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
SUSTAIN 11 was a randomized, parallel-group, phase 3 treatment trial comparing semaglutide with insulin aspart in people with type 2 diabetes receiving metformin and insulin glargine. The registry reports a non-inferiority hypothesis for the primary HbA1c endpoint.
| Feature | SUSTAIN 11 |
|---|---|
| Trial name | SUSTAIN 11 |
| ClinicalTrials.gov identifier | NCT03689374 |
| Phase | Phase 3 |
| Condition | Diabetes Mellitus, Type 2 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 2274 |
| Arms | 2 |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| Start | October 1, 2018 |
| Primary completion | February 22, 2021 |
2. Clinical Question
The registered primary question concerns the change from baseline in glycated haemoglobin (HbA1c) at week 52. The treatment comparison is semaglutide versus insulin aspart, with both interventions used in the setting of metformin and insulin glargine.
Population
People with type 2 diabetes mellitus enrolled in the phase 3 SUSTAIN 11 trial.
Intervention
Semaglutide, taken together with metformin and insulin glargine U100.
Comparator
Insulin aspart, taken together with metformin and insulin glargine U100.
Primary question
How does the change from baseline in HbA1c at week 52 compare between semaglutide and insulin aspart under the prespecified non-inferiority framework?
3. Trial Design
Semaglutide
- Semaglutide
- Metformin
- Insulin glargine U100
Insulin aspart
- Insulin aspart
- Metformin
- Insulin glargine U100
The registry describes the allocation as randomized and the design model as parallel. The trial was not masked. Randomization is important statistically because treatment assignment is established before outcome information is observed, allowing the between-group comparison to retain the principal protection provided by random allocation.
4. Endpoint and Analysis Time Frame
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Change From Baseline in Glycated Haemoglobin (HbA1c) | Change from baseline in HbA1c at week 52. Data are reported for the on-treatment observation period, from the date of first dose of trial product at week 0 to the last date on trial product with a visit window of +7 days. | Baseline (week 0), week 52 |
The registry lists one registered primary endpoint. The posted statistical analysis concerns this endpoint and compares semaglutide with insulin aspart.
5. Statistical Methodology
Analysis population
The reported analysis used the full analysis set, defined as all randomized participants. For this particular outcome measure, the overall number analyzed consists of participants with available data for the outcome.
The analysis therefore retains randomized treatment assignment as the basis of the efficacy comparison. The registry also specifies that the number analyzed for the outcome consists of participants with available data for that outcome measure.
ANCOVA for the HbA1c treatment comparison
The registry analysis notes that responses were analyzed using an analysis of covariance (ANCOVA), with treatment as a fixed factor and the baseline HbA1c value as a covariate. Conceptually, ANCOVA compares treatment groups at the follow-up time while accounting for baseline differences in the outcome.
The treatment term estimates the adjusted between-group difference after accounting for the baseline value. Baseline adjustment can improve precision when baseline HbA1c is predictive of the week 52 measurement.
Missing-data handling
Before analysis, missing data were multiple imputed using observed data from participants within the same group defined by randomized treatment. The regression model used for imputation included randomized treatment group and data from baseline and all previous visits as covariates.
Multiple imputation is designed to preserve uncertainty associated with missing observations rather than replacing every missing value with a single deterministic number. The use of previous visits as predictors allows observed longitudinal information to contribute to the imputation model.
Non-inferiority framework
The prespecified non-inferiority margin was 0.3 percentage points. The registry identifies the hypothesis type as non-inferiority and reports a treatment difference of −0.29 percentage points for semaglutide versus insulin aspart.
Primary treatment difference
95% CI: −0.38 to −0.20 percentage points
P < 0.0001
Prespecified non-inferiority margin: 0.3 percentage points
Reported testing procedure
The registry describes the formal method as a t-distributed test. The analysis notes that the non-inferiority p-value was calculated as two times the one-sided p-value from a t-distributed test.
6. Primary Result: Change From Baseline in HbA1c
The registry posts a formal statistical analysis for the primary endpoint. The reported treatment difference is for semaglutide versus insulin aspart at week 52 over the on-treatment observation period.
| Analysis characteristic | Reported result |
|---|---|
| Endpoint | Change From Baseline in Glycated Haemoglobin (HbA1c) |
| Time frame | Baseline (week 0), week 52 |
| Groups | Semaglutide vs Insulin Aspart |
| Outcome unit | Percentage of HbA1c |
| Analysis population | Full analysis set |
| Reported method | t-distributed test |
| Effect measure | Treatment difference |
| Estimate | −0.29 |
| 95% confidence interval | −0.38 to −0.20 |
| P-value | <0.0001 |
| Hypothesis type | Non-inferiority |
| Non-inferiority margin | 0.3 percentage points |
The estimated treatment difference of −0.29 percentage points means that the adjusted change in HbA1c favored semaglutide by 0.29 percentage points relative to insulin aspart under the reported treatment-difference convention. Because a lower HbA1c value represents a greater reduction from baseline, the negative direction is consistent with a greater reduction for semaglutide in the reported comparison.
The estimate does not mean that every individual patient experienced a 0.29-percentage-point difference. It is a group-level estimated treatment difference from the statistical analysis.
The 95% confidence interval extends from −0.38 to −0.20. This interval describes uncertainty around the estimated treatment difference under the analysis framework; it does not describe the range of individual patient responses.
The p-value of <0.0001 is evidence against the null hypothesis used for the reported test. It is not a measure of the size or clinical importance of the treatment difference. Effect size is described by the estimate and its confidence interval.
For non-inferiority, the direction of the treatment difference and the prespecified margin must be considered together. The margin was 0.3 percentage points, so the relevant question is whether the evidence rules out a loss of more than 0.3 percentage points relative to the comparator under the prespecified one-sided testing framework.
7. Understanding the Non-inferiority Question
Non-inferiority trials are designed differently from conventional superiority trials. The null hypothesis generally represents an unacceptable loss of efficacy for the experimental treatment, while the alternative represents efficacy sufficiently close to the comparator that the prespecified loss threshold is not exceeded.
With a non-inferiority margin of 0.3 percentage points, the margin defines how much worse the experimental treatment could be while still satisfying the prespecified non-inferiority criterion.
The reported estimate was −0.29 percentage points, while the reported 95% confidence interval was −0.38 to −0.20.
The reported non-inferiority p-value was generated from a one-sided t-distributed test and then multiplied by two. This is important when interpreting the p-value because non-inferiority testing is directional: the analysis is concerned with whether the experimental treatment is sufficiently worse than the comparator, not simply whether the two treatments are different in either direction.
8. Why Baseline Adjustment Matters
The analysis used ANCOVA with baseline HbA1c as a covariate. Baseline adjustment is especially useful when the baseline value is strongly related to the follow-up measurement because it can account for predictable variation that is not attributable to treatment.
Treatment factor
The randomized treatment group provides the primary comparison: semaglutide versus insulin aspart.
Baseline covariate
The baseline HbA1c measurement adjusts the comparison for the starting value of the outcome.
Estimated difference
The resulting treatment difference summarizes the adjusted contrast between groups at the analyzed time point.
Precision
Including a prognostic baseline variable can reduce unexplained residual variation and improve precision of the treatment comparison.
Importantly, baseline adjustment does not turn the analysis into an observational comparison. The treatment groups remain defined by randomization; the baseline covariate is used to improve the statistical comparison rather than to create the treatment groups.
9. Missing Data and Multiple Imputation
The registry specifies multiple imputation before the analysis. Participants were grouped according to randomized treatment, and the imputation regression model incorporated randomized treatment together with baseline and all previous visit data.
| Component | Role in the imputation approach |
|---|---|
| Randomized treatment group | Defines the treatment grouping used in the imputation process. |
| Baseline data | Provides information about the participant's starting measurement. |
| Previous visits | Uses observed longitudinal measurements to inform missing values. |
| Regression model | Combines the available predictors to generate imputed observations. |
Multiple imputation is preferable to treating a missing value as if it were known because it recognizes uncertainty about the unobserved value. In a randomized trial, the assumptions underlying the missing-data model still matter: the validity of the final estimate depends in part on whether the observed information used for imputation adequately represents the mechanism generating missing observations.
10. Intention-to-Treat and the Full Analysis Set
The registry states that the full analysis set included all randomized participants. This makes randomization central to the efficacy comparison: participants are classified according to randomized treatment rather than being reassigned according to later treatment behavior.
Randomization helps balance measured and unmeasured baseline factors in expectation. An analysis anchored to randomized assignment therefore preserves the principal causal comparison created by the trial design. The reported outcome-specific analysis still includes participants with available data for the particular endpoint, and missing values are addressed through the specified multiple-imputation procedure.
The registry also describes the analysis text as incorporating an intention-to-treat concept. In practical terms, this emphasizes preservation of the randomized comparison rather than analyzing only participants who adhered perfectly to their assigned treatment.
11. Serious Adverse Events
The registry reports serious adverse events by treatment arm as affected participants divided by participants at risk.
| Safety measure | Semaglutide | Insulin Aspart |
|---|---|---|
| Serious adverse events | 65/874 | 84/864 |
The affected/at-risk counts are descriptive safety data. They should not be interpreted as a formal between-group hypothesis test because the ClinicalTrials.gov record does not provide a corresponding effect estimate, confidence interval, or p-value for serious adverse events.
12. Statistical Methods Explained
Why was ANCOVA used?
ANCOVA was used to compare treatment groups while incorporating baseline HbA1c as a covariate. Because baseline HbA1c is directly related to the outcome being evaluated, adjusting for it can reduce residual variability and make the treatment comparison more statistically precise.
What does a treatment difference of −0.29 mean?
The treatment difference was reported as semaglutide minus insulin aspart. An estimate of −0.29 means the estimated change in HbA1c was 0.29 percentage points lower for semaglutide than for insulin aspart under the reported analysis. It is an estimated group difference, not a guarantee about an individual patient's response.
What does the 95% confidence interval mean?
The 95% confidence interval of −0.38 to −0.20 represents statistical uncertainty around the estimated treatment difference. It is not a probability statement that there is a 95% chance that the true effect for this particular completed trial lies inside the interval, nor does it describe individual patient variability.
Why is the non-inferiority margin important?
The 0.3-percentage-point margin defines the maximum prespecified loss in HbA1c efficacy that could still be considered compatible with non-inferiority. Unlike a superiority test, the non-inferiority question is therefore tied to a clinically and statistically prespecified boundary rather than simply asking whether the two treatments differ.
Why was one-sided testing used?
Non-inferiority is directional. The concern is whether semaglutide is unacceptably worse than insulin aspart by more than the prespecified margin. A one-sided test reflects that directional question. The registry notes that the reported non-inferiority p-value was calculated as two times the one-sided p-value from a t-distributed test.
Why does the p-value not describe the treatment effect size?
A p-value measures the compatibility of the observed result with the null hypothesis under the specified statistical model. It does not tell the reader whether the treatment difference is large, small, clinically important, or precise. Those questions are addressed by the treatment estimate, confidence interval, and prespecified non-inferiority margin.
13. Primary Endpoint: Statistical Interpretation in Context
The primary endpoint combines several statistical ingredients: a randomized comparison, baseline adjustment, an explicit missing-data strategy, a treatment-difference estimate, a confidence interval, and a non-inferiority hypothesis test. Each answers a different part of the statistical question.
| Component | What it contributes |
|---|---|
| Randomization | Defines the treatment comparison and supports causal interpretation under the trial design. |
| ANCOVA | Adjusts the week 52 comparison for baseline HbA1c. |
| Multiple imputation | Addresses missing observations using treatment and longitudinal observed data. |
| Treatment difference | Quantifies the estimated contrast between semaglutide and insulin aspart. |
| 95% confidence interval | Quantifies statistical uncertainty around the treatment estimate. |
| Non-inferiority margin | Defines the prespecified boundary for an unacceptable loss of efficacy. |
| One-sided test | Addresses the directional non-inferiority hypothesis. |
This layered interpretation is more informative than relying on the p-value alone. The estimate describes the observed direction and magnitude, the confidence interval describes precision, and the non-inferiority margin supplies the decision boundary specified before the results were evaluated.
14. Important Limitations and Interpretation Issues
- Endpoint-specific analysis population: although the full analysis set included all randomized participants, the registry states that the number analyzed for this outcome consists of participants with available data for the outcome measure.
- Missing-data assumptions: multiple imputation depends on the model used to generate the missing observations. The validity of the resulting inference therefore depends on how well the imputation model represents the available longitudinal information and the missing-data process.
- Non-inferiority interpretation: the 0.3-percentage-point margin is central to the hypothesis. A non-inferiority p-value should not be interpreted in isolation from the prespecified margin and direction of the treatment difference.
- Two-sided confidence interval versus one-sided testing: the registry reports a two-sided 95% confidence interval while describing the non-inferiority p-value as twice a one-sided p-value. These quantities should not be treated as interchangeable evidence thresholds.
- Outcome measurement: HbA1c is a continuous biomarker, whereas the registry's normalized endpoint description labels the posted statistical-analysis endpoint as "Binary." The reported treatment difference and ANCOVA description provide the substantive statistical framework for interpreting the posted result.
- Safety interpretation: serious adverse-event counts are descriptive in the available registry analysis information. A formal comparative safety inference would require the corresponding effect estimate and uncertainty measure.
- Limited posted statistical analyses: the registry posts one statistical analysis for the primary endpoint. The information available here therefore supports detailed interpretation of the primary HbA1c analysis but does not provide a corresponding formal statistical analysis for the other posted outcome measures.
15. Why This Trial Matters Statistically
SUSTAIN 11 is a useful teaching example because the primary question is not simply whether two treatments are statistically different. Instead, the analysis asks whether the experimental treatment stays within a prespecified acceptable difference from the comparator, while adjusting for baseline HbA1c and addressing missing data.
| Concept | How it appears in SUSTAIN 11 |
|---|---|
| Randomization | The registry describes randomized allocation in a parallel-group phase 3 trial. |
| Intention-to-treat principle | The analysis text identifies the full analysis set as all randomized participants and references intention-to-treat analysis. |
| ANCOVA | Treatment is modeled as a fixed factor with baseline value as a covariate. |
| Missing-data handling | Missing data were multiple imputed using treatment and baseline/previous-visit information. |
| Non-inferiority | The prespecified non-inferiority margin was 0.3 percentage points. |
| One-sided testing | The non-inferiority p-value was based on a one-sided t-distributed test and multiplied by two. |
| Confidence interval | The reported treatment difference has a 95% CI of −0.38 to −0.20. |
| Effect measure | The primary effect was reported as a treatment difference rather than a ratio. |
16. Non-inferiority Versus Superiority
A central statistical lesson from SUSTAIN 11 is that a non-inferiority design asks a different question from a conventional superiority trial.
Superiority question
Is there sufficient evidence that the treatments differ in the prespecified direction?
Non-inferiority question
Is the experimental treatment sufficiently close to the comparator that any loss in efficacy is smaller than the prespecified acceptable margin?
Role of the margin
The margin supplies the boundary against which the estimated treatment difference is evaluated.
Role of the p-value
The p-value describes evidence against the non-inferiority null under the specified testing framework; it does not replace the effect estimate or the margin.
For this reason, a small p-value alone does not communicate the full meaning of a non-inferiority result. Readers need to see the treatment difference, confidence interval, direction of the contrast, and prespecified margin together.
17. Timeline
Trial start
The registry lists October 1, 2018 as the study start date.
Randomized parallel-group evaluation
The study evaluated semaglutide versus insulin aspart in people with type 2 diabetes, with metformin and insulin glargine U100 forming part of the intervention regimens.
Primary completion
The registry lists February 22, 2021 as the primary completion date.
Results posted
The registry reports that results were posted, including one statistical analysis for the primary HbA1c endpoint.
18. Results Summary
The primary posted statistical analysis found a treatment difference of −0.29 percentage points in change from baseline HbA1c for semaglutide versus insulin aspart at week 52. The reported 95% confidence interval was −0.38 to −0.20, and the reported p-value was <0.0001 under the non-inferiority testing framework.
Primary statistical result
95% CI −0.38 to −0.20
P < 0.0001
Non-inferiority margin: 0.3 percentage points
The reported analysis provides an estimated between-treatment difference, its uncertainty interval, and a formal non-inferiority test. The result must be interpreted according to the prespecified non-inferiority framework rather than as a generic test of whether the treatments are identical.
The primary result does not by itself describe individual patient responses, long-term outcomes, all safety outcomes, or the effect of treatment on every other outcome measure. Those questions require their corresponding outcome definitions and analyses.
19. Clinical Biostats Interpretation: Reading the Result Correctly
The statistical story of SUSTAIN 11 is best understood in layers. First, randomization establishes the comparison between semaglutide and insulin aspart. Second, ANCOVA adjusts the week 52 comparison for baseline HbA1c. Third, multiple imputation addresses missing observations using the longitudinal information specified in the registry. Finally, the estimated treatment difference is evaluated against a prespecified non-inferiority margin using a directional hypothesis test.
The estimate of −0.29 is therefore not simply a descriptive difference between two observed means. It is the reported treatment contrast from an adjusted analysis. The 95% confidence interval gives a measure of uncertainty around that estimate, while the non-inferiority margin provides the clinical-statistical boundary specified for the trial.
The reported p-value should be read as evidence under that particular hypothesis-testing framework. It should not be used as a proxy for effect magnitude. The effect magnitude is the treatment difference, while its precision is represented by the confidence interval.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Statistical Calculators
22. Sources
- ClinicalTrials.gov: SUSTAIN 11, NCT03689374.
- Publication: PubMed record, PMID 35546450.
Continue through Clinical Biostats
Connect the statistical methods used in SUSTAIN 11 with deeper tutorials, calculators, and additional clinical-trial analyses.
23. Record Summary
SUSTAIN 11 provides a clear example of how a randomized phase 3 trial can frame efficacy through a non-inferiority question. The primary endpoint was change from baseline in HbA1c at week 52, analyzed using ANCOVA with treatment as a fixed factor and baseline HbA1c as a covariate. Missing data were addressed through multiple imputation using randomized treatment and longitudinal observed data. The reported treatment difference was −0.29 percentage points, with a 95% CI of −0.38 to −0.20 and a p-value of <0.0001, against a prespecified non-inferiority margin of 0.3 percentage points.
The principal statistical lesson is that non-inferiority evidence cannot be reduced to a p-value. The treatment estimate, confidence interval, direction of the comparison, prespecified margin, missing-data strategy, and one-sided testing framework all contribute to the interpretation. The reported serious adverse-event counts provide a separate descriptive view of safety and should be considered independently from the primary efficacy analysis.