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
CHOIR was a randomized, parallel, unmasked phase 4 treatment study evaluating epoetin alfa in a trial population recorded under the condition anemia. The registry records 1432 enrolled participants and a composite primary outcome consisting of mortality, myocardial infarction, stroke, and hospitalization for congestive heart failure, with an exclusion for hospitalizations during which renal replacement therapy occurs.
| Feature | CHOIR |
|---|---|
| Study acronym | CHOIR |
| Full title | Correction of Hemoglobin and Outcomes in Renal Insufficiency (CHOIR) |
| Phase | Phase 4 |
| Status | Terminated |
| Condition | Anemia |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1432 |
| Intervention | Epoetin Alfa (drug) |
| Lead sponsor | Johnson & Johnson Pharmaceutical Research & Development, L.L.C. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT00211120 |
2. Clinical Question
The registry identifies epoetin alfa as the intervention and anemia as the study condition. The primary outcome is a composite of all-cause mortality, myocardial infarction, stroke, and hospitalization for congestive heart failure, excluding hospitalizations during which renal replacement therapy occurs.
Population
The registry lists anemia as the condition under study. The record summarized here does not provide additional population-defining characteristics.
Intervention
Epoetin Alfa, recorded as a drug intervention.
Comparator
The registry information summarized here does not identify the comparator arm or its intervention description.
Primary question
Does randomized assignment involving epoetin alfa affect the time to the first occurrence of the prespecified composite clinical outcome?
3. Trial Design
Participants were allocated randomly according to the registry design.
Participants remained in parallel randomized groups rather than moving through a crossover design described in the record.
The registry records the study as unmasked.
The registry classifies the primary purpose as treatment.
Study status
The ClinicalTrials.gov record lists the study status as terminated. The recorded study start date is 2002-03. Enrollment was 1432.
Why randomization matters
Randomization is the central design feature for estimating a causal treatment contrast. When allocation is randomized, baseline differences between groups arise through the allocation process rather than through deliberate selection of participants into treatment groups. With sufficiently large samples, randomization supports balance in both measured and unmeasured prognostic factors in expectation.
That does not mean that every baseline characteristic will be numerically identical between randomized groups. Nor does randomization by itself guarantee an unbiased estimate if post-randomization events, missing outcome assessments, treatment switching, or other departures from the intended analysis are handled inappropriately.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint | The primary outcome will be a composite consisting of mortality (all cause), myocardial infarction, stroke, and CHF hospitalization (not including those hospitalizations during which RRT occurs) | The registry definition provided does not specify a separate numerical follow-up time frame. |
Understanding the composite
The primary endpoint combines four clinically distinct types of events: death from any cause, myocardial infarction, stroke, and hospitalization for congestive heart failure. The registry definition also makes a specific distinction for CHF hospitalization by excluding hospitalizations during which renal replacement therapy occurs.
For a time-to-event analysis, a composite endpoint is generally operationalized around the first qualifying component event. That produces a single event time for each participant who experiences one of the qualifying components, while participants who do not experience a qualifying event during observation are censored according to the prespecified follow-up rules.
5. Statistical Methodology
The registry identifies a randomized, parallel trial with a composite clinical outcome. No formal statistical analyses are posted in the ClinicalTrials.gov record summarized here. Results have not been posted on ClinicalTrials.gov.
Typical analysis of the primary endpoint
A composite outcome defined by clinical events occurring over follow-up is typically analyzed as a time-to-first-event endpoint. The statistical objective is to compare the distribution of time until the first qualifying event between randomized groups.
A standard analysis would generally begin with Kaplan-Meier estimation of event-free survival and a between-group comparison using a log-rank test. A Cox proportional-hazards model is commonly used to estimate a hazard ratio and its confidence interval.
Here, h(t) represents the instantaneous event rate at time t. A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the numerator group under the fitted model; it does not directly equal a relative reduction in cumulative event probability.
Kaplan-Meier estimation
Kaplan-Meier estimation is useful when participants are followed for different lengths of time or when some participants have not experienced the event by the end of their observed follow-up. Such participants are censored rather than treated as if they had experienced the endpoint.
At each event time, di is the number of events and ni is the number of participants at risk immediately before that time.
Cox proportional-hazards modeling
A Cox model estimates a relative hazard while leaving the baseline hazard unspecified. For a randomized comparison, the treatment indicator is the primary explanatory variable. Additional covariates can be included when prespecified by the statistical analysis plan and appropriate to the estimand.
The hazard ratio associated with a binary treatment indicator is exp(β). The interpretation depends on the proportional-hazards structure of the fitted model.
Log-rank comparison
The log-rank test compares observed and expected event counts over follow-up under the null hypothesis that the time-to-event distributions do not differ between randomized groups. It gives a hypothesis test, while a hazard ratio supplies an effect estimate.
Composite-event analysis
For a composite endpoint, the primary time-to-event analysis generally uses the first qualifying component. This makes the endpoint statistically tractable as one event process, but it also means that frequent components can contribute more events to the composite than rare components. Consequently, a composite result should always be understood in the context of its component definitions.
6. Planned Analysis
The registry's primary endpoint is a composite consisting of all-cause mortality, myocardial infarction, stroke, and CHF hospitalization, excluding hospitalizations during which renal replacement therapy occurs. Because the endpoint consists of clinical events observed over follow-up, the natural primary framework is a time-to-first-event analysis.
| Analytic question | Typical statistical approach | What it estimates |
|---|---|---|
| How quickly do participants experience the composite? | Kaplan-Meier method | Estimated event-free probability over time |
| Do randomized groups differ in time to the composite? | Log-rank test | Evidence against equality of the time-to-event distributions |
| How large is the relative treatment effect? | Cox proportional-hazards model | Hazard ratio with a confidence interval |
| Which component drove the composite? | Component-specific event summaries and time-to-event analyses | Contribution of mortality, myocardial infarction, stroke, and CHF hospitalization |
Why time-to-event analysis is appropriate
A binary analysis at one arbitrary time point discards information about when events occur and how long participants remain event-free. A time-to-event analysis uses the timing of events and accommodates censoring, making it particularly suited to composite clinical outcomes that can occur at different times during follow-up.
What the analysis would need to specify
- Event definition: the exact rules for identifying each qualifying component.
- Time origin: the date from which event-free survival is measured.
- Censoring: rules for participants who have not experienced a qualifying event by the end of observation.
- Analysis population: whether the primary analysis follows intention-to-treat principles.
- Model specification: whether the Cox model is unadjusted or stratified/adjusted and how ties are handled.
- Confidence interval: the prespecified confidence level for the treatment-effect estimate.
- Multiplicity: how the composite endpoint relates to any secondary or component analyses.
The registry information summarized here does not provide these additional statistical-analysis specifications. They should therefore not be inferred as if they were prespecified features of this particular trial.
7. Statistical Methods Explained
Why is a composite endpoint used?
A composite can combine several clinically important outcomes into a single endpoint. This can increase the number of observed events and create one prespecified primary comparison. The tradeoff is interpretability: the components may differ substantially in clinical importance, frequency, and treatment effect.
What does "time to the first event" mean?
Suppose a participant experiences myocardial infarction before any other qualifying event. That myocardial infarction supplies the participant's event time for the composite. If another participant later experiences stroke after first experiencing myocardial infarction, the stroke does not create a second primary composite event under a time-to-first-event analysis.
Why use Kaplan-Meier rather than a simple percentage?
Participants can have different amounts of observable follow-up. Kaplan-Meier estimation incorporates the available information up to censoring and estimates the probability of remaining free of the event through each point in time. A simple percentage at the end of follow-up does not handle unequal observation in the same way.
What does a hazard ratio mean for this trial?
A hazard ratio compares instantaneous event rates between randomized groups under a Cox model. A value below 1 would indicate a lower estimated instantaneous rate in the treatment group named in the numerator. It is not equivalent to saying that the same percentage of participants avoid the composite or that the probability of an event is reduced by exactly the same percentage.
Why does the composite definition matter?
Because the primary endpoint combines mortality, myocardial infarction, stroke, and CHF hospitalization, the overall result represents a mixture of these outcomes. If one component is much more frequent than the others, it can contribute disproportionately to the composite estimate. A composite should therefore be interpreted together with component-specific event information when such information is available.
Why does censoring matter?
Participants who have not experienced a qualifying event by the time their observation ends are generally censored rather than classified as event-free indefinitely. The validity of Kaplan-Meier and Cox analyses depends on assumptions about the censoring process. Informative censoring can bias estimates if participants who leave observation differ systematically in future event risk from those who remain under observation.
8. Interpreting the Primary Endpoint
The endpoint is a composite clinical outcome, not a laboratory measurement or surrogate endpoint. It includes four event categories: all-cause mortality, myocardial infarction, stroke, and CHF hospitalization, with the specified exclusion for hospitalizations during which renal replacement therapy occurs.
If a hazard ratio were reported, it would quantify the relative difference in the instantaneous rate of the first qualifying composite event between the randomized groups under the fitted survival model. It would not directly describe the absolute probability of experiencing the composite by a particular time.
A confidence interval around a hazard ratio would quantify statistical uncertainty around the estimated treatment effect under the model and sampling framework. A narrow interval would indicate greater precision than a wide interval, all else equal. The interval would not describe the range of treatment effects experienced by individual participants.
A p-value addresses the strength of evidence against a specified null hypothesis under the statistical model. It does not tell the reader how large or clinically important the treatment effect is. Effect estimates and confidence intervals are required to describe magnitude and precision.
9. Composite Endpoint Interpretation
The CHOIR primary endpoint illustrates a central issue in clinical-trial statistics: a single summary endpoint can simplify the primary hypothesis test while simultaneously obscuring differences among its component outcomes.
Mortality
All-cause mortality is one of the four qualifying components. It is intrinsically a different type of event from a hospitalization endpoint because death prevents subsequent first-event observation of other components.
Myocardial infarction
Myocardial infarction is included as an independent qualifying component of the composite primary outcome.
Stroke
Stroke is included as a separate qualifying component and therefore contributes directly to the composite event definition.
CHF hospitalization
CHF hospitalization is included, with the registry specifically excluding hospitalizations during which renal replacement therapy occurs.
Why component balance matters
Suppose a composite contains one very frequent component and three uncommon components. The overall treatment estimate will contain much more information about the frequent component than about the uncommon components. That does not make the composite invalid, but it changes how the result should be interpreted.
The same issue applies to clinical importance. A hospitalization and a fatal event are not interchangeable outcomes from a patient perspective simply because both satisfy the mathematical definition of the composite. Statistical combination does not erase their clinical differences.
10. Analysis Population and Randomization
The record identifies CHOIR as randomized, but the registry information summarized here does not provide a detailed analysis-population definition. In a randomized treatment trial, the primary efficacy analysis is commonly based on the intention-to-treat principle: participants are analyzed according to the group to which they were randomized.
| Principle | Statistical role |
|---|---|
| Randomized allocation | Creates the basis for a causal comparison between randomized groups. |
| Intention-to-treat | Preserves the treatment comparison defined by randomization when participants discontinue or deviate from assigned treatment. |
| Time-to-event follow-up | Uses information about both event occurrence and timing while accommodating censoring. |
| Component adjudication | Determines whether an observed clinical event qualifies for the composite endpoint. |
The strength of an intention-to-treat analysis is that it estimates the effect of assignment to a treatment strategy rather than the effect among only those who remain adherent. However, the exact estimand still depends on the protocol-defined treatment, follow-up, censoring, and endpoint rules.
11. Masking and Potential Sources of Bias
The registry records no masking. In an unmasked trial, participants and investigators are not blinded to treatment assignment. That feature can matter differently for different types of endpoints.
Objective events
Events such as mortality can be less susceptible to subjective assessment than patient-reported or investigator-assessed outcomes.
Hospitalization outcomes
Hospitalization-related endpoints can involve clinical decisions and classification processes, making endpoint definitions and adjudication procedures important.
Treatment behavior
Knowledge of treatment assignment can influence clinical management, follow-up, discontinuation, or other post-randomization behavior.
Interpretation
Randomization remains a major source of protection against confounding, but lack of masking can introduce other sources of bias depending on how outcomes and follow-up are determined.
12. Statistical Assumptions
Proportional hazards
If a Cox proportional-hazards model is used, the interpretation of a single hazard ratio relies on a proportional-hazards structure: the relative hazard between treatment groups is assumed to be reasonably stable over time.
When hazards are not proportional, a single hazard ratio can compress a changing treatment effect into one summary number. In that setting, time-specific survival estimates, restricted mean survival time, or other prespecified summaries may provide additional information.
Non-informative censoring
Survival methods generally assume that censoring is independent of the future event process, conditional on the information accounted for by the analysis. This does not mean that censored participants are identical to those remaining under observation; it means that the censoring mechanism must not systematically remove participants in a way that depends on their unobserved future event risk after accounting for relevant information.
Correct endpoint classification
A composite outcome also depends on accurate classification of each component. The exclusion of CHF hospitalizations during which renal replacement therapy occurs illustrates why endpoint definitions matter: changing the event definition can change which observations count as primary events.
13. Multiplicity and Component Analyses
The registry information summarized here identifies one registered primary endpoint: the composite of mortality, myocardial infarction, stroke, and CHF hospitalization with the stated exclusion. It does not provide a multiplicity strategy for additional hypothesis tests.
| Analysis | Statistical role | Interpretive issue |
|---|---|---|
| Primary composite | Prespecified primary outcome | Provides the principal treatment comparison. |
| Individual components | Component-level description | Components can differ in frequency, clinical importance, and treatment effect. |
| Multiple secondary comparisons | Additional hypothesis tests if performed | Repeated testing can increase the probability of false-positive findings unless appropriately controlled. |
A statistically significant composite result would not automatically establish a statistically significant effect on each component. Conversely, a nonsignificant component result would not necessarily contradict a significant composite result. The two analyses address different statistical questions and may have very different amounts of information.
14. Missing Data and Censoring
The registry information summarized here does not provide a detailed missing-data or censoring strategy. For a time-to-event endpoint, the relevant issue is usually not ordinary missingness alone but whether and when a participant's event status becomes unavailable.
Participants who have complete follow-up without an event contribute information through their final observation. Participants who experience a qualifying event contribute their event time. Participants whose follow-up ends before an event is observed may be censored.
15. No Posted Results
The ClinicalTrials.gov record contains no posted statistical analyses for the primary endpoint. The record therefore does not provide a posted hazard ratio, confidence interval, p-value, Kaplan-Meier estimate, event count, or component-specific treatment estimate for the composite outcome in the information summarized here.
Current statistical evidence in the registry
The registered primary endpoint is defined, but no statistical analysis results are posted in the ClinicalTrials.gov record.
This distinction is important because a registered endpoint defines the intended clinical outcome, whereas an estimated treatment effect requires observed event data and a specified statistical analysis. The two should not be treated as interchangeable.
16. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record summarized here does not contain formal statistical results for the primary endpoint.
- Comparator information: the intervention is identified as epoetin alfa, but the comparator intervention is not identified in the registry information summarized here.
- Limited endpoint timing information: the registry definition specifies the composite components but does not provide a separate numerical follow-up time frame in the information summarized here.
- No event counts: the registry information summarized here does not provide the number of participants experiencing the composite endpoint or its individual components.
- No effect estimates: hazard ratios, risk differences, confidence intervals, and p-values are not posted in the summarized registry analysis.
- No baseline table: detailed baseline characteristics by randomized group are not provided in the information summarized here.
- No component-specific results: the registry does not provide separate outcome estimates for mortality, myocardial infarction, stroke, or CHF hospitalization in the information summarized here.
- Unmasked design: the absence of masking can create opportunities for bias in treatment behavior, follow-up, and outcome assessment depending on the endpoint and trial procedures.
- Composite interpretation: even with a formal composite result, the contribution and clinical importance of each component would need to be considered separately.
17. Why This Trial Matters Statistically
CHOIR is a useful statistical teaching case because its registry design combines randomized allocation with an event-based composite outcome. That combination brings together several foundational concepts in clinical-trial statistics without requiring the reader to rely on a single numerical result.
| Concept | How it appears in CHOIR |
|---|---|
| Randomization | The registry classifies allocation as randomized. |
| Parallel design | The design model is recorded as parallel. |
| Unmasked treatment | The registry records no masking. |
| Time-to-event analysis | The composite clinical outcome is naturally suited to analysis of time to first qualifying event. |
| Composite endpoint | Four clinical outcomes are combined into one primary endpoint. |
| Censoring | Time-to-event methods require explicit rules for participants whose event status is not observed through the end of follow-up. |
| Hazard ratio | A Cox model would commonly summarize the relative instantaneous event rate between randomized groups. |
| Kaplan-Meier estimation | Provides an estimate of event-free probability over time. |
| Component analysis | The composite should be distinguished from its individual components when interpreting clinical meaning. |
| Multiplicity | Additional component or secondary analyses require separate consideration of type I error. |
18. Statistical Methods Explained in Context
Why does the randomized design matter if the endpoint is a composite?
The composite definition determines what counts as an event, but randomization determines how treatment groups are constructed. Randomization is what permits the primary comparison to be interpreted as a treatment-assignment comparison rather than merely an association between treatment exposure and outcome.
Why not simply compare the percentage of participants with an event?
A simple percentage ignores when events occurred and can be distorted by unequal follow-up. If one group experiences events earlier than another, a time-to-event method captures that temporal difference rather than reducing all observations to a single endpoint status.
Why might a Cox model be used?
A Cox model provides a compact estimate of the relative event rate while accommodating censored observations and allowing prespecified covariate adjustment or stratification. Its main summary measure is the hazard ratio.
Why is a composite not equivalent to four separate endpoints?
The composite asks one question about the time to the first qualifying event. Four separate endpoint analyses ask four different questions. Combining the components changes the estimand and can increase the information available for the primary analysis while making component-specific interpretation less direct.
What would make a composite result clinically difficult to interpret?
Interpretation becomes more difficult when component frequencies are very different, when component clinical importance differs substantially, or when treatment effects vary materially across components. In such situations, the overall composite can conceal important differences that become visible only after examining the individual components.
Why is the exclusion for renal replacement therapy important?
Because the registry explicitly excludes CHF hospitalizations during which renal replacement therapy occurs, that exclusion is part of the mathematical definition of the primary endpoint. It affects which observations qualify as events and therefore affects the estimand being analyzed.
19. From Registry Endpoint to Statistical Estimand
A useful way to understand a clinical-trial endpoint is to translate the registry wording into the statistical question it creates.
| Layer | CHOIR interpretation |
|---|---|
| Clinical outcome | Mortality, myocardial infarction, stroke, or qualifying CHF hospitalization. |
| Composite rule | Any qualifying component can define the primary event, with the stated CHF hospitalization exclusion. |
| Time component | Time until the first qualifying composite event. |
| Comparison | Difference between randomized groups. |
| Typical effect measure | Hazard ratio from a time-to-event model. |
| Uncertainty | Confidence interval around the treatment-effect estimate. |
| Hypothesis test | Typically a log-rank or model-based test under the prespecified statistical plan. |
This framework separates the clinical definition from the statistical method. The registry defines what constitutes the outcome; the analysis plan determines how the resulting event times are compared and what assumptions are required.
20. Trial Design in Statistical Context
Randomized
Randomization provides the fundamental basis for comparing outcome distributions between assigned treatment strategies.
Parallel
A parallel design means treatment groups are evaluated alongside one another rather than relying on within-participant crossover comparisons.
Unmasked
No masking is recorded, so treatment assignment was not concealed through a blinded design.
Phase 4
The registry classifies the study as phase 4, placing it within the post-approval clinical-development phase category used by the registry.
21. What a Complete Statistical Result Would Contain
For this type of primary endpoint, a complete statistical presentation would normally allow the reader to see both the treatment-effect estimate and the evidence supporting its precision.
| Result element | Purpose |
|---|---|
| Number randomized | Defines the randomized analysis population. |
| Number with primary events | Shows how much event information contributed to the comparison. |
| Kaplan-Meier estimates | Shows event-free probability over time. |
| Median time to event | Provides an interpretable summary when the event-free curve reaches 50%. |
| Hazard ratio | Summarizes the relative event rate under a Cox model. |
| Confidence interval | Describes statistical precision around the estimated treatment effect. |
| P-value | Quantifies evidence against the prespecified null hypothesis under the statistical framework. |
| Component outcomes | Shows whether the composite result is driven primarily by mortality, myocardial infarction, stroke, or CHF hospitalization. |
None of these numerical outcome estimates are posted in the ClinicalTrials.gov record summarized here. The absence of posted estimates is therefore an important characteristic of the currently available registry record, rather than a reason to substitute unrelated estimates from other sources.
22. Why This Endpoint Is Statistically Interesting
The CHOIR primary endpoint illustrates the tension between statistical efficiency and clinical interpretability. Combining several events can produce a single outcome with more total events than any individual component. That can improve the information available for a primary comparison. At the same time, the resulting treatment effect is an average over the components according to how often they occur and how the first-event rule operates.
This is particularly important when mortality and nonfatal events appear in the same composite. A participant can experience myocardial infarction and later die, but a time-to-first-event composite records the first qualifying event. The subsequent death is clinically important but does not create another primary composite event for that participant.
The statistical analysis therefore answers a precise question: how does randomized treatment assignment relate to the time until the first qualifying event under the prespecified composite definition? It does not answer every question about the individual components or the total burden of disease events.
23. Limitations of Interpreting an Unreported Result
Without a posted effect estimate, confidence interval, event count, and p-value, the statistical strength and magnitude of the randomized comparison cannot be characterized numerically from the registry record alone.
This distinction is especially important for educational interpretation. Knowing that a trial was randomized and had a prespecified composite endpoint tells the reader how the study was designed to generate evidence. It does not establish what that evidence ultimately showed.
24. Why This Trial Matters Statistically
CHOIR provides a compact example of how a randomized clinical trial moves from a clinical question to a formal statistical estimand. The intervention is epoetin alfa, the study population is recorded under anemia, and the primary outcome is a defined composite of major clinical events. The randomized parallel design creates the treatment comparison, while survival-analysis methods provide a natural framework for comparing the timing of the first qualifying event.
The most important statistical lesson is that an endpoint definition and an analysis method are separate pieces of a trial. The composite definition determines which observations count as events. Randomization determines the comparison. Kaplan-Meier estimation describes event-free survival over time. The log-rank test provides a comparison of survival distributions. A Cox model can summarize the relative event rate with a hazard ratio. Confidence intervals quantify precision, while p-values address the evidence against a null hypothesis.
Equally important, the composite endpoint must remain clinically interpretable. Mortality, myocardial infarction, stroke, and CHF hospitalization are not interchangeable outcomes merely because they can be represented within one statistical endpoint. A rigorous analysis therefore treats the primary composite as one prespecified estimand while recognizing the distinct clinical meaning of its components.
25. Sources
- ClinicalTrials.gov: NCT00211120 — CHOIR.
Continue through the Clinical Biostats knowledge graph
Explore clinical-trial design, survival analysis, composite endpoints, and the statistical principles used to interpret randomized clinical evidence.
26. Record Summary
CHOIR is recorded as a randomized, parallel, unmasked phase 4 treatment study with 1432 enrolled participants, beginning in 2002-03 and listed as terminated. The intervention is epoetin alfa, and the registered primary outcome is a composite consisting of all-cause mortality, myocardial infarction, stroke, and CHF hospitalization, with hospitalizations during which renal replacement therapy occurs excluded.
From a statistical perspective, the primary endpoint is naturally suited to time-to-first-event analysis. Kaplan-Meier methods can describe event-free survival, the log-rank test can compare randomized groups, and a Cox model can estimate a hazard ratio with a confidence interval. The interpretation must account for censoring, the proportional-hazards assumption when a Cox model is used, and the distinct clinical meaning of the composite components.