A new analysis of traumatic brain injury trials is challenging a common assumption in medical evidence: that results measured with the Glasgow Outcome Scale and its expanded version can be combined as though they were interchangeable. The study, published in Neurocritical Care, examined whether the choice between the Glasgow Outcome Scale (GOS) and Extended Glasgow Outcome Scale (GOSE) is associated with systematically different estimates of treatment benefit. Researchers found that trials using the traditional GOS generally reported larger apparent treatment effects than trials using the GOSE. The finding does not prove that one scale causes stronger or weaker treatment effects, but it raises an important warning for researchers, clinicians and patients who rely on pooled evidence to judge therapies for moderate-to-severe traumatic brain injury.
The investigation was conducted as a meta-epidemiologic analysis, a form of research that studies how features of clinical trials influence their results. Instead of asking whether a particular drug, procedure or intensive-care strategy works, the researchers asked whether the way researchers measured recovery might be linked to the size of the reported effect. They identified randomized controlled trials involving adults with moderate-to-severe traumatic brain injury who received an acute-phase intervention or concurrent control treatment and were assessed at six months using either GOS or GOSE. In total, 84 trials involving more than 20,000 participants met the eligibility criteria. Dichotomous outcome data—typically classifying participants as having a favorable or unfavorable functional result—were available from 78 trials, including 53 GOS observations and 26 GOSE observations.
The GOS is a five-level scale that summarizes global recovery after brain injury, ranging from death to good recovery. The GOSE expands the instrument to eight levels by dividing several GOS categories into more detailed subdivisions. For example, “moderate disability” and “good recovery” can each be separated into upper and lower levels. This added resolution is designed to detect clinically meaningful differences that may be invisible on the shorter scale. However, increasing the number of categories also introduces additional requirements. Assessors must distinguish more finely between levels of independence, social participation and functional limitation, while trial teams must use consistent interviews and scoring procedures. Small differences in how outcomes are elicited or classified may therefore influence the final statistical estimate.
The central comparison was expressed as a ratio of odds ratios, or ROR. An odds ratio above one generally indicates greater odds of a favorable outcome with treatment, depending on how the endpoint is defined. The ROR compares the treatment effect reported in one trial group with the effect reported in another. In this analysis, an ROR below one indicated that GOSE-based trials tended to produce smaller treatment-effect estimates than GOS-based trials. Before accounting for differences between studies, the ROR was 0.718, a result that suggested a statistically significant contrast. After adjustment for publication year, multicenter status, income setting, intervention class, endpoint role and sample size, the ROR moved toward the null value of one, reaching 0.880 with a 95% confidence interval from 0.679 to 1.140. The adjusted association was not statistically significant, with a p value of 0.326.
The pooled results nevertheless showed a clear numerical pattern. GOS-based trials produced an overall odds ratio of 1.47, with a 95% confidence interval from 1.27 to 1.69. GOSE-based trials produced an odds ratio of 1.07, with a 95% confidence interval from 0.90 to 1.27. In practical terms, the conventional-scale studies appeared to report substantially more favorable treatment effects, while the extended-scale studies were, on average, close to showing no clear benefit. Because the confidence interval for the adjusted ROR included one, the investigators emphasized that the difference could reflect characteristics of the trials rather than a direct causal effect of the outcome scale. The outcome instrument may be part of a larger network of factors involving study era, patient selection, intervention type, analytical choices and research quality.
The researchers performed several analyses to test whether the pattern was fragile. Across sensitivity, ordinal-subset and structural analyses, the ROR remained directionally below one, ranging from 0.669 to 0.886. This consistency suggests that the observed contrast was not confined to a single statistical specification or a narrow group of studies. At the same time, the variation across analyses illustrates why a meta-epidemiologic association must be interpreted cautiously. When different trials use different outcome measures, the scale is not randomly assigned. Investigators may choose GOSE because a study is newer, larger, more methodologically rigorous or designed around modern functional-outcome standards. These characteristics can overlap, making it difficult to isolate the independent influence of the scale itself.
One notable finding was that GOSE-based trials had lower risk of bias across several assessed domains. That observation complicates any simple interpretation that the extended scale merely suppresses treatment effects. If GOSE studies were generally conducted with stronger methods, their more conservative results could reflect better control of bias, more careful outcome ascertainment or more realistic estimates of benefit. Older GOS trials may have used less standardized interviews, broader investigator discretion or analytic practices that amplified apparent differences between treatment and control groups. The analysis also examined small-study effects, which can occur when smaller trials produce unusually large estimates because of random variation, selective reporting or publication processes. These checks did not eliminate the broader pattern.
The study further explored whether the statistical method used to analyze the outcome could explain the findings. In a within-trial negative-control analysis, the comparison between binary and ordinal estimators was close to null. This result is important because the GOSE is inherently ordinal: its categories contain ordered information, and methods that use the full ranking can be more statistically efficient than simply dividing participants into favorable and unfavorable groups. An ordinal analysis may detect shifts across several levels of disability even when the proportion above a single threshold changes little. The near-null negative-control result suggested that differences between binary and ordinal estimators alone were unlikely to account for the between-trial contrast. Instead, the researchers point toward broader differences in trial design, populations, intervention classes and outcome ascertainment.
The findings arrive as the field increasingly recognizes that functional recovery after traumatic brain injury cannot always be reduced to survival or a single favorable-outcome threshold. A patient who moves from severe disability to moderate disability may experience a major improvement in daily life, even if both outcomes fall on the same side of a binary cutoff. Conversely, a threshold can make a modest numerical shift appear decisive when participants cluster near the boundary. The GOSE was developed partly to capture these gradations, but its value depends on reliable interviewing, clear definitions and consistent handling of missing data. Differences in how assessors conduct structured interviews, whether they use patient and caregiver information, and how they resolve uncertain classifications may affect both the precision and direction of treatment estimates.
For evidence synthesis, the message is practical rather than revolutionary: researchers should not automatically assume that GOS and GOSE results are fully interchangeable. Meta-analyses combining trials that use both scales may benefit from scale-stratified sensitivity analyses, in which the results are calculated separately and then compared. Review authors should also record whether outcomes were binary or ordinal, whether assessments used structured interviews, how favorable outcomes were defined and whether the measure was a primary endpoint or a secondary analysis. Trial designers, meanwhile, should justify their choice of outcome scale in advance and standardize training, interview procedures and statistical plans. The study does not establish that GOS is inferior or that GOSE is universally superior. Instead, it highlights how a seemingly technical measurement decision can influence the apparent strength of evidence in a field where treatment effects are often modest and long-term recovery is highly variable.
The authors conclude that GOS-based trials yielded larger treatment-effect estimates than GOSE-based trials, but they caution against interpreting the relationship as causal. Residual confounding between trials remains possible, even after adjustment for several study characteristics. The analysis also depended on published reports, which may omit relevant outcome details or reflect selective publication. Still, the results offer a timely challenge to conventional evidence-synthesis habits. In traumatic brain injury research, the scale used to describe recovery is not merely a label attached to the endpoint; it shapes the resolution, classification and statistical behavior of the outcome. As future trials increasingly adopt detailed functional measures, transparent reporting and harmonized assessment procedures may determine whether the next generation of pooled analyses produces clearer answers—or simply combines different kinds of evidence under the same name.
Subject of Research: Outcome-scale differences in treatment-effect estimates from moderate-to-severe traumatic brain injury trials.
Article Title: A Meta-epidemiologic Analysis of Differences in Treatment-Effect Estimates According to Outcome Scale (GOS vs. GOSE) in Moderate-to-Severe Traumatic Brain Injury Trials
Article References: Raccagni, N. G., Dotti, M., Lightfoot, Y. A., et al. “A Meta-epidemiologic Analysis of Differences in Treatment-Effect Estimates According to Outcome Scale (GOS vs. GOSE) in Moderate-to-Severe Traumatic Brain Injury Trials.” Neurocritical Care (2026). Springer Nature.
Image Credits: AI Generated
DOI: 10.1007/s12028-026-02634-9
Keywords: Traumatic brain injury, Glasgow Outcome Scale, Extended Glasgow Outcome Scale, meta-epidemiology, randomized controlled trials, outcome measurement, evidence synthesis, treatment effects, functional recovery
Tags: accuracydifferences between GOS and GOSE in TBI researchGlasgow Outcome Scale vs Extended Glasgow Outcome Scaleimpact of outcome measurement scales on treatment effect estimatesimplications for clinical decision-making in TBIinfluence of outcome scales on evidence synthesismeta-epidemiologic analysis in clinical trialsmethodological considerations in TBI trial analysispooling of TBI trial resultssystematic review challenges in TBI treatment evaluationtraumatic brain injury treatment outcomestreatment effect heterogeneity in TBI trials
