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
CATIE was a randomized, double-blind, phase 4 crossover trial in schizophrenia with an enrollment of 1600 participants. The registry identifies multiple antipsychotic interventions, including perphenazine, olanzapine, quetiapine, risperidone, ziprasidone, clozapine, and fluphenazine decanoate.
| Feature | CATIE |
|---|---|
| Acronym | CATIE |
| ClinicalTrials.gov identifier | NCT00014001 |
| Official title | CATIE- Schizophrenia Trial |
| Condition | Schizophrenia |
| Phase | Phase 4 |
| Status | Completed |
| Allocation | Randomized |
| Design model | Crossover |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 1600 |
| Lead sponsor | National Institute of Mental Health (NIMH) |
| Sponsor type | NIH |
2. Clinical Question
The registry identifies CATIE as a randomized, double-blind crossover treatment trial in schizophrenia. The central statistical question is how outcomes should be compared when participants are randomly assigned to treatment strategies but the study design permits treatment sequences and within-participant comparisons through crossover.
Population
Participants enrolled in a clinical trial for schizophrenia.
Interventions
Perphenazine, olanzapine, quetiapine, risperidone, ziprasidone, clozapine, and fluphenazine decanoate.
Comparator structure
The trial is registered as a crossover design rather than as a simple parallel-group comparison.
Primary question
How do the randomized antipsychotic treatment strategies compare within the crossover framework used by the study?
3. Trial Design
CATIE is registered as a randomized, double-blind, phase 4 crossover trial with treatment as its primary purpose. Crossover designs differ fundamentally from conventional parallel-group trials because treatment exposure occurs in more than one period and the sequence of treatments can become part of the statistical structure.
Registered antipsychotic interventions
- Perphenazine
- Olanzapine
- Quetiapine
- Risperidone
Additional registered treatments
- Ziprasidone
- Clozapine
- Fluphenazine decanoate
4. Trial Characteristics Relevant to Statistical Analysis
5. Registered Endpoints
The ClinicalTrials.gov record does not report registered primary endpoints for NCT00014001. Consequently, there is no registry-defined primary endpoint, endpoint time frame, or endpoint-specific analysis specification available in the record summarized here.
| Endpoint element | ClinicalTrials.gov record |
|---|---|
| Registered primary endpoints | None reported |
| Endpoint definitions | Not reported in the registered primary-endpoint section |
| Endpoint time frames | Not reported in the registered primary-endpoint section |
| Posted statistical analyses | None |
6. Statistical Methodology
Why crossover trials require a different statistical framework
A crossover study produces observations that are not necessarily independent. The same participant can be observed during more than one treatment period, so outcomes from the same individual may be correlated. A statistical analysis that treats every period observation as though it came from a different participant can underestimate uncertainty and produce inappropriate standard errors.
Here, i identifies a participant and j identifies a treatment period. The exact statistical model depends on the endpoint and the prespecified analysis plan.
Within-participant correlation
Repeated observations from the same participant are usually correlated. This correlation is one reason crossover analyses commonly use repeated-measures, mixed-effects, generalized estimating equation, or other methods that explicitly account for participant-level dependence, depending on the outcome.
Treatment-period effects
Period effects occur when outcomes systematically differ between treatment periods for reasons unrelated to the treatment itself. Calendar time, disease progression, learning effects, changing clinical circumstances, or other temporal factors can contribute to a period effect.
Sequence effects
A sequence effect occurs when the order in which treatments are received influences the observed outcome. In a crossover design, treatment A followed by treatment B is not necessarily statistically interchangeable with treatment B followed by treatment A.
Carryover effects
A carryover effect occurs when exposure to a treatment influences outcomes during a later treatment period. This is particularly important when treatment effects do not disappear before the next treatment period begins. The appropriate handling of carryover depends on the treatment, outcome, timing of assessments, and prespecified design.
7. Planned Analysis
The registry does not post statistical analyses for CATIE. No formal endpoint-specific results are posted on ClinicalTrials.gov in the record summarized here.
For a randomized crossover trial, the statistical analysis would normally begin by identifying the endpoint type and the unit of analysis. Continuous outcomes could be analyzed with a crossover-appropriate repeated-measures or mixed-effects model. Binary outcomes could use a model that accounts for within-participant correlation. Time-to-event outcomes require additional consideration because ordinary crossover methods do not automatically apply to survival endpoints.
| Statistical issue | Typical analytical consideration |
|---|---|
| Continuous outcome | Mixed-effects or repeated-measures modeling can account for treatment, period, sequence, and participant-level dependence. |
| Binary outcome | Generalized mixed-effects or other correlated-data methods may be appropriate. |
| Repeated observations | Participant-level correlation should be incorporated into estimation and standard errors. |
| Treatment sequence | Sequence can be included as a design factor when relevant to the prespecified analysis. |
| Period | Period effects should be considered because outcomes can change over time independently of treatment. |
| Carryover | Potential residual treatment effects must be considered when interpreting later periods. |
| Missing observations | The mechanism and amount of missing data should be assessed before selecting an estimation or imputation strategy. |
8. Statistical Methods Explained
Why does crossover design change the analysis?
In a parallel-group trial, treatment is typically a between-participant factor. In a crossover trial, treatment can become a within-participant factor because the same participant may receive different treatments in different periods. That creates correlation between observations and changes the variance calculation.
Why is treatment sequence important?
Sequence describes the order in which treatments are received. If outcomes differ depending on whether a treatment is given first or second, the estimated treatment contrast can become entangled with the order of exposure. Sequence therefore belongs to the design considerations of a crossover analysis rather than being an incidental descriptive variable.
What is a period effect?
A period effect is a systematic change associated with the treatment period rather than with the treatment itself. For example, if outcomes tend to change from the first period to the second period regardless of treatment, an analysis that ignores period could attribute some of that change to treatment.
What is carryover?
Carryover means that the effect of treatment from an earlier period persists into a later period. If treatment A has a lasting effect after the participant switches to treatment B, the observed response during the B period may reflect both current and previous treatment exposure.
Why does participant-level correlation matter?
Repeated observations from one person are generally more similar to one another than observations from different people. Statistical models that recognize this correlation can estimate treatment effects with appropriate uncertainty rather than treating all observations as independent.
Why can a crossover design be statistically efficient?
When appropriate for the clinical setting, a crossover design can allow each participant to provide information under more than one treatment condition. This can reduce some between-person variability because treatment comparisons can incorporate within-participant information. The advantage depends on the disease, treatment, outcome, period duration, and absence of problematic carryover.
When is a crossover design difficult to interpret?
Crossover designs become more complicated when the disease changes materially over time, when treatment effects persist after switching, when periods are too short for treatment effects to stabilize, or when treatment withdrawal itself affects subsequent outcomes. These features can make period, sequence, and carryover effects difficult to separate from treatment effects.
9. Randomization and Causal Interpretation
Randomization is central to the causal interpretation of treatment comparisons. If treatment assignment is randomized, systematic baseline differences are expected to balance in expectation across treatment assignments. This is the principal reason randomized trials can support stronger causal inference than observational comparisons.
What randomization provides
A mechanism for making treatment assignment independent of measured and unmeasured baseline prognostic factors in expectation.
What randomization does not guarantee
It does not guarantee identical baseline characteristics in every realized sample, nor does it eliminate bias caused by post-randomization departures from the intended design.
The crossover structure adds another layer: treatment comparisons must account for repeated observations and the order and timing of treatment exposure. Randomization alone does not make an analysis that ignores these design features statistically appropriate.
10. Double-Blind Design
The registry identifies CATIE as double-blind. Masking is relevant because knowledge of treatment assignment can influence treatment behavior, reporting of subjective outcomes, assessment of symptoms, and decisions surrounding treatment continuation or switching.
| Design feature | Statistical relevance |
|---|---|
| Double-blind | Can reduce differential expectations and assessment behavior associated with treatment assignment. |
| Randomized | Supports causal treatment comparisons by balancing treatment assignment in expectation. |
| Crossover | Creates repeated treatment exposure and makes period, sequence, and carryover important. |
| Phase 4 | Places the trial in the post-approval phase category used by the registry. |
11. Multiple Treatment Comparisons
CATIE includes seven registered drug interventions. A study with multiple treatment strategies can generate many possible pairwise comparisons. The statistical interpretation of those comparisons depends on which comparisons were prespecified and how multiplicity was handled.
With seven interventions, there are 21 unique unordered pairwise comparisons if every treatment is compared with every other treatment. The existence of these possible contrasts does not establish that all 21 comparisons were prespecified or performed.
When many hypotheses are tested, the probability of obtaining at least one apparently positive result by chance increases unless the analysis accounts for multiplicity. Common approaches include hierarchical testing, adjusted significance thresholds, familywise-error procedures, or false-discovery-rate methods, depending on the scientific objective.
12. Analysis Populations and Missing Data
The registry record does not provide endpoint-specific analysis populations or a missing-data analysis plan. These details matter particularly in crossover trials because participants may discontinue before completing later treatment periods.
Early discontinuation
A participant who does not complete a later period contributes a different amount of information from a participant who completes every planned period.
Informative missingness
If the probability of missing an outcome is related to treatment response, tolerability, or disease status, a complete-case analysis can be biased.
Imputation
Imputation should be based on the endpoint, timing of missingness, assumptions about the missing-data mechanism, and the prespecified analysis plan.
Sensitivity analysis
Alternative assumptions about missing outcomes can be used to evaluate how dependent the treatment conclusion is on unverifiable assumptions.
Because endpoint definitions and posted analyses are not reported in the registry record, a specific missing-data method cannot be assigned to CATIE from the registry alone.
13. Safety and Adverse Events
The ClinicalTrials.gov record summarized here does not post serious adverse-event counts by treatment arm. No arm-specific safety estimates are therefore presented.
| Safety element | Registry record |
|---|---|
| Serious adverse events by arm | Not reported in the ClinicalTrials.gov record |
| Arm-specific safety estimates | Not reported |
| Safety analysis population | Not reported |
| Specific adverse-event endpoints | Not reported |
Statistically, safety should be analyzed separately from efficacy because adverse events can be binary, categorical, continuous, recurrent, or time-to-event outcomes. The appropriate method depends on the event definition and observation period.
14. Crossover: The Central Statistical Issue
The crossover designation is the most important methodological feature in the registry record. It means that a simple comparison of all observations as though they arose from independent parallel groups would fail to represent the design.
Why these effects can be confounded
Suppose treatment A is commonly administered before treatment B. If outcomes improve from period 1 to period 2, that improvement could reflect treatment B, natural temporal change, learning or adaptation, or residual effects from treatment A. The statistical design must distinguish these possibilities as far as the data and prespecified model allow.
15. Why This Trial Matters Statistically
CATIE is a useful teaching case because its registry design combines randomization, double masking, multiple treatment interventions, and a crossover structure. Each feature changes a different part of the statistical problem.
| Concept | How it appears in CATIE |
|---|---|
| Randomization | The registry identifies the allocation as randomized. |
| Masking | The registry identifies the trial as double-blind. |
| Crossover | The registered design model is crossover. |
| Multiple treatments | Seven antipsychotic drug interventions are registered. |
| Repeated observations | Crossover treatment periods can produce multiple observations per participant. |
| Within-participant dependence | Repeated measurements from the same participant should not automatically be treated as independent. |
| Period effects | Changes over treatment periods can affect crossover comparisons. |
| Sequence effects | The order of treatment exposure can influence interpretation. |
| Carryover | Residual effects from an earlier treatment can complicate later-period comparisons. |
| Multiplicity | Multiple interventions create the possibility of multiple treatment comparisons. |
16. Important Limitations and Interpretation Issues
- No registered primary endpoints are reported: the registry record does not provide endpoint definitions or endpoint-specific time frames from which a primary efficacy analysis can be specified.
- No posted statistical analyses: the registry does not provide endpoint-level statistical estimates, confidence intervals, or p-values.
- Crossover dependence: observations from the same participant may be correlated, requiring analysis methods appropriate for repeated measurements.
- Period effects: outcomes can change over time independently of treatment assignment.
- Sequence effects: treatment order can influence outcomes in a crossover study.
- Carryover: residual effects from an earlier treatment can complicate interpretation of subsequent treatment periods.
- Missing data: discontinuation before later periods can produce unequal information across treatment sequences and can make assumptions about missing outcomes important.
- Multiple comparisons: seven registered interventions create numerous possible pairwise treatment contrasts, making multiplicity an important consideration if several comparisons are tested.
- Generalizability: the registry identifies schizophrenia as the condition and the trial as phase 4, but the summarized record does not provide the detailed eligibility and baseline characteristics needed to define the represented patient population more narrowly.
17. Statistical Interpretation Without Posted Results
The registry identifies a randomized, double-blind crossover treatment trial with 1600 participants and seven registered antipsychotic interventions. These design features establish the statistical structure of the study but do not themselves establish the magnitude or direction of any treatment effect.
A treatment comparison requires an estimated contrast, such as a mean difference, odds ratio, risk ratio, hazard ratio, or another endpoint-specific measure. Without an endpoint definition and a posted estimate, the numerical magnitude of a treatment effect cannot be described.
When an effect estimate is available, its confidence interval describes statistical uncertainty around the estimate under the specified model and sampling framework. A confidence interval is not a range containing the outcomes of individual participants.
A p-value evaluates compatibility with a specified null hypothesis under the statistical model. It does not describe the size, clinical importance, or individual-level distribution of a treatment effect.
18. What a Complete Crossover Analysis Would Need to Establish
A complete statistical interpretation of CATIE would need to connect each outcome to its corresponding treatment period, treatment sequence, participant-level correlation structure, and analysis population.
| Question | Why it matters |
|---|---|
| What was the primary outcome? | Defines the principal treatment comparison and its statistical scale. |
| When was the outcome measured? | Determines the relevant treatment period and temporal interpretation. |
| How were treatment sequences assigned? | Determines how sequence contributes to the analysis. |
| Was carryover plausible? | Determines whether later-period outcomes can be interpreted as effects of the current treatment alone. |
| How were repeated observations modeled? | Determines whether standard errors appropriately reflect within-participant dependence. |
| How was missingness handled? | Determines how incomplete treatment sequences affect the estimated comparison. |
| How were multiple treatment comparisons handled? | Determines the interpretation of p-values and confidence intervals across multiple hypotheses. |
19. Planned Analysis: Interpreting Treatment Contrasts
For a continuous outcome, an adjusted mean difference can provide an interpretable treatment contrast. For a binary outcome, an odds ratio, risk ratio, or risk difference may be appropriate. For a time-to-event endpoint, a hazard ratio or another survival measure may be used. The correct choice depends on the actual registered or prespecified endpoint.
The scale of the effect depends on the endpoint. A positive difference is not inherently beneficial or harmful without knowing how the endpoint is defined.
In a crossover study, that basic contrast must be interpreted alongside treatment period, sequence, and within-participant dependence. A statistically appropriate model can separate these components when the design and data provide enough information to do so.
20. Limitations of Simple Before-and-After Comparisons
A tempting analysis of a crossover trial is to compare each participant's outcome before and after switching treatments. Although within-participant comparisons can be informative, a simple before-and-after difference can be misleading if time itself changes the outcome.
Temporal change
If the condition changes naturally over time, the second-period outcome may differ even without a treatment effect.
Regression to the mean
Participants entering a treatment period because of unusually high or low measurements may tend to move toward their usual level independently of treatment.
Treatment persistence
An earlier treatment can continue influencing outcomes after the participant crosses over.
Selective completion
Participants who complete later treatment periods may differ systematically from those who discontinue earlier.
21. Registry Status and Evidence Interpretation
The ClinicalTrials.gov record identifies CATIE as a completed phase 4 randomized, double-blind crossover trial with an enrollment of 1600. It does not post statistical analyses in the record summarized here.
| Evidence element | Reported status |
|---|---|
| Trial status | Completed |
| Start date | 2000-12 |
| Enrollment | 1600 |
| Allocation | Randomized |
| Masking | Double |
| Design model | Crossover |
| Primary purpose | Treatment |
| Registered primary endpoints | None reported |
| Posted statistical analyses | None |
22. Sponsor and Trial Context
The lead sponsor is the National Institute of Mental Health (NIMH), and the sponsor type is NIH. The condition listed in the registry is schizophrenia, and the primary purpose is treatment.
Lead sponsor
National Institute of Mental Health (NIMH)
Sponsor type
NIH
Condition
Schizophrenia
Primary purpose
Treatment
23. Related Tutorials
Learn more about the method most directly connected with this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT00014001.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through the Clinical Biostats statistical library
Explore statistical tutorials and calculators that develop the methods used to design, analyze, and interpret clinical trials.
26. Record Summary
CATIE is registered as a completed phase 4 randomized, double-blind crossover trial in schizophrenia with 1600 participants and seven antipsychotic drug interventions: perphenazine, olanzapine, quetiapine, risperidone, ziprasidone, clozapine, and fluphenazine decanoate. From a statistical perspective, the crossover structure is central because treatment period, sequence, carryover, participant-level correlation, missing observations, and multiple treatment comparisons can all influence the analysis.