Abstract
BACKGROUND: Withdrawal of life-sustaining treatment (WLST) is common in clinical trials of patients with acute brain injuries (ABI), but current reporting practices and impact on trial-reported findings are unclear. We evaluated reporting practices of WLST in contemporary clinical trials of patients with ABI and quantified the magnitude of bias on treatment effect estimates in hypothetical trials. METHODS: We conducted a literature review of contemporary ABI randomized clinical trials and a simulation-based analysis. In the literature review, we included two-arm, randomized, superiority trials of adults with ABI (traumatic brain injury, intracranial hemorrhage, subarachnoid hemorrhage, ischemic stroke, or post-cardiac arrest brain injury) published in 10 high-impact journals from January 1, 2015 to December 19, 2024. We extracted WLST characteristics including frequency, timing, reasons, and neuro-prognostication criteria. In the simulation-based analysis, we evaluated the impact of WLST misclassification-defined as WLST occurring in patients who could have survived with a good neurological outcome-on observed treatment effects. For each scenario, we estimated the observed treatment effect after misclassification and calculated bias as the difference between observed and true treatment effects. We assessed both blinded and unblinded trials and binary and ordinal neurologic outcomes. RESULTS: Among 69 trials included in the literature review, 17 trials (24.6%) reported WLST frequency, 9 (13.0%) timing, 10 (14.5%) reasons, and 7 (10.1%) standardized neuro-prognostication criteria. In simulations of blinded trials, WLST misclassification consistently attenuated observed treatment effects. Increasing the fraction of misclassified WLST events led to progressively greater bias, making beneficial treatments appear less effective and harmful treatments appear less harmful. In unblinded trial simulations, the direction of bias varied by the magnitude of the true treatment effect and degree of misclassification. Findings were similar for binary and ordinal neurologic outcomes. Across all simulations, WLST misclassification reversed statistical conclusions in a median of 22.1% (interquartile range 17.4-32.4%) of trials. CONCLUSIONS: WLST is poorly reported in contemporary ABI trials. Misclassification of WLST-related deaths leads to important bias in trial-reported treatment effects, potentially yielding underpowered studies and erroneous trial conclusions. Standardized, transparent WLST reporting is essential to strengthen ABI trial design and interpretation.