LAUSR

In their article in this issue, Droitcourt et al. investigate the risk of some systemic infections leading to hospitalization in children with atopic dermatitis (AD) compared with sexand age-matched comparators in Denmark. Studied outcomes were upper and lower respiratory tract, gastrointestinal tract, musculoskeletal tract, urinary tract, heart and central nervous system infections and sepsis. Infections treated in hospitals were regarded as serious infections, and AD was defined as at least one recorded diagnosis of AD, but an algorithm may have captured more children with AD. The authors performed a cohort study using the nationwide Danish population and health registers applying survival analysis, comparing time from study inclusion to first infection between children with AD and comparators. The authors conclude that children with AD have an increased risk of systemic infections requiring hospitalization. This work is of importance to patients and physicians to raise awareness of a possible increased risk of serious infections. The authors studied incident infections, however, there is no information on exclusion of those with prior infections, and recurrent infections would reflect the real world to a larger extent. The authors maybe censored children at the first infection during follow-up, which makes sense statistically. On the other hand, children with many prior infections might be at higher risk for a new infection. The adjusted and the fully adjusted model for lower respiratory infections yields nonoverlapping confidence intervals, which is unusual. Maybe asthma/hay fever is an intermediator on the pathway between AD and lower respiratory infection rather than a confounder. From the baseline table we see that asthma/hay fever shows an imbalance between children with AD and the comparators, 5% vs. 14% at baseline and 16% vs. 49% at the end of the study, a fact that could explain the difference between the models for lower respiratory infections. Many analyses were performed, which causes the multiple testing problem, i.e. yielding a higher risk of a false significant result than the targeted 5%. The authors used Bonferroni correction to adjust for this, i.e. lowering the significance level by dividing it by the number of tests. When highlighting statistically significant results rather than clinically relevant results, there is a need to discuss the risk of false positives. For example, if 10 tests are performed, the risk of at least one false positive is about 40%. The results might be affected by channelling, reverse causation and surveillance bias. In ‘children treated with immunosuppressants have a thorough assessment that focuses, among other things, on infection risk before they are prescribed, and they also undergo repeated monitoring. These efforts likely minimize the risk in children with AD who are immunosuppressed, and even prevent their use in those who have higher baseline risk of severe infection’, the authors describe channelling, i.e. informed selection of patients receiving a treatment. Selected patients might have lower baseline AD severity, affecting the risk of infections later, and surveillance bias from the repeated monitoring. Moreover, ‘Infections can trigger worsening of atopic dermatitis (AD)’ describes reverse causation, i.e. the study outcome is the cause of the exposure.