This page separates reported trial results from statistical interpretation. The numerical results and trial facts presented here are restricted to the ClinicalTrials.gov record for NCT02030600.
1. Trial at a Glance
SWITCH 2 was a randomized, double-blind, phase 3 crossover trial comparing insulin degludec with insulin glargine in subjects with type 2 diabetes. The registry reports 721 enrolled subjects, two treatment arms, six posted outcome measures, and five posted statistical analyses.
| Feature | SWITCH 2 |
|---|---|
| Trial name | SWITCH 2 |
| Phase | Phase 3 |
| Therapeutic area | Diabetes |
| Condition | Diabetes; Diabetes Mellitus, Type 2 |
| Design | Randomized, double-blind, crossover |
| Primary purpose | Treatment |
| Enrollment | 721 |
| Interventions | Insulin degludec; insulin glargine |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT02030600 |
2. Clinical Question
The SWITCH 2 trial asks a question that is particularly suited to a crossover design: when subjects with type 2 diabetes receive both insulin degludec and insulin glargine in randomized treatment periods, how do the treatments compare with respect to treatment-emergent severe or blood-glucose-confirmed symptomatic hypoglycaemic episodes during the maintenance period?
Population
Subjects with type 2 diabetes enrolled in the randomized phase 3 trial.
Intervention
Insulin degludec (IDeg).
Comparator
Insulin glargine (IGlar).
Primary question
How does the number of treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes compare between IDeg and IGlar during the maintenance periods?
3. Trial Design
SWITCH 2 used a randomized, double-blind crossover design. Unlike a conventional parallel-group trial, a crossover trial exposes subjects to more than one randomized treatment sequence. The registry analysis text explicitly identifies the crossover structure when describing the safety analysis set and the analysis of the hypoglycaemic outcomes.
Insulin Degludec
- Randomized insulin treatment.
- Compared with insulin glargine.
- Hypoglycaemia was assessed during the maintenance period in each treatment period.
Insulin Glargine
- Randomized comparator insulin treatment.
- Compared with insulin degludec.
- Hypoglycaemia was assessed during the maintenance period in each treatment period.
4. Randomization and Analysis Populations
The registry identifies the allocation as randomized and the masking as double. For the primary hypoglycaemia analysis, the registry states that descriptive analysis was based on the safety analysis set, defined as subjects receiving at least one dose of the investigational product or its comparator.
| Analysis population | Registry-supported description |
|---|---|
| Safety analysis set | Subjects receiving at least one dose of the investigational product or its comparator; used for descriptive analysis of the crossover trial. |
| Full analysis set | Used for the posted HbA1c analyses. The registry states that n specifies the number of subjects with available data at the specified timepoint. |
| Endpoint-specific available data | For some secondary outcomes, the registry describes the analysed population in terms of subjects with available data for the endpoint. |
5. Endpoints
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Primary: Number of Treatment Emergent Severe or BG Confirmed Symptomatic Hypoglycaemic Episodes During the Maintenance Period | After 16 weeks of treatment, in each treatment period (Week 16-32 and Week 48-64) | Poisson regression; treatment ratio |
| Secondary: Number of Treatment Emergent Severe or BG Confirmed Symptomatic Nocturnal Hypoglycaemic Episode During the Maintenance Period | After 16 weeks of treatment, in each treatment period (Week 16-32 and Week 48-64) | Poisson regression; treatment ratio |
| Secondary: Change From Baseline in HbA1c (Glycosylated Haemoglobin) | Week 32, Week 64 | MMRM; treatment contrast |
| Secondary: 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) | McNemar test |
Primary hypoglycaemia definition
The registered primary endpoint defined severe or blood glucose (BG) confirmed symptomatic hypoglycaemic episodes as episodes that were severe and/or BG confirmed by a plasma glucose value of <56 mg/dL (3.1 mmol/L), with symptoms consistent with hypoglycaemia. A treatment-emergent hypoglycaemic episode was defined as an event with onset date on or after the first day of exposure to randomized treatment and no later than the registry-defined endpoint window.
6. Statistical Methodology
Poisson regression for hypoglycaemic episode counts
The primary endpoint is a count of hypoglycaemic episodes rather than a simple binary outcome. The registry reports Poisson regression as the primary analysis method and describes the resulting effect measure as a treatment ratio.
For a Poisson-type model, the treatment coefficient is expressed on a logarithmic scale and exponentiating the treatment contrast produces a ratio of expected event rates. The exact additional model terms should be taken from the trial's complete statistical analysis documentation rather than inferred from the registry summary.
MMRM for HbA1c
The HbA1c analyses used a mixed model for repeated measurements (MMRM). This is appropriate when a continuous outcome is observed repeatedly over treatment periods and the analysis needs to account for correlations among measurements from the same subject.
McNemar test for paired binary outcomes
The proportion of subjects with one or more severe hypoglycaemic episodes was analysed with the McNemar test. In a crossover setting, this is conceptually different from an ordinary two-group chi-square comparison because the relevant information is the within-subject pairing of binary outcomes across treatments.
Intention-to-treat principles
The registry analysis metadata identifies intention-to-treat analysis as a concept in the posted statistical analyses. In randomized trials, the intention-to-treat principle preserves the treatment assignment established by randomization. In a crossover study, however, the exact estimand and analysis population still depend on the prespecified statistical model and treatment-period structure.
Multiplicity and hierarchical testing
The registry states that a stepwise hierarchical testing procedure was used for confirmatory endpoints. The first step was the primary hypoglycaemia analysis, followed by nocturnal hypoglycaemia and then the proportion of subjects with one or more severe hypoglycaemic episodes.
| Step | Endpoint | Registry decision rule |
|---|---|---|
| Step 1 | Number of treatment-emergent severe or BG confirmed symptomatic hypoglycaemic episodes | Superiority confirmed if the two-sided 95% confidence interval for the rate ratio (IDeg/IGlar) was entirely below 1.0. |
| Step 2 | Number of treatment-emergent severe or BG confirmed symptomatic nocturnal hypoglycaemic episodes | Superiority confirmed if the two-sided 95% confidence interval for the rate ratio (IDeg/IGlar) was entirely below 1.0. |
| Step 3 | Proportion of subjects with one or more severe hypoglycaemic episodes | Stepwise hierarchical testing procedure; McNemar test reported. |
7. Primary Result: Severe or BG-Confirmed Symptomatic Hypoglycaemic Episodes
The primary endpoint was the number of treatment-emergent severe or BG-confirmed symptomatic hypoglycaemic episodes during the maintenance period after 16 weeks of treatment in each treatment period.
Treatment ratio: IDeg / IGlar
95% CI: 0.61–0.80 · P < 0.0001
Two-sided confidence interval; superiority hypothesis.
A treatment ratio of 0.70 means that the estimated hypoglycaemic episode rate under IDeg was 0.70 times the corresponding rate under IGlar in the reported Poisson analysis. Expressed as a relative comparison, that corresponds to a 30% lower estimated episode rate for IDeg relative to IGlar.
This does not mean that 30% of subjects avoided hypoglycaemia, that each subject experienced exactly 30% fewer events, or that the absolute number of episodes was reduced by a fixed amount for every patient. The endpoint is an event count analysed as a treatment-rate comparison.
The 95% confidence interval of 0.61–0.80 describes uncertainty around the estimated treatment ratio under the statistical model. Because the entire interval is below 1.0, the registry's prespecified superiority criterion for Step 1 was met.
The P-value of <0.0001 addresses evidence against the relevant null hypothesis under the specified testing framework. It does not measure the magnitude or clinical importance of the treatment effect. The magnitude is conveyed by the treatment ratio and its confidence interval.
The crossover structure and the use of Poisson regression are central to interpretation. The reported result should not be treated as a simple comparison of two independent proportions, and the ClinicalTrials.gov record does not provide enough detail to independently assess dispersion assumptions or every model term.
8. Secondary Result: Nocturnal Hypoglycaemic Episodes
The second hierarchical endpoint was the 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.
Treatment ratio: IDeg / IGlar
95% CI: 0.46–0.74 · P < 0.0001
Two-sided confidence interval; superiority hypothesis.
The treatment ratio of 0.58 indicates an estimated nocturnal hypoglycaemic episode rate under IDeg equal to 0.58 times the corresponding rate under IGlar in the reported Poisson analysis. On a relative scale, this is a 42% lower estimated episode rate for IDeg relative to IGlar.
The estimate is a rate comparison, not a statement that 42% of subjects experienced a particular outcome. It also does not establish that every individual patient would experience the same relative difference.
The 95% confidence interval of 0.46–0.74 provides the reported precision around the treatment ratio. The interval is entirely below 1.0, matching the registry's Step 2 criterion for confirming superiority.
The P-value of <0.0001 provides evidence against the null hypothesis within the prespecified hierarchical framework. It should not be interpreted as a measure of the size of the nocturnal hypoglycaemia effect.
9. Secondary Results: HbA1c and Non-Inferiority
HbA1c was analysed at Week 32 and Week 64 using an MMRM. The registry reports treatment contrasts comparing the treatment sequences IDeg/IGlar and IGlar/IDeg.
| Time frame | Comparison | Estimate | 95% CI | Hypothesis |
|---|---|---|---|---|
| Week 32 | IDeg/IGlar vs IGlar/IDeg | 0.09 | -0.04 to 0.23 | Non-inferiority or equivalence |
| Week 64 | IDeg/IGlar vs IGlar/IDeg | 0.06 | -0.07 to 0.18 | Non-inferiority or equivalence |
Week 32
The registry reports an estimated treatment contrast of 0.09 with a two-sided 95% CI of -0.04 to 0.23.
Week 64
The registry reports an estimated treatment contrast of 0.06 with a two-sided 95% CI of -0.07 to 0.18.
The non-inferiority margin
For the HbA1c analysis, the registry states that non-inferiority of IDeg against IGlar was considered confirmed if the upper bound of the two-sided 95% confidence interval was below or equal to 0.40%.
The decision is based on the prespecified margin and confidence interval, rather than simply asking whether a conventional superiority P-value crosses a threshold.
The registry further states that the Week 32 supportive efficacy endpoint was tested for non-inferiority as a prerequisite for testing the primary endpoint. The ClinicalTrials.gov record therefore show that non-inferiority was an explicit component of the hierarchical confirmatory strategy.
The Week 32 estimate of 0.09 indicates that the reported treatment contrast favored a higher HbA1c value by 0.09 percentage points on the stated comparison scale, while the two-sided 95% confidence interval ranged from -0.04 to 0.23. The Week 64 estimate was 0.06, with a 95% confidence interval of -0.07 to 0.18.
These confidence intervals are notably narrower than the non-inferiority margin of 0.40% on their upper ends. Under the registry's stated rule, an upper confidence-limit criterion is what determines non-inferiority; the analysis is not asking whether the treatments are statistically identical.
A non-inferiority conclusion, when established under a prespecified margin, means the data are compatible with the new treatment being no worse than the comparator by more than the allowed margin. It does not mean that the two treatments have exactly the same effect.
10. Secondary Result: Subjects With One or More Severe Hypoglycaemic Episodes
The third step of the hierarchical testing procedure evaluated the proportion of subjects with one or more severe hypoglycaemic episodes during the maintenance period. The registry reports a McNemar test.
McNemar test
Superiority hypothesis; maintenance period after 16 weeks in each treatment period.
The P-value of 0.3458 does not provide evidence of a statistically significant treatment difference under the stated Step 3 superiority analysis. Importantly, this is a P-value for a paired binary comparison; it is not a treatment effect estimate.
The absence of a statistically significant result does not prove that the treatment effects are identical. It means that the reported test did not provide sufficient evidence to reject the null hypothesis at the conventional interpretation of the test.
The ClinicalTrials.gov record does not provide an effect estimate or confidence interval for this endpoint. It would therefore be inappropriate to infer an absolute risk difference, risk ratio, odds ratio, or other magnitude measure from the P-value alone.
11. Safety
The registry data provide serious adverse event counts by treatment arm. These figures should be interpreted as safety counts rather than as measures of treatment efficacy.
| Treatment | Serious adverse events affected / at risk |
|---|---|
| Insulin Degludec (IDeg) | 64 / 671 |
| Insulin Glargine (IGlar) | 65 / 665 |
The ClinicalTrials.gov record does not provide a formal statistical comparison for serious adverse events. The counts should therefore not be converted into a treatment-effect claim. In particular, the difference between the affected-subject counts alone does not establish a difference in safety risk.
12. Statistical Methods Explained
Why was Poisson regression used?
The primary and nocturnal hypoglycaemia endpoints count events. A subject can experience more than one episode, so a binary endpoint such as "at least one event" would discard information about repeated episodes. Poisson regression is a natural model family for count or rate outcomes and produces a treatment ratio that directly compares the modeled event rates.
What does a treatment ratio of 0.70 mean?
A treatment ratio of 0.70 means the estimated event rate under IDeg was 70% of the estimated event rate under IGlar in the reported analysis. The complementary relative interpretation is a 30% lower estimated event rate. It is not an absolute reduction in the number of events and does not imply that 30% of patients benefited.
Why was a McNemar test appropriate?
The McNemar test is designed for paired binary outcomes. A crossover trial can produce paired treatment observations within the same subject, making a paired comparison more appropriate than treating the observations as two independent samples for this endpoint.
Why was an MMRM used for HbA1c?
HbA1c was assessed at multiple timepoints, including Week 32 and Week 64. MMRM allows repeated observations from the same subject to be analysed jointly rather than treating each timepoint as an unrelated analysis. It can also incorporate prespecified covariates and treatment-related terms.
Why is non-inferiority judged against a margin?
A non-inferiority trial asks whether the new treatment is not unacceptably worse than the comparator by more than a prespecified amount. Here, the registry specifies an upper confidence-limit criterion of 0.40% for the HbA1c contrast. The margin therefore defines what degree of possible disadvantage would still satisfy the trial's non-inferiority criterion.
Why does the P-value not measure effect size?
A P-value quantifies the compatibility of the observed data with a specified null hypothesis under the statistical model and testing procedure. It does not tell the reader how large the treatment effect is. The treatment ratio of 0.70, its 95% CI of 0.61–0.80, and the P-value of <0.0001 answer different statistical questions.
Why does crossover design change the interpretation?
In a crossover trial, treatment is compared within a study structure in which subjects receive different treatments across periods. This can make each subject serve as part of the comparison with themselves, but it also means treatment effects must be considered alongside period and sequence structure. The ClinicalTrials.gov record does not provide the complete model specification, so those terms should not be inferred.
13. Crossover Design and Its Statistical Consequences
The crossover design is one of the most important statistical features of SWITCH 2. A conventional parallel trial estimates a treatment contrast by comparing outcomes from distinct groups of subjects. A crossover trial instead observes subjects under more than one randomized treatment condition.
The key statistical point is that the design and analysis should correspond. Treating a crossover trial as though it were simply two unrelated parallel groups can ignore the pairing created by the experimental design.
14. Hierarchical Testing and Multiplicity
SWITCH 2 used a stepwise hierarchical testing procedure for confirmatory endpoints. This is important because several related hypotheses were tested in sequence. A hierarchical strategy can control the interpretation of later hypotheses by requiring earlier hypotheses to satisfy their prespecified criterion first.
| Step | Endpoint | Reported result | Statistical role |
|---|---|---|---|
| Step 1 | Severe or BG-confirmed symptomatic hypoglycaemic episodes | Treatment ratio 0.70; 95% CI 0.61–0.80; P < 0.0001 | Primary analysis; superiority |
| Step 2 | Nocturnal severe or BG-confirmed symptomatic hypoglycaemic episodes | Treatment ratio 0.58; 95% CI 0.46–0.74; P < 0.0001 | Secondary confirmatory analysis; superiority |
| Step 3 | One or more severe hypoglycaemic episodes | P = 0.3458 | Secondary confirmatory analysis; superiority |
The registry explicitly states that superiority for the first two steps was considered confirmed when the two-sided 95% confidence interval for the IDeg/IGlar rate ratio was entirely below 1.0. This is a useful example of why an isolated P-value should not be interpreted without knowing the endpoint hierarchy.
15. Non-Inferiority Analysis of HbA1c
The HbA1c analysis illustrates a different statistical objective from the superiority analysis of hypoglycaemic episodes. The question is not whether IDeg produces a statistically detectable difference from IGlar, but whether any disadvantage remains within a clinically prespecified non-inferiority margin.
| Timepoint | Estimate | Upper 95% CI | NI margin | Registry criterion |
|---|---|---|---|---|
| Week 32 | 0.09 | 0.23 | 0.40% | Upper bound ≤ 0.40% |
| Week 64 | 0.06 | 0.18 | 0.40% | Upper bound ≤ 0.40% |
The Week 32 and Week 64 confidence intervals both have upper bounds below the stated 0.40% margin. That illustrates the mechanics of a non-inferiority assessment: the relevant question is whether the uncertainty interval extends beyond the prespecified clinically unacceptable difference.
The non-inferiority framework should not be confused with equivalence testing. Equivalence generally seeks to show that the entire confidence interval lies within a prespecified two-sided equivalence interval. The ClinicalTrials.gov record specifically describe a non-inferiority criterion based on the upper confidence bound.
16. Confidence Intervals: Reading the Reported Effects
| Endpoint | Estimate | 95% CI | Interpretive scale |
|---|---|---|---|
| Primary hypoglycaemic episodes | 0.70 | 0.61–0.80 | Treatment ratio |
| Nocturnal hypoglycaemic episodes | 0.58 | 0.46–0.74 | Treatment ratio |
| HbA1c, Week 32 | 0.09 | -0.04 to 0.23 | Treatment contrast |
| HbA1c, Week 64 | 0.06 | -0.07 to 0.18 | Treatment contrast |
The scale of the effect measure determines how the confidence interval should be read. For the hypoglycaemia analyses, a ratio of 1.0 represents no treatment-rate difference. For the HbA1c contrasts, a value of 0 represents no treatment contrast on the reported scale.
The primary hypoglycaemia CI of 0.61–0.80 is entirely below the ratio-null value of 1.0. By contrast, the HbA1c CIs are centered around an additive contrast whose null value is 0. The statistical meaning of a confidence interval therefore depends on the effect measure, not simply on whether the interval "contains one particular number."
17. What the Primary Treatment Ratio Does — and Does Not — Mean
The primary treatment ratio of 0.70 indicates an estimated hypoglycaemic episode rate under IDeg that was 70% of the rate under IGlar in the reported Poisson analysis.
Equivalently, the estimated relative difference is a 30% lower episode rate for IDeg. This is a model-based relative measure of event rates.
It does not mean that 30% of subjects avoided hypoglycaemia, that the absolute number of events was reduced by 30%, or that every subject experienced a 30% reduction.
The 95% CI of 0.61–0.80 conveys uncertainty around the estimated treatment ratio. It does not describe the range of effects experienced by individual subjects. The registry's superiority criterion is satisfied because the entire interval is below 1.0.
The P-value of <0.0001 provides evidence against the null hypothesis within the prespecified analysis. It does not tell the reader whether the treatment effect is 0.70, 0.61, or 0.80; those quantities come from the estimate and confidence interval.
18. Limitations
- Incomplete model specification: the registry analysis text does not provide every term in the Poisson or MMRM models, so the complete fitted model should not be reconstructed from partial excerpts.
- Crossover complexity: treatment effects in a crossover study are embedded in a treatment-period and sequence structure that differs from a simple parallel-group comparison.
- Missing-data detail: the ClinicalTrials.gov record identifies available-data rules for some endpoints but do not provide a complete missing-data or imputation strategy.
- Safety denominators differ: the serious adverse-event counts are reported with affected and at-risk denominators of 671 and 665, respectively; they should not be confused with the total enrollment of 721.
- Endpoint-specific populations: the registry uses different analysis-population descriptions for the safety, full analysis, and endpoint-specific analyses.
- Limited effect information for the McNemar analysis: the registry provides a P-value of 0.3458 but no corresponding effect estimate or confidence interval in the ClinicalTrials.gov record.
- No independent reconstruction: the reported estimates are presented exactly as reported in the registry. They have not been recomputed from underlying subject-level data.
19. Why This Trial Matters Statistically
SWITCH 2 is a useful teaching example because several distinct statistical ideas appear in one trial: randomized crossover allocation, count-data modelling, paired binary testing, longitudinal mixed models, non-inferiority, hierarchical multiplicity control, and interpretation of treatment ratios.
| Concept | How it appears in SWITCH 2 |
|---|---|
| Randomization | Randomized treatment allocation between insulin degludec and insulin glargine. |
| Double blinding | The registry identifies masking as double. |
| Crossover design | Subjects receive randomized treatments across treatment periods. |
| Poisson regression | Primary and nocturnal hypoglycaemic episode counts. |
| Treatment ratio | Primary effect measure for hypoglycaemic episode counts. |
| MMRM | Repeated HbA1c measurements at Week 32 and Week 64. |
| McNemar test | Paired binary comparison of subjects with one or more severe hypoglycaemic episodes. |
| Non-inferiority | HbA1c analysis using a 0.40% upper confidence-limit criterion. |
| Multiplicity | Stepwise hierarchical testing across confirmatory endpoints. |
| Confidence intervals | 95% intervals accompany the treatment-ratio and HbA1c estimates. |
| Intention-to-treat concepts | Identified in the posted statistical analysis metadata. |
| Covariate adjustment | Identified in the MMRM HbA1c analysis metadata. |
20. Trial Timeline
SWITCH 2 begins
The registry lists 2014-01-06 as the study start date.
Primary study completion
The registry lists 2015-12-04 as the primary completion date.
Results posted
The trial is listed as completed, with results posted and five statistical analyses reported in the ClinicalTrials.gov record.
21. A Practical Statistical Reading of SWITCH 2
Start with the design
The randomized double-blind crossover structure determines why paired and repeated-measures methods appear in the analysis.
Identify the estimand scale
Hypoglycaemia is reported as a treatment ratio, whereas HbA1c is reported as a treatment contrast.
Check the decision rule
The primary superiority analysis uses a 95% CI entirely below 1.0, while HbA1c non-inferiority uses an upper 95% CI bound of 0.40%.
Separate evidence from magnitude
P-values describe statistical evidence against a null hypothesis; estimates and confidence intervals describe the size and precision of the treatment comparison.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: NCT02030600 — SWITCH 2.
- PubMed: PMID 28672317.
- PubMed: PMID 30097995.
- PubMed: PMID 30362250.
- PubMed: PMID 30891886.
Continue with the statistical methods behind SWITCH 2
Explore the broader Clinical Biostats tutorials and statistical calculators covering crossover trials, repeated-measures models, non-inferiority, categorical testing, count-data regression, and confidence intervals.
25. Record Summary
SWITCH 2 provides a compact but statistically rich example of how clinical-trial design and analysis must fit together. The randomized double-blind crossover structure is paired with Poisson regression for treatment-emergent hypoglycaemic episode counts, MMRM for repeated HbA1c measurements, and McNemar testing for a paired binary endpoint. The trial also illustrates two different hypothesis-testing frameworks: superiority for the hypoglycaemia endpoints and non-inferiority for HbA1c.
The primary treatment ratio of 0.70 had a two-sided 95% confidence interval of 0.61–0.80 and a P-value of <0.0001. The nocturnal hypoglycaemia treatment ratio was 0.58, with a 95% confidence interval of 0.46–0.74 and P < 0.0001. The HbA1c treatment contrasts were 0.09 at Week 32 and 0.06 at Week 64, with upper confidence limits of 0.23 and 0.18, respectively, against the stated non-inferiority margin of 0.40%. The Step 3 McNemar analysis reported P = 0.3458.