Written by Seth Walsh-Blackmore
Spoon Feed
Advanced statistics emulating clinical trial protocols from an ICU database did not identify consistent differences in 30-day mortality when ICU patients were intubated early (within 8 hours of admission) or late (after 48 hours).
Don’t hesitate to wait to intubate.
ICU intubation carries intrinsic risk; therefore, evaluating strategies to reduce the frequency and duration of intubation safely is an important clinical question. However, the general circumstances indicating intubation complicate an ethical RCT.
The authors sought to reduce bias intrinsic to observational data by statistically modeling three sets of eligibility criteria and treatment protocols as in a clinical trial (table below) from the MIMIC-IV ICU database.
For the first two trials, 5,893 patients met the inclusion criteria, and 1,281 patients met the criteria for the third. Intubation occurred in 12.5% (737) of patients eligible for trials 1 and 2 and 20.5% (263) of trial 3 (figure below). All patients had complete follow-up for 30-day mortality.
How was a trial protocol emulated with these data? Augmented inverse probability weighting (AIPW) was used. In AIPW. a propensity matching model (inverse probability weighting) is combined (augmented) with an outcome model, classically a regression. In theory, only one model requires correctly specified parameters to estimate treatment effect accurately.
In this case, the probability of following a treatment strategy and risk of death had one followed said strategy were modeled individually. An adjustment term restricted to those following the strategy of interest was computed from the probability model then applied to the outcome model to estimate its mortality risk with results below.
How will this change my practice?
The study provides great detail of its statistical model at the mathematical level but does not present most of the clinical details/definitions that would be standard in a trial. Thus, I’ll consider this hypothesis-generating. However, it affirms my practice to delay intubation with non invasive supportive measures whenever possible.
Source
Emulating Target Trials Comparing Early and Delayed Intubation Strategies. Chest. 2023 Oct;164(4):885-891. doi: 10.1016/j.chest.2023.04.048. Epub 2023 May 5.
