This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
SUSTAIN-7 was a randomized, parallel, open-label phase 3 trial evaluating semaglutide versus dulaglutide as add-on to metformin in subjects with type 2 diabetes. The registry reports 1201 enrolled subjects across four arms and posts formal statistical analyses for the primary HbA1c endpoint and secondary body-weight endpoint.
| Feature | SUSTAIN-7 |
|---|---|
| Trial name | SUSTAIN-7 |
| Phase | Phase 3 |
| Therapeutic area | Diabetes |
| Condition | Diabetes; Diabetes Mellitus, Type 2 |
| Population | Subjects with type 2 diabetes receiving the study interventions as add-on to metformin |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1201 |
| Interventions | Semaglutide; Dulaglutide |
| Results posted | Yes |
| Primary endpoint analyses | 4 |
| Primary analyses with estimate + confidence interval | 4 |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT02648204 |
2. Clinical Question
The central statistical question was whether semaglutide produced a greater improvement in the registered change-in-HbA1c endpoint than corresponding dulaglutide doses at week 40 when both were studied as add-on to metformin in subjects with type 2 diabetes.
Population
Subjects with type 2 diabetes. The registry-reported brief title specifies that the interventions were studied as add-on to metformin.
Intervention
Semaglutide, represented in the posted analyses by 0.5 mg and 1.0 mg groups.
Comparator
Dulaglutide, represented in the posted analyses by 0.75 mg and 1.5 mg groups.
Primary question
How does the change in HbA1c from week 0 to week 40 compare between the corresponding semaglutide and dulaglutide dose groups?
The registry's formal analyses evaluate two dose-to-dose comparisons: semaglutide 0.5 mg versus dulaglutide 0.75 mg, and semaglutide 1.0 mg versus dulaglutide 1.5 mg. Each comparison is reported twice in the statistical-analysis record: once under a non-inferiority hypothesis and once under a superiority hypothesis.
3. Trial Design
Semaglutide
- Semaglutide 0.5 mg
- Semaglutide 1.0 mg
Dulaglutide
- Dulaglutide 0.75 mg
- Dulaglutide 1.5 mg
The ClinicalTrials.gov record identifies four arms and the two dose-matched comparisons used in the posted primary analyses. It does not provide arm-specific enrollment counts for all four arms, so those counts are not inferred here.
4. Endpoints
The registered primary endpoint was Change in HbA1c, measured at week 0 and week 40. The registry defines the posted HbA1c results as being based on HbA1c data from the on-treatment without rescue medication observation period.
| Endpoint | Time frame | Registry definition / analysis detail |
|---|---|---|
| Change in HbA1c | Week 0, week 40 | Results are based on HbA1c data from on-treatment without rescue medication observation period. The on-treatment observation period was the period where the subject was considered to be exposed to trial product. The on-treatment without rescue medication observation period was a subset of the on-treatment observation period where subjects did not receive any non-investigational antidiabetic. |
The registry reports the primary endpoint type in the posted statistical-analysis record as Binary, while the outcome itself is a change in HbA1c expressed as a percentage of HbA1c and analyzed using a mixed models analysis. Because the ClinicalTrials.gov record contains both fields, this page preserves the registry information rather than replacing the endpoint-type label with an inferred classification.
Secondary endpoint represented in the posted analyses
| Endpoint | Time frame | Population / analysis detail |
|---|---|---|
| Change in Body Weight (kg) | Week 0, week 40 | Analysis based on FAS. Number of subjects analysed = number of subjects with available data for body weight. |
5. Analysis Populations and Estimands
The primary HbA1c analyses were based on the FAS, as stated in the registry. The ClinicalTrials.gov record does not expand the FAS acronym or provide a separate numerical count for the FAS in the posted analyses. For body weight, the registry additionally states that the number of subjects analyzed was the number with available body-weight data.
FAS for HbA1c
The posted primary analyses state that analysis was based on FAS. The ClinicalTrials.gov record does not provide a numerical FAS size.
Available data for body weight
The secondary body-weight analyses used the FAS and specified that subjects analyzed were those with available body-weight data.
This distinction matters because a treatment difference is always conditional on the population represented in the analysis. The randomized enrollment of 1201 describes the trial as a whole; it does not automatically mean that 1201 subjects contributed observations to every posted model.
6. Statistical Methodology
Mixed-effects model
All six posted statistical analyses use a Mixed Models Analysis, normalized in the registry data to a mixed-effects model. The method category is identified as longitudinal / mixed models.
A mixed-effects framework can represent repeated observations within subjects while accounting for correlation among measurements from the same subject. The expression above is conceptual; the ClinicalTrials.gov record does not specify the complete model formula, covariance structure, fixed-effect terms, or estimation method.
The important point for this trial is that the posted treatment effect is a treatment difference from a mixed-model analysis rather than a hazard ratio, odds ratio, or ratio of means. The estimate therefore retains the scale of the analyzed outcome: a difference in percentage of HbA1c for the primary endpoint and a difference in kilograms for body weight.
Why a mixed-effects model fits the endpoint structure
The registered HbA1c endpoint is defined using measurements at week 0 and week 40, and the registry describes the method as a longitudinal / mixed-model analysis. Mixed-effects models are useful in this setting because repeated measurements from one subject are generally not independent. A model can use the longitudinal structure while estimating the treatment contrast of interest.
This also changes how the result should be read. The reported -0.40 and -0.41 estimates are not percentages of subjects responding and are not ratios. They are treatment differences on the HbA1c outcome scale.
Two-sided confidence intervals
All four posted primary analyses report 95% two-sided confidence intervals. A confidence interval gives a range of parameter values compatible with the statistical model and data under the stated confidence procedure. It is a statement about uncertainty around the estimated treatment difference, not a range in which individual patients' changes must fall.
On this orientation, a negative estimate means the modeled change was lower in the semaglutide group than in the corresponding dulaglutide group. The clinical meaning of the difference depends on the direction in which change in the endpoint is considered favorable.
Hypothesis testing
The registry labels the primary analyses under both non-inferiority and superiority hypotheses. The same treatment difference and 95% confidence interval are posted for each corresponding dose comparison, while the hypothesis label changes.
This creates an important statistical teaching point: non-inferiority and superiority are not interchangeable questions. A non-inferiority analysis asks whether the treatment is sufficiently close to the comparator according to a prespecified margin. A superiority analysis asks whether the data provide evidence of a difference in the specified direction.
Non-inferiority margin fields
7. Primary Results: HbA1c
The registry contains four formal primary- endpoint analyses: two dose-matched comparisons, each evaluated under both non-inferiority and superiority hypotheses. The analysis population was FAS in all four records, the method was Mixed Models Analysis, and all four analyses reported a 95% two-sided confidence interval.
Semaglutide 0.5 mg vs Dulaglutide 0.75 mg — Non-inferiority
Treatment difference in change in HbA1c
95% CI: -0.55 to -0.25 · P < 0.0001
Hypothesis type: Non-inferiority · Analysis population: FAS
| Feature | Reported value |
|---|---|
| Outcome | Change in HbA1c |
| Time frame | Week 0, week 40 |
| Comparison | Semaglutide 0.5 mg vs Dulaglutide 0.75 mg |
| Method | Mixed Models Analysis |
| Effect measure | Treatment difference |
| Estimate | -0.40 |
| 95% CI | -0.55 to -0.25 |
| P-value | <0.0001 |
| Hypothesis | Non-inferiority |
| Registry non-inferiority margin | 0.4 |
The estimated treatment difference was -0.40 on the HbA1c percentage scale. With the comparison oriented as semaglutide minus corresponding dulaglutide, the negative value indicates a lower modeled change in HbA1c for semaglutide 0.5 mg than for dulaglutide 0.75 mg under the analysis definition.
The 95% CI of -0.55 to -0.25 describes the statistical precision around the estimated treatment difference. Every value in the reported interval is negative, so the uncertainty interval lies on one side of zero. That is different from saying that every patient experienced the same change or that the interval describes individual patient responses.
The P < 0.0001 value addresses evidence against the null hypothesis used for the corresponding statistical test. It does not measure the size of the treatment effect. The effect size is represented by the treatment difference, while the confidence interval communicates its precision.
For non-inferiority, the key comparison is with the prespecified non-inferiority margin, not simply whether a p-value is small. The ClinicalTrials.gov record gives the margin as 0.4 for this analysis. Because the sign convention and exact non-inferiority decision rule are not fully specified in the ClinicalTrials.gov record, the numerical result is reported without reconstructing a protocol-level decision rule beyond the registry's stated hypothesis label.
Semaglutide 0.5 mg vs Dulaglutide 0.75 mg — Superiority
Treatment difference in change in HbA1c
95% CI: -0.55 to -0.25 · P < 0.0001
Hypothesis type: Superiority · Analysis population: FAS
For the superiority analysis, the same estimated treatment difference of -0.40 is reported, with a 95% CI of -0.55 to -0.25 and P < 0.0001. The negative treatment difference indicates that the modeled HbA1c change was lower in the semaglutide 0.5 mg group under the registry's treatment-difference orientation.
The confidence interval provides substantially more information about the magnitude of the difference than the p-value alone. Its width reflects statistical precision, while its location relative to zero indicates the direction of the estimated contrast.
The result should not be interpreted as a statement that semaglutide lowered HbA1c by exactly 0.40 percentage points in every individual. It is a model-based treatment contrast. Likewise, the p-value does not establish clinical importance; statistical evidence and clinical magnitude are related but distinct concepts.
Semaglutide 1.0 mg vs Dulaglutide 1.5 mg — Non-inferiority
Treatment difference in change in HbA1c
95% CI: -0.57 to -0.25 · P < 0.0001
Hypothesis type: Non-inferiority · Analysis population: FAS
| Feature | Reported value |
|---|---|
| Outcome | Change in HbA1c |
| Time frame | Week 0, week 40 |
| Comparison | Semaglutide 1.0 mg vs Dulaglutide 1.5 mg |
| Method | Mixed Models Analysis |
| Effect measure | Treatment difference |
| Estimate | -0.41 |
| 95% CI | -0.57 to -0.25 |
| P-value | <0.0001 |
| Hypothesis | Non-inferiority |
| Registry non-inferiority margin | 0.04 |
The estimated treatment difference was -0.41, with a 95% CI from -0.57 to -0.25. As with the lower-dose comparison, the negative estimate represents a lower modeled change in HbA1c for semaglutide relative to the corresponding dulaglutide group under the registry's treatment-difference orientation.
The confidence interval is entirely below zero, providing a precise estimate of a negative treatment contrast within the stated model. The interval does not describe the distribution of individual patients' HbA1c changes, nor does it establish that the true effect must be exactly centered at -0.41.
The registry gives a non-inferiority margin of 0.04 for this analysis. That value should be interpreted as a registry-reported design field, not independently reconstructed from the observed estimate. The ClinicalTrials.gov record does not provide the full sign convention or decision rule needed to reproduce the protocol-level non-inferiority conclusion from first principles.
The P < 0.0001 result should again be separated from the effect estimate. A very small p-value indicates strong statistical evidence under the relevant hypothesis test; it does not tell us whether a difference is large, small, or clinically important.
Semaglutide 1.0 mg vs Dulaglutide 1.5 mg — Superiority
Treatment difference in change in HbA1c
95% CI: -0.57 to -0.25 · P < 0.0001
Hypothesis type: Superiority · Analysis population: FAS
The superiority analysis reports the same treatment difference of -0.41 and the same 95% CI of -0.57 to -0.25, with P < 0.0001. The result indicates a negative modeled treatment contrast for change in HbA1c under the registry-reported analysis definition.
The important statistical distinction is the hypothesis label. The numerical estimate itself does not change merely because the analysis is described under a superiority hypothesis. The inferential question changes: superiority concerns evidence for a treatment difference in the specified direction, whereas non-inferiority concerns whether the treatment remains within a prespecified acceptable margin.
Because the registry posts both hypothesis types for each dose comparison, readers should not collapse the four analyses into a single p-value or treat the two dose comparisons as though they were one analysis. They represent two distinct treatment contrasts and two distinct hypothesis labels.
8. Primary HbA1c Results in One View
| Comparison | Hypothesis | Treatment difference | 95% CI | P-value |
|---|---|---|---|---|
| Semaglutide 0.5 mg vs Dulaglutide 0.75 mg | Non-inferiority | -0.40 | -0.55 to -0.25 | <0.0001 |
| Semaglutide 0.5 mg vs Dulaglutide 0.75 mg | Superiority | -0.40 | -0.55 to -0.25 | <0.0001 |
| Semaglutide 1.0 mg vs Dulaglutide 1.5 mg | Non-inferiority | -0.41 | -0.57 to -0.25 | <0.0001 |
| Semaglutide 1.0 mg vs Dulaglutide 1.5 mg | Superiority | -0.41 | -0.57 to -0.25 | <0.0001 |
Two features stand out statistically. First, the two dose-matched comparisons produce very similar treatment-difference estimates. Second, the corresponding confidence intervals remain entirely below zero. These observations describe the posted estimates; they should not be expanded into claims about an interaction between dose levels because the ClinicalTrials.gov record does not report a formal interaction analysis.
9. Secondary Results: Change in Body Weight
The registry posts two secondary analyses for change in body weight from week 0 to week 40. Both use the FAS, restrict the analyzed subjects to those with available body-weight data, and use a mixed models analysis with treatment difference as the effect measure.
Semaglutide 0.5 mg vs Dulaglutide 0.75 mg
Treatment difference in body weight
95% CI: -3.02 to -1.51 · P < 0.0001
Hypothesis type: Superiority · Analysis population: FAS with available body-weight data
The estimated treatment difference was -2.26 kg. With semaglutide minus dulaglutide as the treatment-difference orientation, the negative estimate indicates a lower modeled change in body weight for semaglutide 0.5 mg relative to dulaglutide 0.75 mg.
The 95% CI of -3.02 to -1.51 kg expresses uncertainty around that estimated contrast. It does not mean that individual patients' weight changes must lie between -3.02 kg and -1.51 kg. It describes uncertainty in the treatment comparison estimated by the statistical analysis.
The p-value of <0.0001 is evidence from the corresponding superiority test; it is not a measure of the practical magnitude of the weight difference. The magnitude is described by the treatment difference itself.
Semaglutide 1.0 mg vs Dulaglutide 1.5 mg
Treatment difference in body weight
95% CI: -4.32 to -2.78 · P < 0.0001
Hypothesis type: Superiority · Analysis population: FAS with available body-weight data
The estimated treatment difference was -3.55 kg, with a 95% CI of -4.32 to -2.78 kg. The negative value indicates a lower modeled change in body weight for semaglutide 1.0 mg than for dulaglutide 1.5 mg under the posted treatment-difference orientation.
The interval is entirely negative and relatively narrow compared with the magnitude of the estimate, which indicates that the posted model produced a comparatively precise estimate of the between-group contrast. Precision should not be confused with certainty about individual responses: patients can vary substantially even when a population-level treatment difference is estimated precisely.
The P < 0.0001 value supports the corresponding statistical test but does not quantify how meaningful a 3.55 kg treatment difference is in an individual patient. Clinical interpretation requires context beyond the p-value.
Secondary results together
| Comparison | Estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Semaglutide 0.5 mg vs Dulaglutide 0.75 mg | -2.26 kg | -3.02 to -1.51 | <0.0001 | Superiority |
| Semaglutide 1.0 mg vs Dulaglutide 1.5 mg | -3.55 kg | -4.32 to -2.78 | <0.0001 | Superiority |
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events by arm as affected subjects over subjects at risk. These figures provide an arm-specific safety summary without requiring an inferred percentage.
| Arm | Serious adverse events affected / at risk |
|---|---|
| Semaglutide 0.5 mg | 17/301 |
| Semaglutide 1.0 mg | 23/300 |
| Dulaglutide 0.75 mg | 24/299 |
| Dulaglutide 1.5 mg | 22/299 |
The graphic above is scaled to the largest affected-subject count for visual comparison; it is not a risk estimate. The ClinicalTrials.gov record provides the numerator and denominator for each arm, but this page does not convert those values into additional percentages because the requested data rules specify that numbers should be reported exactly as reported in the registry.
11. Statistical Methods Explained
Why was a mixed-effects model used?
The registry explicitly reports a Mixed Models Analysis and classifies the method as a mixed-effects model within the longitudinal / mixed-model family. This is appropriate to the structure of a clinical trial in which outcomes such as HbA1c can be observed repeatedly over time. A mixed model can represent within-subject correlation rather than treating repeated measurements from the same subject as independent.
The important caution is that the ClinicalTrials.gov record does not identify the exact covariance structure or every fixed and random effect. Therefore, the conceptual model explains why the method is useful without pretending that the complete statistical model has been recovered.
What does a treatment difference of -0.40 mean?
A treatment difference of -0.40 means the model estimated the semaglutide group to have a change in HbA1c that was 0.40 percentage points lower than the corresponding dulaglutide group, using the treatment-difference orientation implied by the posted comparison. It is a group-level model estimate, not an individual-level prediction.
The same distinction applies to body weight. A treatment difference of -2.26 kg means a modeled between-group difference of 2.26 kg in the negative direction; it does not say that every person lost exactly 2.26 kg more.
Why does the confidence interval matter?
The confidence interval communicates uncertainty around the estimated treatment difference. For the semaglutide 0.5 mg versus dulaglutide 0.75 mg HbA1c comparison, the estimate is -0.40 and the 95% CI is -0.55 to -0.25. The interval therefore gives a statistical precision statement that cannot be obtained from the point estimate alone.
A confidence interval is not a prediction interval for individual subjects. It should also not be interpreted as assigning a probability that a fixed parameter lies inside the interval after the data have been observed.
Why does the p-value not measure effect size?
A p-value measures the compatibility of the observed data with a specified null hypothesis under the statistical testing framework. It does not tell us how large the treatment difference is. That is why the reported P < 0.0001 should always be read alongside the treatment estimate and its confidence interval.
Two studies can have the same p-value while having very different effect sizes, or a clinically meaningful estimate can fail to achieve a small p-value in a smaller study. Statistical evidence and effect magnitude answer different questions.
Why is non-inferiority judged against a margin?
Non-inferiority is fundamentally a margin-based question. The objective is not simply to demonstrate that a treatment differs from zero. Instead, the prespecified margin defines how much worse the new treatment could be, under the relevant direction and scale, while still meeting the non-inferiority criterion.
For SUSTAIN-7, the registry fields report a non-inferiority margin of 0.4 for the 0.5 mg comparison and 0.04 for the 1.0 mg comparison. Because the complete protocol-level decision rule and sign convention are not reported, the exact non-inferiority decision procedure should not be reconstructed beyond those registry fields.
What does FAS mean for interpretation?
The posted primary analyses state that the analysis was based on FAS. That means the reported treatment differences are estimates for the registry-defined analysis population, not necessarily simple calculations using every one of the 1201 enrolled subjects.
This matters because the distinction between enrollment, randomization, treatment exposure, and analysis population can affect both precision and interpretation. For body weight, the registry explicitly adds that the number of subjects analyzed was the number with available body-weight data.
Why should the two dose comparisons not be collapsed into one result?
The registry contains two distinct treatment comparisons: semaglutide 0.5 mg versus dulaglutide 0.75 mg and semaglutide 1.0 mg versus dulaglutide 1.5 mg. Their estimates happen to be close, but they are separate contrasts. Treating them as one analysis would obscure the dose-specific comparisons actually reported by the registry.
12. Non-Inferiority and Superiority: Reading the Same Estimate Differently
SUSTAIN-7 is especially useful statistically because the analyses posted on ClinicalTrials.gov explicitly pair non-inferiority and superiority hypotheses. The numerical estimate is not itself a label of non-inferiority or superiority. The inferential claim depends on the hypothesis being tested and the prespecified decision framework.
| Comparison | Estimate | Non-inferiority analysis | Superiority analysis |
|---|---|---|---|
| Semaglutide 0.5 mg vs Dulaglutide 0.75 mg | -0.40 | 95% CI -0.55 to -0.25; P < 0.0001; margin 0.4 | 95% CI -0.55 to -0.25; P < 0.0001 |
| Semaglutide 1.0 mg vs Dulaglutide 1.5 mg | -0.41 | 95% CI -0.57 to -0.25; P < 0.0001; margin 0.04 | 95% CI -0.57 to -0.25; P < 0.0001 |
A superiority test asks whether the data support a difference in the specified direction. A non-inferiority test asks whether the treatment difference remains within a prespecified acceptable boundary. Therefore, a small p-value by itself is not the definition of non-inferiority.
The confidence interval is particularly important in non-inferiority work because its position relative to the non-inferiority margin determines whether the data satisfy the margin-based criterion. The ClinicalTrials.gov record does not provide enough information to reconstruct every detail of that decision rule, so the registry's reported hypothesis labels and margins are preserved without adding assumptions.
13. Missing Data and Rescue Medication
The registered HbA1c definition specifies that results are based on the on-treatment without rescue medication observation period. The registry-reported definition explains that this is a subset of the on-treatment observation period during which subjects did not receive any non-investigational antidiabetic.
This is an important part of the estimand because the analyzed data are not simply an unrestricted collection of all HbA1c observations from enrollment through week 40. The observation-period definition identifies which measurements contribute to the posted result.
On-treatment
The registry defines this as the period during which the subject was considered to be exposed to trial product.
Without rescue medication
This is a subset of the on-treatment period in which subjects did not receive any non-investigational antidiabetic.
The ClinicalTrials.gov record does not describe a separate imputation algorithm, pattern-mixture model, multiple-imputation procedure, or other formal missing-data method. Accordingly, none is attributed to SUSTAIN-7 on this page.
14. Multiplicity and Multiple Comparisons
The registry posts four primary statistical analyses: two dose-matched comparisons, each evaluated under non-inferiority and superiority hypotheses. It also posts two secondary body-weight analyses. These multiple analyses create an inferential structure that deserves attention.
| Analysis family | Number posted | What is known from the ClinicalTrials.gov record |
|---|---|---|
| Primary HbA1c analyses | 4 | Two dose-matched treatment comparisons, each represented under non-inferiority and superiority hypotheses. |
| Secondary body-weight analyses | 2 | One analysis for each dose-matched comparison, both under superiority. |
| Total statistical analyses | 6 | All six use Mixed Models Analysis. |
The ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure, alpha allocation across the hypotheses, a hierarchical testing sequence, or an overall familywise error strategy. Therefore, the individual reported p-values are presented exactly as posted, but this page does not claim that all six analyses represent independent confirmatory tests at a common type I error level.
15. Randomization and Causal Interpretation
Randomization is one of the central statistical features of SUSTAIN-7. The registry records the allocation as randomized and the design model as parallel. Randomization creates the framework for comparing outcomes between treatment groups while reducing systematic allocation differences that would otherwise threaten causal interpretation.
The actual posted estimate is produced by the mixed-model analysis rather than by this simple arithmetic expression. The expression is useful for understanding the target contrast represented by a treatment-difference estimate.
Randomization does not guarantee that every baseline characteristic will be numerically identical between groups. More importantly for this page, the ClinicalTrials.gov record does not provide a baseline-characteristics table, so no numerical baseline balance assessment is attempted.
The open-label design is also relevant. Because masking is recorded as none, the treatment comparison is not insulated from every possible effect of treatment awareness. That does not invalidate the randomized comparison, but it is a design characteristic that should remain visible when interpreting subjective outcomes or treatment behavior.
16. Results Visualization: Treatment Differences
The chart is a visual representation of the two posted point estimates, not a confidence-interval plot. Because the two estimates are on the same outcome scale, the display makes their numerical similarity easy to see. It does not establish that the two treatment effects are statistically identical; such a comparison would require a formal analysis of their difference.
The body-weight display similarly shows the posted point estimates on a common scale. It should not be read as evidence of a dose-response interaction because the ClinicalTrials.gov record does not report a formal test comparing the two treatment contrasts.
17. Limitations
- Registry-level detail: this analysis is limited to the ClinicalTrials.gov record. The complete statistical analysis plan, protocol, and underlying patient-level data are not provided here.
- Non-inferiority margins: the ClinicalTrials.gov record reports different margins, 0.4 and 0.04, for the two dose comparisons. They are reproduced exactly rather than reconciled through an unsupported assumption.
- Incomplete model specification: the registry identifies a mixed-effects model but does not provide the full fixed-effect specification, covariance structure, random-effect structure, or estimation details in the ClinicalTrials.gov record.
- Analysis-population detail: the primary analyses are identified as FAS, but the ClinicalTrials.gov record does not provide a numerical FAS size. Body-weight analyses additionally require available body-weight data.
- Missing-data methods: the ClinicalTrials.gov record defines the HbA1c observation period but does not report a separate imputation strategy. No unreported imputation method is assumed here.
- Multiplicity: four primary analyses and two secondary analyses are posted, but the ClinicalTrials.gov record does not state the complete multiplicity-control strategy or testing hierarchy.
- Open-label design: masking is recorded as none, so treatment awareness is a relevant design consideration.
- Endpoint-type field: the posted primary statistical-analysis record labels the endpoint type as Binary even though the outcome is a change in HbA1c analyzed as a treatment difference. This page preserves the registry-reported field rather than silently correcting it.
- Safety comparisons: serious adverse-event counts are reported by arm, but the ClinicalTrials.gov record does not provide a formal statistical comparison, exposure-adjusted rate, or detailed event definitions.
- No subgroup analysis in the ClinicalTrials.gov record: the provided record does not contain subgroup estimates, so no subgroup conclusions are drawn.
- No baseline table in the ClinicalTrials.gov record: baseline characteristics are not reported and therefore are not reconstructed or inferred.
18. Why This Trial Matters Statistically
SUSTAIN-7 is a useful teaching example because its registry record places several fundamental clinical-trial concepts next to one another: randomization, parallel-group comparison, longitudinal outcomes, mixed-effects modeling, treatment differences, confidence intervals, non-inferiority, superiority, secondary endpoints, and safety summaries.
| Concept | How it appears in SUSTAIN-7 |
|---|---|
| Randomization | The registry records randomized allocation. |
| Parallel design | The design model is recorded as parallel with four arms. |
| Mixed-effects model | All six posted statistical analyses use Mixed Models Analysis. |
| Longitudinal analysis | HbA1c and body weight are assessed from week 0 to week 40. |
| Treatment difference | The primary and secondary posted effect measure is a treatment difference. |
| Confidence interval | All four primary analyses report 95% two-sided confidence intervals. |
| Non-inferiority | Both dose-matched HbA1c comparisons are posted under a non-inferiority hypothesis. |
| Superiority | Both HbA1c comparisons and both body-weight comparisons are posted under superiority hypotheses. |
| Analysis population | Primary analyses are based on FAS; body-weight analyses require available body-weight data. |
| Multiplicity | Four primary analyses and two secondary statistical analyses are posted. |
| Safety | Serious adverse events are reported as affected subjects over subjects at risk by arm. |
The most instructive feature is the relationship between effect estimate, confidence interval, and hypothesis type. A reader who focuses only on p-values misses much of the statistical information. A reader who focuses only on the treatment difference misses the inferential framework. The strongest interpretation reads the estimate, uncertainty interval, hypothesis, analysis population, and model together.
19. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
20. Related Statistical Calculators
21. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The posted analyses estimate negative treatment differences for both dose-matched HbA1c comparisons and both body-weight comparisons. The primary HbA1c analyses report 95% two-sided confidence intervals and P < 0.0001 under both non-inferiority and superiority hypothesis labels.
Clinical interpretation
The numerical estimates describe between-group differences in HbA1c and body weight. Whether a particular magnitude represents an important clinical difference is a separate question from whether the statistical model provides strong evidence against its null hypothesis.
This separation is particularly important when reading a result such as -0.41. The number communicates the estimated treatment contrast. The 95% CI communicates its statistical precision. The p-value communicates evidence under a specified hypothesis test. None of these quantities alone provides a complete description of an individual patient's response.
22. What the Treatment Difference Does — and Does Not — Mean
A treatment difference is a contrast between groups on the outcome scale. In the primary analysis, a value of -0.40 means the modeled change in HbA1c was 0.40 percentage points lower in the semaglutide 0.5 mg group than in the corresponding dulaglutide group, using the comparison orientation represented in the registry record.
It does not mean that every participant experienced a 0.40 percentage-point difference. It is also not a relative percentage reduction, a probability of response, or a statement about an individual participant's future outcome.
The 95% CI of -0.55 to -0.25 describes uncertainty around the estimated treatment difference under the statistical analysis. A narrower interval indicates greater statistical precision than a wider interval, all else being equal. The interval does not describe the range of individual treatment responses.
The p-value of <0.0001 is evidence under the specified hypothesis test. It is not a measure of how large the treatment difference is. The estimate and confidence interval are required to understand the magnitude and precision of the result.
Non-inferiority requires comparison with a prespecified margin. Therefore, the fact that a p-value is small does not by itself establish non-inferiority. The registry margins are 0.4 and 0.04 for the two primary comparisons and are reported here without modification.
23. Longitudinal Trial History
Trial start
The SUSTAIN-7 trial is recorded as starting on 2016-01-06.
Primary completion
The registry records primary completion on 2017-04-10.
Results posted
The trial is recorded as completed, with 28 outcome measures and 6 statistical analyses posted.
The ClinicalTrials.gov record does not provide additional analysis dates, data-cutoff dates, interim analyses, long-term follow-up results, or later efficacy analyses. Those elements are therefore not added to this page.
24. What the Registry Results Allow Us to Conclude Statistically
The posted evidence supports several precise statistical observations without requiring additional assumptions. The primary HbA1c endpoint was analyzed using a mixed-effects model in the FAS. Both dose-matched comparisons produced negative treatment differences, with the corresponding 95% confidence intervals entirely below zero and p-values reported as less than 0.0001 under both the non-inferiority and superiority hypothesis labels.
The secondary body-weight analyses use the same general mixed-model framework and report negative treatment differences of -2.26 kg and -3.55 kg, respectively, with corresponding 95% confidence intervals and p-values of less than 0.0001.
These results describe the statistical estimates contained in the registry. They do not justify additional claims about unreported subgroups, long-term outcomes, baseline balance, dose interactions, or formal multiplicity control. Keeping those boundaries visible is part of a rigorous trial analysis.
25. Sources
- ClinicalTrials.gov: SUSTAIN-7, NCT02648204. Official trial registry record and source for the trial data summarized on this page.
- PubMed: PMID 29397376.
- PubMed: PMID 29246950.
- PubMed: PMID 29764222.
- PubMed: PMID 29557057.
- PubMed: PMID 30615985.
The numerical results presented above are restricted to the ClinicalTrials.gov record. The PubMed links are provided as source navigation for the linked publications; no additional numerical results from those publications have been incorporated into this analysis.
Continue through the Clinical Biostats statistical pathway
Explore the statistical methods connected with randomized clinical trials, longitudinal outcomes, confidence intervals, non-inferiority testing, p-values, and treatment-effect interpretation.
26. Record Summary
SUSTAIN-7 is a randomized phase 3, parallel, unmasked trial with 1201 enrolled subjects and four arms comparing semaglutide with dulaglutide as add-on to metformin in subjects with type 2 diabetes. The registered primary endpoint is change in HbA1c from week 0 to week 40, with the statistical analyses posted on ClinicalTrials.gov based on FAS and using a mixed-effects model.
The four primary analyses consist of two dose-matched treatment comparisons evaluated under both non-inferiority and superiority hypotheses. Semaglutide 0.5 mg versus dulaglutide 0.75 mg produced a treatment difference of -0.40 with a 95% CI of -0.55 to -0.25 and P < 0.0001. Semaglutide 1.0 mg versus dulaglutide 1.5 mg produced a treatment difference of -0.41 with a 95% CI of -0.57 to -0.25 and P < 0.0001. The registry separately reports non-inferiority margins of 0.4 and 0.04 for those comparisons.
The secondary body-weight analyses report treatment differences of -2.26 kg and -3.55 kg for the two dose-matched comparisons, respectively, both with 95% confidence intervals and P-values of less than 0.0001. Serious adverse events are reported as 17/301, 23/300, 24/299, and 22/299 across the four arms.
Statistically, the central lesson is that these results should be interpreted as model-based treatment differences with quantified uncertainty, not as isolated p-values. The non-inferiority and superiority labels represent different hypotheses, the FAS defines the primary analysis population, the observation-period definition affects the HbA1c data used, and the presence of multiple posted analyses means that the complete testing hierarchy would be necessary for a full confirmatory-error assessment.