This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
COBRIT was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating citicoline versus placebo for traumatic brain injury. The primary outcome was a global functional and cognitive measure assessed at 90 days and analyzed with logistic regression.
| Feature | COBRIT |
|---|---|
| Phase | Phase 3 |
| Condition | Traumatic Brain Injury |
| Design | Randomized, parallel-group, quadruple-masked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1,213 |
| Arms | 2 |
| Interventions | Placebo; citicoline |
| Primary endpoint | Functional and Cognitive Outcome at 90 days |
| Primary statistical method | Logistic regression |
| Effect measure | Odds ratio |
| ClinicalTrials.gov | NCT00545662 |
| Lead sponsor | Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) |
2. Clinical Question
The central question was whether participants assigned to citicoline had different odds of achieving a favorable global functional and cognitive outcome at 90 days compared with participants assigned to placebo.
Population
Participants enrolled in a phase 3 study for traumatic brain injury.
Intervention
Citicoline.
Comparator
Placebo.
Primary question
Do the citicoline and placebo groups differ at 90 days on the global functional and cognitive outcome?
The statistical question was framed as a superiority comparison. The registry states the null hypothesis as no difference between the placebo and citicoline groups at 90 days on the Core Battery global statistic.
3. Trial Design
Placebo
- Placebo intervention
- Randomized control group
- Compared with citicoline for the 90-day primary outcome
Citicoline
- Citicoline intervention
- Randomized treatment group
- Compared with placebo for the 90-day primary outcome
4. Primary Endpoint
| Endpoint | Time frame | Definition |
|---|---|---|
| Functional and Cognitive Outcome | 90 days | The primary outcome was analyzed using a global statistic of the Network Core Battery. There were 9 scales: California Verbal Learning Test II (CVLT-II); Controlled Oral Word Association Test (COWAT); Digit Span (DS); Glasgow Outcome Scale Extended (GOSE); Processing Speed Index (PSI); Stroop Test 1 and 2 (ST1&2); and Trail Making Test part A and B (TMT parts A and B). |
Definition of a good outcome on each scale
| Scale | Cut-off for good outcome |
|---|---|
| Glasgow Outcome Scale Extended (GOSE) | GOSE >7 |
| California Verbal Learning Test II (CVLT-II) | CVLT >36 |
| Processing Speed Index (PSI) | PSI >85 |
| Trail Making Test part A | TMT part A <42 |
| Trail Making Test part B | TMT part B <138.1 |
| Digit Span (DS) | DS >7.15 |
| Stroop Test 1 | ST1 <60.29 |
| Stroop Test 2 | ST2 <151.47 |
| Controlled Oral Word Association Test (COWAT) | COWAT >32.5 |
The nine scales were combined into a global statistic rather than analyzed as nine independent primary endpoints. The registry states that correlations among the nine measures were accounted for and that a generalized estimating equation approach was used to account for correlations of the scales.
5. Statistical Methodology
Logistic regression
The primary analysis used logistic regression to estimate the global odds ratio. The outcome was binary: the global functional and cognitive outcome classified participants according to the study's definition of good outcome.
The treatment coefficient is transformed into an odds ratio. In this analysis, the odds in the citicoline group were compared with the odds in the placebo group.
Covariate adjustment
The registry states that the logistic regression models were adjusted for site and injury severity. Adjustment incorporates these prespecified variables into the model while retaining the randomized treatment comparison.
An odds ratio of 1 corresponds to equal odds. Values below 1 indicate lower estimated odds in the citicoline group relative to placebo; values above 1 indicate higher estimated odds.
Correlation among the nine scales
The registry notes that correlations among the nine measures were accounted for. Because the global outcome incorporated multiple correlated scales, treating each scale as completely independent would not represent the analysis described in the registry. The reported approach incorporated the within-outcome correlation structure through generalized estimating equations.
Analysis population
The posted analysis included both participants with complete outcome data and participants with at least one measure. Participants who died were also included in the analysis. This is an important feature of the analysis population because excluding participants who died could otherwise change the composition of the population contributing to the functional and cognitive endpoint.
6. Sample Size and Power Framework
The registry describes a prespecified superiority framework using a two-sided type I error of 0.05 and 85% power. The expected odds ratio for the global statistic was 1.40, and 1,240 participants were required to detect an odds ratio greater than or equal to 1.4 for the global statistic under the stated assumptions.
| Design parameter | Registry value |
|---|---|
| Hypothesis type | Superiority |
| Type I error | Two-sided 0.05 |
| Power | 85% |
| Expected OR | 1.40 |
| Required participants under the stated design assumptions | 1,240 |
| Actual enrollment | 1,213 |
The distinction between the planned sample size and actual enrollment is statistically important. The registry describes 1,240 participants as required under the design assumptions, while enrollment was 1,213. The planned detectable effect was an odds ratio of at least 1.4; it should not be confused with the effect ultimately estimated in the posted analysis.
7. Primary Result: Functional and Cognitive Outcome
Global odds ratio at 90 days
95% CI: 0.83–1.14 · P = 0.76
Citicoline odds compared with placebo odds for the global functional and cognitive outcome.
| Endpoint | Time frame | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Functional and Cognitive Outcome | 90 days | Odds ratio | 0.98 | 0.83–1.14 | 0.76 |
The estimated odds ratio of 0.98 compares the odds of a good global functional and cognitive outcome in the citicoline group with the corresponding odds in the placebo group. An odds ratio of 0.98 is close to 1, meaning that the fitted model estimated very similar odds between the two randomized groups.
The odds ratio does not mean that 98% of participants had a good outcome, nor does it represent a probability, risk ratio, or percentage change in the outcome probability. Odds and probabilities are related but are not interchangeable.
The 95% confidence interval of 0.83–1.14 describes uncertainty around the estimated odds ratio under the statistical model. It includes 1, the value corresponding to equal odds, so the interval is compatible with both modestly lower and modestly higher odds for citicoline relative to placebo.
The P-value of 0.76 measures the compatibility of the observed data with the specified null hypothesis under the analysis framework. It does not measure the size of the treatment effect, the probability that the null hypothesis is true, or the clinical importance of the result.
Because the analysis used logistic regression with adjustment for site and injury severity and accounted for correlation among the nine component scales, the reported odds ratio is a model-based summary of the global endpoint rather than a simple unadjusted comparison of one binary variable.
8. How to Read the Primary Odds Ratio
The most useful reference point for an odds ratio is 1. An estimate of 1 means equal odds between groups. The COBRIT estimate of 0.98 lies very close to that reference value, while the confidence interval extends from 0.83 to 1.14.
What 0.98 means
The model estimated the odds of a good global outcome in the citicoline group to be 0.98 times the odds in the placebo group.
What 0.98 does not mean
It is not a 98% probability of a good outcome and does not mean that citicoline changed outcome probability by 2%.
What 0.83–1.14 means
The 95% confidence interval spans values below and above 1, indicating uncertainty that includes equal odds.
What P = 0.76 means
The reported two-sided P-value is not a measure of effect magnitude or clinical importance.
9. Statistical Methods Explained
Why was logistic regression used?
The primary outcome was represented as a binary good-versus-not-good global functional and cognitive outcome. Logistic regression is designed for binary outcomes and estimates how treatment assignment and covariates relate to the odds of the event or outcome of interest. Here, the treatment effect was summarized as an odds ratio.
Why adjust for site and injury severity?
The registry states that the models were adjusted for site and injury severity. Covariate adjustment can account for systematic variation associated with these factors while estimating the treatment association. Because treatment was randomized, the primary causal comparison remains anchored to the randomized treatment assignment.
What does an odds ratio of 0.98 mean?
An odds ratio of 0.98 means that the estimated odds of a good global outcome with citicoline were 0.98 times the estimated odds with placebo. The comparison is on the odds scale. It should not be translated directly into a 2% difference in the percentage of participants with a good outcome.
Why was the correlation among the nine scales important?
The primary global statistic incorporated nine functional and cognitive scales. Those measures are not necessarily statistically independent. The registry states that correlations among the nine measures were accounted for and that generalized estimating equations were used to account for correlations of the scales. This allows the global analysis to reflect the dependence structure among the component measures.
Why were participants who died included?
The posted analysis explicitly states that patients who died were included. This matters because death is part of the clinical outcome context in traumatic brain injury. Excluding deaths could produce a selected population consisting only of survivors and could therefore change the meaning of the global functional and cognitive comparison.
Why is the confidence interval important?
The confidence interval of 0.83–1.14 communicates the statistical precision of the estimated odds ratio. It provides a range of effect estimates that are compatible with the observed data and model under the stated confidence framework. It is more informative than the point estimate alone because it shows how much uncertainty surrounds 0.98.
Why is the P-value not an effect-size measure?
The P-value of 0.76 describes the evidence against the specified null hypothesis under the analysis model. It does not tell the reader how large the treatment effect is. The effect size is described by the odds ratio, while the confidence interval provides information about its uncertainty.
10. Interpreting the Global Endpoint
The COBRIT primary endpoint is statistically interesting because it combines multiple domains of recovery into a single global outcome. The nine components include cognitive testing, processing speed, executive function, memory, and the Glasgow Outcome Scale Extended. Rather than treating each scale as a separate primary endpoint, the registry describes a global statistic and specifies cut-offs for a good outcome on each scale.
| Component | Domain represented | Good-outcome criterion |
|---|---|---|
| CVLT-II | Verbal learning / memory | CVLT >36 |
| COWAT | Verbal fluency | COWAT >32.5 |
| DS | Digit span | DS >7.15 |
| GOSE | Global functional outcome | GOSE >7 |
| PSI | Processing speed | PSI >85 |
| ST1 | Stroop performance | ST1 <60.29 |
| ST2 | Stroop performance | ST2 <151.47 |
| TMT part A | Processing / visual-motor performance | TMT part A <42 |
| TMT part B | Executive / set-shifting performance | TMT part B <138.1 |
The statistical advantage of a global endpoint is that several clinically relevant domains can contribute to one prespecified comparison. The trade-off is interpretive complexity: an overall odds ratio does not show which individual component drove the result, and the component scales operate on different measurement systems before being incorporated into the global framework.
11. Serious Adverse Events
The registry reports serious adverse events by randomized arm using affected participants and participants at risk.
| Group | Affected | At risk |
|---|---|---|
| Control | 117 | 606 |
| Treatment | 117 | 607 |
Control group
117 participants were affected among 606 at risk for the reported serious adverse-event summary.
Treatment group
117 participants were affected among 607 at risk for the reported serious adverse-event summary.
The reported counts are descriptive safety results. They should be interpreted separately from the primary efficacy odds ratio because safety and efficacy answer different statistical questions and use different outcome definitions.
12. Design Features That Matter Statistically
| Design feature | COBRIT | Statistical relevance |
|---|---|---|
| Randomization | Randomized | Supports comparison of outcomes according to assigned treatment. |
| Parallel design | Parallel | Participants remain in their randomized treatment groups rather than serving as their own controls. |
| Masking | Quadruple | Masking can reduce bias associated with knowledge of treatment assignment. |
| Primary endpoint | Global functional and cognitive outcome | Combines nine component scales into a single primary statistical framework. |
| Regression method | Logistic regression | Appropriate for a binary global outcome and permits covariate adjustment. |
| Covariate adjustment | Site and injury severity | Accounts for these variables in the fitted model. |
| Correlation handling | GEE for scale correlations | Accounts for dependence among the nine measures contributing to the global statistic. |
13. Planned Analysis Versus Posted Result
The registry describes a superiority analysis designed around a two-sided type I error of 0.05, 85% power, an expected odds ratio of 1.40, and a required sample size of 1,240 participants under the stated assumptions. The posted analysis used logistic regression to estimate the global odds ratio, with models adjusted for site and injury severity.
The observed posted estimate was 0.98, with a 95% confidence interval of 0.83–1.14 and a two-sided P-value of 0.76. The planned expected effect of 1.40 and the observed estimate of 0.98 are therefore distinct quantities: the former is a design assumption used for planning, while the latter is the estimated effect from the posted analysis.
14. Limitations
- Global endpoint complexity: the primary outcome combines nine different functional and cognitive scales. A single odds ratio summarizes the global statistic but does not identify the contribution of each component.
- Odds ratio interpretation: the reported effect is on the odds scale. It cannot be interpreted directly as a risk ratio or an absolute percentage-point difference.
- Confidence interval width: the 95% CI of 0.83–1.14 spans both sides of the null value of 1, so the point estimate should not be interpreted independently of its uncertainty.
- Covariate-adjusted model: the reported odds ratio comes from a logistic regression model adjusted for site and injury severity rather than from a simple unadjusted two-group comparison.
- Correlation among component scales: the global outcome requires accounting for correlations among its nine measures; the registry specifically describes GEE for this purpose.
- Analysis population: the posted analysis included participants with complete outcome data and participants with at least one measure, and it also included participants who died. This differs from a complete-case-only analysis.
- Sample-size difference: the registry describes 1,240 participants as required under the design assumptions, while actual enrollment was 1,213.
- Safety interpretation: the serious-adverse-event counts are descriptive affected/at-risk summaries and are not directly interchangeable with the primary efficacy analysis.
15. Why This Trial Matters Statistically
COBRIT is a useful teaching case because its primary endpoint illustrates a different statistical problem from the time-to-event analyses often encountered in phase 3 trials. The study reduced a multidomain functional and cognitive assessment to a prespecified global binary outcome and then used logistic regression to estimate a treatment odds ratio while accounting for covariates and correlations among the component measures.
| Concept | How it appears in COBRIT |
|---|---|
| Randomization | Participants were randomized to placebo or citicoline. |
| Blinding | The study was quadruple-masked. |
| Binary outcome | The global functional and cognitive outcome was represented using good-outcome criteria. |
| Logistic regression | Used to estimate the global odds ratio. |
| Odds ratio | Compared the odds of a good outcome between citicoline and placebo. |
| Covariate adjustment | Models were adjusted for site and injury severity. |
| Correlation | Correlations among the nine scales were accounted for using GEE. |
| Confidence interval | The primary estimate was reported with a two-sided 95% CI of 0.83–1.14. |
| P-value | The primary analysis reported P = 0.76. |
| Sample-size planning | The design used 85% power and an expected OR of 1.40, with 1,240 participants required under the stated assumptions. |
16. Related Tutorials
Learn more about the methods used in this trial:
17. Related Calculators
18. Sources
- ClinicalTrials.gov: NCT00545662 — COBRIT.
- PubMed: PMID 42765966.
- PubMed: PMID 23168823.
Continue through Clinical Biostats
Explore statistical tutorials and calculators related to clinical-trial design, regression, effect measures, and statistical inference.
19. Record Summary
COBRIT provides a useful example of how a multidomain clinical outcome can be converted into a prespecified global binary endpoint and analyzed with logistic regression. The primary analysis estimated an odds ratio of 0.98 for citicoline versus placebo, with a two-sided 95% confidence interval of 0.83–1.14 and a P-value of 0.76. The analysis incorporated adjustment for site and injury severity and accounted for correlations among the nine component scales.