This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results and design details on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
SWITCH 1 was a randomized, double-blind, phase 3 crossover trial comparing insulin degludec (IDeg) with insulin glargine (IGlar), with insulin aspart used as mealtime insulin, in subjects with type 1 diabetes. The registry reports 501 enrolled subjects, two treatment arms, six posted outcome measures, and five posted statistical analyses.
| Feature | SWITCH 1 |
|---|---|
| Trial name | SWITCH 1 |
| NCT ID | NCT02034513 |
| Phase | Phase 3 |
| Condition | Diabetes; Diabetes Mellitus, Type 1 |
| Allocation | Randomized |
| Design model | Crossover |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 501 |
| Interventions | Insulin degludec; insulin glargine; insulin aspart |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| Status | Completed |
| Start | 2014-01-05 |
| Primary completion | 2016-01-11 |
2. Clinical Question
The central statistical question was whether insulin degludec could be shown to be non-inferior to insulin glargine with respect to the number of treatment-emergent severe or blood-glucose-confirmed symptomatic hypoglycaemic episodes during the maintenance period, with the registered hierarchical testing framework also evaluating subsequent endpoints.
Population
Subjects with diabetes mellitus, type 1, enrolled in a phase 3 randomized crossover trial.
Intervention
Insulin degludec (IDeg), with insulin aspart as mealtime insulin.
Comparator
Insulin glargine (IGlar), with insulin aspart as mealtime insulin.
Primary question
Is the treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episode rate with IDeg no worse than the prespecified non-inferiority margin relative to IGlar?
3. Trial Design
Insulin degludec
- Insulin degludec was one of the two randomized treatment interventions.
- Insulin aspart was used as mealtime insulin.
- The primary comparison was IDeg versus IGlar.
Insulin glargine
- Insulin glargine was the comparator randomized treatment.
- Insulin aspart was used as mealtime insulin.
- The primary comparison was IDeg versus IGlar.
The crossover structure is important statistically because each treatment sequence contains exposure to both randomized treatment conditions. The registry's statistical analyses therefore distinguish between the safety analysis set, the full analysis set, and the subjects exposed in the relevant maintenance periods.
4. Trial Timeline and Analysis Structure
Trial start
The registered trial start date was January 5, 2014.
Primary maintenance-period assessment
The primary endpoint was the number of treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes during a 16-week treatment period.
First HbA1c assessment
Change from baseline in HbA1c was evaluated at Week 32 as a supportive efficacy endpoint and was analyzed using MMRM.
Second HbA1c assessment
Change from baseline in HbA1c was also evaluated at Week 64, with the Week 32 values serving as the baseline for the second treatment period.
Primary completion
The registered primary completion date was January 11, 2016.
5. Endpoints
| Endpoint | Time frame | Statistical role |
|---|---|---|
| Number of Treatment Emergent Severe or BG (Blood Glucose) Confirmed Symptomatic Hypoglycaemic Episodes During the Maintenance Period | A 16-week treatment period. | Primary endpoint |
| Number of Treatment Emergent Severe or BG Confirmed Symptomatic Nocturnal Hypoglycaemic Episodes During the Maintenance Period | After 16 weeks of treatment, in each treatment period (Week 16-32 and Week 48-64) | Secondary endpoint |
| Change From Baseline in HbA1c (Glycosylated Haemoglobin) | Week 32, Week 64 | Secondary supportive efficacy endpoint |
| Proportion of Subjects With One or More Severe Hypoglycaemic Episodes During the Maintenance Period | After 16 weeks of treatment, in each treatment period (Week 16-32 and Week 48-64) | Secondary endpoint |
Primary endpoint definition
The registered primary endpoint counted severe or blood-glucose-confirmed symptomatic hypoglycaemic episodes. The registry defines these as episodes that were severe and/or BG confirmed by a plasma glucose value of <3.1 mmol/L (56 mg/dL), with symptoms consistent with hypoglycaemia. A treatment-emergent hypoglycaemic episode was defined as an event with onset on or after the first day of exposure to randomized treatment, within the registry-defined treatment-emergent period.
6. Analysis Populations
| Population / set | Role described in the registry data |
|---|---|
| Full Analysis Set (FAS) | Used for the statistical analysis of the primary endpoint and the reported HbA1c analyses. For the primary endpoint, the FAS included subjects exposed in at least one maintenance period. |
| Safety Analysis Set (SAS) | Descriptive safety analysis was based on subjects receiving at least 1 dose of the investigational product, IDeg or its comparator, IGlar. |
| FAS exposed in both maintenance periods | Used for the statistical analysis of the proportion of subjects with one or more severe hypoglycaemic episodes. |
| HbA1c analysis sets | Both descriptive and statistical analysis were based on the FAS; the reported n specifies subjects with available data at the specified time point. |
7. Statistical Methodology
Poisson regression for hypoglycaemic episode counts
The primary endpoint counts treatment-emergent hypoglycaemic episodes rather than simply classifying each subject as having or not having an event. The registry reports Poisson regression for this analysis and expresses the treatment effect as a treatment ratio.
For count/rate outcomes, a Poisson model relates an expected event count to an underlying event rate and the amount of observation or exposure. The reported treatment ratio compares the estimated event rates between IDeg and IGlar under the fitted analysis.
Non-inferiority framework
The primary analysis used a non-inferiority framework. Non-inferiority was considered confirmed if the upper bound of the two-sided 95% confidence interval for the rate ratio (IDeg/IGlar) was ≤1.10. The registry also describes an equivalent one-sided hypothesis test of H0: RR >1.10 against HA: RR ≤1.10, with a one-sided significance level of 2.5%.
Prespecified non-inferiority margin
For the rate ratio IDeg/IGlar, the upper bound of the two-sided 95% CI had to be ≤1.10 for non-inferiority to be confirmed.
The registry describes the corresponding one-sided test at the 2.5% level.
MMRM for repeated HbA1c measurements
Change from baseline in HbA1c at Week 32 and Week 64 was analyzed using a mixed model for repeated measures (MMRM). The registry-reported analysis text identifies treatment, sex, region, and pre-trial insulin treatment among the model considerations. The registry also identifies covariate adjustment as an analysis concept.
McNemar test for paired binary outcomes
The proportion of subjects with one or more severe hypoglycaemic episodes was analyzed using the McNemar test. This is consistent with a crossover setting in which each subject contributes paired binary treatment-period information. The statistical analysis was based on the FAS subjects exposed in both maintenance periods.
Multiplicity and hierarchical testing
The registry explicitly identifies multiplicity adjustment in the hypoglycaemia analyses and describes a stepwise hierarchical testing procedure for confirmatory endpoints. The primary endpoint was Step 1. If non-inferiority was confirmed, Step 2 evaluated nocturnal hypoglycaemic episodes, and Step 3 evaluated the proportion of subjects with one or more severe hypoglycaemic episodes.
| Step | Endpoint | Statistical framework |
|---|---|---|
| Step 1 | Number of treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes | Poisson regression; non-inferiority |
| Step 2 | Number of treatment-emergent severe or BG-confirmed symptomatic nocturnal hypoglycaemic episodes | Poisson regression; non-inferiority |
| Step 3 | Proportion of subjects with one or more severe hypoglycaemic episodes | McNemar test; superiority |
8. Primary Result: Treatment-Emergent Severe or BG-Confirmed Symptomatic Hypoglycaemic Episodes
The primary endpoint was analyzed using Poisson regression in the FAS. The comparison was insulin degludec versus insulin glargine, with the effect expressed as a treatment ratio.
Treatment ratio for hypoglycaemic episodes
95% CI: 0.85–0.94 · P < 0.0001
Non-inferiority margin: upper confidence-limit criterion of 1.10
A treatment ratio of 0.89 means that the estimated treatment-period hypoglycaemic episode rate under the fitted Poisson analysis was 0.89 times the corresponding rate for insulin glargine. Expressed as a simple relative interpretation, this corresponds to an estimated rate about 11% lower with insulin degludec.
The ratio does not mean that 11% of subjects avoided hypoglycaemia, that every subject had an 11% reduction, or that the absolute number of events differed by a fixed amount for every participant. It is a model-based relative rate measure.
The two-sided 95% confidence interval of 0.85–0.94 describes statistical uncertainty around the estimated treatment ratio under the analysis framework. It does not describe the range of individual patient responses.
The p-value of <0.0001 describes the evidence against the null hypothesis associated with the reported statistical test; it is not a measure of the magnitude or clinical importance of the treatment effect. The non-inferiority conclusion is governed by the prespecified margin and its confidence-interval or one-sided-test criterion, not by the p-value alone.
The registry identifies a hierarchical testing procedure and multiplicity adjustment. That matters because the subsequent confirmatory endpoints were tested only within the prespecified sequence.
9. Secondary Result: Nocturnal Hypoglycaemic Episodes
The second step of the hierarchical confirmatory sequence evaluated treatment-emergent severe or BG-confirmed symptomatic nocturnal hypoglycaemic episodes after 16 weeks of treatment in each treatment period, specifically Week 16-32 and Week 48-64.
Treatment ratio for nocturnal episodes
95% CI: 0.56–0.73 · P < 0.0001
Non-inferiority margin: upper confidence-limit criterion of 1.10
A treatment ratio of 0.64 indicates an estimated nocturnal hypoglycaemic episode rate 0.64 times the corresponding rate under insulin glargine. As a simple relative-rate description, this is an estimated rate about 36% lower with insulin degludec.
The 95% confidence interval of 0.56–0.73 describes precision around the estimated treatment ratio. It does not mean that individual subjects have treatment effects confined to this interval.
The reported P < 0.0001 provides evidence against the null hypothesis associated with the statistical test, but a p-value should not be interpreted as the size of the treatment effect. For the non-inferiority question, the key design feature is the prespecified 1.10 margin and the location of the confidence interval relative to that margin.
Because this endpoint followed the primary endpoint in a stepwise hierarchical procedure, its confirmatory interpretation depends on the preceding step of the hierarchy being successfully passed.
10. Secondary HbA1c Results
Change from baseline in HbA1c was evaluated at Week 32 and Week 64. The comparisons were between the two crossover treatment sequences: IDeg/IGlar versus IGlar/IDeg. MMRM was used for the statistical analysis, with covariate adjustment described in the registry analysis text.
| Time point | Comparison | Analysis sample | Estimate | 95% CI | Hypothesis |
|---|---|---|---|---|---|
| Week 32 | IDeg/IGlar vs IGlar/IDeg | 437; n=220 for IDeg and n=217 for IGlar | 0.03 | -0.10 to 0.15 | Non-inferiority; margin 0.40% |
| Week 64 | IDeg/IGlar vs IGlar/IDeg | 410; n=202 for IDeg and n=208 for IGlar | 0.11 | -0.00 to 0.23 | Non-inferiority; margin 0.40% |
Week 32
The estimated treatment contrast was 0.03, with a two-sided 95% CI of -0.10 to 0.15. The registry states that non-inferiority was confirmed when the upper confidence bound was ≤0.40%.
Week 64
The estimated treatment contrast was 0.11, with a two-sided 95% CI of -0.00 to 0.23. The same ≤0.40% upper-bound criterion was used for non-inferiority.
These HbA1c estimates are treatment contrasts, not ratios. The Week 32 estimate of 0.03 and Week 64 estimate of 0.11 describe the modeled difference in change from baseline between the crossover treatment groups under the MMRM framework.
The confidence intervals provide the uncertainty around those estimated contrasts. For Week 32, the upper bound is 0.15; for Week 64, it is 0.23. Both are below the registry's stated non-inferiority margin of 0.40%.
The fact that the confidence intervals include or approach zero is a reminder that non-inferiority and superiority are different questions. A non-inferiority analysis can support a conclusion that a treatment is not unacceptably worse without establishing superiority.
11. Secondary Result: Subjects With One or More Severe Hypoglycaemic Episodes
The third step of the reported hierarchical confirmatory sequence evaluated the proportion of subjects with one or more severe hypoglycaemic episodes during the maintenance period. The analysis used the McNemar test and was classified in the registry as a superiority hypothesis.
McNemar test
Analysis based on FAS subjects exposed in both maintenance periods
Hypothesis type: Superiority
The reported P = 0.0016 is evidence against the null hypothesis for the McNemar test under the registry's stated superiority analysis. Because the ClinicalTrials.gov record does not provide the paired subject-level cell counts or the treatment-specific percentages, the result cannot be translated from the ClinicalTrials.gov record into an absolute percentage difference.
The McNemar test is specifically suited to paired binary outcomes. In this crossover setting, the relevant information is the pattern of whether an individual subject experienced at least one severe episode under each treatment condition. It is therefore not the same analysis as comparing two independent proportions.
The p-value does not measure the size of the treatment effect. The registry also identifies multiplicity adjustment and places this endpoint at Step 3 of the hierarchical confirmatory sequence.
12. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. These figures are based on the safety information provided for the two randomized treatments.
| Safety measure | Insulin Degludec (IDeg) | Insulin Glargine (IGlar) |
|---|---|---|
| Serious adverse events, affected / at risk | 58 / 454 | 70 / 460 |
The reported serious-adverse-event counts should be kept conceptually separate from the efficacy analyses. The registry describes the safety analysis set as subjects receiving at least 1 dose of IDeg or IGlar. The ClinicalTrials.gov record does not provide a formal between-treatment statistical test for these serious-adverse-event counts, so no comparative p-value or confidence interval is presented here.
13. Statistical Methods Explained
Why was Poisson regression used for the primary endpoint?
The primary outcome is a number of hypoglycaemic episodes, rather than a simple yes/no outcome. Poisson regression is a standard modeling framework for count data and can express the comparison as a rate ratio or treatment ratio. The registry specifically reports Poisson regression and a treatment ratio for the primary analysis.
Why is the non-inferiority margin 1.10 important?
The margin defines the largest relative rate increase that the trial's non-inferiority framework was designed to exclude. For the primary endpoint, non-inferiority was considered confirmed if the upper bound of the two-sided 95% confidence interval for IDeg/IGlar was ≤1.10. A p-value by itself does not define the non-inferiority criterion.
What does a treatment ratio of 0.89 mean?
A treatment ratio of 0.89 means the estimated event rate under IDeg was 0.89 times the corresponding rate under IGlar in the reported Poisson analysis. It is a relative rate measure. It does not state how many events an individual subject would experience or give an absolute risk difference.
Why was MMRM used for HbA1c?
HbA1c was measured at more than one time point, and the registry reports MMRM for the change-from-baseline analysis. A mixed model for repeated measures allows longitudinal observations to be analyzed together while accounting for the repeated-measure structure and incorporating the covariate-adjusted treatment comparison described in the analysis text.
Why use the McNemar test in a crossover trial?
The McNemar test is designed for paired binary outcomes. In a crossover design, the same subject can contribute a binary outcome under each treatment condition. The registry specifies that the severe-hypoglycaemia proportion analysis used subjects exposed in both maintenance periods, making the paired structure central to the analysis.
Why does hierarchical testing matter?
The registry describes a stepwise confirmatory sequence: the primary hypoglycaemia count was Step 1, nocturnal hypoglycaemic episodes were Step 2, and the proportion with one or more severe episodes was Step 3. A hierarchical strategy can preserve the intended confirmatory interpretation across multiple related hypotheses by requiring the earlier step to be successfully established before proceeding to the next.
What does the confidence interval add beyond the p-value?
The confidence interval gives a range of effect estimates compatible with the statistical model and data under the stated confidence framework. For the primary treatment ratio, the interval is 0.85–0.94. The p-value instead summarizes evidence against a specified null hypothesis. Neither quantity describes the effect for every individual subject.
14. Crossover Design: Statistical Implications
SWITCH 1 is registered as a randomized crossover trial. That design changes the statistical structure compared with a simple parallel-group trial because subjects contribute information under different treatment conditions.
15. Non-Inferiority Logic in SWITCH 1
The primary analysis provides a useful example of why non-inferiority should be interpreted through the prespecified margin rather than through a generic question of whether the p-value is "significant."
| Component | SWITCH 1 primary analysis |
|---|---|
| Effect measure | Treatment ratio / rate ratio, IDeg/IGlar |
| Estimate | 0.89 |
| Two-sided confidence interval | 0.85–0.94 |
| Non-inferiority margin | 1.10 |
| Confidence-limit criterion | Upper bound ≤1.10 |
| Reported p-value | <0.0001 |
| One-sided null described by registry | H0: RR >1.10 |
| One-sided alternative described by registry | HA: RR ≤1.10 |
| One-sided significance level described by registry | 2.5% |
The confidence interval is particularly intuitive here: its entire range is below 1.10. The point estimate of 0.89 is also below 1.10. The statistical conclusion therefore depends on where the interval lies relative to the non-inferiority boundary, rather than simply on whether the estimated ratio is below 1.
16. Multiplicity and Confirmatory Hierarchy
The registry explicitly identifies multiplicity adjustment for the hypoglycaemia analyses. The confirmatory endpoints were tested using a stepwise hierarchy.
| Step | Endpoint | Hypothesis type | Analysis |
|---|---|---|---|
| 1 | Severe or BG-confirmed symptomatic hypoglycaemic episodes | Non-inferiority | Poisson regression |
| 2 | Severe or BG-confirmed symptomatic nocturnal hypoglycaemic episodes | Non-inferiority | Poisson regression |
| 3 | One or more severe hypoglycaemic episodes | Superiority | McNemar test |
This sequence matters because multiple confirmatory questions can otherwise create a larger overall chance of declaring an effect simply through repeated testing. The ClinicalTrials.gov record specifically state that the stepwise procedure was used for the confirmatory endpoints.
17. What the Primary Treatment Ratio Does — and Does Not — Mean
The primary treatment ratio of 0.89 means the estimated rate of treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes under IDeg was 89% of the estimated rate under IGlar in the reported Poisson analysis.
The treatment ratio cannot be converted from the ClinicalTrials.gov record into a difference in the number of subjects experiencing an event. It describes a relative rate, not an absolute event probability.
The 95% CI of 0.85–0.94 quantifies uncertainty around the estimated treatment ratio under the analysis framework. It is not a prediction interval for future individual patients.
The reported P < 0.0001 describes statistical evidence relative to a specified hypothesis. It does not quantify clinical importance, probability that a treatment is effective, or the probability that the null hypothesis is true.
The principal non-inferiority question is whether the estimated rate ratio could plausibly be worse than the prespecified margin of 1.10. The confidence interval's upper bound of 0.94 is below that boundary.
18. Limitations and Interpretation Issues
- Crossover structure: treatment sequence and paired observations are integral to the design. Results from crossover analyses should not be interpreted as if the study were simply a conventional parallel-group comparison.
- Different analysis populations: the primary Poisson analysis used the FAS with exposure in at least one maintenance period, while the McNemar analysis used FAS subjects exposed in both maintenance periods. HbA1c analyses also use available observations at the specified time points.
- Non-inferiority interpretation: the primary conclusion depends on the prespecified 1.10 margin and the corresponding confidence-interval or one-sided-test framework.
- Count-model interpretation: the treatment ratio is a model-based relative rate measure. It should not be interpreted as an absolute percentage of subjects with hypoglycaemia.
- Multiplicity: the registry explicitly identifies multiplicity adjustment and a hierarchical confirmatory sequence. Individual p-values should therefore be interpreted within that sequence rather than as unrelated tests.
- HbA1c versus hypoglycaemia: the HbA1c analyses address glycaemic control, while the primary endpoint addresses hypoglycaemic episodes. They measure different clinical and statistical constructs.
- Missing information in the ClinicalTrials.gov record: the ClinicalTrials.gov record do not include all subject-level counts, all model coefficients, or all paired contingency-table cells. Accordingly, some effect sizes cannot be reconstructed from the ClinicalTrials.gov record.
- Safety comparison: serious adverse-event counts are reported by arm, but the ClinicalTrials.gov record does not provide a formal comparative statistical analysis for those counts.
19. Why This Trial Matters Statistically
SWITCH 1 is a useful statistical teaching case because several distinct methods are needed to analyze different aspects of the same randomized crossover study. The primary endpoint is a count outcome requiring Poisson regression; HbA1c is longitudinal and uses MMRM; a binary crossover outcome uses McNemar testing; and the confirmatory questions are organized through a non-inferiority and hierarchical-testing framework.
| Concept | How it appears in SWITCH 1 |
|---|---|
| Randomization | Randomized phase 3 treatment comparison |
| Double blinding | Registry masking classification is double |
| Crossover design | Two treatment sequences: IDeg/IGlar and IGlar/IDeg |
| Non-inferiority | Primary endpoint tested against a 1.10 rate-ratio margin |
| Poisson regression | Primary and nocturnal hypoglycaemic episode counts |
| Rate ratio | Primary treatment effect of 0.89 with 95% CI 0.85–0.94 |
| MMRM | Change from baseline in HbA1c at Week 32 and Week 64 |
| Covariate adjustment | Included among the concepts in the HbA1c analysis text |
| McNemar test | Proportion of subjects with one or more severe hypoglycaemic episodes |
| Multiplicity | Stepwise hierarchical confirmatory testing |
| Confidence intervals | Used directly to evaluate the non-inferiority boundary |
| Paired data | Important for crossover binary outcomes and McNemar analysis |
20. Results Summary
The registry reports a primary treatment ratio of 0.89 for treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes, with a two-sided 95% CI of 0.85–0.94 and P < 0.0001. The prespecified non-inferiority margin was 1.10, and the registry describes non-inferiority as confirmed when the upper confidence bound was ≤1.10.
| Endpoint | Method | Estimate / result | 95% CI | P-value |
|---|---|---|---|---|
| Primary hypoglycaemic episodes | Poisson regression | Treatment ratio 0.89 | 0.85–0.94 | <0.0001 |
| Nocturnal hypoglycaemic episodes | Poisson regression | Treatment ratio 0.64 | 0.56–0.73 | <0.0001 |
| HbA1c change, Week 32 | MMRM | 0.03 | -0.10 to 0.15 | Not provided |
| HbA1c change, Week 64 | MMRM | 0.11 | -0.00 to 0.23 | Not provided |
| One or more severe hypoglycaemic episodes | McNemar test | Superiority hypothesis | Not provided | 0.0016 |
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: SWITCH 1 — NCT02034513.
- Linked publication: PubMed record: PMID 28672316.
- Linked publication: PubMed record: PMID 30097995.
- Linked publication: PubMed record: PMID 30362250.
Continue through the Clinical Biostats statistical methods library
Explore the statistical concepts that connect randomized trial design, repeated measurements, count outcomes, paired categorical data, confidence intervals, and non-inferiority testing.
24. Record Summary
SWITCH 1 provides a compact example of how the statistical method should follow the structure of the endpoint and the design. The primary hypoglycaemic endpoint is a count outcome analyzed with Poisson regression and interpreted against a prespecified non-inferiority margin. Nocturnal hypoglycaemic episodes use the same modeling family within the hierarchical sequence. HbA1c uses MMRM for repeated measurements, while the binary severe-hypoglycaemia endpoint uses the paired McNemar test. Across these analyses, the key interpretive tasks are to distinguish relative rates from absolute risks, confidence intervals from p-values, non-inferiority from superiority, and paired crossover information from independent-group comparisons.