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
CONCLUDE was a randomized, parallel, open-label phase 3 treatment trial comparing insulin degludec with insulin glargine 300 units/mL in subjects with type 2 diabetes mellitus inadequately treated with basal insulin, with or without oral antidiabetic drugs. The primary endpoint quantified severe or blood-glucose-confirmed symptomatic hypoglycaemic episodes during the 36-week Maintenance 2 period.
| Feature | CONCLUDE |
|---|---|
| Phase | Phase 3 |
| Condition | Diabetes mellitus, type 2 |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Allocation | Randomized |
| Enrollment | 1609 |
| Primary endpoint | Number of severe or BG-confirmed symptomatic hypoglycaemic episodes during Maintenance 2 (36 weeks) |
| Primary endpoint type | Count / rate |
| Primary effect measure | Rate ratio |
| Primary hypothesis | Superiority |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| Trial status | Completed |
| Start | March 13, 2017 |
| Primary completion | February 13, 2019 |
2. Clinical Question
The primary statistical question was whether the rate of severe or blood-glucose-confirmed symptomatic hypoglycaemic episodes during the 36-week Maintenance 2 period differed between insulin degludec 200 units and insulin glargine 300 units/mL.
Population
Subjects with type 2 diabetes mellitus inadequately treated with basal insulin, with or without oral antidiabetic drugs.
Intervention
Insulin degludec 200 units, identified in the analysis as IDeg U200.
Comparator
Insulin glargine 300 units/mL, identified in the analysis as IGlar U300.
Primary question
What is the relative rate of severe or BG-confirmed symptomatic hypoglycaemic episodes between the two randomized treatment groups during Maintenance 2?
3. Trial Design
Insulin degludec
- Insulin degludec
- Analysis population: 802 subjects at risk for the reported serious-adverse-event summary
- Primary comparison: severe or BG-confirmed symptomatic hypoglycaemic episode rate
Insulin glargine 300 units/mL
- Insulin glargine 300 units/mL
- Analysis population: 798 subjects at risk for the reported serious-adverse-event summary
- Primary comparison: severe or BG-confirmed symptomatic hypoglycaemic episode rate
The registry describes the overall design as randomized and parallel, with no masking. The study's primary purpose was treatment. The absence of masking is relevant when interpreting outcomes that can be influenced by knowledge of treatment assignment, although the primary endpoint is an event-based hypoglycaemia measure rather than a subjective symptom score alone.
4. Primary Endpoint
| Endpoint | Registry definition | Time frame | Outcome unit |
|---|---|---|---|
| Number of Severe or BG-confirmed Symptomatic Hypoglycaemic Episodes During Maintenance 2 | Severe or BG-confirmed symptomatic hypoglycaemia was evaluated during the Maintenance 2 period. A severe or BG-confirmed symptomatic hypoglycaemic episode was an episode that was severe according to the ADA 2013 classification or blood glucose confirmed by a plasma glucose value <3.1 mmol/L (56 mg/dL) with symptoms consistent with hypoglycaemia. | 36 Weeks | Episodes/(PYE × 100) |
The endpoint is expressed as an observed rate: the number of episodes divided by patient-years of exposure, multiplied by 100. This distinction is important. The endpoint is not simply the percentage of subjects who experienced at least one episode. A subject can contribute more than one episode, and the exposure-time denominator accounts for differences in the amount of time subjects are observed and at risk.
The registry therefore expresses the outcome as episodes per 100 patient-years of exposure rather than as a simple proportion of participants with an event.
5. Statistical Methodology
Analysis population
The posted primary analysis used the safety analysis set (SAS), comprising all subjects who received at least one dose of the investigational product or comparator. This is distinct from an efficacy population defined solely by randomization: the registry specifically identifies the safety analysis set as the population for the posted primary endpoint analysis.
Primary regression analysis
The registry identifies the reported method as Cox regression and normalizes it as a Cox proportional-hazards model. The effect measure is a treatment rate ratio. The detailed analysis notes also state that the number of hypoglycaemic episodes was analyzed using a negative binomial regression model with a log link, with the logarithm of the relevant treatment-emergent exposure period included as an offset.
Covariate adjustment
The negative binomial model included treatment, number of oral antidiabetic drugs, region, sex, and dosing time as fixed factors, with age as a covariate. Thus, the primary estimate is not a crude comparison of observed episode rates alone. It is an adjusted treatment comparison that accounts for the prespecified factors listed in the registry analysis notes.
Missing-data handling
Missing values were imputed through multiple imputation by treatment arm, based on a Poisson model. Multiple imputation acknowledges uncertainty about missing values by generating multiple plausible completed datasets rather than treating a single imputed value as if it were known without uncertainty.
The exposure term acts as an offset, so the model estimates episode rates rather than merely comparing raw event counts.
6. Primary Result
Severe or BG-confirmed symptomatic hypoglycaemic episodes
Treatment rate ratio
95% CI: 0.73–1.06 · P = 0.1715
Insulin degludec 200 vs insulin glargine 300 units/mL
| Primary endpoint | IDeg U200 vs IGlar U300 |
|---|---|
| Endpoint | Number of severe or BG-confirmed symptomatic hypoglycaemic episodes during Maintenance 2 |
| Time frame | 36 Weeks |
| Analysis population | Safety analysis set |
| Statistical model | Cox regression in the registry method field; detailed analysis description specifies negative binomial regression with log link and exposure-time offset |
| Effect measure | Treatment rate ratio |
| Estimate | 0.88 |
| 95% CI | 0.73–1.06 |
| P-value | 0.1715 |
| Hypothesis | Superiority |
The estimated treatment rate ratio of 0.88 means that the modeled episode rate for insulin degludec 200 was estimated at 88% of the corresponding rate for insulin glargine 300 units/mL. Expressed as a relative difference, this corresponds to an estimated 12% lower episode rate for insulin degludec under the fitted model.
The estimate does not mean that 12% fewer subjects experienced hypoglycaemia, nor does it mean that each individual subject had exactly 12% fewer episodes. The endpoint counts episodes and incorporates patient-years of exposure, so it addresses an event rate rather than a simple patient-level risk.
The 95% confidence interval of 0.73–1.06 describes the statistical uncertainty around the estimated rate ratio. Because the interval includes 1, the observed estimate is compatible with both a lower and a higher episode rate under the specified model and sampling framework.
The P-value of 0.1715 is a measure of compatibility between the observed result and the null hypothesis under the specified statistical test. It is not a measure of effect size, clinical importance, or the probability that one treatment is better than the other.
Because the endpoint concerns recurrent episodes and the detailed analysis specifies negative binomial regression with an exposure-time offset, interpretation should focus on the modeled episode rate. The registry also identifies Cox regression as the reported method, so readers should distinguish the registry's method field from the more detailed model description when reconstructing the statistical analysis.
7. How to Interpret the Rate Ratio
A rate ratio is especially useful when an outcome can occur repeatedly and when participants contribute different amounts of observation time. The numerator represents the modeled event frequency and the denominator represents exposure time. This is different from a binary endpoint such as whether a participant ever experienced at least one event.
Rate ratio = 1
A rate ratio of 1 would indicate equal modeled episode rates between the two treatment groups under the specified analysis.
Rate ratio < 1
A value below 1 indicates a lower modeled episode rate for insulin degludec relative to insulin glargine in the direction of the comparison.
Rate ratio > 1
A value above 1 would indicate a higher modeled episode rate for insulin degludec relative to insulin glargine.
Why exposure matters
Two groups can have the same number of episodes but different episode rates if their total patient-years of exposure differ.
This relative interpretation follows from the treatment rate ratio. It does not convert the result into an absolute number of episodes or an individual patient's probability of hypoglycaemia.
8. Statistical Methods Explained
Why analyze episodes as a rate?
The primary endpoint counts severe or BG-confirmed symptomatic hypoglycaemic episodes rather than merely identifying whether a subject had at least one episode. Because repeated episodes can occur, a rate-based approach preserves more information than reducing each subject to a simple yes/no indicator. The registry also specifies patient-years of exposure in the endpoint definition, making exposure time an integral part of the estimand.
Why use an exposure-time offset?
An offset adjusts the model for the amount of time over which episodes could be observed. In a log-link count model, the logarithm of exposure time enters with a fixed coefficient of 1. Conceptually, this changes the model from comparing expected counts to comparing expected counts per unit of exposure.
What does a rate ratio of 0.88 mean?
A rate ratio of 0.88 means the estimated episode rate in the insulin degludec group was 0.88 times the corresponding rate in the insulin glargine group, after accounting for the variables included in the model. It does not mean that 88% of subjects were event-free or that an individual patient's risk was 12% lower.
Why does the confidence interval matter?
The point estimate is only one possible estimate supported by the observed data. The 95% confidence interval of 0.73–1.06 communicates uncertainty around that estimate. Its inclusion of 1 means that equality of the modeled rates is within the interval of values compatible with the statistical framework used for the analysis.
Why is the P-value not an effect-size measure?
The P-value of 0.1715 summarizes evidence against a specified null hypothesis under the statistical model. It depends on the observed data and the variability represented by the analysis. It does not tell the reader how large the treatment effect is. The rate ratio and its confidence interval provide the information about magnitude and precision.
Why does covariate adjustment matter?
The analysis included treatment, number of oral antidiabetic drugs, region, sex, and dosing time as fixed factors and age as a covariate. Adjustment can improve precision and account for prognostic or design-related variables included in the statistical model. The adjusted rate ratio should therefore be interpreted as the treatment comparison generated by that specified model rather than as an unadjusted descriptive ratio.
9. Multiple Imputation and Missing Data
The registry states that missing values were imputed through multiple imputation by treatment arm based on a Poisson model. Missing-data handling matters because the amount of observed information can differ across randomized treatment groups and because incomplete observations can affect estimated event rates.
Single imputation
Replacing each missing value with one fixed value can make the completed dataset appear more certain than the original data justify.
Multiple imputation
Multiple completed datasets can represent uncertainty about the missing observations, with the resulting estimates combined across imputations.
Here, the registry specifically identifies treatment-arm-based multiple imputation and a Poisson model. That detail matters because missing-data assumptions are part of the statistical analysis rather than merely a data-cleaning step.
10. Safety Results
The registry reports serious adverse events by treatment arm using the number affected divided by the number at risk.
| Safety measure | IDeg U200 | IGlar U300 |
|---|---|---|
| Serious adverse events | 167 / 802 | 141 / 798 |
The serious-adverse-event figures are presented descriptively here. The registry does not provide a formal statistical analysis for this safety measure in the posted analysis described for the primary endpoint. The denominators also differ from one another, so the counts alone should not be interpreted as treatment-group rates.
11. Safety and Efficacy Use Different Statistical Questions
The primary endpoint and serious-adverse-event summary illustrate why clinical-trial analysis cannot be reduced to a single number. The primary endpoint evaluates a recurrent hypoglycaemia event rate during a specified 36-week period and uses a model incorporating exposure time and covariates. The serious-adverse-event summary reports affected subjects relative to subjects at risk.
| Feature | Primary hypoglycaemia endpoint | Serious adverse events |
|---|---|---|
| Outcome structure | Repeated episodes | Subjects affected |
| Reported measure | Episodes/(PYE × 100) | Affected / at risk |
| Formal posted analysis | Yes | No formal analysis identified in the posted statistical analysis |
| Effect measure | Treatment rate ratio | Descriptive counts by arm |
| Primary population stated | Safety analysis set | At-risk denominators reported by arm |
12. Statistical Interpretation of the Primary Result
The estimated treatment rate ratio was 0.88. On the relative-rate scale, the estimate corresponds to a 12% lower modeled rate of severe or BG-confirmed symptomatic hypoglycaemic episodes for insulin degludec compared with insulin glargine 300 units/mL.
The 95% confidence interval was 0.73–1.06. The interval spans the null value of 1, indicating that the data support a range of possible relative rate differences that includes no difference.
The two-sided P-value was 0.1715. This is not a measure of the size of the treatment effect. It summarizes the statistical evidence against the specified null hypothesis under the analysis framework.
The rate ratio does not provide the absolute number of episodes avoided, the proportion of subjects who would experience an episode, or the treatment effect for any particular individual. Those quantities require different estimands and corresponding data.
13. Understanding the Analysis Population
The posted statistical analysis used the safety analysis set, defined as all subjects who received at least one dose of the investigational product or comparator. This choice has a direct interpretation: the primary reported hypoglycaemia analysis is anchored to treatment exposure rather than exclusively to randomized assignment.
That distinction is especially relevant for safety-oriented endpoints. In a randomized trial, an intention-to-treat analysis preserves the randomized comparison for many efficacy questions. By contrast, exposure-based safety analyses commonly focus on participants who actually received treatment, because adverse-event observation depends on treatment exposure.
14. Covariate Adjustment
The primary model included several variables beyond treatment assignment: number of oral antidiabetic drugs, region, sex, and dosing time as fixed factors, with age as a covariate.
| Model component | Role in the reported analysis |
|---|---|
| Treatment | Primary comparison between IDeg U200 and IGlar U300 |
| Number of OADs | Fixed factor |
| Region | Fixed factor |
| Sex | Fixed factor |
| Dosing time | Fixed factor |
| Age | Covariate |
| Exposure time | Offset in the detailed negative binomial model |
Adjustment does not turn the analysis into a comparison of perfectly matched individuals. Instead, it specifies how the model accounts for the listed variables while estimating the treatment effect. The resulting rate ratio is therefore conditional on the model structure and covariate specification.
15. Why the Endpoint Is Not a Simple Proportion
Consider two hypothetical studies with identical numbers of hypoglycaemic episodes but different total patient-years of exposure. The study with more exposure time would have a lower observed episode rate. A count alone would therefore fail to represent the amount of opportunity for events to occur.
Binary endpoint
Answers whether a participant experienced an event at least once during a defined period.
Count endpoint
Counts how many events occurred, allowing repeated events to contribute information.
Rate endpoint
Relates the number of events to the amount of observation or exposure time.
CONCLUDE primary endpoint
Uses the number of severe or BG-confirmed symptomatic hypoglycaemic episodes per patient-years of exposure, multiplied by 100.
16. Why This Trial Matters Statistically
CONCLUDE is a useful statistical teaching case because the primary endpoint requires several layers of interpretation at once: recurrent events, exposure time, regression modeling, covariate adjustment, missing-data handling, and an effect measure that is different from a simple risk ratio.
| Concept | How it appears in CONCLUDE |
|---|---|
| Randomization | Subjects were randomly allocated to two parallel treatment groups. |
| Repeated events | The primary endpoint counts severe or BG-confirmed symptomatic hypoglycaemic episodes. |
| Exposure time | The endpoint is expressed per patient-years of exposure. |
| Rate ratio | The reported treatment effect is a rate ratio of 0.88 with a 95% CI of 0.73–1.06. |
| Regression adjustment | The detailed model includes treatment, number of OADs, region, sex, dosing time, and age. |
| Negative binomial regression | The detailed analysis description specifies a negative binomial model with a log link. |
| Offset | The logarithm of the treatment-emergent exposure period is used as an offset. |
| Multiple imputation | Missing values were imputed through multiple imputation by treatment arm based on a Poisson model. |
| Hypothesis testing | The primary hypothesis type is superiority and the reported P-value is two-sided. |
| Safety analysis | The primary posted analysis uses the safety analysis set; serious adverse events are reported by arm. |
17. Limitations and Interpretation Issues
- Rate rather than risk: the primary endpoint measures recurrent episode rates per patient-years of exposure. It should not be interpreted as the proportion of subjects experiencing hypoglycaemia.
- Model dependence: the reported treatment rate ratio depends on the specified regression model, covariates, offset, and missing-data procedure.
- Method-field distinction: the registry identifies Cox regression in its the statistical method, while the detailed analysis notes describe negative binomial regression with a log link and exposure-time offset. These descriptions should not be treated as interchangeable without reference to the exact registry analysis record.
- Missing-data assumptions: multiple imputation depends on the statistical model used to generate plausible values for missing observations. The registry identifies a Poisson model and treatment-arm-specific imputation.
- Analysis population: the posted primary analysis uses the safety analysis set rather than an explicitly stated all-randomized intention-to-treat population.
- Open-label design: the trial was not masked. Knowledge of treatment assignment can matter for outcomes involving clinical recognition or reporting, although the primary endpoint includes objective blood-glucose confirmation as one qualifying pathway.
- Limited posted statistical analyses: the registry reports one formal statistical analysis for the primary endpoint. Other clinical outcomes should not be interpreted as formally tested treatment effects unless a corresponding statistical analysis is reported.
18. Trial Timeline
Trial start
The registry lists March 13, 2017 as the study start date.
Maintenance 2
The primary endpoint was evaluated during the 36-week Maintenance 2 period, with severe or BG-confirmed symptomatic hypoglycaemic episodes summarized as a rate per patient-years of exposure.
Primary completion
The registry lists February 13, 2019 as the primary completion date.
Registry status
The trial is listed as completed, with results and one posted statistical analysis available in the registry.
19. What the Confidence Interval Tells You
The confidence interval is often more informative than the P-value when evaluating the practical size and precision of an estimated treatment effect.
The point estimate is below 1, while the confidence interval extends below and above 1.
The lower bound of 0.73 is compatible with a substantially lower modeled episode rate for insulin degludec, while the upper bound of 1.06 is compatible with a slightly higher modeled rate. The interval therefore communicates substantially more information than the point estimate alone.
Importantly, a confidence interval is not a range containing 95% of individual treatment effects. It describes uncertainty in the estimated population parameter under the specified statistical framework.
20. The P-value in Context
The reported two-sided P-value is 0.1715. Because the trial's primary hypothesis type is listed as superiority, the relevant question is whether the observed treatment comparison provides sufficient statistical evidence against the null hypothesis under the prespecified framework.
What the P-value does
It quantifies the compatibility of the observed result with the null hypothesis under the specified statistical model and testing framework.
What the P-value does not do
It does not measure the size of the rate difference, clinical importance, or the probability that one treatment is superior.
For this reason, the P-value should be read together with the treatment rate ratio and its confidence interval. Those three quantities provide complementary information about the direction, magnitude, and uncertainty of the estimated treatment comparison.
21. Primary Statistical Story
| Question | CONCLUDE result |
|---|---|
| What was measured? | Severe or BG-confirmed symptomatic hypoglycaemic episodes during Maintenance 2 |
| Over what period? | 36 Weeks |
| How was the endpoint expressed? | Episodes/(PYE × 100) |
| Who was analyzed? | Safety analysis set |
| What treatment comparison? | Insulin degludec 200 vs insulin glargine 300 units/mL |
| What effect measure? | Treatment rate ratio |
| What was estimated? | 0.88 |
| How precise? | 95% CI 0.73–1.06 |
| What was the P-value? | 0.1715 |
| What was the hypothesis type? | Superiority |
The central statistical lesson is that the result must be interpreted on the scale on which it was estimated. The primary endpoint is a recurrent-event rate, the effect measure is a rate ratio, and the model adjusts for specified covariates while accounting for exposure time. Treating the 0.88 estimate as if it were a simple risk ratio would change the question being answered.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT03078478, CONCLUDE.
- Linked publication: PubMed PMID 31984443.
- Linked publication: PubMed PMID 32306247.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical trial analyses covering study design, endpoints, regression, survival analysis, and statistical inference.
25. Record Summary
CONCLUDE provides a useful example of how clinical-trial statistics change when the outcome is a recurrent event rate rather than a one-time binary outcome. The primary endpoint counted severe or BG-confirmed symptomatic hypoglycaemic episodes during 36 weeks and expressed them relative to patient-years of exposure. The posted analysis compared insulin degludec 200 with insulin glargine 300 units/mL using a treatment rate ratio of 0.88, with a 95% confidence interval of 0.73–1.06 and a two-sided P-value of 0.1715.
The detailed analysis description adds important statistical structure: negative binomial regression with a log link, an exposure-time offset, adjustment for treatment-related and demographic factors, and multiple imputation by treatment arm based on a Poisson model. These details explain why the reported rate ratio should be interpreted as a model-based comparison of episode rates rather than as a simple percentage difference in patients experiencing hypoglycaemia.