Commentary

Confounding and causality

This report from the NICHD Neonatal Research Network addresses the previously underexplored question of whether receiving a prolonged course of empirical antibiotics increases the risk of NEC or death for ELBW infants. This association is biologically plausible. Evidence exists that selection of a pathogenic intestinal microflora by antibiotics may be a key step in the development of NEC.

The investigators retrospectively categorised ELBW infants in 19 neonatal units who survived beyond the early neonatal period as having been exposed to ≥5 days of empirical antibiotics (an arbitrarily selected cutoff that coincided with the median duration of antibiotic exposure in the cohort). Primary analyses revealed that infants who had longer durations of antibiotic exposure were more likely to develop NEC or die. The report estimates an increase in odds of 4% per day beyond 5 days after birth. This translates as a 26% versus 19% absolute risk. For every 14 infants who had prolonged antibiotics, one extra infant will die or develop NEC.

Those infants who received the prolonged course of antibiotics were also smaller, less mature and (probably) sicker at birth. How then can we know that the relationship of duration of antibiotic usage to NEC and death is causal? The investigators used multivariate analysis to statistically account for the potentially confounding effects of factors such as gestational age and birth weight. Following this adjustment, the association with NEC and death remained.

However, there are problems with this approach. First, the investigators could not adjust their model for some known risk factors for NEC, particularly different feeding strategies and use of breast milk versus formula. Second, only known confounders can be accounted for; there may be other unsuspected factors that affect the risk of developing NEC that also differ between infants with prolonged versus shorter duration of antibiotics. Third, stratifying analyses for confounding variables is a very blunt tool. Differences in the way data are defined and categorised may have resulted in residual confounding.

Another concern with the interpretation of the data is that the distribution of duration of antibiotic exposure in the cohort was bimodal; it appears that one group of infants had about 3–4 days' exposure of antibiotics with a second peak at 7–8 days after birth, suggesting that there are two overlapping populations. The second may reflect infants with new or emerging clinical concerns or signs of illness including suspected late-onset sepsis. These infants may, therefore, represent a group at higher risk of subsequent NEC or death.

Implications for practice and research

The hypothesis is intriguing and relevant to clinical practice. Despite the methodological limitations of the study design, the findings are a reminder of the potential for unintended consequences of what are often considered benign actions. It reinforces the maxim that administrationof antibiotics for preterm infants should start early (when sepsis is suspected) and stop early (when sepsis has been excluded). However, although this association is hypothesis generating, it does not prove causality. Furthermore, a trial of brief versus prolonged antibiotic courses is unlikely to be considered a research priority given that other more plausible interventions—for example, related to feeding strategies, remain to be assessed in pragmatic trials.