Related Experiment Videos
A Meta-epidemiologic Analysis of Differences in Treatment-Effect Estimates According to Outcome Scale (GOS vs. GOSE)
Nicholas Giulio Raccagni1, Marco Dotti2, Yasmin Arabella Lightfoot2
1School of Medicine and Surgery, Università degli Studi di Milano-Bicocca, Piazza dell'Ateneo Nuovo, 1, 20126, Milan, Italy. n.raccagni@campus.unimib.it.
Background:
Traumatic brain injury (TBI) trials commonly assess global functional outcome using the Glasgow Outcome Scale (GOS) or Extended Glasgow Outcome Scale (GOSE). Because GOS- and GOSE-based trials are often pooled or compared in evidence synthesis, whether treatment-effect estimates differ systematically by outcome scale remains unresolved.
Objective:
We aim to estimate the between-trial contrast in treatment-effect estimates comparing GOSE-based with GOS-based randomized controlled trials (RCTs) of moderate-to-severe TBI.
Methods:
We conducted a meta-epidemiologic analysis of RCTs comparing acute-phase interventions with a concurrent control in adults with moderate-to-severe TBI and reporting 6-month global functional outcome using GOS or GOSE. Random-effects meta-regression estimated the ratio of odds ratios (ROR) contrasting GOSE- with GOS-based trials, adjusted for publication year, multicenter status, income setting, intervention class, endpoint role, and sample size. Sensitivity, structural, and small-study-effect analyses were prespecified.
Results:
Eighty-four RCTs (> 20,000 participants) met eligibility criteria. Dichotomous outcomes were available for 78 trials (53 GOS and 26 GOSE observations). The adjusted outcome-scale contrast was ROR 0.880 [95% confidence interval (CI) 0.679-1.140; I2 = 33.6%; p = 0.326], with larger pooled effects in GOS-based (OR 1.47, 95% CI 1.27-1.69) than GOSE-based trials (OR 1.07, 95% CI 0.90-1.27). The contrast attenuated from an unadjusted ROR of 0.718 (p = 0.004) and remained directionally consistent across sensitivity, ordinal-subset, and structural analyses (ROR range 0.669-0.886). Risk of bias was lower in GOSE-based trials across domains, and a within-trial negative-control analysis comparing binary and ordinal estimators was near null.
Conclusions:
GOS-based TBI trials yielded larger treatment-effect estimates than GOSE-based trials. Although directionally consistent across prespecified analyses, this pattern remained compatible with residual between-trial confounding and should not be interpreted causally. GOS- and GOSE-based treatment effects should not be assumed fully interchangeable in evidence synthesis. Meta-analyses spanning both scales should consider scale-stratified sensitivity analyses, and RCTs should justify and standardize outcome-scale ascertainment.
