Neuromuscular blockade (NMB) is commonly used in patients with acute respiratory failure [1,2], particularly in patients with acute respiratory distress syndrome (ARDS) [3], a disease that carries a profound morbidity and mortality [4] and became a leading cause of death worldwide during the COVID pandemic [5]. NMB has been suggested to provide benefit through several mechanisms. Removal of spontaneous breathing activity from the patient allows full control during mechanical ventilation, which could reduce the risk for self-inflicted lung injury [6,7], while removal of dyssynchrony and expiratory muscle activity may reduce lung derecruitment and overdistension [8,9]. Additionally, NMB use has been associated with improved mechanics, decreased inflammatory markers, decreased oxygen consumption, and decreased hemodynamic complications [8,10,11]. Conversely, the use of NMB may result in direct and indirect harm. NMB may cause skeletal and respiratory muscle injury and weakness [12], it may lead to further atelectasis and derecruitment [13], and it may lead to higher sedation requirements and prolonged duration of mechanical ventilation [14], which have independently been found to worsen patient outcomes.

Several randomized trials have investigated the potential risks and benefits of NMB with mixed results. Two small single center trials suggested NMB initiation improved mortality, inflammatory biomarkers, mechanics, and oxygenation [10,15]. The larger subsequent single center ACURASYS trial found improved 90-day mortality in patients with moderate-severe ARDS (−9.1%, 0.95 CI = [−18.9%, 0.7%]) [16]. In the large follow up multi-center ROSE trial, this mortality benefit was no longer found, with similar outcomes between the NMB and control groups (−0.3%, 0.95 CI = [−6.4%, 5.9%]) [17]. Patients were also less active and had more adverse cardiovascular events in the NMB arm, providing some suggestive evidence that the net effects of NMB initiation may be harmful in the absence of mortality benefits.

It is unclear how these mixed findings should be incorporated into practice. The wide confidence intervals from both large studies overlap considerably, and both are consistent with null effects. The control arm of the ACURASYS trial entailed never initiating NMB within 48 h after baseline, and the NMB treatment arm protocols of both studies called for early initiation of NMB immediately after randomization. However, under usual care, clinicians may be unlikely to either immediately initiate NMB or avoid initiation at all costs, but rather monitor their patients and make the decision to initiate based on clinician-specific aspects of evolving time-varying patient state (for example, dyssynchrony level, presence of spontaneous breathing, or severity of illness). Such a monitoring clinician might interpret the (highly uncertain) mortality results from the ACURASYS trial as reflecting harms of never initiating NMB as much as benefits of always initiating. While the ROSE trial did provide “usual care” in its control arm (modified to include light sedation targets), its confidence interval included a broad range of plausible effect sizes (beneficial or harmful), and it only provided weak evidence against universal immediate initiation. It did not provide any direct evidence that strategies more conservative than usual care would be beneficial.

The question remains whether, on average, clinicians should be more aggressive in NMB initiation decisions than they currently are (even if perhaps not as aggressive as in the experimental arms of the ROSE and ACURASYS trials), or whether they should on average be more conservative (even if perhaps not as conservative as in the control arm of the ACURASYS trial). In this paper, we address these questions by applying a recently developed method for estimating effects of so-called incremental interventions [18,19]. This approach has been applied to clinical care in several studies [20,21], including our own application to mechanically ventilated patients in the ICU [22]. An incremental intervention multiplies the odds of initiating NMB in each patient at each time relative to usual care by some factor. An incremental intervention by a factor > 1 shifts usual care to be more aggressive, and an incremental intervention by a factor < 1 shifts usual care to be more conservative. Thus, incremental effect estimates address our questions of interest. A shift by a factor of infinity corresponds to the experimental arms of the ROSE and ACURASYS trials, where everyone initiates NMB, and a shift by a factor of 0 corresponds to the control arm of ACURASYS, where no one initiates NMB. We consider shifts by factors between 0.2 and 5, which reflect more realistic alterations to usual care, in which clinicians continue to exercise their judgement.

While incremental interventions are not directly actionable in that they do not indicate exactly when to initiate NMB, incremental effect estimates can be highly suggestive about how care should change and inform the design of future randomized trials whose results would be actionable. If incremental interventions making NMB initiation less likely than under usual care (given patient history) are estimated to be beneficial, this implies that on average physicians would do better to be more hesitant to initiate. Such findings would also imply that further randomized trials testing specific initiation strategies might identify initiation rules that are both beneficial and directly actionable. If usual care is found to be near optimal in the class of incremental interventions, this suggests that improvements in care will need to come from treatment rules encouraging treatment in particular situations as opposed to broad encouragement or discouragement to treat.