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
onset® 1 was a randomized, double-blind, parallel phase 3 trial in adults with type 1 diabetes. The registry reports 1290 enrolled participants and three arms, with the primary endpoint defined as change from baseline in HbA1c after 26 weeks of randomized treatment.
| Feature | onset® 1 |
|---|---|
| Trial name | onset® 1 |
| ClinicalTrials.gov identifier | NCT01831765 |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Endocrinology & Metabolism |
| Condition | Diabetes; Diabetes Mellitus, Type 1 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 1290 |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
2. Clinical Question
The central question was whether faster-acting insulin aspart could produce a change from baseline in HbA1c after 26 weeks that was non-inferior to insulin aspart in the randomized treatment comparison, when both were used in combination with insulin detemir.
Population
Adults with diabetes mellitus, type 1, as described by the registered brief title.
Intervention
Faster-acting insulin aspart, with insulin detemir as part of the treatment regimen.
Comparator
Insulin aspart, with insulin detemir as part of the treatment regimen.
Primary question
Is the difference in change from baseline in HbA1c after 26 weeks within the prespecified non-inferiority limit of 0.4%?
3. Trial Design
The registry describes the study as randomized, double-blind, and parallel, with treatment as the primary purpose. The statistical analyses posted for the primary endpoint used repeated measurements over several visits rather than analyzing only a single week-26 value.
Randomization separates treatment assignment from prognostic characteristics in expectation.
The registry identifies the study as double-blind.
The three treatment arms were evaluated as parallel groups.
Change from baseline after 26 weeks of randomized treatment.
Three-arm structure
The registry identifies three interventions or treatment arms. The posted primary statistical analyses, however, report two specific comparisons: Faster Aspart (Meal) versus NovoRapid (Meal), and Faster Aspart (Post) versus NovoRapid (Meal). These are the comparisons for which the registry provides formal estimates and confidence intervals.
Faster Aspart (Meal)
- Faster-acting insulin aspart
- Insulin detemir included in the treatment regimen
Faster Aspart (Post)
- Faster-acting insulin aspart
- Insulin detemir included in the treatment regimen
NovoRapid (Meal)
- Insulin aspart
- Insulin detemir included in the treatment regimen
4. Primary Endpoint
| Endpoint | Registry definition | Time frame | Posted method |
|---|---|---|---|
| Change From Baseline in HbA1c (Glycosylated Haemoglobin) | Change from baseline in HbA1c after 26 weeks of randomised treatment. | Week 0, week 26 | MMRM with covariate adjustment |
The registry lists one primary endpoint. Although the registry's normalized endpoint-type field describes it as "Binary," the posted statistical analyses use a mean difference in change from baseline and a longitudinal mixed model for repeated measurements. That analytical structure is consistent with the continuous HbA1c measurement itself.
5. Statistical Methodology
Mixed model for repeated measurements
The primary endpoint was analyzed using a mixed effect model for repeated measurements. The analysis included visits 14, 18, 22, 26, 30, 34 and 36. This is important because the model uses information from repeated HbA1c measurements rather than treating week 26 as an isolated observation.
The posted model included treatment, region and strata as fixed effects; subject as a random effect; baseline HbA1c as a covariate; and interactions between the fixed effects and visit and between the covariate and visit.
Covariate adjustment
Baseline HbA1c was included as a covariate. Region and a prespecified stratum variable were also included as fixed effects. The strata combined bolus adjusting method, basal treatment regimen, continuous glucose monitoring and frequently sampled meal test subgroup.
| Model component | Role in the posted analysis |
|---|---|
| Treatment | Represents the randomized treatment comparison. |
| Region | Included as a fixed effect. |
| Strata | Included as a fixed effect; the strata combined bolus adjusting method, basal treatment regimen, CGM and frequently sampled meal test subgroup. |
| Subject | Included as a random effect to account for repeated observations within subjects. |
| Baseline HbA1c | Included as a covariate. |
| Fixed effect × visit | Included to allow modeled effects to vary across visits. |
| Baseline HbA1c × visit | Included to allow the covariate relationship to vary across visits. |
Analysis population
The posted analysis population was the full analysis set (FAS), which included all randomized subjects. For this endpoint, baseline and week 26 were presented, with week 26 described as the end of trial containing the last available measurement.
Effect measure
The reported effect measure was the mean difference (net). In the context of this endpoint, the estimate represents the modeled difference between treatment groups in change from baseline in HbA1c.
6. Non-Inferiority Framework
The primary comparisons were evaluated using a non-inferiority framework. The registry states that the difference between faster aspart and NovoRapid in change from baseline in HbA1c after 26 weeks was compared with a non-inferiority limit of 0.4%.
Prespecified non-inferiority limit
The assessment compared the treatment difference in change from baseline in HbA1c with this limit.
For a continuous endpoint where a larger positive difference represents a potentially less favorable result for the experimental treatment, the essential statistical question is whether the uncertainty interval around the treatment difference remains within the prespecified non-inferiority boundary. The direction of the difference therefore matters as much as the magnitude.
7. Primary Results: Faster Aspart (Meal) vs NovoRapid (Meal)
The first posted primary analysis compared Faster Aspart (Meal) with NovoRapid (Meal) for change from baseline in HbA1c at week 26.
Mean difference in change from baseline
95% CI: −0.23% to −0.07%
Outcome unit: percentage of glycosylated haemoglobin
| Feature | Reported result |
|---|---|
| Endpoint | Change From Baseline in HbA1c (Glycosylated Haemoglobin) |
| Time frame | Week 0, week 26 |
| Comparison | Faster Aspart (Meal) vs NovoRapid (Meal) |
| Analysis population | Full analysis set; all randomized subjects |
| Method | MMRM |
| Effect measure | Mean Difference (Net) |
| Estimate | −0.15 |
| 95% confidence interval | −0.23 to −0.07 |
| Non-inferiority limit | 0.4% |
The estimated mean difference of −0.15 indicates that the modeled change from baseline in HbA1c was 0.15 percentage points lower with Faster Aspart (Meal) than with NovoRapid (Meal), using the posted MMRM analysis.
The 95% CI of −0.23 to −0.07 describes the uncertainty around the estimated treatment difference under the specified model. The entire interval is well within the stated non-inferiority limit of 0.4% in magnitude, so the observed estimate is compatible with a treatment difference substantially smaller than the prespecified non-inferiority boundary.
The confidence interval does not mean that 95% of individual patients had treatment differences between −0.23 and −0.07. It describes uncertainty around the estimated group-level treatment difference.
A p-value is not reported in this posted analysis. In general, a p-value addresses evidence against a specified null hypothesis; it does not measure the size, clinical importance, or precision of an effect. In a non-inferiority trial, the confidence interval relative to the prespecified margin is particularly important for interpreting the treatment difference.
Because the analysis is based on an MMRM rather than a simple week-26 comparison, the estimate reflects the model's use of the repeated longitudinal measurements and the specified covariate and stratification structure.
8. Primary Results: Faster Aspart (Post) vs NovoRapid (Meal)
The second posted primary analysis compared Faster Aspart (Post) with NovoRapid (Meal) using the same registered HbA1c endpoint and non-inferiority framework.
Mean difference in change from baseline
95% CI: −0.04% to 0.12%
Outcome unit: percentage of glycosylated haemoglobin
| Feature | Reported result |
|---|---|
| Endpoint | Change From Baseline in HbA1c (Glycosylated Haemoglobin) |
| Time frame | Week 0, week 26 |
| Comparison | Faster Aspart (Post) vs NovoRapid (Meal) |
| Analysis population | Full analysis set; all randomized subjects |
| Method | MMRM |
| Effect measure | Mean Difference (Net) |
| Estimate | 0.04 |
| 95% confidence interval | −0.04 to 0.12 |
| Non-inferiority limit | 0.4% |
The estimated mean difference of 0.04 indicates that the modeled change from baseline in HbA1c was 0.04 percentage points higher with Faster Aspart (Post) than with NovoRapid (Meal).
The 95% CI of −0.04 to 0.12 includes zero, so the interval is compatible with a small difference in either direction. At the same time, the entire interval is substantially narrower than the prespecified non-inferiority limit of 0.4% when considered in the relevant direction for the non-inferiority assessment.
The confidence interval is a statement about uncertainty in the estimated group-level treatment difference; it is not a prediction interval for individual HbA1c responses.
A p-value is not reported in this posted analysis. Even when a p-value is available, it would not quantify the size or clinical relevance of the 0.04 estimate. For a non-inferiority question, the relationship between the confidence interval and the prespecified margin is central.
The interpretation also depends on the assumptions underlying the MMRM, including the adequacy of the specified covariance and mean structure and the treatment of incomplete longitudinal measurements within the model.
9. Comparing the Two Primary Analyses
| Primary comparison | Mean difference | 95% CI | Non-inferiority limit |
|---|---|---|---|
| Faster Aspart (Meal) vs NovoRapid (Meal) | −0.15 | −0.23 to −0.07 | 0.4% |
| Faster Aspart (Post) vs NovoRapid (Meal) | 0.04 | −0.04 to 0.12 | 0.4% |
Both posted analyses use the same endpoint, analysis population, longitudinal modeling strategy and non-inferiority limit. The treatment contrasts differ in direction and magnitude: the Faster Aspart (Meal) comparison has an estimate of −0.15, whereas the Faster Aspart (Post) comparison has an estimate of 0.04.
These estimates should not be converted into a single overall treatment effect. They represent distinct randomized comparisons reported separately by the registry.
10. The MMRM in Detail
The mixed model for repeated measurements is particularly relevant because HbA1c was measured longitudinally. Rather than discarding earlier measurements and comparing only two observations, the model incorporated several visits and allowed the treatment and covariate relationships to vary with visit.
| Visit structure | Role |
|---|---|
| Visits 14, 18, 22, 26, 30, 34 and 36 | Included in the repeated-measures analysis |
| Baseline HbA1c | Covariate |
| Treatment, region and strata | Fixed effects |
| Subject | Random effect |
| Fixed effects × visit | Included in the model |
| Baseline HbA1c × visit | Included in the model |
A useful way to think about the model is that each participant contributes a trajectory rather than only a single endpoint value. The model then estimates the treatment difference while accounting for baseline HbA1c, region, prespecified strata and repeated observations within subjects.
The subject-level random effect provides a mechanism for representing within-subject dependence. Without accounting for repeated measurements appropriately, standard errors can be misstated.
11. Covariate Adjustment and Stratification
The posted model included baseline HbA1c as a covariate and incorporated region and the specified strata as fixed effects. This is more informative than simply comparing raw week-26 means because the model uses baseline information and the prespecified design structure.
Baseline HbA1c
Including baseline HbA1c as a covariate can improve precision by accounting for variation in the outcome measure before randomized treatment.
Region
Region was included as a fixed effect, allowing the model to account for systematic differences associated with geographic region.
Strata
The model included strata combining bolus adjusting method, basal treatment regimen, CGM and frequently sampled meal test subgroup.
Visit interactions
Interactions with visit allow modeled relationships to change over the repeated-measurement schedule rather than forcing one constant relationship across all visits.
12. Analysis Population and Missing Measurements
The full analysis set included all randomized subjects. The registry describes baseline and week-26 data for the primary endpoint, with week 26 representing the end of trial and containing the last available measurement.
Because the primary analysis is an MMRM, its use of repeated observations is relevant to how incomplete longitudinal data contribute to estimation. The registry does not provide additional detail here about a separate imputation procedure or a specific missing-data sensitivity analysis.
13. Safety Results
The registry reports serious adverse events by treatment arm as affected subjects over subjects at risk. These counts provide a descriptive safety comparison among the three arms identified in the record.
| Arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Faster Aspart (Meal) | Serious adverse events | 35 / 386 |
| Faster Aspart (Post) | Serious adverse events | 28 / 377 |
| NovoRapid (Meal) | Serious adverse events | 33 / 380 |
The reported quantities are counts of affected participants relative to the number at risk. They should be distinguished from the efficacy estimates, which are model-based mean differences in HbA1c change.
Faster Aspart (Meal)
35 affected participants among 386 at risk were reported for serious adverse events.
Faster Aspart (Post)
28 affected participants among 377 at risk were reported for serious adverse events.
NovoRapid (Meal)
33 affected participants among 380 at risk were reported for serious adverse events.
How to interpret
These are descriptive arm-level safety counts; they are not the same estimand as the primary HbA1c treatment differences.
14. Statistical Methods Explained
Why was an MMRM used?
HbA1c was measured repeatedly over time, and the posted model included visits 14, 18, 22, 26, 30, 34 and 36. An MMRM is designed for this longitudinal structure because it estimates treatment effects using repeated observations while accounting for correlation among measurements from the same participant.
What does a mean difference of −0.15 mean?
For the Faster Aspart (Meal) versus NovoRapid (Meal) comparison, −0.15 is the modeled difference in change from baseline in HbA1c. A negative value means the modeled change was lower in the Faster Aspart (Meal) group by 0.15 percentage points relative to the NovoRapid (Meal) group.
What does a mean difference of 0.04 mean?
For Faster Aspart (Post) versus NovoRapid (Meal), 0.04 means the modeled change from baseline was 0.04 percentage points higher in the Faster Aspart (Post) group than in the NovoRapid (Meal) group. The 95% CI extends from −0.04 to 0.12, so the uncertainty interval includes zero.
Why is the non-inferiority margin important?
The non-inferiority margin of 0.4% defines the maximum difference considered compatible with non-inferiority according to the trial's prespecified framework. The key question is not simply whether the estimated difference is close to zero; it is whether the confidence interval remains compatible with a difference no larger than the prespecified acceptable loss.
Why include baseline HbA1c as a covariate?
Baseline HbA1c is a strong measure of the same outcome domain being studied. Including it as a covariate allows the model to account for baseline variation and can improve the precision of the treatment comparison.
Why include a random subject effect?
Each participant can contribute multiple HbA1c measurements. Those measurements are correlated because they come from the same individual. Treating subject as a random effect helps represent this within-subject dependence in the repeated-measures model.
Why does the confidence interval matter more than a p-value for non-inferiority interpretation?
A p-value addresses a hypothesis-testing question, whereas the confidence interval shows the range of treatment differences compatible with the data and model at the stated confidence level. For non-inferiority, the interval can be directly compared with the prespecified 0.4% margin. The posted primary analyses report confidence intervals but do not report p-values.
15. Interpreting the Confidence Intervals
The 95% CI runs from −0.23 to −0.07. Every value in the interval is negative, so the modeled treatment difference remains below zero throughout the interval. More importantly for the non-inferiority framework, the interval is far inside the stated 0.4% non-inferiority boundary.
The 95% CI runs from −0.04 to 0.12. The interval includes zero, meaning the data are compatible with a small difference in either direction. The interval is nevertheless narrow relative to the prespecified 0.4% non-inferiority limit.
These two intervals illustrate an important distinction between superiority-style inference and non-inferiority inference. An interval that crosses zero may not establish a difference from zero, but it can still provide sufficient evidence for a non-inferiority question if the unfavorable side of the interval remains within the prespecified margin.
16. What the Primary Estimates Do — and Do Not — Mean
The estimates −0.15 and 0.04 describe modeled differences between randomized treatment groups. They do not describe the response of every individual participant.
The 95% confidence intervals quantify uncertainty around the respective estimated treatment differences under the specified MMRM framework. They do not describe the range of individual HbA1c changes.
The posted primary analyses do not report p-values. Even where p-values are used in clinical trials, they quantify evidence against a null hypothesis rather than the magnitude or clinical importance of the treatment difference.
The practical inferential reference point is the prespecified 0.4% non-inferiority limit. The treatment estimate and its confidence interval need to be interpreted in relation to that margin rather than in isolation.
17. Non-Inferiority Logic Using the Posted Results
A useful way to visualize the statistical question is to place each confidence interval beside the 0.4% non-inferiority boundary.
| Comparison | Estimate | 95% CI | Relationship to 0.4% margin |
|---|---|---|---|
| Faster Aspart (Meal) vs NovoRapid (Meal) | −0.15 | −0.23 to −0.07 | The entire interval is within the magnitude of the prespecified margin. |
| Faster Aspart (Post) vs NovoRapid (Meal) | 0.04 | −0.04 to 0.12 | The upper confidence limit remains below the 0.4% margin. |
The second comparison is especially useful educationally: its confidence interval includes zero, but the non-inferiority question does not require the interval to exclude zero. What matters is whether the interval excludes effects worse than the prespecified non-inferiority threshold.
18. Design Features That Affect Interpretation
Randomization
Randomized allocation provides the structural basis for comparing treatment groups while reducing systematic confounding from measured and unmeasured baseline factors.
Double blinding
The registry identifies the study as double-blind, reducing opportunities for knowledge of treatment assignment to influence trial conduct or outcome assessment.
Repeated measurements
Longitudinal HbA1c observations allow the MMRM to use more of the available outcome trajectory than a single endpoint comparison.
Non-inferiority margin
The 0.4% boundary is a prespecified inferential reference point and is essential to interpreting the primary treatment differences.
19. What the Registry Reports About Limitations
The registry's stated limitation is that trial limitations can include small numbers of subjects analysed or technical problems leading to unreliable data.
The posted primary analyses also require the usual interpretation of a longitudinal mixed model: estimates depend on the adequacy of the specified mean structure, covariance structure and assumptions concerning incomplete repeated measurements. The registry describes the model components but does not provide additional sensitivity-analysis results addressing these assumptions.
20. Important Statistical Interpretation Issues
- Three-arm design: the trial contains three arms, while the two posted primary analyses each specify a particular faster-aspart versus NovoRapid comparison. Those two estimates should therefore remain separate rather than being combined into one overall effect.
- Non-inferiority: the 0.4% margin is central to interpretation. A conventional test against zero is not the same statistical question as demonstrating that the treatment difference is within the non-inferiority boundary.
- MMRM dependence: repeated observations from the same subject are correlated, which is why the model includes subject as a random effect.
- Covariate adjustment: baseline HbA1c, region and the specified strata are incorporated into the model rather than relying only on unadjusted group means.
- Analysis population: the posted analysis uses the full analysis set, defined as all randomized subjects.
- Endpoint timing: the registered primary endpoint is explicitly framed as change from baseline after 26 weeks, even though the longitudinal model uses additional visits.
- Safety versus efficacy: serious adverse-event counts are descriptive arm-level safety information and are not directly comparable with the model-based HbA1c estimand.
21. Why This Trial Matters Statistically
onset® 1 is a useful teaching example because it combines several important clinical-trial concepts in a single randomized study: a continuous biomarker endpoint, repeated measurements, covariate adjustment, a three-arm parallel design, double blinding, and a non-inferiority framework.
| Concept | How it appears in onset® 1 |
|---|---|
| Randomization | Randomized allocation in a three-arm parallel phase 3 trial. |
| Blinding | The registry identifies the trial as double-blind. |
| Continuous outcome | HbA1c is analyzed through change from baseline and a mean difference. |
| Repeated measures | Visits 14, 18, 22, 26, 30, 34 and 36 are included in the MMRM. |
| Covariate adjustment | Baseline HbA1c is included as a covariate. |
| Stratification | Region and a composite stratum are included as fixed effects. |
| Non-inferiority | The treatment difference is compared with a prespecified 0.4% limit. |
| Confidence intervals | Both primary analyses report two-sided 95% confidence intervals. |
| Safety analysis | Serious adverse events are reported by arm as affected subjects over subjects at risk. |
22. Statistical Methods Explained: A Worked Conceptual View
The primary quantity is the difference between treatment groups in change from baseline in HbA1c after 26 weeks, estimated through the posted longitudinal model.
Instead of treating only week 26 as informative, the MMRM incorporates measurements from visits 14, 18, 22, 26, 30, 34 and 36.
Treatment, region and strata enter as fixed effects; baseline HbA1c enters as a covariate; subject enters as a random effect; and the specified interactions with visit are included.
The analysis produces a mean difference and a two-sided 95% confidence interval for each posted primary comparison.
The treatment difference is evaluated against the prespecified 0.4% non-inferiority limit rather than relying solely on whether the confidence interval contains zero.
23. Results Summary
| Primary endpoint comparison | Estimate | 95% CI | Method |
|---|---|---|---|
| Faster Aspart (Meal) vs NovoRapid (Meal) | −0.15 | −0.23 to −0.07 | MMRM |
| Faster Aspart (Post) vs NovoRapid (Meal) | 0.04 | −0.04 to 0.12 | MMRM |
The two posted primary analyses therefore provide estimates of −0.15 and 0.04, respectively, with relatively narrow two-sided 95% confidence intervals. Both analyses use the full analysis set and the same longitudinal modeling framework, and both are evaluated against the registered 0.4% non-inferiority limit.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: NCT01831765 — onset® 1.
- PubMed: PMID 28356319.
- PubMed: PMID 29316130.
- PubMed: PMID 29463670.
- PubMed: PMID 30402720.
- PubMed: PMID 30547388.
Continue through the Clinical Biostats knowledge graph
Connect the trial's endpoints and methods to deeper statistical tutorials and practical analysis tools.
27. Record Summary
onset® 1 provides a useful example of how a non-inferiority clinical trial can be analyzed when the primary outcome is a continuously measured biomarker collected repeatedly over time. The posted analyses use an MMRM with baseline HbA1c adjustment, region and composite strata as fixed effects, subject as a random effect, and treatment-by-visit and covariate-by-visit interactions. The two reported primary comparisons produce mean differences of −0.15 and 0.04, with two-sided 95% confidence intervals of −0.23 to −0.07 and −0.04 to 0.12, respectively, against a prespecified non-inferiority limit of 0.4%.
The most important statistical lesson is that the interpretation of a non-inferiority analysis depends on the relationship between the estimated treatment difference, its confidence interval, and the prespecified margin. The MMRM adds another important layer: the reported treatment effect is a model-based longitudinal comparison that uses repeated observations and prespecified covariate and stratification information rather than a simple unadjusted week-26 difference.