Strengths and limitations
The major strengths of our study include the population based data, large sample size, detailed and standardised data on CA, and inclusion of TOPFA, which constitute a large proportion of some CAs.55 Additionally, exposures were mainly prospectively ascertained, and our case-malformed control design limits recall and information bias, especially for the small proportion of retrospectively ascertained exposures. Our study was hypothesis driven, and the use of two control and exposure comparison groups together with sensitivity analysis, allowed us to evaluate the robustness of any associations.
Our study also had some limitations. Exposure to antibiotics (2.36%) was low, compared with the expected 3 - 14% rate of first trimester antibiotic exposure in the European population.10,19,56 This suggests antibiotic exposures were under-ascertained in our data, as was shown also in a study comparing the registry data to linked prescription data.57 The reporting of antibiotic exposures would not, however, have been different between cases and controls and thus should not lead to biased odds ratios. We had very little data about the indication for prescribing, or about untreated infection, so we could not examine confounding by indication in this way. Our sensitivity analysis comparing macrolide use with penicillins obtained similar results to that with the primary exposure comparison. Since macrolides and penicillins are commonly used for the same indication, this suggests that the excess risk we found relates to macrolides rather than the underlying infection. Other studies have used this same approach as a proxy to account for residual confounding by indication.12,19,25 We examined evidence of teratogen non-specificity bias and found our results to be robust; they were similar after excluding two control subgroups which were associated with antibiotics, and similar for the secondary comparison with genetic syndromes which cannot be associated with first trimester medication exposure. Finally, we conducted many tests with some significant findings possible by chance, and, therefore, specified our hypotheses in advance, and recommend confirmation of new (exploratory) findings in independent datasets.