Credible subgroup effects in randomized clinical trials

For nearly a decade following the promising results from the ACURASYS trial, neuromuscular blockers were on the frustratingly short list of beneficial therapies for Acute Respiratory Distress Syndrome (ARDS). That is – until the Revaluation of Systemic Early Neuromuscular Blockade (ROSE) trial showed a mere 0.3% absolute decline in 90-daymortality. Alas, the critical care community is again navigating a complex evidence base for yet another ARDS therapy.

Here, Dr. Zalucky and colleagues re-analyzed data from the ROSE trial to explore whether the benefit of neuromuscular blockade varied with baseline respiratory system elastance. The physiological rationale is that higher elastance (i.e., stiffer lungs and chest wall) increases driving pressures, leading to greater risk of hyperinflation and de-recruitment during spontaneous breathing. But instead of the traditional frequentist method of analyzing potentially promising subgroups and possible treatment interactions, the investigators chose to employ a Bayesian statistical approach. To understand why, let us first examine the general trouble with subgroups.

National Heart, Lung, and Blood Institute

Harm-Jan de Grooth, Nadir Yehya, Michael O Harhay