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
CALGB 9633 was a randomized, double-blind, parallel phase 3 treatment trial evaluating surgery with or without chemotherapy in patients with stage IB non-small cell lung cancer. The registry records overall survival and failure-free survival as the primary endpoints.
| Feature | CALGB 9633 |
|---|---|
| Phase | Phase 3 |
| Population | Patients with stage IB non-small cell lung cancer |
| Design | Randomized, double-blind, parallel |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Primary endpoints | Overall survival (OS) and failure-free survival |
| Enrollment | 500 |
| Number of arms | 2 |
| Lead sponsor | National Cancer Institute (NCI) |
| Sponsor type | NIH |
| ClinicalTrials.gov | NCT00002852 |
2. Clinical Question
The central question was whether adding chemotherapy to surgery affected overall survival and failure-free survival in patients with stage IB non-small cell lung cancer.
Population
The registry lists patients with stage IB non-small cell lung cancer, including adenocarcinoma, adenosquamous cell lung cancer, bronchoalveolar cell lung cancer, large cell lung cancer, and squamous cell lung cancer.
Intervention
The trial interventions listed in the registry include paclitaxel, carboplatin, and therapeutic conventional surgery.
Comparator
The study title describes surgery with chemotherapy versus surgery without chemotherapy.
Primary question
Does the addition of chemotherapy to surgical treatment change overall survival and failure-free survival?
3. Trial Design
Intervention structure
Surgery with chemotherapy
- Therapeutic conventional surgery
- Paclitaxel
- Carboplatin
Surgery without chemotherapy
- Therapeutic conventional surgery
- No chemotherapy described in the study title
The registry identifies the trial as a parallel randomized comparison. Because the registry information does not provide additional treatment-schedule details, the statistical interpretation is appropriately focused on the randomized treatment contrast and the registered time-to-event endpoints.
4. Trial Timeline
Trial initiation
The registry lists October 1996 as the study start date.
Primary completion
The registry lists July 2004 as the primary completion date.
Completed study
The ClinicalTrials.gov record lists CALGB 9633 as completed.
5. Endpoints
The registry specifies two primary endpoints. Both are time-to-event outcomes, but they represent different clinical event definitions.
| Endpoint | Registry time frame | Registry definition / analysis |
|---|---|---|
| Overall survival (OS) | Up to 2.4 years | OS curves will be calculated using the Kaplan-Meier life-table method. Comparison of the survival curve for the two treatment regimens will be performed using the log-rank test for censored data. |
| Failure-free survival | Time between randomization and disease relapse or death, assessed up to 2.4 years | Failure-free survival curves will be calculated using the Kaplan-Meier life-table method. |
Overall survival
Overall survival is the more direct of the two endpoints conceptually: the event is death. A patient who remains alive at the end of available follow-up does not have an observed death event at that time and is handled as censored in a conventional time-to-event analysis.
Failure-free survival
Failure-free survival incorporates either disease relapse or death as the event. This makes it a composite time-to-event endpoint. A patient can therefore experience the endpoint through relapse even if the patient remains alive.
6. Statistical Methodology
Kaplan-Meier estimation
The registry explicitly specifies the Kaplan-Meier life-table method for calculating OS curves and failure-free survival curves. Kaplan-Meier estimation is designed for time-to-event data in which some observations may be censored rather than followed until the event occurs.
Here, di represents the number of events occurring at time ti, while ni represents the number of patients at risk immediately before that event time.
The resulting curve estimates the probability of remaining event-free beyond a given time. For OS, that means remaining alive. For failure-free survival, it means remaining free of both relapse and death.
Log-rank comparison
For OS, the registry specifically states that survival curves for the two treatment regimens will be compared using the log-rank test for censored data.
The log-rank test compares the observed and expected numbers of events between treatment groups across the follow-up period. It is therefore a global comparison of the time-to-event experience rather than a comparison of only one selected time point.
The test uses information accumulated over event times while accounting for the changing number of patients at risk and censored observations.
Time origin
For failure-free survival, the registry explicitly defines the interval as the time between randomization and disease relapse or death. Randomization is therefore the stated starting point for this endpoint.
Censoring
Time-to-event analysis distinguishes an observed event from incomplete observation. If a patient is still being followed without the event at the point observation ends, that patient's follow-up can contribute information up to the censoring time without being counted as an event at that time.
This is one reason Kaplan-Meier analysis is preferable to treating every patient as though identical follow-up were available. It uses the timing of events and the available follow-up rather than requiring every patient to be observed through the full study period.
Typical complementary modeling
For randomized time-to-event trials, a Cox proportional-hazards model is commonly used when a relative hazard estimate is desired. Such a model estimates a hazard ratio while allowing different baseline hazard levels over time. The registry, however, specifically names Kaplan-Meier estimation and a log-rank comparison for OS; it does not list a Cox model in the registered endpoint descriptions.
7. Planned Analysis
No formal statistical analyses were posted to ClinicalTrials.gov. The registry does, however, specify the planned measurement and analysis framework for the two primary endpoints.
Overall survival
The planned OS analysis is based on Kaplan-Meier life-table estimation, with comparison of the survival curves for the two treatment regimens using the log-rank test for censored data. The registry specifies a time frame of up to 2.4 years.
Failure-free survival
The planned failure-free survival analysis uses Kaplan-Meier life-table curves. The endpoint is defined as the time between randomization and disease relapse or death, assessed up to 2.4 years.
| Endpoint | What is measured | Planned statistical approach |
|---|---|---|
| Overall survival | Time-to-death outcome | Kaplan-Meier life-table method; log-rank comparison of survival curves |
| Failure-free survival | Time from randomization to disease relapse or death | Kaplan-Meier life-table method |
What a future numerical analysis would require
A numerical survival analysis would ordinarily require the observed event and censoring information, or published estimates sufficiently detailed to support a valid secondary analysis. A single enrollment count is not enough to calculate a Kaplan-Meier curve, median survival, a hazard ratio, or a confidence interval.
Accordingly, the statistical value of the registry record here is primarily its prespecified analytical structure: randomized comparison, time-to-event endpoints, Kaplan-Meier estimation, and an explicitly stated log-rank comparison for OS.
8. Statistical Methods Explained
Why was Kaplan-Meier analysis used?
Kaplan-Meier estimation is suited to outcomes measured as time until an event when follow-up can end before the event occurs. In CALGB 9633, both primary endpoints are time-to-event outcomes, so the method allows patients with different amounts of follow-up to contribute appropriately to the estimated survival curves.
What does censoring mean in this trial?
Censoring means that the exact event time is not observed for a patient during the available observation period. For OS, for example, a patient can be known to be alive through the last observation even though a later death time is not available. The patient contributes information through that observed follow-up period.
Why is the log-rank test appropriate for OS?
The log-rank test is designed to compare two survival distributions while accounting for censored observations. Rather than asking only whether the proportion alive at one particular time differs, it evaluates the accumulated separation between the treatment-group survival experiences over the observed follow-up.
How is failure-free survival different from overall survival?
OS has death as its event. Failure-free survival has a broader event definition: disease relapse or death. Consequently, a patient can remain alive and still experience the failure-free survival endpoint if disease relapse occurs.
Why does randomization matter statistically?
Randomization establishes the treatment assignment before the outcome occurs and provides the foundation for comparing the treatment groups. Under proper randomization, systematic differences in measured and unmeasured baseline factors are expected to be balanced on average across repeated trials. The subsequent time-to-event comparison is therefore anchored to a randomized treatment contrast.
Why should OS and failure-free survival be interpreted separately?
The endpoints represent different event processes. Failure-free survival can occur earlier because relapse is an event in addition to death, while OS remains event-free until death. Reporting the two endpoints separately preserves the distinction between disease-control timing and survival timing.
9. Understanding the Kaplan-Meier Curve
A Kaplan-Meier curve begins with all patients who are available for analysis and decreases whenever an event occurs. The size of each downward step depends on the number of events and the number of patients still at risk at that time.
Vertical position
The height of the curve at a particular time represents the estimated probability of remaining free of the specified event beyond that time.
Downward steps
Each observed event produces a downward change in the estimated survival probability.
Censoring marks
Censored patients remain part of the risk set until their censoring time, after which they no longer contribute to subsequent risk sets.
Risk-set size
As follow-up progresses, the number of patients contributing information generally declines because of events and censoring.
For CALGB 9633, the OS curve would describe the probability of remaining alive over time, while the failure-free survival curve would describe the probability of remaining free from both relapse and death.
10. Interpreting a Log-Rank Test
The log-rank test evaluates the null hypothesis that the survival experience is the same between the two treatment groups. At each relevant event time, the method compares the events actually observed in each group with the events expected if the groups had the same underlying survival distribution.
The log-rank test addresses whether there is statistical evidence that the two time-to-event curves differ over the follow-up period. It is a hypothesis test about the survival distributions, not a measure of how large or clinically important a treatment effect is.
A log-rank test does not by itself provide an absolute difference in survival, a median survival time, or a hazard ratio. Those quantities require additional estimation from the underlying time-to-event data.
The meaning of a statistically compared curve depends on the endpoint. For OS, the event is death. For failure-free survival, the event is disease relapse or death. The same statistical machinery can therefore be applied to different endpoints while answering different clinical questions.
11. Analysis Population and Randomization
The registry identifies CALGB 9633 as a randomized phase 3 trial with 500 enrolled participants and 2 parallel arms. The ClinicalTrials.gov record does not provide separate arm-specific sample sizes or detailed definitions of efficacy analysis populations.
| Design element | Registry information |
|---|---|
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Enrollment | 500 |
| Number of arms | 2 |
| Primary purpose | Treatment |
Randomization and parallel assignment are particularly important for interpreting the primary comparison. Because treatment is assigned before outcome assessment, the difference in time-to-event experience between the randomized groups is the principal quantity of interest rather than a comparison constructed from patients selected after treatment has begun.
12. Population and Disease Definition
The registry lists several lung cancer histologies and identifies stage IB non-small cell lung cancer as a condition of interest.
| Condition listed in the registry |
|---|
| Adenocarcinoma of the Lung |
| Adenosquamous Cell Lung Cancer |
| Bronchoalveolar Cell Lung Cancer |
| Large Cell Lung Cancer |
| Squamous Cell Lung Cancer |
| Stage IB Non-small Cell Lung Cancer |
These disease categories provide the clinical context for the randomized treatment comparison. The statistical endpoint definitions then translate that clinical question into measurable time-to-event outcomes.
13. Why Two Time-to-Event Endpoints?
Using both OS and failure-free survival allows a trial to distinguish the timing of disease relapse from the ultimate survival outcome.
Statistically, failure-free survival can contain more events than OS because relapse is included in its event definition. This can provide a different view of the treatment course. Clinically, however, the two endpoints should not be conflated: preventing or delaying relapse is not identical to extending overall survival.
The registry's use of Kaplan-Meier estimation for both endpoints creates a consistent framework for displaying these different event processes while preserving their distinct definitions.
14. What a Hazard Ratio Would Mean
A hazard ratio is a common effect measure for time-to-event outcomes, although the CALGB 9633 registry entry does not report a hazard-ratio estimate.
A hazard ratio below 1 would indicate a lower estimated instantaneous event rate in the numerator group under the specified model; a value above 1 would indicate a higher estimated instantaneous event rate.
The hazard is not the same as the probability of an event over the entire follow-up period. Consequently, a hazard ratio should not be translated directly into a percentage of patients who survive or relapse.
A relative hazard measure summarizes the instantaneous event-rate relationship under a model. It does not tell us the absolute probability of survival at a particular time, the median survival time, or the experience of every individual patient.
For CALGB 9633, the registry specifies Kaplan-Meier curves and a log-rank comparison for OS but does not post a hazard-ratio result. Any numerical hazard ratio would therefore require information beyond the registry record.
15. Confidence Intervals and Statistical Precision
Confidence intervals are commonly used with estimated treatment effects to express statistical uncertainty. For survival analyses, they can be attached to quantities such as survival probabilities, median survival times, or model-based hazard ratios.
A confidence interval should not be interpreted as a range containing the true effect with a fixed probability after the data have been observed. Rather, it describes the performance of the statistical interval procedure under repeated sampling and the assumptions of the analysis.
Estimate
A point estimate summarizes the observed treatment contrast or survival quantity.
Confidence interval
The interval describes statistical uncertainty around an estimated quantity.
P-value
A p-value addresses evidence against a specified null hypothesis; it is not a measure of effect size.
Clinical meaning
Statistical precision and clinical importance are related but distinct questions.
No confidence intervals or p-values are posted for CALGB 9633 in the registry information considered here. The primary endpoint descriptions establish the intended statistical framework, but they do not provide numerical estimates from which precision can be quantified.
16. Statistical Interpretation of the Trial Design
| Statistical feature | Role in CALGB 9633 |
|---|---|
| Randomization | Creates the treatment-group comparison before outcome follow-up. |
| Parallel design | Patients remain assigned to one of 2 treatment arms for the randomized comparison. |
| Kaplan-Meier estimation | Estimates the OS and failure-free survival curves while accommodating censoring. |
| Log-rank test | Used by the registered plan to compare the OS curves for the two treatment regimens. |
| Time-to-event framework | Uses both event timing and available follow-up rather than relying only on a fixed-time proportion. |
| Composite endpoint | Failure-free survival treats disease relapse or death as the endpoint event. |
The combination of randomization and time-to-event methods is well aligned with the stated clinical question. Randomization establishes the comparison, while Kaplan-Meier estimation and the log-rank test account for the fact that patients may experience events at different times and may have different amounts of observed follow-up.
17. Important Limitations and Interpretation Issues
- No posted statistical results: the ClinicalTrials.gov record does not provide numerical estimates, confidence intervals, or p-values for the registered primary endpoints.
- No event-level information: the registry information does not provide the individual event and censoring times needed to reconstruct Kaplan-Meier curves.
- No arm-specific enrollment counts: the registry reports total enrollment of 500 and 2 arms but does not provide arm-specific enrollment in the information considered here.
- Failure-free survival is a composite endpoint: relapse and death are combined into one event definition, so the endpoint should not be interpreted as equivalent to either component alone.
- Time horizon: both registered primary endpoints are assessed within a timeframe extending up to 2.4 years, so interpretations should remain tied to that registered assessment period.
- Kaplan-Meier assumptions: standard Kaplan-Meier estimation relies on appropriate handling of censoring, including the usual assumption that censoring is sufficiently non-informative with respect to the event process conditional on relevant information.
- Log-rank interpretation: the log-rank test evaluates differences in survival distributions; it does not quantify the magnitude of a treatment effect by itself.
- Generalizability: conclusions from the randomized study population should be considered in the context of the eligibility characteristics defining patients with the listed stage and histologic conditions.
18. Censoring: The Key Statistical Feature of Survival Data
Censoring is central to the interpretation of CALGB 9633 because both primary endpoints are time-to-event outcomes. A patient who has not experienced the endpoint by the end of observation does not necessarily have an endpoint time of "infinity." Instead, the analysis records that the event-free state was known through the patient's last observed time.
Observed event
The endpoint occurs during observed follow-up, such as death for OS or relapse/death for failure-free survival.
Censored observation
The patient is known to be event-free through a particular observation time, but the event time is not observed during that follow-up.
Kaplan-Meier estimation uses this information efficiently. A censored patient contributes to the risk set while under observation and is then removed from later risk sets once censoring occurs.
This distinction is important because simply calculating the percentage of patients who experienced an event would discard the timing of the events and could mishandle patients with incomplete follow-up.
19. Why This Trial Matters Statistically
CALGB 9633 provides a compact example of how a randomized clinical question is translated into formal survival-analysis endpoints.
| Concept | How it appears in CALGB 9633 |
|---|---|
| Randomization | The trial uses randomized allocation between 2 treatment arms. |
| Parallel design | The registry identifies a parallel study design. |
| Time-to-event analysis | Both registered primary endpoints are time-to-event outcomes. |
| Kaplan-Meier estimation | Specified for OS and failure-free survival. |
| Log-rank testing | Specified for comparison of the OS curves. |
| Censoring | The OS analysis explicitly refers to censored data, and survival analysis generally requires appropriate treatment of incomplete follow-up. |
| Composite endpoint | Failure-free survival combines disease relapse or death into the event definition. |
| Clinical interpretation | OS and failure-free survival describe different aspects of the treatment course and should be interpreted separately. |
20. Statistical Story of CALGB 9633
The statistical structure of the study can be summarized as a sequence from treatment assignment to time-to-event estimation.
This sequence illustrates an important principle in clinical-trial statistics: the analysis method should follow the structure of the endpoint. Because the outcomes are defined by time until an event, the analysis preserves both whether the event occurred and when the event occurred.
21. What the Registry Establishes — and What It Does Not
| Question | Registry information |
|---|---|
| Was the study randomized? | Yes. |
| Was it a parallel trial? | Yes. |
| How many participants were enrolled? | 500. |
| How many arms were specified? | 2. |
| What were the primary endpoints? | Overall survival and failure-free survival. |
| How was OS planned to be analyzed? | Kaplan-Meier life-table method and log-rank comparison for censored data. |
| How was failure-free survival planned to be analyzed? | Kaplan-Meier life-table method. |
| What was the registered OS timeframe? | Up to 2.4 years. |
| What was the registered failure-free survival timeframe? | Time between randomization and disease relapse or death, assessed up to 2.4 years. |
| Are numerical primary endpoint results posted? | No formal statistical analyses are posted to ClinicalTrials.gov. |
This distinction is important when interpreting a registry-based statistical record. The registered design and endpoint methodology can be examined in detail even when numerical outcome analyses are not posted in the registry.
22. Sources
- ClinicalTrials.gov: CALGB 9633 (NCT00002852).
Continue exploring Clinical Biostats
Explore additional statistical tutorials, calculators, and clinical-trial analyses for deeper study of clinical research methods.
23. Record Summary
CALGB 9633 is a randomized phase 3 trial of surgery with or without chemotherapy in patients with stage IB non-small cell lung cancer. The registry records 500 enrolled participants, 2 parallel treatment arms, and a double-blind design. Its two primary endpoints are overall survival and failure-free survival, both assessed using Kaplan-Meier life-table methods. For overall survival, the registered analysis additionally specifies comparison of the two treatment-regimen survival curves using the log-rank test for censored data.
From a statistical perspective, the study illustrates the distinction between randomized treatment comparison, time-to-event estimation, censoring, and hypothesis testing. It also demonstrates why endpoint definitions matter: overall survival records death, whereas failure-free survival records the earlier of disease relapse or death.