SURPASS-5: Complete Statistical Analysis of Tirzepatide in Type 2 Diabetes

1. Trial at a Glance

SURPASS-5 was a randomized, double-blind, parallel phase 3 trial evaluating three tirzepatide doses against placebo in participants with type 2 diabetes inadequately controlled on insulin glargine with or without metformin. The registry reports 475 enrolled participants, four treatment arms, 11 posted outcome measures, and 21 posted statistical analyses.

2. Clinical Question

The central statistical question was whether adding tirzepatide at 5 mg, 10 mg, or 15 mg produced a greater improvement in HbA1c at Week 40 than placebo in participants with type 2 diabetes inadequately controlled on insulin glargine with or without metformin.

3. Trial Design

The design is statistically informative because the three active-dose comparisons create several parallel treatment-versus-placebo questions within the same randomized trial. The registry classifies the hypothesis type for the posted analyses as superiority.

4. Endpoints

The registry identifies one registered primary endpoint, with two posted primary statistical comparisons: 10 mg versus placebo and 15 mg versus placebo. The 5 mg HbA1c comparison is posted as a secondary analysis.

5. Analysis Populations

The registry-defined populations for the posted analyses are not simply the full enrollment total. The primary HbA1c analyses use participants who were randomized, received at least one dose of study drug, and had a baseline and at least one post-baseline value, with an exclusion for patients discontinuing study drug due to inadvertent enrollment.

This distinction matters because the enrolled population of 475 is not automatically identical to every analysis population. Missing baseline or post-baseline measurements, treatment exposure, and the registry's specified exclusions can change the number of observations contributing to an individual analysis.

6. Statistical Methodology

Mixed-model repeated measures

The primary endpoint uses a mixed-model repeated measures framework. The registry states that the least-squares mean was determined with baseline HbA1c, baseline metformin use, pooled country, treatment, time, and treatment-by-time interaction included in the model, with Type III sums of squares.

Least-squares mean difference

The reported primary effect measure is the LS Mean Difference. This is a model-adjusted contrast between treatment groups. It should not automatically be interpreted as the simple difference between two unadjusted sample means because the model incorporates the covariates and repeated-measure structure specified by the registry.

Logistic regression

Binary outcomes such as achieving an HbA1c target of <7%, achieving an HbA1c target of <5.7%, and achieving at least 5% weight loss were analyzed using logistic regression. The effect measure reported by the registry is an odds ratio.

Confidence intervals

The registry reports two-sided 95% confidence intervals for the posted treatment effects. For a mean difference, the interval describes uncertainty around the estimated difference. For an odds ratio, the interval describes uncertainty around the estimated ratio of odds.

P-values

The primary comparisons have P-values of <0.001. A P-value measures the compatibility of the observed data with a specified null hypothesis under the statistical model and testing framework. It does not quantify the size, clinical importance, or probability of the treatment effect.

7. Primary Results: HbA1c at Week 40

The registered primary endpoint was change from baseline in HbA1c at Week 40 for the 10 mg and 15 mg comparisons. Both posted analyses used mixed models and reported least-squares mean differences with two-sided 95% confidence intervals.

10 mg Tirzepatide vs Placebo

15 mg Tirzepatide vs Placebo

8. Secondary HbA1c Result: 5 mg Tirzepatide

The 5 mg tirzepatide comparison was posted as a secondary endpoint using the same general mixed-model approach.

The estimate of -1.30 represents the model-based difference in HbA1c change between 5 mg tirzepatide and placebo. Its two-sided 95% confidence interval extends from -1.52 to -1.07, and the reported P-value is <0.001.

Because this is a secondary endpoint rather than the registered primary endpoint, its inferential role is different from the two primary comparisons. The statistical estimate can be interpreted directly, but the broader evidentiary hierarchy depends on the trial's prespecified multiplicity strategy, which is not reported in the ClinicalTrials.gov record.

9. Secondary Results: Body Weight

Body weight was analyzed from baseline to Week 40 using mixed models. The registry reports LS mean differences for all three tirzepatide doses versus placebo.

10. Secondary Results: HbA1c Target <7%

The registry reports logistic-regression analyses for the percentage of participants achieving an HbA1c target value of <7% at Week 40.

11. Secondary Results: Fasting Serum Glucose

Fasting serum glucose was analyzed as a change-from-baseline outcome at Week 40 using mixed models.

The negative estimates indicate lower model-estimated change in fasting serum glucose relative to placebo. The reported confidence intervals for all three comparisons remain below zero, and all three P-values are <0.001.

12. Secondary Results: Weight Loss ≥5%

The binary endpoint for achieving weight loss of at least 5% was analyzed using logistic regression.

The odds ratios are large relative measures, but they should not be converted directly into probability differences without the underlying event probabilities. The confidence intervals also demonstrate substantial uncertainty around the precise magnitude of the odds ratios, despite the direction of every interval being above 1.

13. Secondary Results: Daily Mean Insulin Glargine Dose

The registry reports percentage change from baseline in daily mean insulin glargine dose at Week 40. For these comparisons, the record does not name a statistical method, so none is inferred here.

14. Secondary Results: HbA1c Target <5.7%

The registry also reports logistic-regression analyses for the percentage of participants achieving an HbA1c target value of <5.7% at Week 40.

These results use the same odds-ratio framework as the <7% endpoint. The odds-ratio scale can produce very large numbers when the outcome is substantially more common in one group than another, so interpretation should remain on the odds scale unless the underlying event probabilities are also reported.

15. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm. These counts should be interpreted as affected participants divided by the number at risk in each arm.

The ClinicalTrials.gov record does not report a formal statistical comparison of serious adverse-event rates. Accordingly, the counts are presented descriptively rather than converted into an inferred treatment effect or P-value.

16. Multiplicity and Multiple Comparisons

SURPASS-5 contains several treatment comparisons and several outcome measures. The registry data identifies the primary endpoint as the change from baseline in HbA1c for the 10 mg and 15 mg comparisons, while additional HbA1c, body-weight, glucose, target-achievement, and insulin-dose analyses are posted as secondary outcomes.

Multiplicity is important because multiple hypothesis tests create more opportunities for statistically significant findings to occur by chance. The ClinicalTrials.gov record does not provide a multiplicity-adjustment procedure, alpha-allocation scheme, hierarchical testing sequence, or interim-analysis plan. Therefore, no such procedure is inferred.

17. Stratification and Covariate Adjustment

The primary endpoint definition provides unusually useful detail about the covariates incorporated into the MMRM. The model included:

• Baseline HbA1c
• Baseline Metformin Use (Yes, No)
• Pooled Country
• Treatment
• Time
• Treatment × Time

This is different from simply comparing the arithmetic mean change in HbA1c between arms. The least-squares means are model-adjusted estimates that account for the covariates specified in the analysis.

18. Missing Data and Repeated Measures

The primary analysis population requires a baseline and at least one post-baseline value. This is important because a mixed-model repeated-measures analysis uses available longitudinal measurements rather than requiring every participant to contribute an observation at every scheduled time point.

However, the ClinicalTrials.gov record does not specify the missing-data assumption, covariance structure, imputation sensitivity analyses, or estimand framework used for the MMRM beyond the model terms explicitly reported in the endpoint definition.

19. Statistical Methods Explained

Why was a mixed-effects model used for HbA1c?

HbA1c is measured longitudinally, with the registered primary endpoint defined from baseline to Week 40 and the model including time and treatment-by-time interaction. A mixed-model repeated-measures framework is suited to repeated observations because it models the longitudinal outcome while estimating treatment contrasts at the relevant assessment time.

What does an LS mean difference of -1.66 mean?

It means the model-estimated change from baseline in HbA1c was 1.66 percentage points lower for 10 mg tirzepatide than for placebo at the specified Week 40 comparison. It is an adjusted model contrast, not an individual prediction and not necessarily the same as the difference between two unadjusted observed means.

What does an odds ratio of 100.07 mean?

It means the estimated odds of achieving the HbA1c target of <7% were 100.07 times the odds in the placebo group for the 10 mg comparison under the posted logistic-regression analysis. It does not mean that 100.07% of participants achieved the target and should not be interpreted as a probability ratio.

Why are confidence intervals important when the P-value is <0.001?

The P-value addresses evidence against the null hypothesis, whereas the confidence interval describes the statistical precision of the estimated treatment effect. For example, the 10 mg primary HbA1c estimate is -1.66 with a 95% CI of -1.88 to -1.43. The interval tells the reader considerably more about the range of model-compatible effect estimates than the P-value alone.

Why doesn't a P-value measure effect size?

A P-value depends on the estimated effect, its uncertainty, and the amount of information contributing to the analysis. A small P-value can accompany a modest effect in a large, precise study, while a larger effect can fail to achieve a small P-value when uncertainty is substantial. Effect estimates and confidence intervals should therefore be read alongside the P-value.

Why distinguish primary from secondary endpoints?

Primary endpoints define the main confirmatory questions of a trial. Secondary endpoints provide additional evidence but can involve many additional statistical comparisons. Without knowing the prespecified multiplicity strategy, one should not automatically treat every statistically significant secondary result as having the same confirmatory status as a primary endpoint.

Why does randomization matter for interpretation?

Randomization creates the trial's intended basis for comparing treatment groups. It helps balance measured and unmeasured factors in expectation and supports causal interpretation of the randomized treatment contrast, subject to the trial's design, conduct, analysis population, and missing-data assumptions.

20. Understanding the Primary HbA1c Effect

21. Comparing Dose Groups Without Overinterpreting Them

The posted results include three active-dose comparisons with placebo, but the presence of three estimates does not by itself create a formal dose-response test.

The pattern across the posted point estimates can be described, but a formal claim that one dose is statistically superior to another would require a direct dose-to-dose comparison and an appropriate inferential framework. The ClinicalTrials.gov record does not provide such a formal comparison for these outcomes.

22. Why This Trial Matters Statistically

SURPASS-5 is a useful teaching example because it combines randomized treatment allocation, multiple active doses, repeated continuous outcomes, binary responder endpoints, model-adjusted mean contrasts, odds ratios, confidence intervals, and multiple secondary analyses within one phase 3 trial.

23. Important Limitations and Interpretation Issues

• Registry-level reporting: the analysis is constrained to the statistical and numerical information from ClinicalTrials.gov. Some protocol or statistical-analysis-plan details may not be represented in the registry fields provided here.
• Analysis population differs from enrollment: the enrolled population was 475, but individual analyses use populations requiring treatment exposure and baseline/post-baseline observations.
• MMRM assumptions: the primary analysis uses a repeated-measures model, but the ClinicalTrials.gov record does not specify every covariance, missing-data, or sensitivity-analysis assumption.
• Multiple comparisons: numerous dose-versus-placebo and secondary endpoint analyses are reported. The ClinicalTrials.gov record does not specify a multiplicity-adjustment strategy.
• Secondary endpoints: statistically significant secondary analyses should not automatically be interpreted as having the same confirmatory status as the registered primary endpoint.
• Odds ratios: large odds ratios are not equivalent to large absolute probability differences. Event probabilities are needed to translate an odds ratio into an absolute risk perspective.
• No inferred method: the daily mean insulin glargine dose analyses report estimates and confidence intervals but no named statistical method. A specific statistical model is therefore not assigned to those results.
• No dose-to-dose confirmation: differences between point estimates across 5 mg, 10 mg, and 15 mg groups do not by themselves establish formal statistical differences between doses.
• Safety is descriptive here: serious adverse events are reported as affected participants divided by those at risk, without a registry-provided formal comparative analysis in the ClinicalTrials.gov record.

24. What the Odds Ratio Does — and Does Not — Mean

25. Planned Analysis Features Not Reported in the Supplied Data

Several design features often discussed on clinical-trial statistical pages are not provided in the trial data for SURPASS-5. They are therefore not reconstructed here.

26. Related Tutorials

Learn more about the methods used in this trial:

27. Related Calculators

28. Sources

• ClinicalTrials.gov: SURPASS-5, NCT04039503.
• PubMed: PMID 41961454.
• PubMed: PMID 39531161.
• PubMed: PMID 37668888.
• PubMed: PMID 37526908.
• PubMed: PMID 35210595.

29. Record Summary

SURPASS-5 provides a useful statistical example of how a randomized phase 3 trial can combine repeated-measures modeling with categorical responder analyses. The primary HbA1c endpoint was evaluated using a mixed-model repeated-measures framework incorporating baseline HbA1c, baseline metformin use, pooled country, treatment, time, and treatment-by-time interaction. The two primary treatment comparisons produced LS mean differences of -1.66 for 10 mg tirzepatide versus placebo and -1.65 for 15 mg tirzepatide versus placebo, each with a two-sided 95% confidence interval of -1.88 to -1.43 and a P-value of <0.001.

The secondary analyses extend the statistical story across body weight, fasting serum glucose, HbA1c targets, weight-loss response, and insulin glargine dose. Continuous outcomes were generally reported as LS mean differences, while binary outcomes were analyzed with logistic regression and expressed as odds ratios. The registry also reports serious adverse events descriptively by arm.