Drug treatment for a preterm heart-duct problem was linked to higher death risk

The ductus arteriosus is a blood vessel every baby is born with. In the womb it shunts blood past the not-yet-working lungs; in a term baby it closes within days. In very preterm infants it often stays open — a patent ductus arteriosus (PDA) — and for decades neonatologists have tried to close it with drugs such as ibuprofen, indomethacin or paracetamol, on the reasoning that the extra blood flow strains the heart and lungs. A Bayesian meta-analysis published in The Lancet Child & Adolescent Health on October 30 asks a blunter question: does treating it help the babies, or harm them? [s1]

Why the question is open

A run of recent randomised trials and meta-analyses has already undercut the case for early PDA treatment, raising the possibility that closing the duct increases the risk of death and of bronchopulmonary dysplasia — the chronic lung injury of prematurity — rather than preventing them [s1]. One large randomised comparison of expectant management against medication reported no benefit from treating [s3]. What remained unclear was whether the risk differs by how premature the baby is [s1].

What the analysis did

The authors searched PubMed, Embase and the Cochrane CENTRAL register for randomised controlled trials that enrolled infants under 32 weeks' gestation with an echocardiographically confirmed PDA and assigned them either to pharmacological treatment or to placebo or expectant management [s1]. The pre-specified primary outcome was all-cause mortality before 36 weeks' postmenstrual age; secondary outcomes were moderate-to-severe bronchopulmonary dysplasia and the composite of death or that lung disease [s1]. The protocol was registered on PROSPERO (CRD42024622527) [s1].

Rather than a conventional pooled estimate, the team used a Bayesian model-averaged approach. That lets them report not just a risk ratio but the posterior probability of harm — the probability that treatment increases mortality — and a Bayes factor weighing the evidence for harm against the evidence for benefit [s1]. A risk ratio above 1 indicates harm.

Ten trials including 2,365 infants (1,133 female and 1,232 male) were analysed; 1,045 had a gestational age below 26 weeks [s1].

What it found

Pharmacological treatment was associated with higher mortality in every stratum. In infants under 26 weeks the risk ratio was 1.12 (95% credible interval 0.90-1.41), with an 85% posterior probability of harm and a Bayes factor of 5.48 [s1]. In those at 26 weeks or more it was 1.22 (0.90-1.77), probability of harm 89%, Bayes factor 7.82 [s1]. Across the whole population the risk ratio was 1.23 (95% credible interval 1.00-1.54), with a 97% probability of harm and a Bayes factor of 36.04 — strong evidence favouring harm over benefit [s1].

Crucially, there was no sign of a trade-off. For bronchopulmonary dysplasia and for the death-or-lung-disease composite, analyses that explicitly allowed for the possibility of no effect found the data more consistent with no effect than with either benefit or harm [s1]. In other words, the higher mortality was not bought in exchange for healthier lungs in the survivors.

How to read the credible intervals

Two of the three mortality intervals cross 1, so this is not proof that treatment kills. But the Bayesian framing is the point. A conventional "not significant" would stop the conversation; reporting a 97% posterior probability of harm overall, with no compensating lung benefit, reframes the burden of proof. The authors conclude that the available randomised evidence is consistent with an increased mortality risk from pharmacological PDA treatment, above and below the 26-week threshold, and that it points towards more selective, evidence-based use rather than routine treatment [s1].

A linked commentary in the same issue goes further, framing the pragmatic PDA treatment trials of the past decade as the end of an era in which the open duct was treated reflexively [s2].

The limits

This is a meta-analysis, not a new trial, and it inherits the heterogeneity of ten studies that used different drugs, doses, timing and comparators — some against placebo, some against watchful waiting [s1]. A credible interval that just touches 1.00 is not a mortality verdict. And an analysis pooling trials cannot tell an individual clinician which specific baby, with a large duct and failing lungs, might still benefit.

What to watch

Whether neonatal guideline bodies move from "treat the duct" toward reserving drugs for selected infants; whether the next generation of trials tests targeted treatment of hemodynamically significant PDAs rather than the duct's mere presence; and whether the mortality signal holds as individual-patient-data analyses mature.

This article describes a meta-analysis of existing trials and is informational only. It is not medical advice, and nothing here should be used to start, stop or change any treatment in a newborn.