← Clinical Trials
Type 2 Diabetes Phase 3 Randomized NCT03882970

SURPASS-3: Complete Statistical Analysis of Tirzepatide in Type 2 Diabetes

An independent statistical review of the randomized phase 3 SURPASS-3 trial comparing tirzepatide with insulin degludec in participants with type 2 diabetes, focusing on HbA1c, body weight, fasting serum glucose, treatment satisfaction, and the statistical methods used for these endpoints.

Trial start: April 2019  ·  Primary completion: December 2020  ·  Enrollment: 1,444
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are taken from the ClinicalTrials.gov record. The registry record is the official source for the trial's design and posted analyses.

Registry note: 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

SURPASS-3 was a completed randomized, parallel, open-label phase 3 trial evaluating tirzepatide versus insulin degludec in participants with type 2 diabetes mellitus. The registry reports 1,444 enrolled participants, four arms, one registered primary endpoint, nine posted outcome measures, and 24 posted statistical analyses.

1,444
Enrollment
Randomized trial
4
Arms
Parallel design
3
Tirzepatide doses
5 mg, 10 mg, 15 mg
1
Comparator
Insulin degludec
FeatureSURPASS-3
Trial nameSURPASS-3
PhasePhase 3
ConditionType 2 Diabetes Mellitus
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment1,444
InterventionsTirzepatide; insulin degludec
Primary endpoints registered1
Outcome measures posted9
Statistical analyses posted24
Trial statusCompleted
Lead sponsorEli Lilly and Company

2. Clinical Question

The principal statistical question was whether tirzepatide produced a favorable change in HbA1c at Week 52 compared with insulin degludec in participants with type 2 diabetes mellitus. The registry reports separate primary comparisons for the 10 mg and 15 mg tirzepatide doses.

Population

Participants with Type 2 Diabetes Mellitus enrolled in the randomized phase 3 SURPASS-3 trial.

Intervention

Tirzepatide, evaluated in 5 mg, 10 mg, and 15 mg dose groups.

Comparator

Insulin degludec.

Primary question

What is the difference in change from baseline in HbA1c at Week 52 between tirzepatide and insulin degludec?

3. Trial Design

01
Randomize1,444 enrolled
02
4 armsParallel design
03
TreatmentTirzepatide or insulin degludec
04
Week 52HbA1c and other outcomes
05
AnalysisMixed models, logistic regression, ANCOVA
ARMS 1–3 · TIRZEPATIDE

Tirzepatide

  • 5 mg tirzepatide
  • 10 mg tirzepatide
  • 15 mg tirzepatide
ARM 4 · COMPARATOR

Insulin degludec

  • Insulin degludec
  • Comparator arm for the reported statistical analyses
Allocation
Randomized
Model
Parallel
Masking
None
Primary purpose
Treatment

Study timing

April 2019

Trial start

The registry reports a study start date of 2019-04-01.

December 2020

Primary completion

The registry reports a primary completion date of 2020-12-11.

Completed

Registry status

The study is listed as completed, with results posted on ClinicalTrials.gov.

4. Endpoints

The registry lists one primary endpoint: change from baseline in HbA1c for the 10 mg and 15 mg tirzepatide comparisons. The posted analyses also cover body weight, fasting serum glucose, HbA1c target attainment, weight-loss attainment, and components of the Diabetes Treatment Satisfaction Questionnaire, Change Version.

EndpointTime frameAnalysis reportedEffect measure
Change From Baseline in Hemoglobin A1c (HbA1c) (10 mg and 15 mg) Baseline, Week 52 Mixed Models Analysis LS Mean Difference
Change From Baseline in HbA1c (5 mg) Baseline, Week 52 Mixed Models Analysis LS Mean Difference
Change From Baseline in Body Weight Baseline, Week 52 Mixed Models Analysis LS Mean Difference
Change From Baseline in Fasting Serum Glucose Baseline, Week 52 Mixed Models Analysis LS Mean Difference
Percentage of Participants Achieving an HbA1c Target Value of <7% Week 52 Logistic regression Odds Ratio
Percentage of Participants Who Achieved Weight Loss ≥5% Week 52 Logistic regression Odds Ratio
Diabetes Treatment Satisfaction measured by DTSQc Week 52 ANCOVA LS Mean Difference
Registry definition for the primary endpoint: HbA1c is the glycosylated fraction of hemoglobin A. HbA1c is measured primarily to identify average plasma glucose concentration over prolonged periods of time. The registry states that the least squares mean was determined by a mixed-model repeated-measures model with covariates including baseline, pooled country, baseline oral antihyperglycemic medication use, treatment, and time. The registry text is truncated after “Treatment + Time +”, so no additional covariates are added here.
Endpoint-labeling nuance: the registry data posted on ClinicalTrials.gov for the two primary HbA1c analyses labels the endpoint type as “Binary,” while the outcome itself is a continuous HbA1c measure and the reported analysis is a mixed-effects model producing an LS mean difference. This page therefore follows the reported statistical method and effect measure rather than treating HbA1c as a binary outcome.

5. Statistical Methodology

Mixed-effects model for repeated HbA1c measurements

The primary HbA1c analyses used a mixed-model repeated-measures framework. The registry describes the LS mean as being determined from a model incorporating baseline, pooled country, baseline oral antihyperglycemic medication use, treatment, and time. This approach is designed for longitudinal data in which outcome measurements are collected over time rather than treating the Week 52 measurement as an isolated observation.

Conceptual treatment contrast
LS Mean Difference = LS Meantirzepatide − LS Meaninsulin degludec

A negative value indicates a lower modeled change in the tirzepatide group relative to insulin degludec for the reported outcome.

Logistic regression

Binary Week 52 outcomes were analyzed using logistic regression. The reported effect measure was the odds ratio, including analyses of achieving an HbA1c target value of <7% and achieving weight loss ≥5%.

Odds-ratio interpretation
OR = odds of outcome with tirzepatide ÷ odds of outcome with insulin degludec

An odds ratio above 1 means higher odds of the specified outcome in the tirzepatide group relative to insulin degludec. It is not the same as a risk ratio or a difference in percentages.

ANCOVA

ANCOVA was used for the reported Diabetes Treatment Satisfaction Questionnaire, Change Version outcomes. The reported effect measure was an LS mean difference. ANCOVA provides a framework for comparing adjusted group means while incorporating baseline information or other specified covariates.

Non-inferiority and superiority

The primary HbA1c analyses are identified in the registry as non-inferiority analyses, while the posted secondary analyses are identified as superiority analyses. The registry's non-inferiority comment states that the study was powered for superiority in HbA1c and that the sample size provided >99% power to show non-inferiority assuming a 0.3% non-inferiority boundary, a 0.35% greater mean reduction in tirzepatide doses compared with insulin degludec, 1:1:1:1 randomization, a common SD of 1.1%, and a one-sided significance level of 0.0125.

Non-inferiority logic

A non-inferiority analysis asks whether the data exclude a prespecified loss of efficacy beyond the non-inferiority margin. The registry reports a 0.3% boundary for the HbA1c design.

Superiority logic

Once superiority is the relevant hypothesis, the estimate and confidence interval are interpreted for evidence of a difference in the specified direction rather than simply asking whether an unacceptable loss has been excluded.

6. Results

The ClinicalTrials.gov record contains two formal primary-endpoint analyses and multiple secondary analyses. The primary comparisons are reported below exactly as reported in the registry.

Primary HbA1c: 10 mg Tirzepatide vs Insulin Degludec

LS mean difference in HbA1c change

−0.86

95% CI: −1.00 to −0.72   ·   P < 0.001

Time frame: Baseline, Week 52   ·   Hypothesis type: Non-inferiority

Clinical Biostats interpretation

The reported LS mean difference of −0.86 percentage points of HbA1c represents the modeled difference in change from baseline between 10 mg tirzepatide and insulin degludec, with the comparison defined as tirzepatide minus insulin degludec. The negative estimate therefore indicates a more negative modeled HbA1c change for the tirzepatide group under the reported model.

The 95% confidence interval extends from −1.00 to −0.72. This interval describes statistical uncertainty around the estimated group difference; it is not a range containing the individual treatment effects experienced by patients.

The P < 0.001 value addresses the statistical evidence under the specified hypothesis-testing framework. It does not measure the size or clinical importance of the effect. The magnitude is conveyed by the LS mean difference and its confidence interval.

For the non-inferiority framework, the registry reports a 0.3% non-inferiority boundary and a one-sided significance level of 0.0125. Non-inferiority is judged against the prespecified margin rather than by treating an ordinary superiority P-value as the definition of non-inferiority. The registry also states that the study was powered for superiority in HbA1c.

Primary HbA1c: 15 mg Tirzepatide vs Insulin Degludec

LS mean difference in HbA1c change

−1.04

95% CI: −1.17 to −0.90   ·   P < 0.001

Time frame: Baseline, Week 52   ·   Hypothesis type: Non-inferiority

Clinical Biostats interpretation

The reported LS mean difference of −1.04 percentage points of HbA1c represents the modeled difference in change from baseline between 15 mg tirzepatide and insulin degludec, again defined as tirzepatide minus insulin degludec. The negative estimate indicates a more negative modeled HbA1c change for the tirzepatide group.

The 95% confidence interval of −1.17 to −0.90 quantifies uncertainty around the estimated group difference under the reported model. Its width provides information about precision, while the location of the interval provides information about the direction and magnitude of the estimated contrast.

The P < 0.001 result indicates strong statistical evidence under the reported testing framework, but a P-value is not an effect-size measure. It does not say that the probability of the treatment effect being true is greater than 99.9%, nor does it describe the magnitude of benefit for an individual participant.

The non-inferiority interpretation should use the prespecified 0.3% boundary reported by the registry and the stated one-sided significance level of 0.0125. The registry separately states that the study was powered for superiority in HbA1c.

7. Secondary HbA1c Analysis

The registry also reports the Week 52 HbA1c comparison for 5 mg tirzepatide versus insulin degludec using the same mixed-model framework.

ComparisonLS Mean Difference95% CIP-valueHypothesis
5 mg Tirzepatide vs Insulin Degludec −0.59 −0.73 to −0.45 <0.001 Non-inferiority
10 mg Tirzepatide vs Insulin Degludec −0.86 −1.00 to −0.72 <0.001 Non-inferiority
15 mg Tirzepatide vs Insulin Degludec −1.04 −1.17 to −0.90 <0.001 Non-inferiority

The three reported LS mean differences are all negative. Because the registry defines the contrast as the tirzepatide group versus insulin degludec, the direction of these estimates indicates increasingly negative modeled HbA1c differences across the reported tirzepatide dose comparisons. These values should be interpreted as model-based LS mean differences, not as direct comparisons of individual patients.

8. Secondary Body-Weight Results

Change from baseline in body weight at Week 52 was analyzed using the same mixed-model family and reported as an LS mean difference. The registry identifies these analyses as superiority tests.

ComparisonLS Mean Difference, kg95% CIP-value
5 mg Tirzepatide vs Insulin Degludec−9.8−10.8 to −8.8<0.001
10 mg Tirzepatide vs Insulin Degludec−13.0−14.0 to −11.9<0.001
15 mg Tirzepatide vs Insulin Degludec−15.2−16.2 to −14.2<0.001
How to read these estimates

The reported differences are tirzepatide minus insulin degludec. A value of −13.0 kg, for example, represents a modeled LS mean difference of 13.0 kg in the negative direction for 10 mg tirzepatide relative to insulin degludec.

The confidence intervals describe uncertainty around the corresponding LS mean differences. They do not describe the distribution of individual weight changes. The P-values indicate statistical evidence under the reported superiority framework, not the magnitude of the weight difference.

9. Secondary Fasting Serum Glucose Results

Fasting serum glucose was assessed from baseline to Week 52 and analyzed using a mixed-model approach. The registry identifies these comparisons as superiority analyses.

ComparisonLS Mean Difference, mg/dL95% CIP-value
5 mg Tirzepatide vs Insulin Degludec7.52.4 to 12.50.004
10 mg Tirzepatide vs Insulin Degludec0.8−4.3 to 5.90.751
15 mg Tirzepatide vs Insulin Degludec−3.6−8.7 to 1.50.168
Why the confidence intervals matter

The three estimates illustrate why an effect estimate should be read together with its confidence interval. For 5 mg tirzepatide, the reported estimate is 7.5 mg/dL with a 95% CI of 2.4 to 12.5 and P = 0.004. For 10 mg, the interval ranges from −4.3 to 5.9, and for 15 mg it ranges from −8.7 to 1.5. The latter intervals include zero, indicating that the reported data do not provide the same statistical evidence for a nonzero difference under the stated superiority analyses.

The direction of an estimate should also be checked against the endpoint definition. Here, positive and negative values describe the tirzepatide-minus-insulin-degludec LS mean difference; they should not be converted into a different clinical metric without additional information.

10. HbA1c Target Achievement

The registry reports logistic-regression analyses of the percentage of participants achieving an HbA1c target value of <7% at Week 52.

ComparisonOdds Ratio95% CIP-valueHypothesis
5 mg Tirzepatide vs Insulin Degludec3.452.38 to 5.01<0.001Superiority
10 mg Tirzepatide vs Insulin Degludec7.024.55 to 10.84<0.001Superiority
15 mg Tirzepatide vs Insulin Degludec10.796.65 to 17.48<0.001Superiority

Odds ratios are not risk ratios

An odds ratio of 7.02 means that the modeled odds of achieving the specified HbA1c target were 7.02 times the corresponding odds under insulin degludec for the reported 10 mg comparison. It does not mean that 7.02 times as many participants achieved the target, and it cannot be translated into a risk ratio without knowing the underlying event probabilities.

Odds versus probability
Odds = p / (1 − p)

Because odds and probability are different quantities, an odds ratio can appear substantially larger than the corresponding ratio of probabilities, particularly when the outcome is common.

11. Weight-Loss ≥5% Results

The registry reports logistic-regression analyses for the percentage of participants who achieved weight loss ≥5% by Week 52.

ComparisonOdds Ratio95% CIP-value
5 mg Tirzepatide vs Insulin Degludec29.7818.35 to 48.35<0.001
10 mg Tirzepatide vs Insulin Degludec79.8847.56 to 134.17<0.001
15 mg Tirzepatide vs Insulin Degludec110.7764.73 to 189.55<0.001
Statistical interpretation

These odds ratios quantify the relative odds of crossing the prespecified ≥5% weight-loss threshold. They do not quantify the average amount of weight lost; that question is addressed separately by the continuous body-weight analysis.

The large odds ratios are accompanied by confidence intervals that remain well above 1. This provides the statistical evidence reported by the registry for superiority on this binary endpoint. The confidence intervals also show that the estimates themselves remain uncertain; a point estimate such as 110.77 should not be treated as an exact population parameter.

12. Diabetes Treatment Satisfaction

The registry reports ANCOVA analyses of the Diabetes Treatment Satisfaction Questionnaire, Change Version (DTSQc), including hyperglycemia, hypoglycemia, and treatment satisfaction scores. All analyses use Week 52 as the time frame.

Hyperglycemia

ComparisonLS Mean Difference95% CIP-value
5 mg Tirzepatide vs Insulin Degludec−0.26−0.57 to 0.050.096
10 mg Tirzepatide vs Insulin Degludec−0.25−0.57 to 0.060.113
15 mg Tirzepatide vs Insulin Degludec−0.47−0.78 to −0.160.003

Hypoglycemia

ComparisonLS Mean Difference95% CIP-value
5 mg Tirzepatide vs Insulin Degludec−0.41−0.74 to −0.080.014
10 mg Tirzepatide vs Insulin Degludec−0.18−0.51 to 0.150.280
15 mg Tirzepatide vs Insulin Degludec−0.26−0.59 to 0.070.129

Treatment Satisfaction Score

ComparisonLS Mean Difference95% CIP-value
5 mg Tirzepatide vs Insulin Degludec3.012.26 to 3.75<0.001
10 mg Tirzepatide vs Insulin Degludec2.902.15 to 3.65<0.001
15 mg Tirzepatide vs Insulin Degludec2.992.24 to 3.74<0.001
Reading the satisfaction analyses

The treatment-satisfaction estimates are LS mean differences from ANCOVA, not odds ratios or raw percentage differences. The positive estimates for the Treatment Satisfaction Score indicate a higher modeled score for tirzepatide relative to insulin degludec, whereas the hyperglycemia and hypoglycemia components have negative estimates for the reported comparisons.

For example, the 5 mg treatment-satisfaction comparison has an LS mean difference of 3.01 with a 95% CI of 2.26 to 3.75. By contrast, the 5 mg hyperglycemia comparison has an estimate of −0.26 with a 95% CI of −0.57 to 0.05. These are different components of the questionnaire and should not be combined into a single overall effect.

13. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk. These figures should be interpreted as counts and denominators exactly as reported.

ArmSerious adverse eventsAffected / at risk
5 mg TirzepatideSerious adverse events29 / 358
10 mg TirzepatideSerious adverse events20 / 360
15 mg TirzepatideSerious adverse events26 / 359
Insulin DegludecSerious adverse events22 / 360

The denominators in this safety table are the registry-reported at-risk counts and should not be substituted for the overall enrollment figure of 1,444. Safety analyses can use an analysis population defined by treatment exposure, whereas enrollment describes the trial as a whole.

Do not infer a comparative safety conclusion from these counts alone. The ClinicalTrials.gov record identifies serious adverse-event counts by arm but do not provide a formal between-group statistical analysis of serious adverse events. The appropriate interpretation is therefore descriptive rather than a claim of superiority or inferiority in safety.

14. Analysis Population and Missing Data

The statistical analyses posted on ClinicalTrials.gov use an analysis population defined as all randomized participants who received at least one dose of study drug and had a baseline and at least one post-baseline value, excluding participants discontinuing study drug due to inadvertent enrollment. This definition is repeated across the reported primary and secondary analyses.

FeatureReported registry information
RandomizationRandomized
Minimum exposureAt least one dose of study drug
Baseline requirementBaseline value required
Post-baseline requirementAt least one post-baseline value required
Explicit exclusionParticipants discontinuing study drug due to inadvertent enrollment

The presence of a baseline and post-baseline requirement means that the formal analyses are not simply calculations over every enrolled participant. For the mixed-model outcomes, the longitudinal framework is intended to use available repeated measurements under the model, but the ClinicalTrials.gov record does not provide a separate detailed missing-data or imputation specification. No additional imputation method is therefore attributed to the trial here.

15. Non-Inferiority Analysis in SURPASS-3

The primary HbA1c analyses are identified as non-inferiority analyses. The registry comment provides unusually useful design information: the study was powered for superiority in HbA1c, while the sample size provided >99% power to show non-inferiority under specified assumptions.

Design quantityRegistry value
Non-inferiority boundary0.3%
Assumed greater mean reduction in tirzepatide doses0.35%
Randomization assumption1:1:1:1
Common standard deviation1.1%
One-sided significance level0.0125
Power to show non-inferiority>99%

Why the margin matters

Suppose the treatment difference is defined as tirzepatide minus insulin degludec and the non-inferiority margin is 0.3% in the unfavorable direction. The central statistical question is whether the confidence interval excludes differences worse than that boundary. This is conceptually different from asking whether the P-value is below a generic 0.05 threshold.

Conceptual non-inferiority rule
Lower confidence limit > −0.3%

The exact testing procedure should follow the prespecified statistical analysis plan. The expression above illustrates the directional logic implied by a 0.3% non-inferiority margin when the treatment difference is defined as tirzepatide minus insulin degludec.

For both reported primary comparisons, the lower limits of the 95% confidence intervals are −1.00 and −1.17, respectively. However, the registry identifies the analyses as non-inferiority analyses while also reporting P < 0.001 and stating that the study was powered for superiority. Those features should not be collapsed into one generic “significant” result; the estimand, margin, confidence interval, and prespecified testing hierarchy all matter.

16. Multiplicity and Multiple Comparisons

SURPASS-3 contains multiple dose comparisons and multiple outcome analyses. The ClinicalTrials.gov record reports one registered primary endpoint but provide two primary statistical analyses, corresponding to 10 mg and 15 mg tirzepatide versus insulin degludec. Numerous secondary comparisons are also posted.

Analysis familyReported structureStatistical interpretation
Primary HbA1c10 mg vs comparator; 15 mg vs comparatorPrimary non-inferiority analyses
5 mg HbA1c5 mg vs comparatorSecondary non-inferiority analysis
Body weightThree dose comparisonsSecondary superiority analyses
Fasting serum glucoseThree dose comparisonsSecondary superiority analyses
HbA1c <7%Three dose comparisonsSecondary superiority analyses using logistic regression
Weight loss ≥5%Three dose comparisonsSecondary superiority analyses using logistic regression
DTSQcThree dose comparisons across three componentsSecondary ANCOVA analyses

The existence of many reported P-values means that each result should be interpreted in the context of the trial's prespecified multiplicity strategy. The ClinicalTrials.gov record does not provide a complete multiplicity-adjustment algorithm for all posted secondary endpoints, so this page does not assign an unreported familywise error interpretation to those P-values.

17. Statistical Methods Explained

Why was a mixed-effects model used for HbA1c?

HbA1c is measured longitudinally, so participants can contribute multiple measurements over time. A mixed-model repeated-measures approach can model the trajectory while accounting for the correlation created by repeated observations within the same participant. In SURPASS-3, the registry reports baseline, pooled country, baseline oral antihyperglycemic medication use, treatment, and time among the model covariates.

What does an LS mean difference of −1.04 mean?

It is the modeled difference between the least-squares mean changes for 15 mg tirzepatide and insulin degludec, with tirzepatide minus insulin degludec defining the direction. The negative sign therefore indicates a more negative modeled change in HbA1c for tirzepatide. It is not a statement that every participant experienced a 1.04-point change.

Why does the non-inferiority margin matter?

A non-inferiority claim depends on a clinically or scientifically prespecified acceptable loss, not simply on whether a P-value is below a conventional threshold. The registry reports a 0.3% non-inferiority boundary for the HbA1c design. The confidence interval is therefore interpreted relative to that boundary.

What does an odds ratio of 10.79 mean?

For the reported 15 mg comparison, the odds ratio of 10.79 means that the modeled odds of achieving an HbA1c target of <7% were 10.79 times the corresponding odds under insulin degludec. It does not mean that the probability was 10.79 times higher.

Why can an odds ratio be very different from a risk ratio?

Odds are calculated as probability divided by one minus probability. When an outcome is common, the numerical distance between odds and probability can become substantial. Consequently, an odds ratio should not automatically be described as a percentage increase in probability.

What does a P-value of <0.001 tell us?

It indicates strong statistical evidence against the relevant null hypothesis under the specified analysis and testing framework. It does not quantify effect size, clinical importance, the probability that the null hypothesis is true, or the probability that a patient will benefit.

Why use ANCOVA for treatment satisfaction?

ANCOVA provides a linear-model framework for comparing treatment groups on a continuous outcome while accounting for specified covariates. In the ClinicalTrials.gov record, the DTSQc outcomes were analyzed using ANCOVA and expressed as LS mean differences with 95% confidence intervals.

18. Confidence Intervals: The Precision of the Estimates

The reported results contain both very narrow and relatively wider intervals. The interval should always be read alongside the point estimate because the point estimate alone hides sampling uncertainty.

EndpointComparisonEstimate95% CI
HbA1c change10 mg vs insulin degludec−0.86−1.00 to −0.72
HbA1c change15 mg vs insulin degludec−1.04−1.17 to −0.90
Body weight10 mg vs insulin degludec−13.0 kg−14.0 to −11.9
Fasting serum glucose10 mg vs insulin degludec0.8 mg/dL−4.3 to 5.9
HbA1c <7%10 mg vs insulin degludec7.024.55 to 10.84
Weight loss ≥5%10 mg vs insulin degludec79.8847.56 to 134.17
A confidence interval is not a patient-level prediction interval

A 95% confidence interval describes uncertainty in an estimated population contrast under the statistical model and sampling framework. It does not mean that 95% of individual participants will have treatment effects inside that numerical interval.

19. Continuous and Binary Endpoints Answer Different Questions

SURPASS-3 is particularly useful statistically because the same clinical domain is evaluated using both continuous and binary outcomes.

Continuous HbA1c

The mixed-model LS mean difference asks how the modeled average change in HbA1c differs between treatment groups.

HbA1c <7%

Logistic regression asks how the odds of crossing a specific Week 52 threshold differ between treatment groups.

Continuous body weight

The LS mean difference describes the modeled difference in average weight change.

Weight loss ≥5%

The odds ratio describes the relative odds of achieving a predefined categorical threshold.

These analyses are complementary rather than interchangeable. A continuous endpoint preserves more information about the distribution of changes, whereas a binary threshold provides a clinically recognizable classification but discards information about how far participants lie above or below the threshold.

20. Interpretation of the Primary Findings

The two primary HbA1c comparisons report LS mean differences of −0.86 for 10 mg tirzepatide and −1.04 for 15 mg tirzepatide versus insulin degludec. Their 95% confidence intervals are entirely below zero, and both analyses report P < 0.001.

What the primary estimates mean

The estimates quantify the modeled between-group difference in change from baseline in HbA1c at Week 52. Because the contrast is tirzepatide minus insulin degludec, negative values indicate a more negative modeled change in HbA1c for tirzepatide.

What the primary estimates do not mean

They do not mean that every participant's HbA1c changed by exactly −0.86 or −1.04 percentage points. They also do not provide a probability that a randomly selected participant will experience a particular response.

Why the confidence interval matters

The 95% confidence intervals of −1.00 to −0.72 and −1.17 to −0.90 provide information about the precision of the estimated group contrasts. Their widths are substantially narrower than the corresponding range of plausible individual responses would be expected to be, because these are population-level estimates rather than patient-level prediction intervals.

Why the P-value does not measure effect size

P < 0.001 indicates strong statistical evidence under the specified testing framework. It does not tell us whether the estimated difference is large, small, clinically meaningful, or important to a particular participant. Those questions require the effect estimate, its uncertainty, the endpoint scale, and clinical context.

21. Limitations

22. Why This Trial Matters Statistically

SURPASS-3 is a useful teaching case because it places several core clinical-trial methods in the same randomized study. The primary endpoint uses a longitudinal mixed-model framework, while secondary endpoints demonstrate logistic regression and ANCOVA. The trial also illustrates the difference between non-inferiority and superiority questions and between continuous and binary estimands.

ConceptHow it appears in SURPASS-3
RandomizationRandomized phase 3 parallel-group design
Multiple treatment doses5 mg, 10 mg, and 15 mg tirzepatide compared with insulin degludec
Mixed-effects modelPrimary and several secondary continuous outcomes
Repeated measuresHbA1c analyzed from baseline through Week 52 using MMRM
LS mean differencePrimary HbA1c and continuous secondary outcomes
Logistic regressionHbA1c <7% and weight loss ≥5%
Odds ratioBinary Week 52 outcomes
ANCOVADTSQc hyperglycemia, hypoglycemia, and treatment satisfaction analyses
Non-inferiorityPrimary HbA1c comparisons with a reported 0.3% boundary
SuperiorityBody weight, fasting serum glucose, HbA1c target, weight-loss, and selected DTSQc analyses
Confidence intervals95% intervals accompany the reported effect estimates
Safety analysisSerious adverse events reported as affected / at-risk counts by arm

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

Continue with the statistical methods behind SURPASS-3

Explore the underlying methods used to analyze continuous outcomes, binary endpoints, confidence intervals, and non-inferiority questions in randomized clinical trials.

26. Record Summary

SURPASS-3 provides a compact example of how several statistical estimands can coexist within a randomized phase 3 trial. The primary HbA1c endpoint was evaluated with a mixed-model repeated-measures approach and reported as an LS mean difference for 10 mg and 15 mg tirzepatide versus insulin degludec. The same framework was used for continuous secondary outcomes including body weight and fasting serum glucose, while logistic regression was used for binary Week 52 thresholds and ANCOVA for the reported treatment-satisfaction outcomes.

The primary HbA1c estimates were −0.86 for 10 mg and −1.04 for 15 mg tirzepatide, each with a 95% confidence interval entirely below zero and P < 0.001. The registry identifies these as non-inferiority analyses and reports a 0.3% non-inferiority boundary, while also stating that the study was powered for superiority in HbA1c. That distinction illustrates why the statistical interpretation of a clinical trial depends on the prespecified hypothesis, estimand, confidence interval, and testing framework—not simply on whether a P-value is small.

Clinical Biostats methodology: A trial-results page should not merely repeat reported outcomes. The goal is to reconstruct the statistical story of the trial while clearly separating registry-reported evidence from educational interpretation and avoiding conclusions that are not supported by the available analysis data.