (Higgins and others, 2019).

Studies with penicillin as a comparator were initially prioritised over the studies with comparator groups unexposed to antibiotics. This decision was made with the objective to review research with a reduced presence of confounding by indication.

The presence of this type of confounding can compromise the quality of observational studies on the effects of medication. However, on closer review of the studies, only a small number used penicillin as a comparator group (Fan, 2020; Damkier, 2019; Muanda, 2017a; Muanda, 2017b). Therefore, studies with unexposed pregnancies were also included in the review.

The objectives were to:

• evaluate the effect of erythromycin, azithromycin and clarithromycin use in the first trimester of pregnancy on the risk of major congenital malformations compared to penicillin exposure or pregnancies not exposed to antibiotics
• evaluate the effect of erythromycin, azithromycin and clarithromycin use in the first trimester of pregnancy on the risk of congenital cardiovascular malformations compared to penicillin exposure or pregnancies not exposed to antibiotics
• evaluate the effect of erythromycin, azithromycin and clarithromycin use in the first trimester of pregnancy on the risk of miscarriage compared to penicillin exposure or pregnancies not exposed to antibiotics

The studies that addressed these objectives qualified for further review and assessment of the risk of bias using the ROBINS-I tool for non-randomised studies of interventions (Sterne and others, 2016).

Data on study design, study period, location, participants characteristics (age, health, pregnancy characteristics, and so on), and intervention characteristics (erythromycin, penicillin or control) were extracted and recorded on a pre-defined form.

The quality of the studies was independently evaluated by 2 MHRA reviewers using the ROBINS-I tool (Sterne and others, 2016). The ROBINS-I approach is based on the evaluation of bias in relation to a target trial, which is designed to allow an unbiased unconfounded study of the association without concern over feasibility or ethical considerations.

Each study was assessed across 7 bias domains from the ROBINS-I tool, adapted below from Sterne and others (2016).

Pre-intervention bias domains:

Domain 1 (D1) considers bias arising from confounding, including selection bias, allocation bias and channeling bias. This is explained as 'baseline confounding occurs when one or more prognostic variables (factors that predict the outcome of interest) also predicts the intervention received at baseline. ROBINS-I can also address time-varying confounding, which occurs when individuals switch between the interventions being compared and when post-baseline prognostic factors affect the intervention received after baseline.'

Domain 2 (D2) considers bias arising from participant selection, including selection bias, inception bias and immortal time bias. This is explained as 'when exclusion of some eligible participants, or the initial follow up time of some participants, or some outcome events, is related to both intervention and outcome, there will be an association between interventions and outcome even if the effects of the interventions are identical. This form of selection bias is distinct from confounding.

At intervention bias domains:

Domain 3 (D3) considers bias arising from intervention classification, including misclassification bias, information bias, recall bias and observer bias. This is explained as bias 'introduced by either differential or non-differential misclassification of intervention status. Non-differential misclassification is unrelated to the outcome and will usually bias the estimated effect of intervention towards the null. Differential misclassification occurs when misclassification of intervention status is related to the outcome or the risk of the outcome and is likely to lead to bias.'

Post-Intervention Bias Domains:

Domain 4 (D4) considers bias arising from deviation from intended interventions, including time-varying confounding. This bias is explained as arising 'when there are systematic differences between experimental intervention and comparator groups in the care provided, which represent a deviation from the intended intervention(s). Depends on the type of effect of interest (assignment to intervention or adherence to intervention.'

Domain 5 (D5) considers bias arising from missing data, including attrition bias and selection bias. This is explained as bias arising 'when later follow-up is missing for individuals initially included and followed (for example differential loss to follow-up that is affected by prognostic factors); bias due to exclusion of individuals with missing information about intervention status or other variables such as confounders.'

Domain 6 (D6) considers bias arising from measurement of outcomes, including recall bias, information bias, misclassification bias, observer bias, and measurement bias. This bias is explained as 'introduced by either differential or non-differential errors in measurement of outcome data. Such bias can arise when outcome assessors are aware of intervention status, if different methods are used to assess outcomes in different intervention groups, or if measurement errors are related to intervention status or effects.'

Domain 7 (D7) considers bias arising from selection of the reported result, including outcome reporting bias and analysis reporting bias. This is explained as 'selective reporting of results in a way that depends on the findings.'

The judgement on the risk of bias was recorded as low, moderate, serious, critical or no information for each of the bias domains:

1. Low risk of bias: the study is comparable to a well-performed randomised trial
2. Moderate risk of bias: the study is sound for a non-randomised study but cannot be considered comparable to a well-performed randomised trial
3. Serious risk of bias: the study has some important problems
4. Critical risk of bias: the study is too problematic to provide any useful evidence on the effects of intervention; and
5. No information on which to base a judgement about risk of bias for this domain.

The overall risk of bias was determined by combining the outcomes from 7 domains with priority given to the most serious risk of bias. The final decision on the risk of bias was reached through discussion between the reviewers, with discrepancies in judgement being resolved by reciprocal consultation between the reviewers.

Results

Of the 38 observational studies selected for review, 12 were considered to satisfy the criteria of the research question. See Annex 2 for a list of the 38 studies including the rationale for exclusion.

The judgments on the risk of bias are presented for each of the 7 domains and overall using a rating system of low, medium, or high. It is suggested these are considered alongside the forest plots available in the full PDF of this public assessment report.

Absolute rates were not available for most of the studies, and no information on risk in a comparator group was available for about 30% of the studies to enable the calculation (see table at Annex 3).

Erythromycin

A total of 11 studies investigated the effect of erythromycin exposure in the first trimester of over 24,000 pregnancies on 1 or more outcomes, jointly covering a period between 1980 and 2019 in 8 countries. Several studies evaluated more than 1 outcome of interest.

Major congenital malformations

Of 7 studies that investigated the effect of erythromycin on major congenital malformations:

• 5 were judged to be at serious risk of bias (due to selection bias introduced by inclusion of live births only)
• 2 at moderate risk of bias (Romoren 2012 and Czeizel 1999) as these studies included stillbirths and induced abortions in their analysis
• only 1 study reported a statistically significant association (Table 1) however this study was judged to be at serious risk of bias overall (Fan and others, 2020)

While most studies are not suggestive of an increased risk of major congenital malformations with erythromycin exposure, the association or the absence of association remains uncertain and may require further investigation using a more robust study design.

Table 1. Effect estimates for studies on erythromycin and major congenital malformations (MCM). Note: number of cases from case-control study design exposed marked with * and presented, where available over the total number of exposed, while for cohort study design the number of cases is given for the exposed cohort.

Study	Design	Setting	Comparator	Erythromycin n/N	Effect estimate (95%CI)
Fan 2020	cohort	UK, 1990-2016	penicillin	53/7,987	1.5 (1.13-1.99)
Damkier 2019	cohort	Denmark, 2000-2015	penicillin	283/5,563	0.94 (0.80-1.11)
Muanda 2017b	cohort	Quebec, 1998-2009	penicillin	64/697	1.02 (0.78-1.34)
Romoren 2012	cohort	Norway, 2004-2007	unexposed	90/1,785	1.02 (0.77-1.35)
Cooper 2009	cohort	US, Tennessee, 1995-2000	unexposed	55/2,2128	0.86 (0.62-1.18)
Kallen 2014	cohort	Sweden, 1996-2011	unexposed	99/2,531	1.18 (0.96-1.44)
Czeizel 1999	case-control	Hungary 1990-1996	unexposed	113*	1.1 (0.9-1.4)

Accompanying forest plots can be found in the full PDF of this public assessment report.

Table 2. Risk of bias assessment for studies on erythromycin and major congenital malformations, Bias domains adapted from Sterne and others, 2016.

Study	D1	D2	D3	D4	D5	D6	D7	Overall
Fan 2020	low	serious	low	low	-	moderate	low	serious
Damkier 2019	serious	serious	moderate	low	low	moderate	moderate	serious
Muanda 2017b	low	serious	moderate	low	low	moderate	low	serious
Romoren 2012	moderate	moderate	moderate	low	moderate	low	low	moderate
Cooper 2009	serious	serious	low	low	moderate	moderate	moderate	serious
Kallen 2014	serious	serious	low	low	low	moderate	low	serious
Czeizel 1999	moderate	moderate	moderate	low	serious	low	moderate	serious

Key: D1 bias due to confounding; D2 bias due to selection of participants; D3 bias in classification of interventions; D4 bias due to deviations from intended interventions; D5 bias due to missing data; D6 bias in measurement of outcomes; D7 bias in selection of the reported result. '-' indicates not where bias was not assessable

Cardiovascular malformations

Of the 9 studies that investigated congenital heart defects in relation to erythromycin exposure Table 4 shows that 8 of the studies were judged to be at serious risk of bias and one at moderate risk (Romoren and others, 2012).

Of the 2 Swedish studies by Källén and others, the 2005 study reported an increased risk of cardiovascular malformations (OR 1.84, 1.29 to 2.62). In the 2014 study based on data from 2004 to 2011 (n=9), after a change in Swedish guidelines advising on using erythromycin in pregnancy, the risk of cardiovascular malformations was non-significant due to the small number of cases (n=9) (RR 1.71, 0.78 to 3.25).

The change in significance was attributed by Källén (2014) to a decline in prescription numbers. However, the risk estimate based on the period covering both studies (1996–2011) remained significant: (OR 1.70, 1.26 to 2.29).

Based on the reviewed evidence, which is generally of low quality, it is not possible to rule out a small increase in the risk of cardiovascular malformations in association with maternal use of erythromycin during the first trimester.

Table 3. Effect estimates for studies on erythromycin and cardiovascular malformations (CVM). Note: number of exposed cases from case-control study design marked with * and presented, where available over the total number of cases, while for cohort study design the number of cases is given for the exposed cohort.

Study	Design	Setting	Comparator	Erythromycin n/N	Effect estimate (95%CI)
Fan 2020	cohort	UK, 1990-2016	penicillin	19/1935	1.48 (0.92-2.37)
Damkier 2019	cohort	Denmark, 2000-2015	penicillin	46/5563	0.94 (0.69-1.28)
Muanda 2017b	cohort	Quebec, 1998-2009	penicillin	15/697	1.09 (0.64-1.86)
Kallen 2005	cohort	Sweden, 1996-2003	unexposed	34/1844	1.84 (1.29-2.62)
Kallen 2014	cohort	Sweden, 1996-2011	unexposed	43/2531	1.7 (1.26-2.29)
Romoren 2012	cohort	Norway, 2004-2007	unexposed	21/1785	1.16 (0.75-1.78)
Crider 2009	case-control	US, 1997-2003	unexposed	81*	1.0 (0.7-1.3)
Lin 2013	case-control	US, Canada, 1994-2008	unexposed	18/4132	1.3 (0.6-2.6)
Cooper 2009	cohort	US, Tennessee, 1995-2000	unexposed	9/2128	0.93 (0.45-1.91)

Accompanying forest plots can be found in the full PDF of this public assessment report.

Table 4. Risk of bias assessment for studies on erythromycin and cardiovascular malformations. Bias domains adapted from Sterne and others, 2016.

Study	D1	D2	D3	D4	D5	D6	D7	Overall
Fan 2020	low	serious	low	low	-	moderate	low	serious
Damkier 2019	moderate	serious	moderate	low	low	moderate	moderate	serious
Muanda 2017b	low	serious	moderate	low	low	moderate	low	serious
Kallen 2005	serious	serious	serious	low	low	moderate	low	serious
Kallen 2014	serious	serious	moderate	low	low	moderate	low	serious
Romoren 2012	moderate	moderate	moderate	low	moderate	low	low	moderate
Crider 2009	moderate	serious	serious	low	moderate	low	moderate	serious
Lin 2013	moderate	serious	serious	low	moderate	low	moderate	serious
Cooper 2009	serious	serious	low	low	moderate	moderate	moderate	serious

Key: D1 bias due to confounding; D2 bias due to selection of participants; D3 bias in classification of interventions; D4 bias due to deviations from intended interventions; D5 bias due to missing data; D6 bias in measurement of outcomes; D7 bias in selection of the reported result. '-' indicates not where bias was not assessable

Miscarriage

A total of 2 studies evaluated the effect of erythromycin prescription on miscarriage (Muanda and others, 2017a; Andersen and others, 2013). Neither study reported an association (table 5). Muanda's study was judged to be at moderate risk of bias and Andersen's at serious risk of bias.

There was an agreement of no association between erythromycin and the risk of miscarriage in 2 studies available for review. However, it was not possible to conclude an absence of risk due to the small number of publications and their poor quality.

Table 5. Effect estimates for studies on erythromycin and miscarriage. Note: number of exposed cases from case-control study design marked with * and presented, where available over the total number of cases, while for cohort study design the number of cases is given for the exposed cohort.

Study	Design	Setting	Comparator	Erythromycin n/N	Effect estimate (95%CI)
Andersen 2013	cohort	Denmark, 1997-2007	unexposed	n/a	1.03 (0.94-1.13)
Muanda 2017a	case-control	Quebec, 1998-2009	penicillin	15/697	0.82 (0.56-1.19)

Accompanying forest plots can be found in the full PDF of this public assessment report.

Table 6. Risk of bias assessment for studies on erythromycin and miscarriage. Bias domains adapted from Sterne and others, 2016.

Study	D1	D2	D3	D4	D5	D6	D7	Overall
Andersen 2013	serious	moderate	moderate	low	moderate	low	moderate	serious
Muanda 2017a	low	low	low	low	low	low	moderate	moderate

Key: D1 bias due to confounding; D2 bias due to selection of participants; D3 bias in classification of interventions; D4 bias due to deviations from intended interventions; D5 bias due to missing data; D6 bias in measurement of outcomes; D7 bias in selection of the reported result.

Clarithromycin

For clarithromycin, 4 studies investigated the effect of clarithromycin exposure in the first trimester of over 1,400 pregnancies on one or more outcomes, jointly covering a period between 1990 and 2016 in 3 countries. 3 studies evaluated more than one outcome of interes...