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
TRIUMPH-2 was a randomized, double-blind, parallel-group phase 3 study evaluating retatrutide in participants with type 2 diabetes mellitus who have obesity or overweight. The registry lists 1152 participants, 4 arms, and 2 primary endpoints assessed at baseline and Week 80.
| Feature | TRIUMPH-2 |
|---|---|
| Study acronym | TRIUMPH-2 |
| Phase | Phase 3 |
| Population | Participants with type 2 diabetes mellitus who have obesity or overweight |
| Additional condition listed | Obstructive sleep apnea |
| Design | Randomized, double-blind, parallel-group |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1152 |
| Number of arms | 4 |
| Interventions | Retatrutide; placebo |
| Status | Completed |
| Study start | 2023-07-11 |
| Primary completion date | 2026-06-16 |
| Lead sponsor | Eli Lilly and Company |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT05929079 |
2. Clinical Question
The central clinical question is how retatrutide compares with placebo in participants with type 2 diabetes mellitus who have obesity or overweight, with the primary assessment focused on change in body weight at Week 80 and change in apnea-hypopnea index for the GSA2 subset at Week 80.
Population
Participants with type 2 diabetes mellitus who have obesity or overweight. Obstructive sleep apnea is also listed among the study conditions.
Intervention
Retatrutide.
Comparator
Placebo.
Primary questions
What is the percent change from baseline in body weight at Week 80, and what is the change from baseline in apnea-hypopnea index for the GSA2 subset at Week 80?
3. Trial Design
The four-arm structure is important statistically because the overall trial contains more than two randomized groups, while the registry information identifies the interventions at the study level as retatrutide and placebo. The available registry record does not specify the individual arm assignments or the retatrutide regimens, so the four arms should not be treated as four interchangeable two-group comparisons.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint | Percent Change from Baseline in Body Weight | Baseline, Week 80 |
| Primary endpoint | Change from Baseline in Apnea-Hypopnea Index (AHI) for GSA2 Subset | Baseline, Week 80 |
Percent Change from Baseline in Body Weight
The first primary endpoint is expressed as a percentage change from baseline rather than as an absolute change in the original measurement scale. Conceptually, for a participant with baseline body weight \(W_0\) and Week 80 body weight \(W_{80}\), percent change is calculated as:
A negative value represents a decrease from baseline body weight, while a positive value represents an increase.
Because the registry endpoint is explicitly defined as a percent change, the statistical comparison should preserve that scale rather than silently replacing it with an absolute weight difference.
Change from Baseline in Apnea-Hypopnea Index for the GSA2 Subset
The second primary endpoint is the change from baseline in apnea-hypopnea index for the GSA2 subset at Week 80. Unlike the body-weight endpoint, this endpoint is defined as an absolute change from baseline in AHI rather than a percentage change.
The distinction matters because an absolute change and a percent change have different interpretations. A treatment-group difference in mean change answers a different question from a ratio or percentage comparison, even when both are based on measurements collected at the same time point.
5. Statistical Methodology
The ClinicalTrials.gov record identifies the endpoints and their assessment time frames but does not post statistical analysis results or detailed statistical-model specifications for these endpoints. Consequently, the most useful statistical interpretation is to distinguish the registry-defined estimands from the standard methods that are commonly used for this type of randomized continuous outcome.
Randomized comparison of continuous outcomes
Both primary endpoints are continuous measurements evaluated at baseline and Week 80. A typical analysis compares the randomized treatment groups using an analysis of covariance (ANCOVA), a mixed model for repeated measures (MMRM), or another prespecified longitudinal model, depending on the statistical analysis plan.
For the body-weight endpoint, the response variable would ordinarily be the percent change from baseline at the prespecified assessment time. For the AHI endpoint, the response would ordinarily be the change from baseline in AHI for the specified GSA2 subset.
Why baseline adjustment is useful
When a continuous outcome is measured at baseline and again after treatment, adjusting for baseline can improve precision by accounting for some of the variability already present before treatment. An ANCOVA-style model can express the treatment comparison after accounting for baseline measurement.
The actual model used for TRIUMPH-2 would depend on the prespecified statistical analysis plan. The registry information available here does not provide the model coefficients, covariance structure, missing-data assumptions, or multiplicity procedure.
Why repeated measurements can change the model choice
If multiple post-baseline measurements are incorporated into the analysis rather than using only the Week 80 value, an MMRM or another longitudinal model can use information across visits. Such a model also requires assumptions about the covariance among repeated measurements and how incomplete outcome histories are handled.
Analysis of a subset
The AHI endpoint is specifically restricted to the GSA2 subset. This is statistically important because the relevant analysis population for that endpoint is not necessarily the full enrollment population of 1152 participants. The subset definition and its sample size are not specified in the registry information presented here.
Confidence intervals
A treatment-group difference should ordinarily be accompanied by a confidence interval. The interval communicates statistical precision around the estimated treatment contrast and provides information that a p-value alone does not provide.
Multiplicity
There are 2 registered primary endpoints. When multiple primary endpoints are tested, the statistical analysis plan normally specifies how the overall type I error rate is controlled or how success is defined across the endpoints. The registry information presented here does not specify the multiplicity strategy.
6. Planned Analysis
ClinicalTrials.gov lists 2 primary endpoints, both assessed from baseline to Week 80, but no formal statistical analyses are posted on the registry.
| Primary endpoint | Typical statistical approach for this endpoint type | Key quantity |
|---|---|---|
| Percent Change from Baseline in Body Weight | Typically analyzed as a continuous outcome using ANCOVA, MMRM, or another prespecified longitudinal model, with baseline weight or an appropriate baseline-derived covariate considered in the model. | Adjusted treatment-group difference in percent change, with a confidence interval and prespecified hypothesis test. |
| Change from Baseline in AHI for GSA2 Subset | Typically analyzed as a continuous change-from-baseline outcome using ANCOVA, MMRM, or another prespecified model appropriate for the repeated measurements and subset population. | Adjusted treatment-group difference in change from baseline, with a confidence interval and prespecified hypothesis test. |
What a complete Week 80 analysis would normally report
For the body-weight endpoint, a complete statistical report would ordinarily identify the treatment-group estimates at Week 80, the estimated between-group contrast, its confidence interval, the hypothesis-test result, the analysis population, and the handling of missing Week 80 measurements.
For the AHI endpoint, the report would additionally identify the GSA2 subset used for analysis and clearly distinguish the subset analysis from the full randomized population. This distinction is essential because an endpoint defined within a subset has a different analysis population from an endpoint defined for all randomized participants.
7. Statistical Methods Explained
Why use percent change for body weight?
Percent change places the change on a relative scale. For example, a decrease from baseline is represented as a negative percentage. This can facilitate comparison across participants with different baseline body weights, although the statistical properties of a percent-change outcome still depend on the distribution of baseline and follow-up measurements.
Why is the AHI endpoint a change from baseline rather than percent change?
The registry defines the AHI endpoint as change from baseline. That means the primary quantity is the Week 80 AHI minus the baseline AHI. This is conceptually different from dividing the change by baseline AHI and expressing it as a percentage.
Why might ANCOVA be used?
ANCOVA can compare treatment groups at the analysis time point while accounting for baseline outcome values. This can reduce residual variability when baseline measurements are predictive of follow-up measurements, improving precision without abandoning the randomized treatment comparison.
Why might an MMRM be used?
An MMRM can incorporate multiple post-baseline measurements rather than discarding all information except one time point. It models the correlation among repeated measurements and can be useful when the trial collects longitudinal outcomes. The precise covariance and missing-data assumptions must be prespecified.
Why does the GSA2 subset matter?
The AHI endpoint is explicitly limited to the GSA2 subset. A treatment effect estimated within that subset describes the randomized comparison among the participants who satisfy the subset definition, not automatically the entire population of 1152 enrolled participants.
Why are confidence intervals important?
A confidence interval gives a range of values compatible with the observed data under the specified statistical framework. It conveys precision around the estimated treatment contrast. A narrow interval indicates greater statistical precision than a wide interval, all else equal.
Why do two primary endpoints require attention to multiplicity?
Testing more than one primary hypothesis can increase the chance of obtaining at least one statistically significant result under a global null hypothesis if each test is treated independently at the same nominal significance level. Confirmatory trials therefore commonly prespecify an error-control strategy, such as a hierarchical testing procedure or an allocation of the type I error rate.
8. Interpreting the Two Primary Endpoints
Body weight
The endpoint asks how much body weight changed relative to each participant's baseline, expressed as a percentage, at Week 80.
Apnea-hypopnea index
The endpoint asks how AHI changed from baseline to Week 80 specifically within the GSA2 subset.
Between-group contrast
The central inferential quantity would ordinarily be the difference between the randomized treatment-group estimates, rather than the change within one group considered alone.
Uncertainty
A point estimate without a confidence interval gives an incomplete picture of statistical precision.
Within-group change is not the randomized treatment effect
A common statistical mistake is to examine whether body weight or AHI changed significantly within each treatment group and then conclude that the treatment groups differ. The randomized comparison is based on the between-group contrast. A statistically detectable change from baseline in one group and a nondetectable change in another does not by itself establish a statistically significant difference between the groups.
Percent change is not a percent of participants
A percent change in body weight is a continuous measurement. It should not be interpreted as the percentage of participants who lost weight or as the probability that an individual participant will experience a particular outcome.
Change in AHI is not automatically a clinical responder rate
A continuous change in AHI summarizes the magnitude of change on the AHI scale. It is distinct from a responder endpoint that would classify participants according to whether they crossed a prespecified clinical threshold.
9. Analysis Population and Randomization
TRIUMPH-2 enrolled 1152 participants and used randomized allocation in a double-blind, parallel-group design. Randomization is the principal design mechanism supporting comparison of the intervention and placebo groups because, in expectation, treatment assignment is independent of both measured and unmeasured baseline characteristics.
| Statistical concept | Role in this trial |
|---|---|
| Randomization | Creates the basis for comparing retatrutide with placebo. |
| Double-blinding | Reduces the potential for treatment knowledge to influence participant behavior, outcome assessment, or other trial processes. |
| Parallel design | Participants are assigned to treatment groups rather than being intentionally crossed between treatments. |
| Four arms | The registry identifies 4 randomized arms, requiring attention to the prespecified contrasts and analysis population for each primary endpoint. |
| Primary endpoint timing | Both registered primary endpoints use baseline and Week 80 measurements. |
The registry information available here does not specify the randomization ratio, the number of participants in each arm, the individual arm regimens, or the stratification factors. Those details should therefore not be inferred from the existence of 4 arms.
10. Missing Data and Estimands
Longitudinal trials commonly encounter participants who do not provide a Week 80 measurement. The statistical consequences depend on why measurements are missing and on the estimand being targeted.
Observed data
The analysis uses measurements that were actually collected, with a statistical model defining how those observations contribute to estimation.
Missing outcomes
Missing Week 80 measurements require a prespecified strategy. The appropriate approach depends on the estimand, missingness assumptions, and available longitudinal information.
Intercurrent events
Events occurring after treatment initiation can affect the scientific question being estimated and may need explicit handling in the estimand framework.
Sensitivity analysis
Alternative assumptions can be used to examine whether conclusions are sensitive to plausible departures from the primary missing-data assumptions.
The ClinicalTrials.gov information presented here does not specify the missing-data method, estimand definition, imputation method, treatment-discontinuation strategy, rescue-medication strategy, or sensitivity-analysis framework for TRIUMPH-2.
11. Multiplicity and the Two Primary Endpoints
TRIUMPH-2 has 2 registered primary endpoints. This creates a fundamental multiplicity question: how should statistical evidence for body-weight change and AHI change be combined into the confirmatory decision framework?
| Question | Why it matters |
|---|---|
| Are both endpoints required? | A trial can define success as requiring both primary endpoints to meet their prespecified criteria, or it can use another prespecified testing strategy. |
| Is alpha shared? | If multiple hypotheses contribute to the confirmatory conclusion, the type I error allocation must be specified. |
| Is there a hierarchy? | A hierarchical testing strategy can test one endpoint first and pass statistical significance to another endpoint according to a prespecified rule. |
| Are the endpoints correlated? | Body weight and AHI are different continuous outcomes, and their correlation can affect the efficiency and interpretation of certain multiplicity procedures. |
The registry information presented here does not state which multiplicity strategy was used. Consequently, the presence of 2 primary endpoints should be recognized as a design feature rather than treated as evidence that either endpoint was statistically significant.
12. Safety and Adverse Events
The ClinicalTrials.gov record identifies the interventions, study design, enrollment, conditions, and primary endpoints. It does not provide serious adverse-event counts by treatment arm or other numerical safety results.
| Safety information | Registry information available here |
|---|---|
| Serious adverse events by arm | Not reported in the ClinicalTrials.gov record information. |
| Arm-specific adverse-event rates | Not reported in the ClinicalTrials.gov record information. |
| Numerical safety comparisons | No posted numerical results. |
Safety is statistically distinct from the primary efficacy endpoints. A randomized trial can evaluate continuous efficacy outcomes and adverse events using different analysis populations, denominators, event definitions, and statistical methods. Without arm-specific safety results, no numerical treatment-group safety comparison can be made from the registry information presented here.
13. Timeline
Study start
The registered study start date was 2023-07-11.
Randomized treatment study
The study was registered as a phase 3 randomized, double-blind, parallel-group treatment study with 4 arms.
Primary endpoint assessment window
Both registered primary endpoints are assessed from baseline to Week 80.
Primary completion
The registered primary completion date was 2026-06-16.
14. What a Treatment Effect Would Mean Statistically
If a treatment-group estimate of percent change were lower than the corresponding placebo estimate, that would indicate a greater reduction in body weight on the percentage scale. The between-group difference, rather than either within-group change alone, would be the principal randomized treatment contrast.
If the retatrutide group had a more negative change from baseline in AHI than the placebo group, that would indicate a larger reduction in AHI over the specified period. The interpretation would apply specifically to the GSA2 subset defined for the endpoint.
For either endpoint, the confidence interval would describe uncertainty around the estimated treatment contrast. It would not describe the range of individual participant responses or guarantee that every participant would experience an effect in the direction of the group estimate.
A p-value, if reported, would address the compatibility of the observed data with a specified null hypothesis under the statistical model. It would not measure the size or clinical importance of the treatment effect.
15. Why This Trial Matters Statistically
TRIUMPH-2 is a useful teaching case because it combines a randomized, double-blind phase 3 design with two continuous primary endpoints measured over time. The statistical questions are therefore centered on baseline adjustment, longitudinal measurement, subset analysis, multiplicity, and interpretation of continuous treatment effects.
| Concept | How it appears in TRIUMPH-2 |
|---|---|
| Randomization | Participants were randomized to 4 parallel arms. |
| Double-blinding | The study was registered as double-blind. |
| Continuous outcome analysis | Both registered primary endpoints are continuous measurements or continuous changes. |
| Baseline adjustment | Both primary endpoints are defined relative to baseline, making baseline handling central to the analysis. |
| Longitudinal assessment | Primary endpoints use baseline and Week 80 measurements. |
| Subset analysis | The AHI endpoint is restricted to the GSA2 subset. |
| Multiplicity | There are 2 registered primary endpoints. |
| Missing data | Week 80 outcomes in a longitudinal randomized trial require prespecified handling if measurements are unavailable. |
| Effect estimation | The principal comparison would ordinarily be a treatment-group contrast with an associated confidence interval. |
16. Important Limitations and Interpretation Issues
- No posted statistical results: the ClinicalTrials.gov record does not provide treatment-effect estimates, confidence intervals, p-values, or numerical outcome results for the registered primary endpoints.
- Four-arm structure: the registry identifies 4 arms but does not specify the individual arm regimens or enrollment counts in the information presented here. The arm-specific comparisons therefore cannot be reconstructed.
- AHI subset: the AHI primary endpoint is defined for the GSA2 subset, but the subset size and detailed definition are not specified in the available registry information.
- Multiplicity: 2 primary endpoints are registered, but the type I error-control strategy is not specified in the available information.
- Missing data: the registry information does not specify how missing Week 80 measurements are handled.
- Model specification: the registry information does not identify the precise statistical model, covariance structure, covariates, or estimation method for the primary endpoints.
- Analysis population: the enrollment count of 1152 does not establish that all participants contribute evaluable observations to both primary endpoints.
- Continuous endpoints: mean differences, adjusted means, medians, distributions, and responder rates answer different questions. The registered endpoint definitions should not be silently replaced with a different outcome scale.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The study was designed as a randomized, double-blind, parallel-group phase 3 comparison with 2 registered continuous primary endpoints assessed from baseline to Week 80. The formal treatment-effect estimates and inferential results are not posted in the registry information available here.
Clinical interpretation
The registered questions concern change in body weight and change in AHI for the specified GSA2 subset. Clinical interpretation of the magnitude of those changes requires the actual Week 80 results and their associated uncertainty measures.
18. Statistical Questions a Complete Results Report Should Answer
| Question | Why it matters |
|---|---|
| What was the adjusted body-weight change in each randomized group? | Provides the treatment-group estimates for the first primary endpoint. |
| What was the between-group difference in percent change? | Provides the principal randomized treatment contrast for body weight. |
| What was the confidence interval? | Quantifies statistical precision around the treatment contrast. |
| What was the Week 80 AHI change in the GSA2 subset? | Provides the estimate for the second primary endpoint. |
| How large was the treatment-group difference in AHI change? | Provides the randomized comparison for the subset endpoint. |
| How were missing measurements handled? | Determines how incomplete follow-up contributes to the estimand and inference. |
| What multiplicity strategy was prespecified? | Determines how the two primary endpoints contribute to confirmatory inference. |
| What analysis population was used? | Defines which participants contributed to each primary endpoint and prevents inappropriate comparison of different populations. |
19. Record Summary
TRIUMPH-2 is a completed phase 3 randomized, double-blind, parallel-group study enrolling 1152 participants across 4 arms. The registered interventions are retatrutide and placebo, and the 2 primary endpoints are percent change from baseline in body weight and change from baseline in apnea-hypopnea index for the GSA2 subset, both assessed at Week 80.
From a statistical perspective, the central issues are the analysis of continuous change-from-baseline outcomes, baseline adjustment, the distinction between the full randomized population and the GSA2 subset, missing Week 80 measurements, and multiplicity across 2 primary endpoints. These features determine how the eventual treatment estimates should be interpreted once the formal statistical results are available.
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical-trial analyses and the statistical concepts used to interpret randomized evidence.
20. Sources
- ClinicalTrials.gov: TRIUMPH-2, NCT05929079.