To the Editor We read with great interest the study by Weisz et al1 that evaluated the association of patent ductus arteriosus (PDA) ligation with death or neurodevelopmental impairment in preterm infants born prior to 28 weeks’ gestation.1 The authors were correctly aware that their estimate of the effect of PDA ligation on mortality (adjusted odds ratio, 0.09; 95% CI, 0.04-0.21) was likely biased by survivor treatment selection (immortal time bias).2 Immortal time bias results from misclassification of the follow-up period when the outcome (mortality) cannot occur in exposed (ligated) patients, who must by default survive until ligation. We are concerned the method used to account for it, a subcohort analysis, was undersensitive. As we have shown, day-specific mortality in infants at risk for PDA decreases rapidly in the first postnatal weeks.3 Given the authors’ uniquely rich data set and ability to control for confounding by indication, we suggest applying time-dependent Cox regression, a more advanced and validated method4 that considers treatment as a time-dependent exposure. Using a postnatal day time-scale, infants with hemodynamically significant PDA would enter the life table as treated on the day they received their first nonsteroidal anti-inflammatory drug (NSAID) dose. They would continue accruing person-time as medically treated until they died, were discharged, were lost to follow-up, or received surgical ligation. The day that an infant received ligation, they would be withdrawn from the NSAID group and begin to accrue person-time as a ligated patient until death, discharge, or loss to follow-up. Infants ligated without prior NSAID treatment would only accrue person-time in the ligated group, beginning on the ligation day.
Slaughter JL, Klebanoff MA. Neurodevelopmental Outcome in Relation to Treatment of Patent Ductus Arteriosus. JAMA Pediatr. 2017;171(10):1017–1018. doi:10.1001/jamapediatrics.2017.2597
Coronavirus Resource Center
Customize your JAMA Network experience by selecting one or more topics from the list below.