In this nationwide cohort study, we found that the use of macrolides in pregnancy was not associated with an increased risk of major birth defects. We included all live birth pregnancies in Denmark where pregnant women had used macrolides during the study period (1997-2016) and compared them with women who had used penicillin and with other comparative groups where antibiotics were not used during pregnancy. Also, our data did not provide evidence of an increased risk of any of the 12 organ specific subgroups of birth defects in women who used macrolides during pregnancy, including defects of the heart.
Interpretation and comparison with previous studies
This study was prompted by the findings of a recent cohort study which suggested an increased risk of any major birth defect associated with the use of macrolides compared with the use of penicillin in pregnancy (risk ratio 1.55; 95% confidence interval 1.19 to 2.03), and specifically cardiovascular defects (1.62; 1.05 to 2.51).7 The study used a clinical database of 6.9% of the population of the United Kingdom and included 2170 live birth pregnancies where macrolides were used in the first trimester. Our analyses included 13 017 pregnancies during which macrolides had been used matched with pregnancies during which penicillin had been used in the first trimester, derived from a nationwide cohort with individual level data obtained from routine clinical care. Our findings did not confirm these previous results and we found no evidence of significant differences in the risk of major birth defects.
Also, we found no associated risk for any of the 12 specific subgroups of birth defects with the use of macrolides in pregnancy. Based on the upper limit of the 95% confidence intervals from these analyses, a more than twofold relative increase in the risk for nine of the 12 categories can likely be excluded. Also, the results were inconsistent with a more than 26% relative difference in the risk of heart defects between the use of macrolides and penicillin in pregnancy. Our analyses on the risk of individual categories of heart defects did not show any significant associations. The overall prevalence of heart defects in our study, and in pregnancies where macrolides were used in the first trimester in the UK study,7 were in line with the reported prevalence of about 0.9-1.0% in similar cohorts from the Nordic countries and the US.10202122 Differences in the representative samples (probability sampling v nationwide cohort) and other methodological differences between our study and the UK study, such as data collection, might contribute to the different findings. For example, information on birth defects was collected from general practitioner records at the age of three in the UK study (where diagnoses are manually entered into the patient record23) whereas we obtained data on diagnosed birth defects from inpatient and outpatient hospital care. In Denmark, all diagnoses are automatically recorded in the registries.
Our findings are in line with a recent meta-analysis that included 6082 women who used macrolides during pregnancy from four previously published studies and found no increased risk of birth defects overall (odds ratio 1.03; 95% confidence interval 0.86 to 1.22).8 The meta-analysis did not find any associations with specific subgroups of birth defects, except for gastrointestinal defects (odds ratio 1.56, 95% confidence interval 1.05 to 2.32). This estimate was based on data from one observational cohort study, however, which included 2332 women who used macrolides during pregnancy (most of the results of the meta-analyses on the risk of specific subgroups of birth defects were driven by this one study).11 Our study had a much larger cohort of pregnant women who used macrolides than all of these reports combined, thus expanding on previous results by providing estimates with a high precision.
Our data might not be conclusive for defects that occur rarely, however, and some of our subgroup analyses had limited power. But even if the use of macrolides in pregnancy was associated with a rare defect, given the upper limit of the 95% confidence interval from the analysis of any major birth defects, our findings are inconsistent with an increase in the relative risk of more than 8% in pregnant women who used macrolides compared with those who used penicillin. This finding in absolute terms means that, with a reported rate of defects of 37.0 per 1000 pregnancies where penicillin was used, an excess of defects of more than 2.7 per 1000 pregnancies where macrolides were used can likely be ruled out.
Most studies of the associated risk of birth defects with the use of macrolides investigated pregnant women who used erythromycin. Results based on Swedish registry data suggested an association between the use of erythromycin in pregnancy and major birth defects of the heart,2124 but this finding has not been replicated in other populations.81025262728 The results of our study add to previous data as we investigated the associated risk of major birth defects overall, and also of specific subgroups of birth defects, with the use of azithromycin, clarithromycin, erythromycin, and roxithromycin individually. Also, because of the large number of pregnant women who used macrolides in our cohort, we were able to study the risk of individual categories of heart defects. Lastly, our sensitivity analyses, restricted to women who used antibiotics in gestational weeks 3 to 8, and including defects in induced abortions, did not change the results. To our knowledge, these results have not been previously reported.
We believe that our data provide reassurance about the risk of major birth defects when treatment with macrolide antibiotics is needed during pregnancy. Our findings could help inform clinicians, patients, and drug regulatory authorities.
Strengths and limitations
Our study cohort was based on all pregnancies in Denmark during the 20 year study period, with individual level data obtained from various nationwide registries, which increased the generalisability of the study findings, and the risk of selection and information bias and loss to follow-up was minimised. In Denmark, all healthcare services are free. The analyses were based on pregnancies resulting in live births only (except for the sensitivity analysis which included defects in induced abortions). Registrations of major birth defects in the National Patient Registry have a high validity, with positive predictive values of 88% overall and 90% for heart defects.2930
Data on the specific indication for a filled prescription and adherence are not available from Danish registries. Our definition of use of antibiotics implied that a filled prescription was equivalent to the use of the drug. Thus we cannot exclude differences in adherence in the active comparison analyses, and low adherence in women who used macrolides would bias the results towards the null for all analyses. Although a potential source of confounding for observational studies of drug safety are when the underlying indication being treated increases the risk of an outcome rather than the treatment itself (that is, confounding by indication), we found no significant increased risks in any of our analyses, including compared with other groups who did not use antibiotics during pregnancy. Hence we believe it is unlikely that the indications confounded the associations. To minimise the risk of confounding by indication, however, we used an active comparative design for the main analysis by comparison with the use of phenoxymethylpenicillin during pregnancy, an antibiotic with a similar spectrum of use.
Although a wide range of potential covariates was controlled for in the propensity score, we lacked information on some potentially important confounders from a clinical point of view (such as alcohol consumption, fever, and folic acid supplements) and therefore unmeasured and residual confounding cannot be fully excluded, particularly if unadjusted factors were differently distributed between comparative groups and not controlled for through proxies included in the propensity score. Given the sample size of our study, inherent unadjusted factors masking a true association would have to be common or strongly associated with the use of macrolides and inversely associated with major birth defects among live birth pregnancies. Comparison with other pregnancy groups where antibiotics were not used did not show any significant associations. In summary, we believe the presence of unmeasured confounding factors was limited or unlikely.