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
DEVOTE was a double-blind, randomized, event-driven phase 3 trial designed to show that the long-acting basal insulin degludec (IDeg) was not associated with more major adverse cardiovascular events than insulin glargine (IGlar) in people with type 2 diabetes at high cardiovascular risk.
| Feature | DEVOTE |
|---|---|
| Official title | A Trial Comparing Cardiovascular Safety of Insulin Degludec Versus Insulin Glargine in Subjects With Type 2 Diabetes at High Risk of Cardiovascular Events |
| Phase | Phase 3 |
| Conditions | Diabetes; Diabetes Mellitus, Type 2 |
| Design | Randomized, parallel-group, double-masked, two arms |
| Primary purpose | Treatment |
| Primary endpoint | Time to first EAC-confirmed 3-component MACE (cardiovascular death, non-fatal myocardial infarction, non-fatal stroke) |
| Primary hypothesis | Non-inferiority of IDeg to IGlar (HR margin 1.3) |
| Enrollment | 7637 |
| Study period | 2013-10-29 to 2016-10-16 (primary completion) |
| ClinicalTrials.gov | NCT01959529 |
| Funding | Novo Nordisk A/S (industry) |
2. Clinical Question
DEVOTE was a cardiovascular safety trial. Its central question was not whether insulin degludec reduces cardiovascular events, but whether it could be shown, with a prespecified degree of confidence, not to increase the hazard of major adverse cardiovascular events by more than an acceptable amount relative to insulin glargine. Severe hypoglycaemia, a clinically important harm of insulin therapy, was examined as a secondary question under a superiority hypothesis.
Population
Adults with type 2 diabetes at high risk of cardiovascular events.
Intervention
Insulin degludec (IDeg), a long-acting basal insulin, administered double-blind.
Comparator
Insulin glargine (IGlar), an established long-acting basal insulin, administered double-blind.
Primary question
Is the hazard ratio for first EAC-confirmed MACE with IDeg versus IGlar credibly below 1.3, establishing cardiovascular non-inferiority?
3. Trial Design
Insulin degludec
- Long-acting basal insulin (IDeg)
- Double-masked administration
- Followed to individual end-of-trial date
Insulin glargine
- Long-acting basal insulin (IGlar)
- Double-masked administration
- Followed to individual end-of-trial date
Analysis population
| Analysis population | Definition / role |
|---|---|
| Full analysis set (FAS) | All randomised subjects; used for the primary MACE analysis and both severe hypoglycaemia analyses. |
| Serious adverse event summary | Participants at risk: 3818 (IDeg) and 3819 (IGlar), which together account for all 7637 enrolled. |
Using the full analysis set in a non-inferiority trial deserves comment. In superiority trials, an analysis of all randomised participants is usually conservative, because protocol deviations and non-adherence tend to blur differences toward the null. In a non-inferiority trial, blurring toward "no difference" makes non-inferiority easier to claim. Regulators and methodologists therefore often want to see a per-protocol or on-treatment analysis alongside the full-set analysis. The ClinicalTrials.gov record reports only the FAS analysis for the primary endpoint.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary: time to first MACE | Time from randomisation to first occurrence of an event adjudication committee (EAC)-confirmed 3-component major adverse cardiovascular event: cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke. Events with EAC-confirmed onset date between randomisation and individual end of trial were included. | From randomisation to individual end of trial date (maximum patient year observation: 2.75 years) |
| Secondary: number of severe hypoglycaemic episodes | Number of EAC-confirmed severe hypoglycaemic episodes | From randomisation to individual end of trial (maximum patient year observation: 2.75 years) |
| Secondary: any severe hypoglycaemic episode | Occurrence of at least one EAC-confirmed severe hypoglycaemic episode within a subject (yes/no) | From randomisation to individual end of trial date (maximum patient year observation: 2.75 years) |
The two secondary endpoints look at the same clinical phenomenon from two different statistical angles. The first counts all episodes, so a participant with repeated episodes contributes each of them; the second reduces each participant to a binary indicator of whether at least one episode occurred. These produce different effect measures (a rate ratio and an odds ratio) and answer different questions: how often severe hypoglycaemia happens overall, versus how many people experience it at all.
Central adjudication by an event adjudication committee is an important design feature. Cardiovascular death, myocardial infarction and stroke, as well as "severe" hypoglycaemia, all require clinical judgement to classify. Blinded, standardised adjudication reduces misclassification and ensures that both arms are assessed against the same criteria.
5. Statistical Methodology
Non-inferiority framework
The primary hypothesis was non-inferiority. The registry states that non-inferiority of IDeg to IGlar was considered confirmed if the upper limit of the two-sided 95% confidence interval for the hazard ratio was below 1.3, or equivalently if the p-value for the one-sided test of H0: HR ≥ 1.3 against Ha: HR < 1.3 was less than 2.5%.
The null hypothesis is that IDeg is unacceptably worse. Rejecting it at a one-sided 2.5% level is the same decision as observing an upper two-sided 95% confidence limit below 1.3. A margin of 1.3 is a common choice in cardiovascular safety trials of glucose-lowering therapies: it rules out a relative increase in the MACE hazard of 30% or more.
Cox regression for time to first MACE
The hazard ratio for IDeg versus IGlar was estimated with Cox regression, with investigational medicinal product as the only factor in the primary analysis. This is an unadjusted, unstratified treatment comparison: no baseline covariates or stratification factors enter the model according to the registry description. Participants who did not experience a confirmed MACE were censored at their individual end-of-trial date.
The baseline hazard h0(t) is left unspecified; only the ratio of hazards between arms is estimated, under the assumption that this ratio is constant over follow-up.
Negative binomial regression for episode counts
The number of EAC-confirmed severe hypoglycaemic episodes was analysed with negative binomial regression, with treatment (IDeg vs IGlar) as a fixed factor, fitted in the FAS. The effect measure is a rate ratio. Superiority was considered confirmed if the upper limit of the two-sided 95% CI for the rate ratio was below 1.0, or equivalently if the one-sided p-value for H0: RR ≥ 1.0 was less than 2.5%.
Logistic regression for any episode
Whether a participant experienced at least one EAC-confirmed severe hypoglycaemic episode was analysed by logistic regression with treatment as a fixed factor, fitted in the FAS. The effect measure is an odds ratio, with superiority confirmed if the upper 95% confidence limit was below 1.0, or equivalently if the one-sided p-value for H0: OR ≥ 1.0 was less than 2.5%. (The registry describes the model as logistic regression "with log-link function"; the reported effect measure is an odds ratio, which corresponds to the standard logit link.)
| Endpoint | Role | Method | Effect measure | Hypothesis / threshold |
|---|---|---|---|---|
| Time to first MACE | Primary | Cox regression | Hazard ratio | Non-inferiority; upper 95% CI < 1.3 |
| Number of severe hypoglycaemic episodes | Secondary | Negative binomial regression | Rate ratio | Superiority; upper 95% CI < 1.0 |
| Any severe hypoglycaemic episode | Secondary | Logistic regression | Odds ratio | Superiority; upper 95% CI < 1.0 |
The registry does not describe an interim analysis plan, a formal multiplicity strategy across the primary and secondary endpoints, or the handling of missing follow-up beyond censoring at the individual end of trial. These elements are therefore not discussed further here.
6. Primary Endpoint Result: Time to First MACE
Hazard ratio for first EAC-confirmed MACE (IDeg vs IGlar)
Two-sided 95% CI: 0.781–1.055 · P < 0.001 (non-inferiority)
Cox regression, treatment as only factor · Full analysis set, all randomised subjects
The upper limit of the 95% confidence interval, 1.055, lies well below the prespecified margin of 1.3. Non-inferiority of insulin degludec to insulin glargine for major adverse cardiovascular events was therefore established under the trial's prespecified criterion. The ClinicalTrials.gov record does not report the number of participants with a first MACE in each arm, Kaplan-Meier cumulative incidence, or results for the individual MACE components.
What the estimate means. An HR of 0.908 means that, under the Cox model and over the observed follow-up, the estimated hazard of a first MACE was about 9% lower with insulin degludec than with insulin glargine. The point estimate is on the favourable side of 1, but the size of the estimated difference is modest.
What it does not mean. It does not show that insulin degludec reduces cardiovascular events. The confidence interval of 0.781–1.055 includes 1.0, so the data are compatible both with a moderate reduction in hazard and with a small increase of up to roughly 5.5%. The trial answers the safety question (no excess of 30% or more is plausible) rather than a benefit question.
Precision. The interval is fairly narrow for a cardiovascular outcomes trial, reflecting a large event-driven sample. Its upper end, not its centre, is what matters for the non-inferiority decision: the whole interval sits below 1.3 with a clear gap.
Why the p-value is not an effect size. The reported P < 0.001 tests H0: HR ≥ 1.3, not H0: HR = 1. It indicates that an HR of 1.3 or higher is very incompatible with the data; it says nothing about whether IDeg is better than IGlar, and it would be a misreading to cite it as evidence of cardiovascular benefit.
Cautions. A single hazard ratio summarises the comparison only if hazards are approximately proportional over the follow-up period; the registry does not report a test of this assumption. The analysis used the full analysis set, which in a non-inferiority setting can bias toward similarity if adherence was imperfect. The comparison is also with an active insulin, so the result says that IDeg is not meaningfully worse than IGlar, not that either is cardiovascularly neutral relative to no insulin.
7. Secondary Endpoint Results: Severe Hypoglycaemia
Number of EAC-confirmed severe hypoglycaemic episodes
Rate ratio for severe hypoglycaemic episodes (IDeg vs IGlar)
Two-sided 95% CI: 0.476–0.759 · P < 0.001 (superiority)
Negative binomial regression, treatment as fixed factor · FAS
A rate ratio of 0.601 means the estimated rate of EAC-confirmed severe hypoglycaemic episodes with insulin degludec was about 60% of the rate with insulin glargine, a relative reduction of roughly 40%. The entire 95% CI (0.476–0.759) lies below 1.0, meeting the prespecified superiority criterion. Because this is a ratio of rates, it reflects the total burden of episodes, including repeat episodes in the same participants, and it does not by itself give the absolute number of episodes avoided; the registry analysis does not report the underlying episode rates per patient-year. The negative binomial model is well suited to such counts because it allows for overdispersion, the tendency for a minority of patients to have many episodes. As a secondary endpoint, this result is best read as supportive, particularly as the registry does not describe how type I error was controlled across endpoints.
Participants with at least one EAC-confirmed severe hypoglycaemic episode
Odds ratio for any severe hypoglycaemic episode (IDeg vs IGlar)
Two-sided 95% CI: 0.600–0.866 · P < 0.001 (superiority)
Logistic regression, treatment as fixed factor · FAS
An odds ratio of 0.729 means the estimated odds of experiencing at least one severe hypoglycaemic episode were about 27% lower with insulin degludec. The upper confidence limit of 0.866 is below 1.0, so superiority was also met on this person-level measure. An odds ratio is not a risk ratio: when an outcome is uncommon the two are similar, but the odds ratio moves further from 1 as the outcome becomes more frequent. The effect on the person-level odds (0.729) being smaller in magnitude than the effect on the episode rate (0.601) is consistent with the reduction in total episodes being driven partly by fewer repeat episodes, not only by fewer people having any episode. This binary analysis also ignores differences in follow-up time between participants, which the count model can accommodate.
| Endpoint | Effect measure | Estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|---|
| Time to first MACE | Hazard ratio | 0.908 | 0.781–1.055 | <0.001 | Non-inferiority (margin 1.3) |
| Number of severe hypoglycaemic episodes | Rate ratio | 0.601 | 0.476–0.759 | <0.001 | Superiority |
| Any severe hypoglycaemic episode | Odds ratio | 0.729 | 0.600–0.866 | <0.001 | Superiority |
8. Safety: Serious Adverse Events
| Arm | Participants with serious adverse events / at risk |
|---|---|
| Insulin degludec | 1473 / 3818 |
| Insulin glargine | 1517 / 3819 |
The numbers of participants with at least one serious adverse event were broadly similar between arms, with 44 fewer in the insulin degludec arm out of nearly identical denominators. These are descriptive counts; no formal statistical comparison of serious adverse events is reported in the registry, and a count of participants with any serious event does not capture the type, severity, or recurrence of events.
9. Statistical Methods Explained
Why is non-inferiority judged against the margin rather than against 1?
The trial's question was whether IDeg is unacceptably worse than IGlar, defined as a hazard ratio of 1.3 or more. The relevant test therefore places the null hypothesis at the margin (HR ≥ 1.3). A confidence interval that crosses 1.0, as this one does, is entirely consistent with a successful non-inferiority result: it only means superiority has not been shown.
What does an HR of 0.908 with an upper limit of 1.055 tell us?
The best single estimate is a slightly lower hazard with IDeg, but the plausible range extends from about a 22% lower hazard to about a 5.5% higher hazard. The key reassurance is that a 30% increase, the prespecified threshold for concern, lies far outside the interval.
Why does a two-sided 95% CI correspond to a one-sided 2.5% test?
A two-sided 95% interval leaves 2.5% in each tail. Non-inferiority only concerns one direction (whether the upper limit is below the margin), so checking the upper limit of a two-sided 95% CI is mathematically equivalent to a one-sided test at the 2.5% level. This is why the registry describes the two criteria as equivalent.
Why was a Cox model used for MACE?
Participants were followed for different lengths of time, up to 33.1 months, and many did not have a MACE before their end-of-trial date. Cox regression uses the timing of events and handles censored follow-up correctly, giving a hazard ratio without needing to specify the shape of the baseline hazard.
Why negative binomial rather than Poisson regression for episode counts?
Severe hypoglycaemia tends to cluster in susceptible individuals. A Poisson model assumes the variance equals the mean, which such clustered counts usually violate; the result would be confidence intervals that are too narrow. The negative binomial model adds a dispersion parameter, producing more honest uncertainty for the rate ratio.
Why report both a rate ratio and an odds ratio for severe hypoglycaemia?
They answer complementary questions. The rate ratio (0.601) describes the total episode burden, while the odds ratio (0.729) describes how many people are affected at least once. Seeing both helps distinguish fewer people affected from fewer repeat episodes among those affected.
10. Limitations
- Non-inferiority and the analysis set: the primary analysis used the full analysis set of all randomised subjects. In non-inferiority trials, non-adherence can bias toward similarity, and no supportive per-protocol analysis is reported in the registry.
- Active comparator: non-inferiority to insulin glargine is a relative statement; it does not establish absolute cardiovascular safety against placebo.
- Proportional hazards: the single HR assumes a roughly constant hazard ratio over follow-up; diagnostic checks are not reported in the registry.
- Relative measures only: the posted analyses give hazard, rate and odds ratios but not the event counts or episode rates needed to judge absolute differences.
- Secondary endpoints and multiplicity: the severe hypoglycaemia analyses are secondary, and the registry does not describe a hierarchy or other procedure controlling type I error across endpoints.
- Unadjusted models: each model included treatment as the only factor. Randomisation makes this a valid comparison, but covariate adjustment could have improved precision.
- Follow-up length: with a maximum of 2.75 patient-years per participant, the trial cannot speak to cardiovascular effects that might emerge over longer exposure.
- Generalizability: the population was selected for high cardiovascular risk, so estimates may not transfer directly to lower-risk people with type 2 diabetes.
11. Why This Trial Matters Statistically
DEVOTE is a clear teaching example of a cardiovascular safety trial, in which the key question is ruling out harm rather than demonstrating benefit, combined with a secondary efficacy-style question on hypoglycaemia analysed using two different models for the same clinical outcome.
| Concept | How it appears in DEVOTE |
|---|---|
| Non-inferiority design | HR margin of 1.3 for first MACE |
| One-sided testing | H0: HR ≥ 1.3 tested at the one-sided 2.5% level |
| Confidence intervals | Upper limit of 1.055 compared with the margin; CI includes 1.0 |
| Cox proportional-hazards model | Unadjusted treatment-only model for time to first MACE |
| Hazard ratio | 0.908 for IDeg versus IGlar |
| Negative binomial regression | Overdispersed counts of severe hypoglycaemic episodes |
| Rate ratio | 0.601 for severe hypoglycaemic episodes |
| Logistic regression and odds ratio | 0.729 for any severe hypoglycaemic episode |
| Event-driven design | Planned maximum 60.5 months; actual duration 35.6 months |
| Central adjudication | EAC-confirmed MACE and severe hypoglycaemia |
Statistical interpretation
The upper 95% confidence limit for the MACE hazard ratio was below the 1.3 margin, establishing non-inferiority. Both severe hypoglycaemia analyses met their superiority criteria, with confidence intervals entirely below 1.0.
What remains uncertain
Whether IDeg changes cardiovascular risk at all relative to IGlar is not resolved: the interval spans both a reduction and a small increase in hazard. Absolute differences cannot be judged from the ratio measures alone.
12. Related Tutorials
Learn more about the methods used in this trial:
13. Related Calculators
14. Sources
- ClinicalTrials.gov: NCT01959529 — A Trial Comparing Cardiovascular Safety of Insulin Degludec Versus Insulin Glargine in Subjects With Type 2 Diabetes at High Risk of Cardiovascular Events.
- PubMed: PMID 27595693.
- PubMed: PMID 28605603.
- PubMed: PMID 28913575.
- PubMed: PMID 28913543.
- PubMed: PMID 29713962.
Continue exploring Clinical Biostats
Connect the endpoints and methods of this trial to deeper statistical tutorials and practical calculators.
Record Summary
DEVOTE illustrates how a cardiovascular safety question is answered statistically: a prespecified non-inferiority margin, a one-sided hypothesis judged through the upper limit of a two-sided confidence interval, and a Cox model for time to first adjudicated event in an event-driven design. Its secondary analyses show how the same clinical outcome, severe hypoglycaemia, can be summarised as a rate ratio for total episodes and an odds ratio for any occurrence. Reading the trial well means keeping the questions separate: non-inferiority for MACE, superiority for severe hypoglycaemia, and descriptive serious adverse event counts.