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
ADJUNCT ONE™ was a randomized, double-blind, parallel phase 3 trial evaluating liraglutide as adjunct therapy to insulin in the treatment of type 1 diabetes. The registry reports six arms, enrollment of 1398 participants, three registered primary endpoints, and formal statistical analyses for the primary endpoints and a secondary hypoglycemia endpoint.
| Feature | ADJUNCT ONE™ |
|---|---|
| Trial name | ADJUNCT ONE™ |
| ClinicalTrials.gov identifier | NCT01836523 |
| 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 | 1398 |
| Start | 2013-11 |
| Primary completion | 2015-06 |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
2. Clinical Question
The central statistical question was whether liraglutide, when used as adjunct therapy to insulin in people with type 1 diabetes, could achieve the prespecified HbA1c objective while also producing differences in body weight and total daily insulin dose at week 52. The posted analyses used both non-inferiority and superiority hypotheses, depending on the endpoint and comparison.
Population
Participants enrolled in the phase 3 study of liraglutide as adjunct therapy to insulin in the treatment of type 1 diabetes.
Intervention
Liraglutide at 0.6 mg, 1.2 mg, or 1.8 mg, with placebo also represented among the six trial arms.
Comparator
Placebo.
Primary questions
How did liraglutide compare with placebo for change from baseline in HbA1c, body weight, and total daily insulin dose at week 52?
3. Trial Design
Liraglutide
- Liraglutide 0.6 mg
- Liraglutide 1.2 mg
- Liraglutide 1.8 mg
Placebo
- Placebo was included as the comparator.
- The posted primary analyses compare each liraglutide dose with placebo.
4. Randomization, Stratification, and Analysis Populations
The posted MMRM analyses incorporated a four-level stratification variable formed by crossing baseline HbA1c with baseline BMI. The four categories were HbA1c <8.5% and ≥8.5%, each intersected by BMI ≤27 kg/m2 and >27 kg/m2. Country and treatment were also included as fixed factors, and the corresponding baseline value was included as a covariate.
| Analysis feature | Registry description |
|---|---|
| Primary HbA1c population | Full analysis set; subjects with any post-baseline HbA1c data. |
| Primary body-weight population | Full analysis set; subjects with any post-baseline body-weight data. |
| Primary insulin-dose population | Full analysis set; subjects with any post-baseline total insulin daily dose data. |
| Safety population | All randomized subjects exposed to at least one dose of liraglutide or placebo. |
| Stratification variable | Four levels defined by HbA1c <8.5% versus ≥8.5% and BMI ≤27 kg/m2 versus >27 kg/m2. |
| Covariate adjustment | Corresponding baseline value included as a covariate in the MMRM analyses. |
5. Registered Primary Endpoints
| Endpoint | Time frame | Registry definition | Analysis |
|---|---|---|---|
| Change From Baseline in HbA1c (Glycosylated Haemoglobin) | Week 0, week 52 | Change from baseline in HbA1c at week 52. Missing values were handled by using a mixed model for repeated measurements (MMRM). | MMRM |
| Change From Baseline in Body Weight | Week 0, week 52 | Change from baseline in body weight at week 52. Missing values were handled by using a MMRM. | Mixed-effects model / MMRM framework |
| Change From Baseline in Total Daily Insulin Dose | Week 0, week 52 | Change from baseline represented as the ratio of total daily insulin dose at week 52 to total daily insulin dose at baseline. Missing values were handled by using a MMRM. | Mixed-effects model / MMRM framework |
The registry reports three primary endpoints, with nine posted primary endpoint analyses: three dose comparisons for each of the three endpoints.
6. Statistical Methodology
Mixed model for repeated measures
The principal longitudinal method for the primary endpoints was a mixed model for repeated measures. All post-baseline measurements from planned visits through week 52, obtained no later than one day after withdrawal from treatment, were entered as the dependent variable. Visit, treatment, country, and the four-level stratification variable were included as fixed factors, with the corresponding baseline measurement as a covariate.
The important feature is that the model uses the repeated post-baseline observations rather than reducing each participant to only one observed value at week 52.
Covariate adjustment
Baseline values were included as covariates. This adjusts the estimated treatment comparison for the corresponding baseline measurement and can improve precision when the baseline covariate is related to the outcome.
Stratified analysis
The MMRM included the four-level baseline HbA1c/BMI stratification variable as a fixed factor. Country was also included as a fixed factor. Thus, the reported estimates are not simple unadjusted differences between observed treatment-group means.
Analysis of total daily insulin dose
The insulin-dose measurements were log-transformed before analysis. The treatment effect was then expressed as a treatment ratio rather than as an absolute difference in insulin dose.
A ratio of 0.92 indicates an estimated treatment level 92% of the comparator level under the fitted model, whereas a ratio of 1.00 indicates no estimated multiplicative difference.
Negative binomial regression
The number of treatment-emergent symptomatic hypoglycaemic episodes was analyzed using negative binomial regression with a log-link function. The logarithm of the time period during which a hypoglycaemic episode was considered treatment emergent was used as an offset.
The offset accounts for differences in the amount of time during which participants could contribute treatment-emergent episodes.
7. Results: HbA1c
The primary HbA1c endpoint was assessed as change from baseline at week 52. All three dose comparisons used MMRM and a two-sided 95% confidence interval. The registry specifies a non-inferiority criterion under which non-inferiority was concluded if the upper bound of the 95% confidence interval was <0.3.
| Comparison | Treatment difference | 95% CI | Hypothesis | P-value |
|---|---|---|---|---|
| Liraglutide 1.8 mg vs placebo | -0.20 | -0.32 to -0.07 | Non-inferiority | Not reported |
| Liraglutide 1.2 mg vs placebo | -0.15 | -0.27 to -0.03 | Non-inferiority | Not reported |
| Liraglutide 0.6 mg vs placebo | -0.09 | -0.21 to 0.03 | Non-inferiority | Not reported |
HbA1c treatment differences
Estimated treatment differences for 1.8 mg, 1.2 mg, and 0.6 mg versus placebo, respectively.
All three upper confidence limits are below the prespecified non-inferiority boundary of 0.3.
The treatment difference of -0.20 for liraglutide 1.8 mg versus placebo means that the model-estimated change in HbA1c was 0.20 percentage points lower for the liraglutide comparison at week 52, under the specified MMRM. The corresponding 95% CI was -0.32 to -0.07.
The 95% confidence interval quantifies uncertainty around the estimated treatment difference; it does not describe the range of individual patient responses. The interval's upper bound of -0.07 is also below the registry's non-inferiority margin of 0.3. The same logic applies to the 1.2 mg comparison, whose upper bound was -0.03, and the 0.6 mg comparison, whose upper bound was 0.03.
The p-value is not an effect-size measure and was not reported for these posted analyses. The non-inferiority conclusion is instead tied to the prespecified confidence-interval criterion. Because this is a repeated-measures model, the estimate also depends on the model specification, included covariates and repeated observations through week 52.
8. Results: Body Weight
Body weight was analyzed as change from baseline at week 52 using a mixed-effects model within the repeated-measures framework. All three comparisons were prespecified as superiority analyses.
| Comparison | Treatment difference (kg) | 95% CI | P-value |
|---|---|---|---|
| Liraglutide 1.8 mg vs placebo | -4.90 | -5.65 to -4.16 | <0.0001 |
| Liraglutide 1.2 mg vs placebo | -3.55 | -4.29 to -2.81 | <0.0001 |
| Liraglutide 0.6 mg vs placebo | -2.19 | -2.91 to -1.47 | <0.0001 |
Largest reported body-weight difference
95% CI: -5.65 to -4.16 kg · P < 0.0001
Liraglutide 1.8 mg versus placebo.
The estimated treatment difference of -4.90 kg for liraglutide 1.8 mg versus placebo indicates that the model-estimated change from baseline in body weight was 4.90 kg lower in the liraglutide comparison at week 52.
The 95% CI of -5.65 to -4.16 kg gives the statistical uncertainty around that model-based difference. Because the entire interval is below zero, the estimated direction is consistently toward lower body weight in the liraglutide group. The p-value of <0.0001 addresses the statistical evidence against the specified null hypothesis; it does not measure the size or clinical importance of the weight difference.
The 1.2 mg and 0.6 mg comparisons were also negative, with confidence intervals entirely below zero. The results should still be understood as adjusted longitudinal model estimates rather than simple differences between two observed week-52 averages.
9. Results: Total Daily Insulin Dose
Total daily insulin dose was represented as the ratio of the week-52 dose to the baseline dose. The measurements were log-transformed before the mixed-effects analysis, and the resulting treatment effects were reported as ratios.
| Comparison | Treatment ratio | 95% CI | P-value |
|---|---|---|---|
| Liraglutide 1.8 mg vs placebo | 0.92 | 0.88 to 0.96 | <0.0001 |
| Liraglutide 1.2 mg vs placebo | 0.95 | 0.91 to 0.99 | 0.0148 |
| Liraglutide 0.6 mg vs placebo | 1.00 | 0.96 to 1.04 | 0.9615 |
Insulin-dose treatment ratio
95% CI: 0.88–0.96 · P < 0.0001
Liraglutide 1.8 mg versus placebo.
A treatment ratio of 0.92 means that the model-estimated week-52-to-baseline insulin-dose ratio was 0.92 times the corresponding ratio for placebo. Because the analysis was performed after log transformation, the ratio is a multiplicative treatment effect rather than an absolute insulin-dose difference.
The 95% CI of 0.88 to 0.96 indicates uncertainty around the ratio estimate. Because the entire interval is below 1.00, the estimated direction favors a lower insulin-dose ratio for the 1.8 mg comparison. The p-value of <0.0001 describes evidence against the superiority null hypothesis; it does not quantify the magnitude of the treatment effect.
For 1.2 mg, the ratio was 0.95 with a 95% CI of 0.91 to 0.99 and P=0.0148. For 0.6 mg, the ratio was 1.00 with a 95% CI of 0.96 to 1.04 and P=0.9615. The latter interval includes 1.00, so the posted estimate does not provide evidence of a difference from the placebo comparison under the specified superiority analysis.
10. Secondary Endpoint: Treatment-Emergent Symptomatic Hypoglycaemic Episodes
The registry reports three secondary analyses of the number of treatment-emergent symptomatic hypoglycaemic episodes over weeks 0-52. The safety analysis set included all randomized subjects exposed to at least one dose of liraglutide or placebo. Negative binomial regression was used with treatment, country, stratification group, and baseline HbA1c in the model.
| Comparison | Rate ratio | 95% CI | P-value |
|---|---|---|---|
| Liraglutide 1.8 mg vs placebo | 1.31 | 1.07 to 1.59 | 0.0081 |
| Liraglutide 1.2 mg vs placebo | 1.27 | 1.03 to 1.55 | 0.0219 |
| Liraglutide 0.6 mg vs placebo | 1.17 | 0.97 to 1.43 | 0.1079 |
Highest reported hypoglycemia rate ratio
95% CI: 1.07–1.59 · P = 0.0081
Liraglutide 1.8 mg versus placebo.
A rate ratio of 1.31 indicates an estimated treatment-emergent symptomatic hypoglycaemic episode rate 1.31 times that of placebo under the fitted negative binomial model. Equivalently, the estimated rate was approximately 31% higher in relative terms.
The 95% CI of 1.07 to 1.59 describes uncertainty around that rate ratio. It does not describe the range of episode counts for individual participants. The p-value of 0.0081 provides evidence against the specified superiority null hypothesis, but it does not indicate that the increase was exactly 31% in every patient.
The 1.2 mg comparison had a rate ratio of 1.27, while the 0.6 mg comparison had a rate ratio of 1.17. For 0.6 mg, the confidence interval included 1.00 and the reported P-value was 0.1079. These results also illustrate why count endpoints are often analyzed as rates rather than simply comparing the mean number of episodes without accounting for exposure time.
11. How the Negative Binomial Model Changes the Question
Hypoglycaemic episodes are counts, and participants can contribute different amounts of observation time. A negative binomial model is useful when the variability in counts is greater than would typically be accommodated by a simple Poisson model.
Count outcome
The endpoint counts treatment-emergent symptomatic hypoglycaemic episodes rather than classifying each participant simply as having or not having an event.
Exposure time
The logarithm of the relevant treatment-emergent observation period was included as an offset, allowing the model to estimate rates.
Rate ratio
The reported effect measure compares estimated episode rates rather than cumulative proportions of participants experiencing an episode.
Covariate adjustment
The model included country, stratification group, and baseline HbA1c in addition to treatment.
12. Safety: Serious Adverse Events
The registry reports serious adverse events by treatment group as affected subjects over subjects at risk. These figures are presented directly rather than converted into derived percentages.
| Group | Serious adverse events affected / at risk |
|---|---|
| Liraglutide 0.6 mg | 35/350 |
| Liraglutide 1.2 mg | 36/348 |
| Liraglutide 1.8 mg | 29/347 |
| Placebo | 38/348 |
These serious-adverse-event figures should be interpreted as the number of affected participants relative to the corresponding number at risk. They are not episode rates and should not be substituted for the negative-binomial analysis of symptomatic hypoglycaemic episodes.
The denominator is important because the four groups have different numbers at risk. Comparing only the numerators would therefore not be a valid statistical comparison of serious-adverse-event occurrence.
13. Missing Data and Repeated Measurements
The registry explicitly states that missing values for the three primary endpoints were handled using MMRM. For the HbA1c and body-weight analyses, all post-baseline measurements from planned visits through week 52 and obtained no later than one day after withdrawal from treatment were entered as dependent observations. The insulin-dose analysis used the same repeated-measures framework, with log-transformed measurements.
14. Statistical Methods Explained
Why was an MMRM used?
The primary endpoints were measured repeatedly over time, with observations from planned visits through week 52. An MMRM allows those repeated measurements to contribute to a longitudinal treatment comparison while accounting for visit and baseline covariates. It also provides a model-based approach to handling missing post-baseline observations rather than restricting the analysis to participants with complete data.
What does a treatment difference of -0.20 mean for HbA1c?
It means the estimated change from baseline was 0.20 percentage points lower for liraglutide 1.8 mg than placebo under the specified MMRM. The estimate is a between-treatment model contrast; it is not the percentage of participants who achieved a particular HbA1c target.
Why is non-inferiority judged against the confidence-interval margin?
For the HbA1c endpoint, the registry states that non-inferiority was concluded if the upper bound of the 95% confidence interval was below 0.3. This criterion directly compares the uncertainty interval with the prespecified maximum clinically acceptable difference used for the non-inferiority hypothesis.
What does a treatment ratio of 0.92 mean?
For total daily insulin dose, the analysis used log-transformed measurements and reported a treatment ratio. A ratio of 0.92 means the estimated week-52-to-baseline dose ratio was 92% of the corresponding placebo ratio under the fitted model. It is a multiplicative comparison, not a 0.92-unit reduction.
Why was negative binomial regression used for hypoglycaemic episodes?
The endpoint is a count of symptomatic hypoglycaemic episodes and the amount of time contributing to the endpoint can vary. Negative binomial regression provides a count-model framework that can accommodate overdispersion, while the offset accounts for the relevant observation period.
What does a rate ratio of 1.31 mean?
A rate ratio of 1.31 indicates an estimated episode rate 1.31 times the comparator rate. It does not mean that 31% more participants necessarily experienced an episode, because the endpoint is based on episode counts and the analysis incorporates exposure time.
15. Non-Inferiority Versus Superiority
ADJUNCT ONE™ illustrates two different hypothesis-testing frameworks within the same trial. The HbA1c analyses were designated as non-inferiority analyses, while the body-weight and insulin-dose analyses were designated as superiority analyses.
| Endpoint | Primary comparisons | Hypothesis type | Decision quantity |
|---|---|---|---|
| HbA1c | 0.6 mg, 1.2 mg, 1.8 mg vs placebo | Non-inferiority or equivalence | Upper 95% CI bound relative to 0.3 |
| Body weight | 0.6 mg, 1.2 mg, 1.8 mg vs placebo | Superiority | Treatment difference and P-value |
| Total daily insulin dose | 0.6 mg, 1.2 mg, 1.8 mg vs placebo | Superiority | Treatment ratio and P-value |
16. Confidence Intervals and P-values
The primary results provide a useful illustration of why confidence intervals and p-values answer related but different questions.
Confidence interval
The 95% CI communicates statistical uncertainty around the estimated treatment effect under the specified model. Its width gives information about precision.
P-value
A p-value measures evidence against a specified null hypothesis under the statistical model. It does not measure effect size, clinical importance, or the probability that the null hypothesis is true.
HbA1c
The posted HbA1c analyses report confidence intervals and the non-inferiority margin, but no p-values are reported for those analyses.
Body weight and insulin dose
These superiority analyses report both confidence intervals and p-values, allowing the estimate, precision, and null-hypothesis evidence to be considered together.
17. Dose-Specific Statistical Pattern
The three dose comparisons show how an analysis can reveal different statistical patterns across related treatment contrasts without reducing the results to a single overall number.
| Endpoint | 0.6 mg vs placebo | 1.2 mg vs placebo | 1.8 mg vs placebo |
|---|---|---|---|
| HbA1c treatment difference | -0.09 | -0.15 | -0.20 |
| HbA1c 95% CI | -0.21 to 0.03 | -0.27 to -0.03 | -0.32 to -0.07 |
| Body-weight treatment difference | -2.19 kg | -3.55 kg | -4.90 kg |
| Body-weight 95% CI | -2.91 to -1.47 kg | -4.29 to -2.81 kg | -5.65 to -4.16 kg |
| Insulin-dose treatment ratio | 1.00 | 0.95 | 0.92 |
| Insulin-dose 95% CI | 0.96 to 1.04 | 0.91 to 0.99 | 0.88 to 0.96 |
Descriptively, the posted estimates become progressively more negative for HbA1c and body weight across the three liraglutide doses, while the insulin-dose ratio moves from 1.00 at 0.6 mg to 0.92 at 1.8 mg. These patterns describe the observed model estimates; they do not by themselves establish a formal dose-response hypothesis unless such a hypothesis and its statistical test are prespecified.
18. Secondary Hypoglycemia Results in Context
| Comparison | Rate ratio | 95% CI | P-value | Statistical reading |
|---|---|---|---|---|
| Liraglutide 1.8 mg vs placebo | 1.31 | 1.07–1.59 | 0.0081 | CI above 1.00 |
| Liraglutide 1.2 mg vs placebo | 1.27 | 1.03–1.55 | 0.0219 | CI above 1.00 |
| Liraglutide 0.6 mg vs placebo | 1.17 | 0.97–1.43 | 0.1079 | CI includes 1.00 |
The secondary endpoint is structurally different from the primary continuous endpoints. A rate ratio compares event rates, while the body-weight analysis reports an absolute treatment difference and the insulin-dose analysis reports a multiplicative treatment ratio. These measures should not be placed on a common numerical scale or interpreted as if they represented the same estimand.
19. Statistical Interpretation of the Primary Results
The three liraglutide comparisons produced treatment differences of -0.20, -0.15, and -0.09 for 1.8 mg, 1.2 mg, and 0.6 mg, respectively. Their upper 95% confidence limits were -0.07, -0.03, and 0.03, all below the registry's 0.3 non-inferiority margin.
The confidence intervals indicate the precision of the model-based treatment differences. The 0.6 mg interval extends above zero, whereas the 1.2 mg and 1.8 mg intervals remain below zero. That distinction describes the location of the confidence intervals; it does not change the stated non-inferiority criterion.
The estimated treatment differences were -2.19 kg, -3.55 kg, and -4.90 kg for 0.6 mg, 1.2 mg, and 1.8 mg, respectively. Each 95% confidence interval was entirely below zero, and each reported P-value was <0.0001.
The estimates describe average model-based treatment contrasts, not the response of every individual participant. A confidence interval describes uncertainty around the estimated contrast and does not establish a distribution of individual-level treatment responses.
The treatment ratios were 1.00, 0.95, and 0.92 for 0.6 mg, 1.2 mg, and 1.8 mg, respectively. The 1.8 mg estimate had a 95% CI of 0.88 to 0.96 and P<0.0001, while the 1.2 mg estimate had a 95% CI of 0.91 to 0.99 and P=0.0148. The 0.6 mg estimate had a 95% CI of 0.96 to 1.04 and P=0.9615.
Because the endpoint was log-transformed before analysis, these are multiplicative comparisons. A ratio near 1.00 represents little estimated multiplicative separation from placebo; a ratio below 1.00 represents a lower estimated ratio under the model.
20. Important Statistical Limitations
- Model dependence: MMRM estimates depend on the specified longitudinal model, including treatment, visit, country, stratification variables, and baseline covariates.
- Missing-data assumptions: the registry states that missing primary-endpoint values were handled using MMRM. The validity of that approach depends on the assumptions underlying the model and missing-data mechanism.
- Different estimands: HbA1c, body weight, insulin dose, and hypoglycaemic episodes use different effect measures and therefore answer different statistical questions.
- Non-inferiority margin: interpretation of HbA1c depends on the prespecified 0.3 margin. The margin is not interchangeable with a conventional superiority null value of zero.
- Multiple dose comparisons: each primary endpoint was analyzed for three liraglutide dose comparisons against placebo. The posted information describes the individual analyses but does not provide a multiplicity-adjustment strategy for treating all nine primary comparisons as one family.
- Secondary endpoint: the hypoglycemia analyses are secondary rather than primary endpoint analyses and should be interpreted in that context.
- Count-data interpretation: a rate ratio for hypoglycaemic episodes is not equivalent to a risk ratio or a difference in the proportion of participants experiencing an event.
- Serious adverse events: the registry reports affected subjects and subjects at risk by group, but those figures alone do not constitute a formal comparative statistical analysis.
21. What These Results Do and Do Not Mean
A treatment difference is not an individual response
The reported body-weight and HbA1c differences are population-level model estimates. They do not imply that every participant experienced the same change.
A confidence interval is not a patient range
A 95% confidence interval quantifies uncertainty around the estimated treatment effect. It is not the range in which 95% of individual treatment effects must fall.
A p-value is not effect size
A small p-value can accompany a modest effect when precision is high, while a larger p-value can accompany a potentially meaningful estimate when uncertainty is greater.
A rate ratio is not a risk ratio
The hypoglycemia rate ratios compare modeled episode rates and incorporate the exposure-time offset. They do not directly compare the proportion of participants with at least one episode.
22. Why This Trial Matters Statistically
ADJUNCT ONE™ is a useful teaching example because it combines longitudinal continuous outcomes, a non-inferiority framework, multiplicative analysis of a dose measure, and count-data modeling within one randomized trial.
| Concept | How it appears in ADJUNCT ONE™ |
|---|---|
| Randomization | Randomized parallel phase 3 design with six arms. |
| Double blinding | The registry identifies the study as double-masked. |
| MMRM | Primary HbA1c, body-weight, and insulin-dose analyses use repeated post-baseline measurements through week 52. |
| Covariate adjustment | Corresponding baseline measurements were included as covariates. |
| Stratified analysis | Country and a four-level HbA1c/BMI stratification variable were incorporated into the primary longitudinal models. |
| Non-inferiority | HbA1c non-inferiority was assessed using a 0.3 upper confidence-bound criterion. |
| Superiority | Body weight and total daily insulin dose were analyzed under superiority hypotheses. |
| Log transformation | Total daily insulin-dose measurements were log-transformed before mixed-model analysis. |
| Negative binomial regression | Treatment-emergent symptomatic hypoglycaemic episodes were analyzed as count/rate data. |
| Rate ratio | Hypoglycemia results were expressed as rate ratios with 95% confidence intervals. |
| Missing data | The registry states that missing primary-endpoint values were handled using MMRM. |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Statistical Calculators
25. Sources
- ClinicalTrials.gov: NCT01836523.
- PubMed: PMID 27506222.
- PubMed: PMID 39717993.
- PubMed: PMID 34463425.
Continue through the Clinical Biostats knowledge graph
Connect the endpoints and statistical methods in this trial to deeper biostatistics tutorials and practical statistical tools.
26. Record Summary
ADJUNCT ONE™ provides a compact example of several important clinical-trial statistical methods. The primary endpoints were analyzed with repeated-measures models that incorporated treatment, visit, country, a four-level HbA1c/BMI stratification variable, and the corresponding baseline value. HbA1c was evaluated using a non-inferiority framework with a 0.3 upper confidence-bound criterion, while body weight and total daily insulin dose were evaluated under superiority hypotheses. Insulin dose was analyzed after log transformation and reported as a treatment ratio. Treatment-emergent symptomatic hypoglycaemic episodes were analyzed separately with negative binomial regression and reported as rate ratios.
The most informative reading of the results therefore combines the estimated treatment effect, confidence interval, hypothesis type, analysis population, and model specification. For HbA1c, the key statistical question is the position of the upper confidence limit relative to the non-inferiority margin. For body weight and insulin dose, the direction and precision of the treatment estimate are interpreted under superiority hypotheses. For hypoglycemia, the rate ratio must be understood as a modeled rate comparison rather than a simple comparison of participant proportions.