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
inTandem3 was a randomized, parallel-group, phase 3 trial evaluating sotagliflozin versus placebo in patients with type 1 diabetes who had inadequate glycemic control with insulin therapy alone. The trial enrolled 1405 participants and evaluated a binary composite primary endpoint at Week 24.
| Feature | inTandem3 |
|---|---|
| Phase | Phase 3 |
| Therapeutic area | Endocrinology & Metabolism |
| Population | Patients with type 1 diabetes who have inadequate glycemic control with insulin therapy alone |
| Design | Randomized, parallel-group, quadruple-masked, treatment study |
| Allocation | Randomized |
| Enrollment | 1405 |
| Interventions | Sotagliflozin 400 mg versus placebo |
| Primary endpoint | Percentage of participants with A1C <7.0% at Week 24 and no episode of severe hypoglycemia and no episode of diabetic ketoacidosis after randomization |
| Lead sponsor | Lexicon Pharmaceuticals |
| Status | Completed |
2. Clinical Question
The central question was whether treatment with sotagliflozin, compared with placebo, increased the percentage of participants who achieved the prespecified glycemic target while avoiding severe hypoglycemia and diabetic ketoacidosis after randomization.
Population
Patients with type 1 diabetes who had inadequate glycemic control with insulin therapy alone.
Intervention
Sotagliflozin 400 mg.
Comparator
Placebo.
Primary question
Does sotagliflozin increase the probability of meeting the composite Week 24 efficacy and safety endpoint compared with placebo?
3. Trial Design
Sotagliflozin
- Sotagliflozin 400 mg
- Background insulin therapy
- Primary endpoint assessed at Week 24
Placebo
- Placebo
- Background insulin therapy
- Primary endpoint assessed at Week 24
The trial was randomized, parallel, and quadruple-masked. These design features are important statistically because randomization establishes the basis for a between-group causal comparison, while masking can reduce differential assessment or treatment behavior during follow-up.
4. Trial Timeline
Trial start
The inTandem3 study began in September 2015.
Primary completion
The registry reports primary completion in April 2017.
Primary endpoint assessment
The primary composite endpoint was assessed at Week 24.
5. Primary Endpoint
| Endpoint | Registry definition | Type | Time frame |
|---|---|---|---|
| Composite glycemic efficacy and safety endpoint | Percentage of participants with A1C <7.0% at Week 24 and no episode of severe hypoglycemia and no episode of diabetic ketoacidosis after randomization | Binary | Week 24 |
The endpoint combines three clinically distinct conditions: reaching an A1C below 7.0%, avoiding severe hypoglycemia, and avoiding diabetic ketoacidosis after randomization. A participant therefore had to satisfy all components to count as having achieved the composite endpoint.
The registry states that blood samples were used to assess Hemoglobin A1C and that a central blinded adjudication process determined whether participants experienced diabetic ketoacidosis or severe hypoglycemia.
6. Statistical Methodology
Primary binary endpoint: Cochran-Mantel-Haenszel test
The primary efficacy analysis compared placebo with sotagliflozin 400 mg using the Cochran-Mantel-Haenszel test. The analysis was based on the modified Intent-to-Treat (mITT) population.
The registry reports the effect as a percentage difference and the effect measure as a risk difference. A positive value therefore indicates a higher percentage of participants achieving the composite endpoint in the sotagliflozin group than in the placebo group.
Repeated-measures endpoints: MMRM
The secondary analyses of A1C, body weight, systolic blood pressure, and mean daily bolus insulin dose used mixed models for repeated measures (MMRM). These models are designed for longitudinal data in which participants can contribute measurements at multiple study visits.
The registry describes treatment, randomization strata, study week, and treatment-by-time interaction as fixed categorical effects, together with a baseline measure-by-time interaction as a covariate. The exact baseline covariate depended on the endpoint.
Analysis population
The primary efficacy analysis used the modified Intent-to-Treat population. For the secondary outcomes, the registry states that analyses included participants from the mITT population who were evaluable for the respective outcome measure.
Multiplicity
The registry describes the secondary endpoint testing as occurring according to a hierarchical procedure. This matters because a sequence of formally tested endpoints can be structured so that statistical significance is evaluated in a prespecified order rather than treating every p-value as an independent hypothesis test.
7. Primary Result
The primary endpoint favored sotagliflozin over placebo. The analysis used the Cochran-Mantel-Haenszel test in the mITT population and reported a two-sided 95% confidence interval for the risk difference.
Primary composite endpoint
95% CI: 8.97–17.81 · P < 0.001
Placebo vs sotagliflozin 400 mg · Week 24 · mITT population
| Primary endpoint | Comparison | Effect estimate | 95% CI | P-value |
|---|---|---|---|---|
| A1C <7.0% at Week 24 with no severe hypoglycemia and no DKA after randomization | Placebo vs Sotagliflozin 400 mg | 13.4 percentage points | 8.97 to 17.81 | < 0.001 |
The reported risk difference of 13.4 percentage points indicates that the percentage of participants meeting the entire composite endpoint was estimated to be 13.4 percentage points higher with sotagliflozin 400 mg than with placebo at Week 24.
This is an absolute between-group difference for the composite endpoint. It does not mean that every participant experienced a 13.4-percentage-point change, nor does it describe the effect of sotagliflozin on A1C alone. Because the endpoint requires all three conditions to be met, its interpretation depends on the joint occurrence of the glycemic and safety components.
The 95% confidence interval of 8.97 to 17.81 percentage points describes statistical uncertainty around the estimated between-group difference. It does not describe the range of effects experienced by individual patients.
The p-value of < 0.001 addresses evidence against the null hypothesis under the specified testing framework; it is not a measure of the size or clinical importance of the effect. The magnitude of the effect is conveyed by the risk difference, while the confidence interval describes its precision.
The analysis was a superiority analysis using a categorical-data method. Because the endpoint is binary and assessed at a fixed time point, a hazard ratio or proportional-hazards assumption is not involved in this primary analysis.
8. Secondary Endpoint Results
The registry reports four secondary statistical analyses. These analyses used the mITT population, with participants required to be evaluable for the respective outcome measure. The reported MMRM estimates are least-squares mean differences between placebo and sotagliflozin 400 mg.
| Endpoint | Time frame | Method | Effect estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Change from baseline in A1C | Baseline to Week 24 | MMRM | -0.46 | -0.54 to -0.38 | < 0.001 |
| Absolute change from baseline in body weight | Baseline to Week 24 | MMRM | -2.98 kg | -3.31 to -2.66 | < 0.001 |
| Change from baseline in systolic blood pressure in participants with baseline SBP ≥130 mmHg | Baseline to Week 16 | MMRM | -3.5 mmHg | -5.7 to -1.3 | = 0.002 |
| Percent change from baseline in mean daily bolus insulin dose | Baseline to Week 24 | MMRM | -12.32 | -18.17 to -6.48 | < 0.001 |
A1C
The estimated least-squares mean difference in change from baseline in A1C was -0.46, with a 95% confidence interval of -0.54 to -0.38 and P < 0.001. Because the comparison is placebo versus sotagliflozin 400 mg as reported, the negative estimate indicates a lower modeled A1C change in the sotagliflozin group relative to placebo under the specified analysis.
Body weight
The least-squares mean difference in absolute change from baseline in body weight was -2.98 kg, with a 95% confidence interval of -3.31 to -2.66 kg and P < 0.001. The estimate represents the adjusted between-group difference in longitudinal change rather than a simple comparison of raw endpoint means.
Systolic blood pressure
Among participants with baseline systolic blood pressure ≥130 mmHg, the least-squares mean difference in change from baseline to Week 16 was -3.5 mmHg, with a 95% confidence interval of -5.7 to -1.3 mmHg and P = 0.002.
Mean daily bolus insulin dose
The least-squares mean difference in percent change from baseline in mean daily bolus insulin dose was -12.32, with a 95% confidence interval of -18.17 to -6.48 and P < 0.001. Because the outcome itself is expressed as percent change, the reported estimate should be interpreted as a between-group difference in that percent-change outcome.
9. Why Was a Cochran-Mantel-Haenszel Test Used?
The primary endpoint is binary: each participant either did or did not meet the complete Week 24 composite criterion. The Cochran-Mantel-Haenszel framework is designed for comparing categorical outcomes while allowing the analysis to account for stratification variables when such strata are part of the prespecified design.
For a simple binary endpoint, the central quantity is the difference in the probability of success between treatment groups. The reported risk difference of 13.4 percentage points is therefore more directly interpretable for this endpoint than a continuous mean difference would be.
The composite endpoint requires every condition to be satisfied. Failure of any one component means the participant does not count as a primary-endpoint success.
10. Why Was MMRM Used for the Secondary Outcomes?
The secondary outcomes were measured longitudinally, with analyses spanning baseline to Week 24 or baseline to Week 16. MMRM is suited to this setting because it models repeated observations rather than reducing the longitudinal record to a single unadjusted change score.
The registry's MMRM specifications included treatment, study week, treatment-by-time interaction, and randomization strata as fixed categorical effects. A baseline-by-time interaction was also included as a covariate, with the baseline variable corresponding to the endpoint being analyzed.
| Outcome | Baseline covariate interaction | Time frame |
|---|---|---|
| A1C | Baseline A1C-by-time interaction | Baseline to Week 24 |
| Body weight | Baseline weight-by-time interaction | Baseline to Week 24 |
| Systolic blood pressure | Baseline SBP-by-time interaction | Baseline to Week 16 |
| Mean daily bolus insulin dose | Baseline mean daily bolus insulin dose-by-time interaction | Baseline to Week 24 |
11. Covariate Adjustment in the MMRM Analyses
The registry specifies three randomization strata in the MMRM analyses: BMI at screening, Week -2 A1C, and use of continuous subcutaneous insulin infusion (CSII) at screening.
| Stratification variable | Categories used in the model |
|---|---|
| BMI at Screening | <25 kg/m2; ≥25 kg/m2 |
| Week -2 A1C | ≤9.0%; >9.0% |
| Use of CSII at Screening | Yes; No |
| Study week | Included as a fixed categorical effects, and Baseline SBP-by-time interaction as a covariate. |
| Treatment-by-time | Included as a fixed categorical interaction |
Covariate adjustment can improve precision by accounting for prespecified variables associated with outcome measurements. Importantly, the adjusted least-squares mean difference is a model-based estimate; it is not simply the arithmetic difference between two observed raw means.
12. What Does a Least-Squares Mean Difference Mean?
A least-squares mean difference from an MMRM represents the model-based difference between treatment groups after accounting for the variables specified in the model. For the A1C analysis, the reported estimate of -0.46 summarizes the adjusted treatment-group difference in change from baseline over the modeled longitudinal framework.
The estimate is therefore different conceptually from an unadjusted comparison of two simple endpoint averages. It reflects the fitted statistical model and its covariate and time structure.
13. What Does the A1C Estimate of -0.46 Mean?
The reported A1C estimate is a mean difference, not a risk difference. The negative value indicates that the modeled A1C change was lower for sotagliflozin relative to placebo under the registry's stated comparison.
The 95% confidence interval from -0.54 to -0.38 describes uncertainty around the estimated difference. Because the interval is entirely below zero, the direction of the estimated difference is consistently negative across the interval.
The p-value of < 0.001 provides evidence against a zero mean difference within the specified hierarchical testing framework. It does not quantify the magnitude of the A1C effect; the estimate and confidence interval provide that information.
14. How Should the Composite Endpoint Be Interpreted?
Composite endpoints can be statistically efficient because multiple clinically relevant conditions are incorporated into one binary outcome. But they also require careful interpretation: the reported 13.4-percentage-point difference applies to the probability of satisfying all components, not to each component separately.
Glycemic component
A participant had to have an A1C value below 7.0% at Week 24.
Hypoglycemia component
The participant had to have no episode of severe hypoglycemia after randomization.
DKA component
The participant had to have no episode of diabetic ketoacidosis after randomization.
Composite success
All three conditions had to be satisfied for the participant to meet the primary endpoint.
This distinction is important when translating the result into plain language. A positive composite result does not by itself establish that the treatment had the same magnitude of effect on A1C, severe hypoglycemia, and DKA considered separately.
15. Safety Results
The registry reports serious adverse events by randomized treatment arm. The reported counts are 23 affected participants among 703 at risk in the placebo group and 48 affected participants among 699 at risk in the sotagliflozin 400 mg group.
| Safety measure | Placebo | Sotagliflozin 400 mg |
|---|---|---|
| Serious adverse events | 23/703 | 48/699 |
These are counts of affected participants relative to the number at risk reported for each arm. They should be kept conceptually separate from the efficacy endpoint: the primary endpoint requires absence of severe hypoglycemia and DKA in addition to achieving the A1C target, whereas serious adverse events represent a broader safety category.
16. Statistical Methods Explained
Why is the primary endpoint analyzed as a binary outcome?
Each participant either satisfies the complete composite criterion at Week 24 or does not. That creates a binary endpoint, making categorical-data methods such as the Cochran-Mantel-Haenszel test appropriate for the primary comparison.
What does a risk difference of 13.4 mean?
A risk difference of 13.4 means that the estimated percentage of participants meeting the composite endpoint was 13.4 percentage points higher in the sotagliflozin group than in the placebo group. It is an absolute rather than relative effect measure.
Why is the confidence interval important?
The 95% confidence interval of 8.97 to 17.81 gives a measure of the statistical precision of the primary risk-difference estimate. A narrower interval would indicate greater precision, while a wider interval would indicate more uncertainty around the estimated difference.
Why use MMRM for A1C and other continuous outcomes?
MMRM is designed for repeated longitudinal measurements. Rather than analyzing only one change score, the model incorporates study week and the treatment-by-time relationship, allowing the treatment effect to be estimated within the longitudinal structure specified by the analysis.
What is the purpose of covariate adjustment?
The MMRM analyses included prespecified randomization strata and endpoint-specific baseline-by-time interactions. Adjustment can account for important measured variables and improve the precision of the estimated treatment difference.
Why does multiplicity matter?
When several endpoints are tested, repeatedly evaluating hypotheses can increase the chance of a false-positive finding if no control strategy is used. The registry states that the secondary endpoint analyses were tested according to a hierarchical procedure, so their p-values should be interpreted within that sequence.
17. Risk Difference Versus Mean Difference
The inTandem3 results illustrate two different statistical effect measures because the primary endpoint is binary while the secondary outcomes analyzed with MMRM are continuous or quantitatively measured outcomes.
| Endpoint type | Effect measure | Interpretation |
|---|---|---|
| Primary composite endpoint | Risk difference | Absolute difference in the percentage achieving the binary endpoint |
| A1C | Mean difference | Model-based difference in change from baseline |
| Body weight | Mean difference | Model-based difference in absolute change from baseline |
| Systolic blood pressure | Mean difference | Model-based difference in change from baseline |
| Mean daily bolus insulin dose | Mean difference | Model-based difference in percent change from baseline |
These estimates should not be placed on a single common scale. A 13.4-percentage-point risk difference and a -2.98 kg mean difference answer different questions about different outcomes.
18. Interpreting the Secondary Confidence Intervals
The estimate of -0.46 has a 95% CI of -0.54 to -0.38. The interval represents uncertainty around the modeled mean difference in A1C change.
The estimate of -2.98 kg has a 95% CI of -3.31 to -2.66 kg. This describes uncertainty around the adjusted between-group difference in absolute weight change.
The estimate of -3.5 mmHg has a 95% CI of -5.7 to -1.3 mmHg in the prespecified subset with baseline SBP ≥130 mmHg.
The estimate of -12.32 has a 95% CI of -18.17 to -6.48, representing the modeled between-group difference in percent change from baseline.
19. Why the P-values Do Not Measure Effect Size
The primary p-value of <0.001 and the secondary p-values describe statistical evidence against their respective null hypotheses under the specified analysis framework. They do not tell the reader how large the treatment effect is.
Effect size is communicated by the reported estimate: the 13.4 percentage-point risk difference for the primary composite endpoint, -0.46 for A1C, -2.98 kg for body weight, -3.5 mmHg for systolic blood pressure, and -12.32 for percent change in mean daily bolus insulin dose.
The confidence intervals then add information about precision. Reading the estimate, confidence interval, and p-value together gives a more complete statistical description than any one of the three alone.
20. Statistical Interpretation Versus Clinical Interpretation
Statistical interpretation
The randomized comparison produced a positive risk difference for the primary binary composite endpoint, with a two-sided 95% confidence interval of 8.97 to 17.81 percentage points and P < 0.001. The secondary MMRM analyses also produced reported estimates with two-sided 95% confidence intervals and hierarchical testing.
Clinical interpretation
The primary endpoint incorporates both glycemic control and the absence of severe hypoglycemia and DKA. The secondary outcomes provide additional information about A1C, body weight, systolic blood pressure, and mean daily bolus insulin dose. These endpoints address different dimensions of treatment effect and should not be collapsed into one statistic.
21. Important Limitations and Interpretation Issues
- Composite endpoint: the primary result applies to a combined endpoint requiring A1C <7.0%, no severe hypoglycemia, and no DKA after randomization. It does not separately quantify the treatment effect on each component.
- Modified ITT population: the primary efficacy analysis was based on the mITT population rather than a description of all enrolled participants.
- Evaluable secondary outcomes: the registry states that secondary analyses included mITT participants evaluable for the respective outcome measure.
- Model dependence: MMRM estimates depend on the prespecified longitudinal model, including treatment, time, treatment-by-time interaction, randomization strata, and baseline-by-time covariates.
- Subset analysis: the systolic blood pressure endpoint was evaluated in participants with baseline SBP ≥130 mmHg, so it does not describe all randomized participants.
- Multiplicity: the secondary endpoints were tested according to a hierarchical procedure. Their p-values therefore require interpretation within that testing strategy.
- Different estimands: the primary risk difference and secondary mean differences describe different outcome types and cannot be interpreted as interchangeable measures of treatment effect.
- Safety versus efficacy: serious adverse events are a broader safety outcome than the specific severe hypoglycemia and DKA components included in the primary composite endpoint.
22. Why This Trial Matters Statistically
inTandem3 is useful as a teaching case because it combines a binary composite endpoint with longitudinal continuous outcomes and demonstrates how the statistical method changes according to the structure of the outcome.
| Concept | How it appears in inTandem3 |
|---|---|
| Randomization | Randomized comparison of sotagliflozin 400 mg and placebo |
| Masking | Quadruple-masked design |
| Binary endpoint | Primary composite endpoint assessed at Week 24 |
| Cochran-Mantel-Haenszel test | Primary efficacy comparison |
| Risk difference | 13.4 percentage-point primary effect estimate |
| Confidence interval | 95% uncertainty interval around each reported effect estimate |
| MMRM | Longitudinal analyses of A1C, body weight, systolic blood pressure, and bolus insulin dose |
| Covariate adjustment | Randomization strata and baseline-by-time interactions included in MMRM models |
| Multiplicity | Secondary endpoints tested according to a hierarchical procedure |
| mITT analysis | Primary efficacy analysis based on the modified Intent-to-Treat population |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT02531035 — inTandem3.
- PubMed: PubMed record for PMID 28899222.
Continue through Clinical Biostats
Connect the statistical methods used in this trial with deeper tutorials, calculators, and additional clinical trial analyses.
26. Record Summary
inTandem3 provides a useful example of how statistical methodology follows endpoint structure. The primary endpoint was a binary composite assessed at Week 24 and analyzed with a Cochran-Mantel-Haenszel test in the mITT population, producing a reported risk difference of 13.4 percentage points with a two-sided 95% CI of 8.97 to 17.81 and P < 0.001.
The secondary analyses addressed longitudinal changes in A1C, body weight, systolic blood pressure, and mean daily bolus insulin dose using MMRM models with prespecified treatment, time, treatment-by-time, randomization-stratum, and baseline-by-time terms. Their reported estimates ranged from -0.46 for A1C to -12.32 for percent change in mean daily bolus insulin dose, with the corresponding confidence intervals and p-values reported above.
The statistical story is therefore best understood by keeping the outcome scales distinct: the primary result is an absolute difference in the probability of meeting a composite endpoint, while the secondary results are model-based differences in longitudinal quantitative outcomes. The confidence intervals provide precision information, the p-values provide hypothesis-testing evidence, and the hierarchical testing procedure provides the context for interpreting the secondary analyses.