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™ 2 was a randomized, double-blind, parallel phase 3 trial evaluating once-weekly semaglutide against once-daily sitagliptin as add-on therapy to metformin and/or TZD in subjects with type 2 diabetes. The registry reports two primary analyses of change in HbA1c from baseline to week 56, with non-inferiority assessed using a prespecified 0.3 percentage-point margin.
| Feature | SUSTAIN™ 2 |
|---|---|
| Trial phase | Phase 3 |
| Condition | Diabetes mellitus, type 2 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 1231 |
| Arms | 4 |
| Primary endpoint | Change in HbA1c from baseline until week 56 |
| Primary analyses posted | 2 |
| Analysis method | Mixed-effects model / mixed models analysis |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT01930188 |
2. Clinical Question
The primary statistical question was whether the change in HbA1c from baseline to week 56 with semaglutide was non-inferior to the change observed with sitagliptin when the treatments were used as add-on therapy to metformin and/or TZD.
Population
Subjects with type 2 diabetes enrolled in the randomized phase 3 SUSTAIN™ 2 trial.
Intervention
Semaglutide once weekly, with analyses reported separately for 1.0 mg and 0.5 mg treatment comparisons.
Comparator
Sitagliptin 100 mg once daily, with semaglutide placebo used in the corresponding blinded comparison.
Primary question
Is the change in HbA1c with each semaglutide dose non-inferior to the change with sitagliptin at week 56?
3. Trial Design
Semaglutide 1.0 mg + sitagliptin placebo
- Semaglutide 1.0 mg
- Sitagliptin placebo
- Serious adverse events: 30/409
Semaglutide 0.5 mg + sitagliptin placebo
- Semaglutide 0.5 mg
- Sitagliptin placebo
- Serious adverse events: 30/409
Sitagliptin 100 mg + semaglutide placebo
- Sitagliptin 100 mg
- Semaglutide placebo
- Serious adverse events: 13/204
Sitagliptin 100 mg + semaglutide placebo
- Sitagliptin 100 mg
- Semaglutide placebo
- Serious adverse events: 16/203
4. Enrollment, Analysis Population, and Trial Timing
The trial enrolled 1231 subjects. The primary endpoint definition specifies a full analysis set of 1225 subjects, consisting of randomized subjects who had received at least one dose of randomized semaglutide or sitagliptin.
The distinction between enrollment and the full analysis set is important. The primary endpoint definition explicitly identifies 1225 subjects for the FAS, rather than treating the enrollment count of 1231 as the analysis population.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame | Analysis population |
|---|---|---|---|
| Change in HbA1c (Glycosylated Haemoglobin) From Baseline | Change in HbA1c from baseline until week 56 | Week 0, week 56 | Full analysis set (FAS = 1225) |
The endpoint is a change-from-baseline measure of glycosylated haemoglobin. The registry's posted statistical analysis uses a treatment difference as the effect measure, so the treatment contrast is expressed on the HbA1c percentage scale rather than as a ratio or hazard ratio.
6. Statistical Methodology
Mixed model for repeated measurements
The registry reports a mixed models analysis for the primary endpoint. Post-baseline responses were analyzed using a mixed model for repeated measurements, with treatment and country as fixed factors and baseline value as a covariate, all nested within visit.
The registry describes treatment and country as fixed factors and baseline HbA1c as a covariate, with the analysis structured within visit.
Covariate adjustment
Baseline HbA1c was included as a covariate. Conceptually, this allows the treatment comparison at follow-up to account for the measured baseline value rather than comparing week-56 values without regard to where subjects started.
Repeated measurements
The model is explicitly described as a mixed model for repeated measurements. This is relevant because the endpoint concerns longitudinal measurements rather than an isolated measurement without regard to the sequence of visits.
Missing data
The registry states that missing data were imputed from the mixed model for repeated measurements. Thus, the primary analysis incorporates the model-based handling of missing post-baseline information rather than simply restricting the analysis to subjects with an observed week-56 measurement.
Country adjustment
Country was included as a fixed factor. In a multicountry randomized trial, including country in the model can account statistically for systematic differences associated with country while preserving the treatment comparison within the specified model.
7. Non-Inferiority Framework
The primary analyses were conducted under a non-inferiority or equivalence hypothesis framework. For both semaglutide comparisons, the registry specifies a non-inferiority margin of 0.3 percentage points for the treatment difference in HbA1c.
Prespecified non-inferiority rule
Non-inferiority was concluded if the upper limit of the two-sided 95% confidence interval for the estimated treatment difference between semaglutide and sitagliptin was below the prespecified non-inferiority margin.
This criterion is fundamentally different from asking only whether a conventional null-hypothesis p-value is below a threshold. The non-inferiority conclusion depends on the entire confidence interval relative to the prespecified margin.
Why the margin matters
The 0.3 percentage-point margin defines how much difference could be tolerated while still meeting the registry's stated non-inferiority criterion.
Why the upper bound matters
The upper confidence-limit rule asks whether the plausible range of the treatment difference remains below the prespecified margin.
8. Primary Results: Semaglutide 1.0 mg vs Sitagliptin
The first posted primary analysis compares semaglutide 1.0 mg plus sitagliptin placebo with sitagliptin plus semaglutide placebo for change in HbA1c from baseline to week 56.
Treatment difference in HbA1c change
95% CI: -1.21% to -0.91% · P < 0.0001
Two-sided 95% confidence interval; non-inferiority margin: 0.3 percentage points.
| Analysis element | Semaglutide 1.0 mg vs sitagliptin |
|---|---|
| Endpoint | Change in HbA1c from baseline to week 56 |
| Analysis population | Full analysis set; missing data imputed from mixed model for repeated measurements |
| Method | Mixed Models Analysis |
| Effect measure | Treatment difference |
| Estimate | -1.06 percentage points |
| 95% CI | -1.21 to -0.91 percentage points |
| P-value | <0.0001 |
| Non-inferiority margin | 0.3 percentage points |
The estimated treatment difference was -1.06 percentage points, meaning that the model estimated a 1.06-percentage-point greater reduction in HbA1c for semaglutide 1.0 mg relative to sitagliptin under the specified analysis framework.
The negative direction is important because the endpoint is change from baseline: a more negative treatment difference corresponds to a greater reduction in HbA1c in the semaglutide group relative to the comparator.
The 95% confidence interval from -1.21 to -0.91 describes statistical uncertainty around the estimated treatment difference. It does not mean that individual patients' HbA1c changes are confined to that interval.
The P < 0.0001 value addresses the statistical evidence under the corresponding hypothesis test; it does not measure the magnitude or clinical importance of the treatment effect. The estimate and its confidence interval are needed to understand the size and precision of the difference.
For the non-inferiority question, the key feature is that the upper confidence limit of -0.91% is below the prespecified 0.3% margin. The registry therefore states the non-inferiority criterion in terms of the confidence interval rather than relying on the p-value alone.
9. Primary Results: Semaglutide 0.5 mg vs Sitagliptin
The second posted primary analysis compares semaglutide 0.5 mg plus sitagliptin placebo with sitagliptin plus semaglutide placebo for the same change-from-baseline HbA1c endpoint at week 56.
Treatment difference in HbA1c change
95% CI: -1.21% to -0.62% · P < 0.0001
Two-sided 95% confidence interval; non-inferiority margin: 0.3 percentage points.
| Analysis element | Semaglutide 0.5 mg vs sitagliptin |
|---|---|
| Endpoint | Change in HbA1c from baseline to week 56 |
| Analysis population | Full analysis set; missing data imputed from mixed model for repeated measurements |
| Method | Mixed Models Analysis |
| Effect measure | Treatment difference |
| Estimate | -0.92 percentage points |
| 95% CI | -1.21 to -0.62 percentage points |
| P-value | <0.0001 |
| Non-inferiority margin | 0.3 percentage points |
The estimated treatment difference was -0.92 percentage points. Under the specified mixed-model analysis, the estimated change in HbA1c therefore favored a greater reduction with semaglutide 0.5 mg relative to sitagliptin by 0.92 percentage points.
The 95% confidence interval extends from -1.21 to -0.62 percentage points. This interval represents uncertainty around the estimated treatment difference; it is not an interval describing the individual-level response of patients.
The P < 0.0001 value provides evidence against the corresponding null hypothesis, but a p-value is not an effect-size measure. The treatment difference and confidence interval provide the information about magnitude and precision.
For non-inferiority, the relevant comparison is again between the confidence interval and the prespecified 0.3-percentage-point margin. The upper confidence limit of -0.62% is below that margin, satisfying the registry's stated non-inferiority criterion.
10. Primary Results Side by Side
| Comparison | Estimate | 95% CI | P-value | NI margin |
|---|---|---|---|---|
| Semaglutide 1.0 mg vs sitagliptin | -1.06% | -1.21% to -0.91% | <0.0001 | 0.3% |
| Semaglutide 0.5 mg vs sitagliptin | -0.92% | -1.21% to -0.62% | <0.0001 | 0.3% |
Both posted analyses use the same endpoint, time frame, model family, and non-inferiority framework. The difference between the analyses is the semaglutide dose being compared with sitagliptin.
The bars above are a visual representation of the two reported point estimates; they should not be interpreted as displaying the full confidence intervals or the distribution of individual HbA1c responses.
11. How to Interpret a Negative Treatment Difference
Because the endpoint is change in HbA1c from baseline, the sign of the treatment difference carries directional information. A negative value means that the estimated change under semaglutide was lower than the estimated change under sitagliptin.
The value -1.06% does not mean that every participant experienced exactly a 1.06-percentage-point reduction, nor does it describe the full range of individual responses. It is a treatment-group contrast estimated by the specified statistical model.
The interval -1.21% to -0.91% gives a measure of precision around the estimated treatment difference under the model and statistical framework. A narrower interval generally conveys greater precision than a wider interval, although precision and clinical importance are distinct concepts.
12. Statistical Methods Explained
Why was a mixed-effects model used?
The registry specifies a mixed model for repeated measurements because the HbA1c endpoint is evaluated over visits and the analysis is explicitly structured within visit. This framework allows the treatment comparison to be estimated using the longitudinal measurements while incorporating the specified fixed factors and baseline covariate.
Why was baseline HbA1c included as a covariate?
Baseline HbA1c was included to adjust the model-based treatment comparison for the measured baseline value. This is different from simply comparing the two groups' observed follow-up measurements without accounting for baseline.
What does a treatment difference of -1.06% mean?
It means the estimated change in HbA1c was 1.06 percentage points lower with semaglutide 1.0 mg than with sitagliptin under the posted model. The negative sign describes the direction of the contrast; it is not a probability and is not a hazard ratio.
Why is the confidence interval central to non-inferiority?
Non-inferiority is defined relative to a prespecified margin. Here, the registry states that non-inferiority was concluded when the upper limit of the two-sided 95% confidence interval was below 0.3 percentage points. The confidence interval therefore directly determines whether the estimated uncertainty remains within the allowed margin.
Why does the p-value not measure effect size?
The p-value describes the statistical evidence associated with a hypothesis test. It does not tell the reader whether a treatment difference is large or small. The treatment estimate provides the effect magnitude, while the confidence interval communicates uncertainty around that estimate.
Why does missing-data handling matter?
The primary analysis states that missing data were imputed from the mixed model for repeated measurements. Because longitudinal clinical-trial datasets can contain measurements that are not observed at every visit, the way missing information is handled can influence the resulting estimate. The registry's posted analysis therefore incorporates missing-data handling as part of the model-based primary analysis.
13. Missing Data and Model-Based Imputation
The primary analysis population is defined as the FAS, while the analysis description states that missing data were imputed from a mixed model for repeated measurements. This is an important part of the statistical specification rather than an administrative detail.
The model incorporates treatment, country, baseline value, and visit structure, with the registry specifying model-based handling of missing data.
A model-based analysis can use information from the longitudinal structure of the trial rather than treating the week-56 observation as the only relevant measurement. The validity of such an approach depends on the statistical assumptions underlying the model and its missing-data framework.
14. Non-Inferiority: Margin Versus Estimate
The non-inferiority margin in both posted primary analyses is 0.3%. The treatment estimates and their upper confidence limits are substantially below that boundary.
| Comparison | Estimate | Upper 95% CI | Non-inferiority margin |
|---|---|---|---|
| Semaglutide 1.0 mg vs sitagliptin | -1.06% | -0.91% | 0.3% |
| Semaglutide 0.5 mg vs sitagliptin | -0.92% | -0.62% | 0.3% |
The important statistical distinction is between estimating a treatment difference and testing whether that difference is compatible with the non-inferiority margin. A point estimate alone cannot establish non-inferiority because uncertainty must also be considered.
Point estimate
Summarizes the estimated treatment difference under the specified statistical model.
Confidence interval
Describes uncertainty around the treatment-difference estimate and determines the non-inferiority conclusion under the registry's stated rule.
Margin
Defines the prespecified boundary used to judge non-inferiority.
P-value
Provides hypothesis-testing evidence but does not replace the confidence-interval comparison required by the stated non-inferiority criterion.
15. Safety Results
The registry reports serious adverse events by treatment arm. The available safety summaries are counts of affected subjects divided by subjects at risk for serious adverse events.
| Treatment arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Semaglutide 1.0 mg + sitagliptin placebo | 30 | 30/409 |
| Semaglutide 0.5 mg + sitagliptin placebo | 30 | 30/409 |
| Sitagliptin 100 mg + semaglutide placebo | 13 | 13/204 |
| Sitagliptin 100 mg + semaglutide placebo | 16 | 16/203 |
The serious-adverse-event summaries should be interpreted as safety counts within the four registry arms. They are not part of the primary HbA1c hypothesis test and should not be combined with the efficacy estimate into a single statistical measure.
16. Design Features That Matter Statistically
| Design feature | How it appears in SUSTAIN™ 2 | Statistical relevance |
|---|---|---|
| Randomization | Randomized allocation | Provides the basis for the comparative treatment analysis. |
| Blinding | Double | Reduces opportunities for knowledge of treatment assignment to influence trial conduct and assessment. |
| Parallel design | Parallel | Subjects remain in their randomized treatment groups rather than serving as sequential treatment comparisons. |
| Four arms | 4 treatment arms | Allows the posted comparisons to evaluate two semaglutide doses against the sitagliptin comparison. |
| Non-inferiority | 0.3% margin | Requires interpretation of the confidence interval relative to a prespecified boundary. |
| Mixed-effects model | Mixed model for repeated measurements | Accounts for the longitudinal analysis structure specified by the registry. |
| Covariate adjustment | Baseline value included | Adjusts the model-based comparison for the baseline HbA1c value. |
| Missing-data handling | Mixed-model imputation | Defines how missing post-baseline information enters the primary analysis. |
17. Why Randomization and Blinding Matter
Randomization establishes the treatment assignment mechanism underlying the comparative analysis. The statistical estimate is therefore interpreted as a between-treatment comparison within a randomized trial rather than as an observational association.
Double blinding adds another layer of protection against treatment knowledge affecting trial behavior or assessments. For an HbA1c endpoint, the laboratory measurement itself is quantitative, but blinding remains a feature of the broader trial conduct.
The mixed-effects model does not replace randomization. The model is the method used to estimate the treatment difference; randomization is the design feature that supports the causal interpretation of the randomized comparison.
18. Understanding the Two Primary Comparisons
The registry contains two formal primary statistical analyses of the same endpoint and time frame. One evaluates semaglutide 1.0 mg against sitagliptin, while the other evaluates semaglutide 0.5 mg against sitagliptin.
| Dimension | Semaglutide 1.0 mg comparison | Semaglutide 0.5 mg comparison |
|---|---|---|
| Endpoint | Change in HbA1c from baseline | Change in HbA1c from baseline |
| Time frame | Week 0, week 56 | Week 0, week 56 |
| Comparator | Sitagliptin | Sitagliptin |
| Model | Mixed model for repeated measurements | Mixed model for repeated measurements |
| Estimate | -1.06% | -0.92% |
| 95% CI | -1.21% to -0.91% | -1.21% to -0.62% |
| P-value | <0.0001 | <0.0001 |
| Non-inferiority margin | 0.3% | 0.3% |
The two analyses should be kept distinct. Their estimates have different treatment comparisons even though they share the same endpoint and analysis framework. The registry's posted results support separate interpretation of each dose comparison.
19. Confidence Intervals in the SUSTAIN™ 2 Analysis
The two-sided 95% confidence intervals are particularly informative because the trial uses a non-inferiority framework. The intervals quantify uncertainty around the treatment-difference estimates and are directly connected to the prespecified non-inferiority rule.
1.0 mg comparison
The estimated difference is -1.06%, with a 95% CI from -1.21% to -0.91%.
0.5 mg comparison
The estimated difference is -0.92%, with a 95% CI from -1.21% to -0.62%.
Neither interval comes close to the upper non-inferiority boundary of 0.3%. That relationship is the central statistical feature of the posted non-inferiority analyses.
20. P-Values and Effect Size
Both posted primary analyses report P < 0.0001. This provides strong statistical evidence against the corresponding null hypothesis, but the p-value should not be used as a substitute for the estimated treatment difference or its confidence interval.
The estimate describes the magnitude and direction of the treatment contrast; the confidence interval describes uncertainty around that estimate; the p-value describes evidence against a specified null hypothesis.
For this trial, the non-inferiority conclusion adds a fourth essential element: the prespecified 0.3% margin. The statistical interpretation is therefore not complete from the p-value alone.
21. Interpretation of the Primary Endpoint
The posted primary analyses estimate negative treatment differences in change in HbA1c for both semaglutide 1.0 mg and semaglutide 0.5 mg relative to sitagliptin. Both two-sided 95% confidence intervals have upper limits below the prespecified 0.3-percentage-point non-inferiority margin.
The treatment-difference estimates do not describe the response of every individual subject. They also do not convert the HbA1c result into a probability of an individual patient achieving a particular outcome.
The estimates arise from a mixed model for repeated measurements with treatment and country as fixed factors and baseline value as a covariate, nested within visit. Consequently, the numbers should be interpreted within that prespecified analytical framework rather than as simple unadjusted differences.
22. Limitations
The registry's limitations field reports no limitations or caveats. Even so, several statistical interpretation points follow directly from the structure of the posted analyses and are important when reading the results.
- Model dependence: the treatment estimates are generated from a mixed model for repeated measurements and therefore depend on the specified longitudinal model.
- Missing-data assumptions: the primary analysis uses model-based imputation for missing data, so the interpretation depends on the assumptions underlying that approach.
- Non-inferiority margin: the clinical meaning of a non-inferiority result depends on the prespecified 0.3-percentage-point margin and the rationale for that margin.
- Multiple primary comparisons: the registry posts two primary analyses comparing two semaglutide doses with the sitagliptin comparison. Their interpretation should remain tied to the prespecified statistical framework rather than treating the two estimates as one combined effect.
- Analysis population: the primary endpoint uses the FAS of 1225 rather than the total enrollment of 1231.
- Effect measure: the treatment difference is a group-level HbA1c contrast and should not be interpreted as an individual-level treatment response.
- Safety and efficacy are distinct: serious adverse-event counts provide safety information but are not incorporated into the HbA1c efficacy estimate.
23. Why This Trial Matters Statistically
SUSTAIN™ 2 is a useful statistical teaching case because the primary endpoint is analyzed with a longitudinal mixed-effects framework while the treatment question is explicitly framed as non-inferiority. That combination requires readers to understand the endpoint, model, confidence interval, missing-data handling, and non-inferiority margin together.
| Concept | How it appears in SUSTAIN™ 2 |
|---|---|
| Randomization | Subjects were randomized in a phase 3 parallel-group trial. |
| Blinding | The trial was double-blind. |
| Multiple treatment arms | Four arms were included. |
| Repeated measurements | The primary analysis used a mixed model for repeated measurements. |
| Covariate adjustment | Baseline HbA1c was included as a covariate. |
| Country adjustment | Country was included as a fixed factor. |
| Missing data | Missing data were imputed from the mixed model for repeated measurements. |
| Non-inferiority | The upper limit of the two-sided 95% CI was compared with a 0.3% margin. |
| Confidence intervals | Both primary treatment differences were reported with two-sided 95% CIs. |
| P-values | Both posted primary analyses reported P < 0.0001. |
| Safety analysis | Serious adverse events were reported separately by treatment arm. |
24. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: NCT01930188.
- PubMed: PMID 28385659.
- PubMed: PMID 29748996.
- PubMed: PMID 29766634.
- PubMed: PMID 29764222.
- PubMed: PMID 29862621.
Continue through the Clinical Biostats statistical library
Connect this trial's design and analysis methods with deeper tutorials and statistical tools.
27. Record Summary
SUSTAIN™ 2 provides a clear example of a randomized phase 3 non-inferiority analysis in which the primary endpoint is a longitudinal change-from-baseline measurement. The registry's two posted primary analyses use mixed models for repeated measurements with treatment and country as fixed factors and baseline value as a covariate, together with model-based handling of missing data. The resulting treatment differences are -1.06% for semaglutide 1.0 mg versus sitagliptin and -0.92% for semaglutide 0.5 mg versus sitagliptin, with two-sided 95% confidence intervals of -1.21% to -0.91% and -1.21% to -0.62%, respectively. Both analyses report P < 0.0001, and both use a prespecified non-inferiority margin of 0.3%.