VAST-D: Complete Statistical Analysis of Augmentation and Switching Treatments in Major Depressive Disorder

1. Trial at a Glance

VAST-D was a randomized, parallel, phase 3 treatment trial in major depressive disorder. It compared three strategies: switching treatment to bupropion-SR, augmenting an antidepressant with bupropion-SR, and augmenting an antidepressant with aripiprazole.

2. Clinical Question

The primary clinical question was whether the probability of achieving protocol-defined remission during the 12-week acute treatment phase differed among switching and augmentation strategies for major depressive disorder.

3. Trial Design

The three reported safety denominators sum to the overall enrollment of 1,522. They provide the arm-specific denominators used for the serious adverse-event counts reported in the registry.

4. Primary Endpoint

The endpoint is binary at the participant level: each participant either meets the protocol remission definition or does not. The requirement for a QIDS-C16 score of ≤ 5 at two consecutive visits makes the endpoint more stringent than a single-visit threshold because remission must be sustained across consecutive assessments.

5. Statistical Methodology

Logistic regression for remission and response

The registry reports logistic regression for the binary remission and response endpoints. The resulting effect measure is an odds ratio. For the primary remission analyses, the model was stratified by participating medical center.

Stratified analysis

The primary remission comparisons were stratified by participating medical center. This accounts for the fact that the analysis compares treatment groups while allowing the statistical comparison to respect the participating-center structure.

Log-rank analysis for relapse

Protocol relapse after acute-phase remission was analyzed as a time-to-event endpoint using the log-rank method. The analysis population was the cohort of participants who met the criteria for protocol remission during the first 12 weeks of follow-up.

Intention-to-treat analysis

The registry specifically identifies the whole randomized cohort by intent-to-treat as the analysis population for the Clinical Global Impression improvement response endpoint. Under this principle, participants are analyzed according to their randomized treatment assignment.

Superiority hypotheses

The posted primary and secondary analyses are identified as superiority hypotheses. Thus, the statistical question is whether one treatment strategy produces a different outcome from the comparison strategy, rather than whether a treatment meets a prespecified non-inferiority margin.

6. Multiplicity and the Primary Testing Sequence

The primary endpoint was evaluated through three pairwise comparisons among the three treatment strategies. The registry identifies a Hochberg approach for the two comparisons involving switching and describes a conditional third comparison.

The ordering and conditional structure matter because three p-values cannot automatically be interpreted as three independent tests at the same nominal threshold. Multiplicity adjustment changes the decision threshold used for the specified family of hypotheses; it does not change the underlying odds ratio or its confidence interval.

7. Primary Results: Protocol Remission

Antidepressant + Bupropion-SR vs Switching to Bupropion-SR

The posted estimate represents the relative odds of remission for antidepressant + bupropion-SR augmentation versus switching to bupropion-SR. An OR of 1.31 corresponds to estimated odds that were 31% higher in the augmentation group than in the switching group.

Antidepressant + Aripiprazole vs Switching to Bupropion-SR

The estimate represents the relative odds of remission for antidepressant + aripiprazole augmentation versus switching to bupropion-SR. An OR of 1.42 corresponds to estimated odds that were 42% higher in the aripiprazole-augmentation group than in the switching group.

Antidepressant + Aripiprazole vs Antidepressant + Bupropion-SR

The estimate represents the relative odds of remission for antidepressant + aripiprazole augmentation versus antidepressant + bupropion-SR augmentation. The registry states that this comparison was performed at the 0.05 significance level if either of the first two co-primary hypothesis tests was significant.

8. Secondary Endpoint: Protocol Relapse After Remission

The registry defines the secondary endpoint as the Rate of Protocol Relapse of Symptoms of Major Depression After Achieving Remission in the Acute Phase. The time frame is within 36 weeks after randomization (initiation of treatment).

The analysis population is the cohort of participants who met the criteria for protocol remission during the acute phase, defined as the first 12 weeks of follow-up. Because relapse is observed over time, the registry uses the log-rank test and reports hazard ratios.

Antidepressant + Bupropion-SR vs Switching

Antidepressant + Aripiprazole vs Switching

Antidepressant + Aripiprazole vs Antidepressant + Bupropion-SR

9. Secondary Endpoint: Protocol Response by QIDS-C

The registry defines this endpoint as the Rate of Protocol Response as Reduction in Symptoms of Major Depression (≥ 50% Reduction in QIDS-C), measured during the acute phase up to 12 weeks.

Antidepressant + Bupropion-SR vs Switching

Antidepressant + Aripiprazole vs Switching

Antidepressant + Aripiprazole vs Antidepressant + Bupropion-SR

10. Secondary Endpoint: Clinical Global Impression Improvement Response

The registry defines this endpoint as the Rate of Protocol Response Measured as a Change in Clinical Global Impression (CGI) - Improvement Scale, during the acute phase up to 12 weeks. The analysis population is the whole randomized cohort by intent-to-treat.

Antidepressant + Bupropion-SR vs Switching

Antidepressant + Aripiprazole vs Switching

Antidepressant + Aripiprazole vs Antidepressant + Bupropion-SR

11. Consolidated Statistical Results

The table shows an important distinction between the primary remission endpoint and the secondary response measures. Several secondary response comparisons have larger odds ratios than the corresponding remission comparisons, while the relapse analyses use hazard ratios because relapse is defined over time among participants who first achieved remission.

12. Statistical Methods Explained

Why was logistic regression used for protocol remission?

Protocol remission is a binary outcome: a participant either met the definition of sustained QIDS-C16 ≤ 5 for two consecutive visits or did not. Logistic regression is designed for binary outcomes and estimates treatment effects on the odds scale. In VAST-D, the primary remission analyses were stratified by participating medical center, so the reported odds ratios account for that stratification rather than representing simple unadjusted odds ratios.

What does an odds ratio of 1.42 mean?

An OR of 1.42 means that the estimated odds of the outcome in the numerator group are 1.42 times the odds in the comparison group. In the primary remission comparison, the numerator was antidepressant + aripiprazole and the comparison was switching to bupropion-SR. It does not mean that remission probability increased by 42 percentage points, or even necessarily by 42%.

Why was a log-rank test used for relapse?

Relapse is not simply a yes-or-no endpoint at one fixed time. The registry specifies relapse within 36 weeks after randomization and identifies a time-to-event analysis population consisting of participants who achieved remission during the first 12 weeks. The log-rank test is designed to compare event-time distributions while accounting for differing follow-up and censoring.

What does a hazard ratio of 0.96 mean?

An HR of 0.96 means that the estimated instantaneous relapse rate in the numerator group is 0.96 times that of the comparison group under the reported analysis. It does not mean that 96% of participants avoided relapse, nor does it establish identical relapse probabilities over the full follow-up period.

Why does the confidence interval matter?

The confidence interval communicates statistical precision around the estimated effect. For example, the primary remission OR of 1.42 has a 95% CI of 1.06–1.89, while the comparison of aripiprazole augmentation with bupropion-SR augmentation has a 95% CI of 0.84–1.48. These intervals convey substantially more information than their point estimates alone because they show the range of effect estimates compatible with the specified statistical framework.

Why was multiplicity adjustment important?

Three primary pairwise comparisons were evaluated. The registry describes a Hochberg approach for the first two co-primary hypotheses and a conditional test of the two augmentation strategies if either initial hypothesis was significant. This structure controls interpretation across the prespecified family more carefully than treating every p-value as an unrelated test at 0.05.

Why does the relapse analysis use a different population?

Relapse after remission can only be assessed meaningfully among participants who first achieved the protocol remission criterion. The registry therefore defines the relapse analysis population as participants who met the remission criteria during the acute phase. This is fundamentally different from the whole randomized cohort used for the CGI improvement response analysis.

13. Interpreting Odds Ratios and P-values Together

VAST-D illustrates why effect estimates, confidence intervals, and p-values should be read together.

For example, the primary remission OR of 1.31 has a P-value of 0.076, while the aripiprazole-versus-switching OR is 1.42 with P = 0.018. The difference between these p-values should not be interpreted as a direct measure of how much more clinically important one treatment effect is. The effect estimates and their confidence intervals describe magnitude and precision; the p-values address evidence against the corresponding null hypotheses.

14. Safety

The registry reports serious adverse events by randomized treatment arm using the following affected/at-risk counts.

The registry provides affected/at-risk counts rather than a posted comparative statistical analysis of serious adverse events. Consequently, the safety information is best read as descriptive arm-level reporting rather than as an independently tested treatment contrast.

15. What the Primary Results Do — and Do Not — Establish

16. Statistical Interpretation vs Clinical Interpretation

17. Important Limitations and Interpretation Issues

• Open-label design: the registry identifies masking as none. For outcomes involving clinician-rated symptom or global-improvement assessments, lack of masking can be relevant to interpretation because treatment assignment is not concealed from participants and investigators.
• Multiple primary comparisons: three pairwise remission comparisons were evaluated, and the registry specifies a Hochberg-based testing structure. Individual p-values should therefore be interpreted within that framework.
• Odds-ratio interpretation: odds ratios are not probability ratios. Converting an OR directly into a percentage-point treatment difference would require additional information about baseline probabilities.
• Different analysis populations: the relapse analysis is restricted to participants who achieved acute-phase protocol remission, whereas the CGI improvement response analysis uses the whole randomized cohort by intent-to-treat.
• Time-to-event assumptions: hazard ratios summarize relative event rates over time and rely on the statistical structure of survival analysis. They should not automatically be interpreted as constant relative risks at every point in follow-up.
• Confidence intervals: several relapse estimates have broad intervals, illustrating the uncertainty that can arise when an analysis is restricted to the subgroup of participants who achieved remission.
• Descriptive safety reporting: the registry reports serious adverse events by arm but does not provide a posted comparative statistical analysis for those counts in the information summarized here.
• Registry detail: the ClinicalTrials.gov record provides the registered endpoint definitions and posted statistical analyses, but not every possible detail of a full statistical analysis plan.

18. Why This Trial Matters Statistically

VAST-D is a useful statistical case study because its three-arm randomized structure generates several related pairwise comparisons while also requiring different methods for binary and time-to-event outcomes.

19. A Statistical Walk-Through of the Trial

20. Reading the Confidence Intervals Across Endpoints

The confidence intervals provide a useful way to compare statistical precision across the posted analyses without reducing the evidence to p-values alone.

The contrast between the remission and relapse intervals also illustrates why the same trial can contain estimates with very different precision. The relapse analysis begins with a restricted population of participants who achieved remission, while the primary remission analysis begins with the randomized treatment comparison.

21. What the P-values Do — and Do Not — Tell You

The VAST-D registry record reports p-values ranging from 0.87 to <0.0001 across the posted analyses. These values should be interpreted in conjunction with the effect estimates, confidence intervals, endpoint definitions, analysis populations, and multiplicity structure.

A smaller p-value does not automatically mean a larger treatment effect. For example, the QIDS-C response comparison with OR 1.74 and P < 0.0001 addresses a different endpoint from the primary remission comparison with OR 1.42 and P = 0.018. The numerical magnitude of the odds ratios, not their p-values alone, describes the relative effect on the odds scale.

22. Limitations of the Statistical Evidence

Several features should remain visible when interpreting the full set of analyses rather than focusing on one reported statistic.

• Endpoint hierarchy matters: the primary endpoint was protocol remission during the acute phase, while response and relapse were secondary endpoints.
• Outcome definitions matter: remission required sustained QIDS-C16 ≤ 5 for two consecutive visits, whereas QIDS-C response required at least a 50% reduction in QIDS-C.
• Measurement scale matters: logistic regression produces odds ratios, while the log-rank analysis produces hazard ratios. They cannot be compared as if they were the same effect measure.
• Population matters: the relapse analysis is conditional on having achieved remission, whereas the CGI improvement response analysis uses the whole randomized cohort by intent-to-treat.
• Multiplicity matters: the primary remission endpoint generated three pairwise comparisons under a prespecified testing sequence.
• Center stratification matters: the primary logistic analyses account for participating medical center through stratification.
• Open-label design matters: masking was listed as none, which is relevant when outcomes involve clinician-rated assessments.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

• ClinicalTrials.gov: VAST-D, NCT01421342.
• PubMed: PMID 27479536.
• PubMed: PMID 28697253.
• PubMed: PMID 40742171.
• PubMed: PMID 37859623.
• PubMed: PMID 35366613.

26. Record Summary

VAST-D provides a useful example of how a three-arm randomized clinical trial can combine several statistical frameworks within a single protocol. The primary endpoint was a binary remission outcome analyzed with stratified logistic regression and odds ratios, while relapse after remission was treated as a time-to-event outcome analyzed with log-rank testing and hazard ratios. Secondary response endpoints used logistic regression, including analyses explicitly identified as intent-to-treat. The primary remission comparisons also incorporated a Hochberg-based multiplicity structure.

The statistical interpretation is therefore best built from several components: the exact endpoint definition, the treatment comparison being made, the effect measure, the confidence interval, the p-value, the analysis population, and the multiplicity framework. Reading those elements together provides substantially more information than any individual odds ratio, hazard ratio, or p-value viewed in isolation.