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
STEP-BD was a randomized, double-blind treatment study in bipolar disorder. The registry reports an enrollment of 5,000 participants and lists multiple drug and behavioral interventions.
| Feature | STEP-BD |
|---|---|
| Acronym | STEP-BD |
| Full title | Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD) |
| Condition | Bipolar Disorder |
| Phase | NA |
| Status | COMPLETED |
| Start date | 1998-09 |
| Allocation | RANDOMIZED |
| Masking | DOUBLE |
| Primary purpose | TREATMENT |
| Enrollment | 5,000 |
| Lead sponsor | National Institute of Mental Health (NIMH) |
| Sponsor type | NIH |
| ClinicalTrials.gov | NCT00012558 |
2. Clinical Question
The registry identifies bipolar disorder as the condition under study and treatment as the primary purpose. The broad statistical question is how treatment strategies represented by the randomized study compare within the study's treatment framework.
Population
Participants enrolled in a treatment study for bipolar disorder.
Interventions
The registry lists lithium, valproate, bupropion, paroxetine, lamotrigine, risperidone, inositol, tranylcypromine, Cognitive Behavioral Therapy, Family-focused Therapy, and Interpersonal and Social Rhythms Therapy.
Comparator
The registry identifies randomized allocation but does not provide a single comparator intervention in the record summarized here.
Primary question
What treatment comparisons were evaluated within a randomized treatment program for bipolar disorder?
3. Trial Design
STEP-BD is registered as randomized and double-blind, with treatment identified as the primary purpose. The enrollment is 5,000 participants. The registry lists both pharmacologic and behavioral interventions, indicating that the program encompasses more than a single two-arm treatment comparison.
Drug treatments
- Lithium
- Valproate
- Bupropion
- Paroxetine
- Lamotrigine
- Risperidone
- Inositol
- Tranylcypromine
Psychosocial treatments
- Cognitive Behavioral Therapy
- Family-focused Therapy
- Interpersonal and Social Rhythms Therapy
4. Interventions and the Statistical Structure of the Study
The registry lists eleven interventions spanning drug and behavioral approaches. That structure is statistically important because a program with multiple interventions cannot automatically be summarized using the same analysis as a simple two-arm parallel-group trial.
| Intervention | Type |
|---|---|
| Lithium | Drug |
| Valproate | Drug |
| Bupropion | Drug |
| Paroxetine | Drug |
| Lamotrigine | Drug |
| Risperidone | Drug |
| Inositol | Drug |
| Tranylcypromine | Drug |
| Cognitive Behavioral Therapy | Behavioral |
| Family-focused Therapy | Behavioral |
| Interpersonal and Social Rhythms Therapy | Behavioral |
For statistical interpretation, the critical question is not simply how many interventions appear in the registry. It is how participants were assigned to those interventions, which comparisons were prespecified, and which outcome was used to evaluate each comparison. Those details determine whether an analysis should use a two-group comparison, a multi-group model, repeated-measures methods, a time-to-event approach, or another framework.
5. Endpoints
The ClinicalTrials.gov record does not list any registered primary endpoints for STEP-BD. Consequently, there is no endpoint definition or time frame in the registry record summarized here that can be presented as the trial's registered primary outcome.
6. Planned Analysis
The registry establishes a randomized, double-blind treatment study but does not list registered primary endpoints or posted statistical analyses. The appropriate statistical method therefore depends on the outcome that was prespecified for a particular treatment comparison.
| Potential endpoint structure | Typical statistical approach | Key issue |
|---|---|---|
| Continuous outcome | Comparison of means or a regression model such as ANCOVA | Baseline adjustment, distribution, and missing observations |
| Binary outcome | Risk difference, risk ratio, odds ratio, or regression for binary outcomes | Definition of the response window and analysis population |
| Repeated outcome | Repeated-measures or mixed-effects model | Correlation among measurements within a participant |
| Time-to-event outcome | Kaplan-Meier estimation, log-rank testing, or a Cox model | Censoring and the interpretation of the hazard ratio |
These are general endpoint-specific frameworks rather than statistical methods identified in the STEP-BD registry entry. A formal analysis of a particular STEP-BD outcome requires the outcome definition and prespecified analysis plan for that comparison.
7. Statistical Methodology
Randomization
Randomized allocation is one of the most important statistical design features reported for STEP-BD. In a randomized study, treatment assignment is determined by the trial allocation mechanism rather than by the participant's or investigator's preference. This supports causal comparison when the trial is properly conducted and analyzed according to assignment.
Randomization is intended to make systematic baseline differences between treatment groups less likely than under nonrandomized treatment selection.
Double masking
The registry identifies the masking as double. Masking can reduce the influence of treatment knowledge on participant behavior, investigator conduct, outcome assessment, and other aspects of trial execution. Its effectiveness depends on whether the intervention can realistically remain masked throughout the study.
Analysis population
The registry entry does not specify an analysis population such as intention-to-treat, per-protocol, or as-treated in the information summarized here. These distinctions matter because analyzing participants according to randomized assignment preserves the principal comparison created by randomization, whereas analyses based on treatment actually received can introduce differences related to adherence, discontinuation, or treatment switching.
Multiplicity
Multiple interventions and multiple possible comparisons create a potential multiplicity issue. If several hypotheses are tested, the probability of obtaining at least one apparently positive result by chance can increase unless the analysis plan accounts for the number and hierarchy of tests. The registry entry does not specify a multiplicity strategy.
8. Statistical Methods Explained
Why does randomization matter?
Randomization creates the basis for a causal treatment comparison by separating treatment assignment from the characteristics that participants bring into the study. In expectation, measured and unmeasured baseline characteristics are balanced across randomized groups. Randomization does not guarantee identical groups in a particular sample, but it provides the statistical foundation for the comparison.
What does double masking add?
Masking reduces opportunities for knowledge of treatment assignment to affect behavior or assessment. This is especially relevant when outcomes can be influenced by participant reporting or investigator judgment. The degree of protection depends on whether participants and investigators can determine which treatment was received.
Why can't every intervention be analyzed with one simple t-test?
A t-test is designed for a particular two-group continuous-outcome setting. STEP-BD lists multiple pharmacologic and behavioral interventions. If several treatment groups or multiple outcomes are involved, the analysis needs to reflect the actual comparison structure and endpoint type rather than applying a generic two-group test.
Why does the endpoint definition matter?
The same participants can generate very different statistical conclusions depending on whether the endpoint is continuous, binary, repeated over time, or time-to-event. The endpoint definition determines the estimand, the appropriate model, how missing observations are handled, and how the final effect estimate should be interpreted.
Why is multiplicity important?
Suppose several independent null hypotheses are tested at the same nominal significance level. Even when every null hypothesis is true, the chance that at least one test appears statistically significant can become larger as the number of tests increases. Prespecified hierarchy, adjustment procedures, or another error-control strategy can address this issue.
Why does the analysis population matter?
An intention-to-treat analysis evaluates participants according to randomized assignment and therefore maintains the comparison established by randomization. An as-treated analysis instead groups participants according to treatment received. Those approaches answer different questions and can be affected differently by treatment discontinuation or switching.
9. What a Treatment Effect Would Mean Statistically
A treatment effect is a numerical comparison between outcomes under different treatment assignments. Its interpretation depends on the estimand: a mean difference, risk difference, ratio, odds ratio, hazard ratio, or another measure does not represent the same quantity.
A confidence interval communicates statistical uncertainty around an estimated effect under the assumptions of the chosen method. A narrow interval indicates greater statistical precision than a wide interval, but precision and clinical importance are separate concepts.
A p-value measures the compatibility of the observed data with a specified null hypothesis under the statistical model. It does not measure the size of an effect, the probability that a treatment works, or the probability that the null hypothesis is true.
The clinical importance of an effect depends on the endpoint, its magnitude, its uncertainty, the duration of follow-up, adverse outcomes, and the population represented by the study. Statistical significance alone does not establish clinical importance.
10. Results Status
The ClinicalTrials.gov record does not contain posted statistical analyses for STEP-BD. The registry also does not list registered primary endpoints in the record summarized here. Accordingly, there are no registry-posted treatment-effect estimates, confidence intervals, or p-values to interpret for a primary endpoint.
11. Safety and Adverse Events
The registry information summarized here does not provide serious adverse-event counts by intervention. It therefore does not establish an arm-level serious adverse-event comparison that can be presented as a numerical safety result.
Statistically, safety should be analyzed separately from efficacy. The appropriate denominator, exposure period, event definition, severity classification, and handling of recurrent events can materially affect the interpretation of an adverse-event comparison.
Risk measures
For a binary safety endpoint, investigators may report event proportions, risk differences, risk ratios, or odds ratios depending on the analysis plan.
Event timing
If time to an adverse event is clinically important, survival methods may be more informative than a simple proportion.
12. Limitations and Interpretation Issues
- No registered primary endpoints in the record: the registry entry does not identify a primary outcome definition or time frame for STEP-BD.
- No posted statistical analyses: the record does not provide treatment-effect estimates, confidence intervals, or p-values from formal statistical analyses.
- Multiple interventions: the registry lists eleven interventions, so the statistical comparison structure cannot be reduced to a single generic treatment-versus-control contrast.
- Multiplicity: multiple potential treatment comparisons and endpoints can require prespecified control of type I error. The registry entry does not specify such a strategy.
- Analysis population: the registry information summarized here does not identify an intention-to-treat, per-protocol, or as-treated analysis population.
- Endpoint-specific assumptions: the appropriate model depends on the outcome definition. A time-to-event analysis, for example, has different assumptions and interpretation from a continuous-outcome analysis.
- Masking: the registry identifies double masking, but masking can be more difficult to maintain for some interventions than others.
- Interpretation of nonreported results: the absence of a numerical result in the registry should not be interpreted as a numerical null effect.
13. Why This Trial Matters Statistically
STEP-BD is a useful statistical teaching case because its registry structure highlights an issue that is often overlooked: a large enrollment does not by itself tell us what statistical comparison should be made. The design, treatment assignment structure, endpoint definition, analysis population, and multiplicity plan determine how evidence should be evaluated.
| Concept | How it appears in STEP-BD |
|---|---|
| Randomization | The registry identifies RANDOMIZED allocation. |
| Masking | The registry identifies DOUBLE masking. |
| Primary purpose | The registry identifies TREATMENT as the primary purpose. |
| Multiple interventions | Eight drug interventions and three behavioral interventions are listed. |
| Sample size | Enrollment is 5,000. |
| Endpoint specification | No registered primary endpoints are listed in the record summarized here. |
| Formal statistical results | No posted statistical analyses are listed. |
| Multiplicity | Multiple interventions make prespecified comparison structure important; the registry does not specify a multiplicity strategy. |
| Analysis population | The summarized registry information does not specify an ITT, per-protocol, or as-treated population. |
14. Statistical Interpretation of the Study Size
An enrollment of 5,000 is statistically substantial, but sample size should never be interpreted independently of the effect being estimated. Statistical power depends on the number of participants contributing to a particular comparison, outcome variability or event frequency, the magnitude of the effect of interest, the significance threshold, and the analysis design.
A total enrollment figure is not equivalent to the effective sample size for every individual hypothesis.
This distinction is especially important for a study that lists multiple interventions. If the total enrollment is distributed across several treatment comparisons, the number of participants informing each comparison can be substantially smaller than the overall enrollment. The registry entry does not provide the allocation counts needed to quantify those individual comparison sizes.
15. Randomization Versus Treatment Received
The randomized assignment is the defining feature for causal inference. Once participants are randomized, treatment discontinuation, switching, or deviations from the assigned intervention can create analytic challenges. The preferred analysis depends on the estimand being targeted.
Assignment-based question
What is the difference between outcomes associated with being assigned to one treatment strategy versus another?
Exposure-based question
What is the difference between outcomes among participants according to the treatment they actually received?
Why the distinction matters
Assignment is determined by randomization, whereas treatment received can be related to adherence, discontinuation, clinical events, or other post-randomization factors.
Interpretive consequence
Two analyses can therefore answer different scientific questions even when they use the same participants.
16. Multiple Interventions and Multiplicity
The intervention list creates an important statistical question: which comparisons constitute the confirmatory family? A study can contain multiple treatments without requiring every possible pairwise comparison to be treated as a primary hypothesis. The prespecified analysis plan should establish which contrasts are confirmatory and which are supportive or exploratory.
| Issue | Statistical implication |
|---|---|
| Several interventions | More than one treatment comparison may be possible. |
| Several endpoints | Repeated testing can increase the familywise false-positive rate. |
| Prespecified hierarchy | Can establish which hypotheses are tested first and which depend on earlier results. |
| Adjusted testing | Can control an error criterion across a defined family of hypotheses. |
| Exploratory comparisons | Can generate evidence for future study without necessarily carrying confirmatory error-control status. |
The registry record does not specify which of these multiplicity strategies, if any, was used. That information would be necessary before assigning confirmatory significance to a collection of statistical tests.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The registry establishes a randomized, double-blind treatment study with 5,000 enrolled participants and multiple interventions. It does not provide registered primary endpoints or posted statistical analyses from which numerical treatment effects can be evaluated.
Clinical interpretation
The study concerns treatment strategies for bipolar disorder, but a clinical comparison requires a defined endpoint, a defined treatment contrast, and corresponding outcome data. Those elements are not specified numerically in the registry information summarized here.
18. Longitudinal Trial History
Study begins
The registry lists September 1998 as the study start date.
Study status
The registry identifies STEP-BD as COMPLETED.
19. Reading a Randomized Treatment Comparison
For readers interpreting a randomized treatment study, the most useful sequence is to begin with the estimand rather than the p-value. First identify exactly what is being compared. Then identify the outcome, the observation window, the analysis population, and the effect measure. Only after those elements are established should the confidence interval and hypothesis test be interpreted.
This sequence helps prevent the common mistake of interpreting statistical significance before understanding what was actually estimated.
For STEP-BD, the registry supplies the first part of that sequence: the condition, treatment purpose, randomized allocation, double masking, enrollment, and intervention list. It does not supply the registered primary endpoint or posted statistical analysis needed to complete an endpoint-specific numerical interpretation.
20. What Can and Cannot Be Concluded From the Registry Record
| Question | Registry record |
|---|---|
| What condition was studied? | Bipolar Disorder |
| What was the primary purpose? | TREATMENT |
| Was allocation randomized? | Yes; RANDOMIZED |
| Was masking reported? | Yes; DOUBLE |
| How many participants were enrolled? | 5,000 |
| What interventions were listed? | Eight drugs and three behavioral interventions |
| What was the registered primary endpoint? | No registered primary endpoints are listed |
| What was the treatment-effect estimate? | No posted statistical analysis is listed |
| What was the confidence interval? | No posted statistical analysis is listed |
| What was the p-value? | No posted statistical analysis is listed |
21. Important Statistical Lessons
STEP-BD illustrates several principles that generalize well beyond this particular study.
- Large enrollment is not a substitute for endpoint definition. A sample size cannot be interpreted without knowing the hypothesis it is intended to test.
- Randomization defines the treatment comparison. The statistical analysis should preserve the structure created by random assignment when the estimand calls for an assignment-based comparison.
- Multiple treatments require an explicit comparison structure. The presence of several interventions does not tell us which pairwise contrasts are confirmatory.
- Endpoint type determines methodology. Continuous, binary, repeated, and time-to-event outcomes require different statistical frameworks.
- Confidence intervals provide more information than p-values alone. They communicate the precision and range of plausible effect estimates under the chosen model.
- Missing statistical results are not equivalent to null results. A missing estimate cannot be interpreted as evidence that the treatment effect was zero.
22. Why the Absence of a Posted Endpoint Matters
A clinical-trial result is interpretable only when the statistical question is sufficiently specified. An endpoint needs a definition, a population, a time frame, and an analysis method. Without those elements, a numerical result can be difficult to connect to a meaningful scientific question.
The absence of a registered primary endpoint in the ClinicalTrials.gov record means that the record itself does not identify the principal outcome against which a primary confirmatory treatment comparison should be evaluated. This is a limitation of what can be concluded from the registry record, not evidence about whether the underlying study produced positive or negative findings.
23. Sources
- ClinicalTrials.gov: NCT00012558 — STEP-BD.
Continue through the Clinical Biostats clinical trial library
Clinical trial analysis becomes more informative when study design, endpoints, estimands, and statistical methods are considered together.
24. Record Summary
STEP-BD is registered as a completed, randomized, double-blind treatment study of bipolar disorder with an enrollment of 5,000 participants. The registry lists eight drug interventions—lithium, valproate, bupropion, paroxetine, lamotrigine, risperidone, inositol, and tranylcypromine—and three behavioral interventions: Cognitive Behavioral Therapy, Family-focused Therapy, and Interpersonal and Social Rhythms Therapy.
From a statistical perspective, the most important feature of the registry record is the distinction between study design information and posted statistical results. The record establishes randomized allocation, double masking, treatment purpose, enrollment, and the intervention set, but it does not list registered primary endpoints or posted statistical analyses. Consequently, an endpoint-specific treatment effect cannot be interpreted from the registry record alone.