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
COMBINE 1 was a randomized, open-label, phase 3 trial evaluating weekly IcoSema versus weekly insulin icodec in people with type 2 diabetes. The primary endpoint was change in glycated haemoglobin (HbA1c) from baseline at week 52, analyzed using ANCOVA.
| Feature | COMBINE 1 |
|---|---|
| Trial name | COMBINE 1 |
| ClinicalTrials.gov identifier | NCT05352815 |
| Phase | Phase 3 |
| Condition | Diabetes Mellitus, Type 2 |
| Therapeutic area | Endocrinology & Metabolism |
| Status | Completed |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1,291 |
| Arms | 2 |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| Start | June 1, 2022 |
| Primary completion | March 19, 2024 |
2. Clinical Question
The central statistical question was whether IcoSema produced a different change in HbA1c from baseline to week 52 compared with insulin icodec in people with type 2 diabetes.
Population
People with Diabetes Mellitus, Type 2 enrolled in the phase 3 COMBINE 1 study.
Intervention
IcoSema, the weekly combination of insulin icodec and semaglutide.
Comparator
Weekly insulin icodec.
Primary question
What is the difference between IcoSema and insulin icodec in change in HbA1c from baseline to week 52?
3. Trial Design
IcoSema
- Weekly IcoSema
- Combination of insulin icodec and semaglutide
Insulin icodec
- Weekly insulin icodec
Randomization is important statistically because it creates the basis for comparing outcomes according to assigned treatment rather than according to characteristics that may have influenced treatment selection. The parallel design means the two randomized groups are followed as separate treatment groups rather than being intentionally exposed to both interventions.
4. Endpoints
| Endpoint | Time frame | Definition | Analysis |
|---|---|---|---|
| Change in Glycated Haemoglobin (HbA1c) | Baseline (Week 0), Week 52 | Change from baseline (week 0) to week 52 in HbA1c. The outcome measure was evaluated based on the data from in-study period, where all data from randomisation until the last date of specified participant or investigator contact, withdrawal of informed consent, loss to follow-up, or death were considered. | ANCOVA |
The registry identifies the primary endpoint as a binary endpoint type, while the posted statistical analysis reports a treatment difference in percentage points of HbA1c using ANCOVA. For the statistical interpretation, the posted ANCOVA analysis and its treatment-difference estimate are the relevant quantities.
5. Statistical Methodology
Analysis population
The primary analysis used the Full Analysis Set (FAS), defined as including all randomized participants. The overall number of participants analyzed was described as participants with available data.
The distinction between the randomized population and the participants with available outcome data is important. Randomization establishes the treatment comparison, while the observed HbA1c measurements determine which participants contribute directly to a particular analysis.
ANCOVA
The primary analysis used an analysis of covariance (ANCOVA) model. The model included region and randomized treatment as fixed factors and baseline HbA1c as a covariate.
The statistical purpose of including baseline HbA1c is to account for baseline differences in the outcome when estimating the treatment difference at the prespecified follow-up time.
Missing HbA1c values
Missing HbA1c values at week 52 were handled using multiple imputation. Each imputed dataset was analyzed separately, and the resulting estimates were combined using Rubin's rules.
This approach separates two tasks that are sometimes conflated: creating plausible completed datasets under the specified imputation framework, and then carrying out the prespecified ANCOVA within each completed dataset. Rubin's rules combine the resulting estimates while accounting for both within-imputation and between-imputation variability.
Hypothesis type
The registry classifies the primary hypothesis as superiority. The reported effect measure is a treatment difference, with IcoSema compared with insulin icodec.
6. Primary Result: Change in HbA1c
The primary endpoint was change in HbA1c from baseline at week 52. The posted analysis compared IcoSema with insulin icodec using ANCOVA in the Full Analysis Set.
Treatment difference in HbA1c change
95% CI: −0.76 to −0.57 · P < 0.0001
ANCOVA; two-sided 95% confidence interval; IcoSema vs insulin icodec.
| Primary endpoint | IcoSema vs insulin icodec |
|---|---|
| Outcome | Change in HbA1c |
| Time frame | Baseline (Week 0), Week 52 |
| Outcome unit | Percentage point of HbA1c |
| Analysis population | Full Analysis Set |
| Statistical method | ANCOVA |
| Effect measure | Treatment difference |
| Estimate | −0.66 |
| 95% CI | −0.76 to −0.57 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
The estimated treatment difference of −0.66 percentage points means that the modeled change in HbA1c from baseline to week 52 was lower with IcoSema than with insulin icodec by 0.66 percentage points, according to the posted ANCOVA analysis.
The negative sign identifies the direction of the comparison; it does not mean that every individual participant experienced exactly the same difference. A treatment difference is a population-level comparison of the modeled outcome, not an individual treatment effect.
The 95% confidence interval of −0.76 to −0.57 describes statistical uncertainty around the estimated treatment difference. It is relatively narrow compared with the point estimate, indicating that the estimated difference is reasonably precise within the assumptions of the analysis.
The P-value < 0.0001 addresses the compatibility of the observed result with the null hypothesis under the specified statistical model and testing framework. It does not measure the size of the treatment effect, clinical importance, or the probability that the treatment works.
The analysis also depends on the handling of missing week 52 HbA1c values. Multiple imputation was used, and the imputed analyses were combined with Rubin's rules. Consequently, interpretation should account for the assumptions underlying that missing-data approach.
7. Interpreting the Treatment Difference
For a continuous outcome such as HbA1c change, the treatment difference is naturally expressed on the original measurement scale. This makes the estimate directly interpretable as a difference in percentage points rather than as a ratio or relative risk.
What −0.66 means
The estimated difference in change from baseline between IcoSema and insulin icodec was −0.66 percentage points at week 52.
What it does not mean
It does not mean that every participant had a 0.66-percentage-point difference, nor does it describe the distribution of individual responses.
What the CI adds
The 95% CI of −0.76 to −0.57 shows the statistical uncertainty around the estimated treatment difference.
What the P-value adds
The P-value provides evidence concerning the null hypothesis; it is not an effect-size measure and should not be used as a substitute for the treatment difference or confidence interval.
A useful statistical reading therefore starts with the treatment difference, then considers the confidence interval, and only afterward considers the hypothesis-test result. Looking only at the P-value would discard much of the information contained in the analysis.
8. Why Was ANCOVA Used?
ANCOVA is appropriate for comparing a continuous outcome between randomized treatment groups while accounting for a prespecified continuous baseline covariate. Here, the outcome was HbA1c change and baseline HbA1c was included as a covariate.
Baseline adjustment can improve precision when baseline values are related to the follow-up outcome. Instead of comparing the two groups only through an unadjusted difference, ANCOVA estimates the treatment difference conditional on the baseline covariate and the other fixed factor specified in the model.
Covariate: baseline HbA1c
The treatment coefficient provides the estimated treatment difference on the HbA1c-change scale after accounting for the model's specified baseline and regional terms.
9. Why Include Region?
Region was included as a fixed factor in the ANCOVA model. Statistically, a categorical factor such as region allows the model to account for systematic differences associated with regional groups rather than leaving those differences entirely in the residual error.
The important point is that the treatment comparison remains the central estimand. Region is a model adjustment factor; it is not itself the treatment effect.
10. Multiple Imputation and Rubin's Rules
Missing HbA1c values at week 52 were imputed using multiple imputation. Instead of replacing each missing value with one fixed number, multiple completed datasets are analyzed separately.
Rubin's rules incorporate two sources of uncertainty: the variability within each imputed dataset and the variability between the separate imputed datasets. The resulting confidence interval therefore reflects uncertainty introduced by the missing-data process as represented by the specified multiple-imputation procedure.
11. Intention-to-Treat and the Full Analysis Set
The posted primary analysis used the Full Analysis Set, which included all randomized participants, while the number analyzed for the outcome consisted of participants with available data. This structure reflects an important distinction between treatment assignment and availability of a particular measurement.
Randomization protects the treatment comparison at the point of assignment. An analysis based on randomized treatment assignment is therefore conceptually different from a comparison constructed only from participants who completed treatment or follow-up.
| Concept | COMBINE 1 application |
|---|---|
| Randomization | Participants were randomized to the two parallel treatment groups. |
| Full Analysis Set | Included all randomized participants. |
| Outcome availability | Participants with available data contributed directly to the corresponding outcome analysis. |
| Missing-data handling | Missing week 52 HbA1c values were handled using multiple imputation. |
12. Results Beyond the Primary Endpoint
The registry reports 10 outcome measures and one posted statistical analysis. The formal statistical analysis posted for the primary endpoint is the ANCOVA analysis of change in HbA1c described above.
The registry also reports serious adverse events by treatment arm. The affected and at-risk counts were 69 of 644 for insulin icodec and 59 of 644 for IcoSema.
13. Safety: Serious Adverse Events
| Safety measure | Insulin icodec | IcoSema |
|---|---|---|
| Serious adverse events | 69 / 644 | 59 / 644 |
| At-risk population | 644 | 644 |
These are descriptive safety counts, not the result of a posted inferential comparison. The appropriate interpretation is therefore to describe the number of affected participants relative to the stated at-risk population rather than treating the difference as a formal test of a treatment effect.
The fact that both arms have the same stated at-risk population also makes the raw counts directly comparable descriptively within the registry's reported safety population.
14. Statistical Methods Explained
Why was ANCOVA used?
ANCOVA was used because the primary outcome was a continuous change in HbA1c and the analysis included baseline HbA1c as a covariate. The model also included region and randomized treatment as fixed factors. This allows the treatment difference to be estimated while accounting for baseline HbA1c and regional effects represented in the model.
What does a treatment difference of −0.66 mean?
The treatment difference is expressed in percentage points of HbA1c. A value of −0.66 for IcoSema versus insulin icodec means that the modeled change from baseline to week 52 was 0.66 percentage points lower in the IcoSema group.
Why is the confidence interval important?
The 95% confidence interval of −0.76 to −0.57 shows the statistical uncertainty around the point estimate. It provides information about precision that a P-value alone cannot provide. The interval concerns the population-level treatment difference under the specified analysis, not the range of individual participant responses.
Why does the P-value not measure effect size?
The P-value addresses the evidence against a null hypothesis under the statistical model and testing framework. It does not tell us how large the treatment difference is. The treatment difference of −0.66 and its confidence interval contain the information about the estimated magnitude and its precision.
Why use multiple imputation?
Week 52 HbA1c values can be missing for some participants. Multiple imputation represents uncertainty about those missing values through multiple completed datasets. Each dataset is analyzed separately, and Rubin's rules combine the results so that uncertainty from the imputation process is incorporated into the final inference.
Why does randomization matter to the ANCOVA?
Randomization creates the primary basis for comparing treatment groups. ANCOVA then adjusts the analysis for the prespecified model terms. The adjustment does not replace randomization; it uses additional baseline and regional information to improve the statistical analysis of the randomized comparison.
15. Confidence Interval vs P-value
| Quantity | COMBINE 1 result | Primary interpretation |
|---|---|---|
| Point estimate | −0.66 | Estimated treatment difference in HbA1c change |
| 95% CI | −0.76 to −0.57 | Statistical uncertainty and precision around the treatment difference |
| P-value | <0.0001 | Evidence concerning the null hypothesis under the specified testing framework |
The three quantities answer different questions. The point estimate describes the estimated effect. The confidence interval describes uncertainty around that estimate. The P-value describes the compatibility of the observed result with a specified null hypothesis.
A common mistake is to treat a very small P-value as evidence that an effect must be large. COMBINE 1 illustrates why the treatment estimate and confidence interval should remain central: the estimated treatment difference is −0.66 percentage points, with a 95% CI of −0.76 to −0.57. The P-value is additional evidence about the statistical hypothesis; it is not a replacement for the estimated treatment effect.
16. Superiority Testing
The registry identifies the primary hypothesis as superiority. In a superiority framework, the statistical question is whether the treatment groups differ in the prespecified direction or, more generally, whether the treatment effect differs from the null value specified for the comparison.
Here, the reported treatment difference is negative for IcoSema versus insulin icodec, and the two-sided 95% confidence interval is entirely below zero. The P-value is reported as less than 0.0001.
95% CI = −0.76 to −0.57
P < 0.0001
The confidence interval and P-value should be interpreted together with the prespecified superiority question and the ANCOVA model.
17. Missing Data and Analysis Assumptions
The registry explicitly states that missing HbA1c values at week 52 were handled through multiple imputation. This is statistically important because the final treatment comparison depends not only on observed measurements but also on the assumptions used to represent missing measurements.
Observed information
Participants with available HbA1c data contribute directly to the analysis of the corresponding outcome.
Imputed information
Missing week 52 HbA1c values are represented through the multiple-imputation procedure.
Repeated analyses
Each imputed dataset is analyzed separately using the ANCOVA model.
Combined inference
Rubin's rules combine the estimates and account for uncertainty between imputed datasets.
Consequently, the confidence interval should be understood as reflecting both the variability in the outcome analysis and the uncertainty introduced through the multiple-imputation procedure.
18. What the Primary Result Does — and Does Not — Establish
The posted ANCOVA analysis estimates a treatment difference of −0.66 percentage points in change in HbA1c from baseline to week 52 for IcoSema versus insulin icodec, with a 95% CI of −0.76 to −0.57.
The primary estimate does not describe every participant's individual response, does not provide a distribution of participant-level treatment effects, and does not by itself summarize the entire clinical benefit-risk profile of the trial.
The result is conditional on the specified ANCOVA model, the baseline HbA1c covariate, region and treatment factors, the Full Analysis Set framework, and the multiple-imputation procedure used for missing week 52 HbA1c values.
19. Limitations
- Open-label design: the registry identifies the masking as none. Consequently, treatment assignment was not masked.
- Primary endpoint interpretation: the registry identifies the endpoint type as binary, while the posted analysis uses ANCOVA for a continuous HbA1c change and reports a treatment difference in percentage points. The posted analysis is the relevant statistical description of the reported result.
- Missing outcome data: missing week 52 HbA1c values were handled by multiple imputation, so inference depends in part on the assumptions underlying that procedure.
- Available-data analysis: although the Full Analysis Set included all randomized participants, the registry states that the overall number analyzed consisted of participants with available data.
- Safety interpretation: serious adverse-event counts are descriptive in the posted information and should not be treated as a formal hypothesis test.
- Single posted formal analysis: the registry's posted formal statistical analysis for the primary endpoint is the ANCOVA comparison described above; additional outcome measures should not automatically be interpreted as having the same inferential status.
- External validity: the applicability of the findings to populations outside the trial depends on how closely those populations resemble the enrolled study population and treatment setting.
20. Why This Trial Matters Statistically
COMBINE 1 is a useful teaching example because it combines randomized treatment allocation with a baseline-adjusted continuous-outcome analysis. The statistical story is not simply a P-value: it involves the treatment difference, confidence interval, analysis population, covariate adjustment, missing-data strategy, and distinction between efficacy and safety summaries.
| Statistical concept | How it appears in COMBINE 1 |
|---|---|
| Randomization | Participants were randomized to IcoSema or insulin icodec in a parallel-group phase 3 design. |
| Full Analysis Set | The FAS included all randomized participants. |
| ANCOVA | The primary HbA1c change analysis used ANCOVA with region and randomized treatment as fixed factors and baseline HbA1c as a covariate. |
| Treatment difference | The primary effect measure was a difference in HbA1c change, reported as −0.66 percentage points. |
| Confidence interval | The 95% CI was −0.76 to −0.57. |
| P-value | The primary P-value was <0.0001. |
| Superiority | The registry classifies the primary hypothesis as superiority. |
| Multiple imputation | Missing week 52 HbA1c values were imputed, with estimates combined using Rubin's rules. |
| Safety analysis | Serious adverse events were reported descriptively by treatment arm. |
21. Trial Timeline
Trial start
The COMBINE 1 study began as a phase 3 randomized parallel-group trial in people with type 2 diabetes.
Primary measurement begins
The primary endpoint is defined from baseline HbA1c at week 0 through the week 52 assessment.
Primary endpoint assessment
Change in HbA1c from baseline to week 52 is evaluated using ANCOVA.
Primary completion
The registry records March 19, 2024 as the primary completion date.
22. A Practical Reading of the COMBINE 1 Analysis
A statistically literate reading of the primary result can be organized into five questions.
- What was compared? IcoSema versus insulin icodec.
- What outcome was measured? Change in HbA1c from baseline at week 52.
- What was the estimated difference? −0.66 percentage points.
- How precise was the estimate? The 95% CI was −0.76 to −0.57.
- What did the hypothesis test show? The P-value was <0.0001 under the posted ANCOVA framework.
This sequence is preferable to beginning with the P-value because it keeps the magnitude and uncertainty of the treatment comparison in view. The statistical method then explains how the estimate was obtained and how missing week 52 values were handled.
23. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The posted ANCOVA estimated a treatment difference of −0.66 percentage points, with a two-sided 95% CI of −0.76 to −0.57 and P < 0.0001.
Clinical interpretation
Clinical meaning requires considering the magnitude of the HbA1c difference together with the trial population, treatment context, other outcomes, and safety information rather than relying on statistical significance alone.
The statistical result answers a specific question about the randomized comparison of HbA1c change. It should not be expanded into conclusions about outcomes that are not represented by the primary endpoint.
24. Key Statistical Takeaways
| Question | Answer |
|---|---|
| What was the primary endpoint? | Change in HbA1c from baseline (week 0) to week 52. |
| What was the analysis method? | ANCOVA. |
| What factors were included? | Region and randomized treatment as fixed factors; baseline HbA1c as a covariate. |
| What analysis population was used? | Full Analysis Set, including all randomized participants; participants with available data contributed to the analysis. |
| How were missing week 52 HbA1c values handled? | Multiple imputation, with estimates combined using Rubin's rules. |
| What was the estimated treatment difference? | −0.66 percentage points. |
| What was the 95% confidence interval? | −0.76 to −0.57. |
| What was the P-value? | <0.0001. |
| What was the hypothesis type? | Superiority. |
| What serious adverse-event counts were reported? | 69/644 for insulin icodec and 59/644 for IcoSema. |
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Calculators
27. Sources
- ClinicalTrials.gov: NCT05352815 — COMBINE 1.
- PubMed: Publication indexed under PMID 42557064.
- PubMed: Publication indexed under PMID 42306490.
- PubMed: Publication indexed under PMID 40482671.
Continue through the Clinical Biostats knowledge graph
Connect this trial's statistical methods to deeper tutorials and practical statistical tools.
28. Record Summary
COMBINE 1 provides a clear example of a randomized phase 3 comparison using a continuous biomarker endpoint and a baseline-adjusted linear-model analysis. The primary endpoint was change in HbA1c from baseline to week 52, analyzed using ANCOVA with region and randomized treatment as fixed factors and baseline HbA1c as a covariate. Missing week 52 HbA1c values were addressed through multiple imputation, with estimates combined using Rubin's rules.
The resulting treatment difference for IcoSema versus insulin icodec was −0.66 percentage points, with a two-sided 95% confidence interval of −0.76 to −0.57 and a P-value <0.0001. The statistical interpretation is strongest when these three quantities are considered together: the estimate describes the magnitude and direction of the modeled difference, the confidence interval describes its precision, and the P-value addresses the prespecified superiority hypothesis under the analysis framework.
The trial also illustrates why analysis population and missing-data methodology matter. The Full Analysis Set included all randomized participants, while the registry states that participants with available data were included in the overall number analyzed. Multiple imputation was used for missing week 52 HbA1c values. Safety information should be considered separately from the primary efficacy analysis; the registry reports serious adverse events in 69 of 644 participants receiving insulin icodec and 59 of 644 receiving IcoSema.