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
JMDB was a randomized phase 3 trial comparing pemetrexed and cisplatin with gemcitabine and cisplatin in patients with locally advanced or metastatic non-small cell lung cancer. The registry lists overall survival as the primary endpoint.
| Feature | JMDB |
|---|---|
| Acronym | JMDB |
| ClinicalTrials.gov identifier | NCT00087711 |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Non Small Cell Lung Carcinoma |
| Enrollment | 1713 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Lead sponsor | Eli Lilly and Company |
| Sponsor type | Industry |
| Start date | 2004-07 |
| Primary completion date | 2007-01 |
2. Clinical Question
The central statistical question is whether the randomized comparison of the two treatment regimens can determine whether overall survival differs between patients receiving pemetrexed and cisplatin and those receiving gemcitabine and cisplatin.
Population
Patients with locally advanced or metastatic non-small cell lung cancer.
Intervention
Pemetrexed and cisplatin.
Comparator
Gemcitabine and cisplatin.
Primary question
Does overall survival differ between the randomized treatment groups?
3. Trial Design
JMDB was a phase 3, randomized, parallel-group trial with two treatment arms. The registry identifies the primary purpose as treatment and the masking status as none.
Pemetrexed + cisplatin
- Pemetrexed
- Cisplatin
Gemcitabine + cisplatin
- Gemcitabine
- Cisplatin
4. Trial Timeline
Trial start
The registry lists 2004-07 as the study start date.
Primary completion
The registry lists 2007-01 as the primary completion date.
Study status
The ClinicalTrials.gov record lists the study status as completed.
5. Primary Endpoint
| Endpoint | Registry definition | Statistical structure |
|---|---|---|
| Overall survival | Baseline to date of death from any cause | Time-to-event endpoint |
Overall survival is defined in the registry as the time from baseline to the date of death from any cause. This definition has two important statistical features: the endpoint is measured over time, and patients who have not died by the end of their available follow-up are generally represented as right-censored observations in a conventional survival analysis.
6. Statistical Methodology
The registry identifies overall survival as the primary endpoint but does not post statistical analyses for the endpoint. For an overall-survival endpoint of this type, a conventional analysis would use time-to-event methods rather than treating survival time as an ordinary continuous measurement.
Kaplan-Meier estimation
A Kaplan-Meier estimator is commonly used to describe the distribution of overall survival over time. It accommodates right censoring and produces an estimated survival function, which can be displayed as a survival curve.
Here, di represents the number of deaths at an event time and ni represents the number of patients at risk immediately before that time.
Log-rank comparison
For a randomized two-arm survival comparison, a log-rank test is a standard method for testing whether the survival experience differs between treatment groups across follow-up. It uses the ordering of event times and accounts for censoring.
Cox proportional-hazards model
A Cox proportional-hazards model is another standard approach for estimating a relative treatment effect for an overall-survival endpoint. The primary effect measure is commonly expressed as a hazard ratio.
An HR below 1 would indicate a lower estimated instantaneous rate of death in the numerator group under the fitted model; an HR above 1 would indicate a higher estimated instantaneous rate.
Confidence intervals
A confidence interval around a hazard ratio quantifies statistical uncertainty about the estimated relative treatment effect under the model and sampling framework. A narrow interval indicates greater statistical precision than a wide interval, all else being equal.
Statistical testing
A formal hypothesis test for overall survival would typically compare the treatment groups under a prespecified null hypothesis. The p-value would quantify how compatible the observed data are with that null hypothesis under the specified statistical model or test; it would not itself quantify the magnitude or clinical importance of an effect.
7. Planned Analysis
The registry says that overall survival will be measured from baseline to the date of death from any cause. Because this is a time-to-event endpoint, the natural analysis framework is survival analysis, typically including Kaplan-Meier estimation and a comparison of survival between randomized groups using an appropriate time-to-event test.
What would be estimated?
A survival function for each randomized treatment group could describe the probability of remaining alive through successive points in follow-up.
What would be compared?
The randomized treatment groups could be compared using a time-to-event test such as the log-rank test, with a hazard ratio providing a relative effect estimate from a Cox model.
How would censoring enter?
Patients who remain alive at the end of their available follow-up would contribute information until their censoring time rather than being treated as deaths.
What does the registry report?
The registry identifies the primary endpoint and its time frame, but no formal statistical analyses are posted to ClinicalTrials.gov.
8. Statistical Methods Explained
Why is overall survival a time-to-event endpoint?
Overall survival is not simply a yes-or-no outcome because the timing of death matters. Two patients can both experience the event while having very different survival times. A time-to-event framework preserves this timing information and also accommodates patients whose follow-up ends before death.
Why is Kaplan-Meier useful for this trial?
The Kaplan-Meier method is designed for survival data in which some observations may be censored. Instead of assigning a single average survival time to every patient, it estimates survival probabilities across follow-up while accounting for the number of patients still at risk at each event time.
What would a hazard ratio add?
A hazard ratio summarizes the relative instantaneous event rate between treatment groups under a proportional-hazards model. It provides a compact measure of treatment effect, but it is not an absolute survival probability and should not be interpreted as the percentage of patients who benefit.
Why is a p-value not enough?
A p-value addresses evidence against a specified null hypothesis. It does not describe the size of the treatment effect, the precision of that effect, or its clinical importance. A complete survival analysis therefore normally considers the effect estimate together with its confidence interval and the underlying survival curves.
What does censoring mean in overall survival?
If a patient has not experienced death by the last time they are known to be alive, the observation is generally censored at that point. Censoring means that the exact event time is not observed within the available follow-up; it does not mean that the patient is treated as having experienced the event.
What assumption matters for a Cox hazard ratio?
The conventional Cox proportional-hazards interpretation assumes that the relative hazard is reasonably stable over time. If the treatment hazards change substantially relative to one another during follow-up, a single hazard ratio can become a less complete description of the treatment effect, making the survival curves and time-specific measures especially important.
9. Interpreting the Statistical Design
10. Why Randomization Matters Statistically
The value of randomization is not simply that it produces two numerical groups. Its principal statistical role is to make treatment assignment independent of baseline prognostic factors in expectation, allowing differences in subsequent outcomes to be attributed to treatment assignment more directly than in an uncontrolled comparison.
With 1713 enrolled patients and two randomized treatment arms, the JMDB design provides a large randomized framework for evaluating the prespecified overall-survival endpoint. The absence of masking is a separate design characteristic and should be considered when evaluating endpoints that involve judgment or treatment-dependent behavior.
11. What an Overall-Survival Analysis Would Show
| Statistical output | What it tells the reader |
|---|---|
| Kaplan-Meier curve | How estimated survival changes over follow-up in each randomized treatment group. |
| Median overall survival | The time by which the estimated survival probability reaches 0.50, when that point is observed within follow-up. |
| Hazard ratio | The estimated relative instantaneous event rate between treatment groups under a Cox model. |
| 95% confidence interval | The statistical uncertainty surrounding the estimated hazard ratio. |
| Log-rank p-value | The evidence against the null hypothesis of equivalent survival experience under the specified test. |
| Number of deaths | The number of observed primary endpoint events contributing to the analysis. |
These outputs answer different questions. A hazard ratio is relative; a survival probability is absolute at a specified time; a median is a time-based summary; and a p-value concerns statistical evidence against a null hypothesis. A rigorous interpretation should not substitute one for another.
12. Limitations
- No formal statistical analyses in the registry: the ClinicalTrials.gov record identifies overall survival as the primary endpoint but does not post statistical analyses for that endpoint.
- Limited registry detail: the record represented here does not provide the full statistical analysis plan, detailed censoring rules, stratification factors, model specifications, or missing-data procedures.
- No posted effect estimate: without a posted hazard ratio, confidence interval, or p-value in the registry, the treatment effect cannot be characterized numerically from this record alone.
- Open-label design: masking is listed as none, so knowledge of treatment assignment is a design consideration for outcomes susceptible to assessment or behavioral influence.
- Endpoint-specific interpretation: overall survival is an objective endpoint, but its statistical analysis still depends on follow-up completeness, censoring, event timing, and prespecified analytic choices.
- Analysis-plan uncertainty: the registry information available here does not specify whether a particular survival model, stratification scheme, interim procedure, or multiplicity adjustment was used.
13. What Is Known and What Is Not Reported in the Registry
| Element | Registry record |
|---|---|
| Trial acronym | JMDB |
| NCT identifier | NCT00087711 |
| Phase | 3 |
| Enrollment | 1713 |
| Primary endpoint | Overall survival |
| Endpoint time frame | Baseline to date of death from any cause |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Posted statistical analyses | None |
| Primary endpoint hazard ratio | Not reported in the ClinicalTrials.gov statistical-analysis record |
| Primary endpoint confidence interval | Not reported in the ClinicalTrials.gov statistical-analysis record |
| Primary endpoint p-value | Not reported in the ClinicalTrials.gov statistical-analysis record |
This distinction is important statistically. Trial registration establishes the prespecified research question and design framework, whereas an analysis result requires observed event and follow-up information together with a defined statistical procedure.
14. Why This Trial Matters Statistically
JMDB is a useful teaching example because it illustrates the basic structure of a large randomized phase 3 time-to-event trial without requiring the statistical interpretation to be reduced to a single number.
| Concept | How it appears in JMDB |
|---|---|
| Randomization | The allocation is randomized. |
| Phase 3 evaluation | The trial is listed as phase 3. |
| Parallel-group design | The design model is parallel. |
| Two-arm comparison | The registry lists 2 arms comparing two treatment regimens. |
| Time-to-event endpoint | Overall survival is measured from baseline to death from any cause. |
| Kaplan-Meier estimation | A natural descriptive method for the registered overall-survival endpoint. |
| Hazard ratio | A standard relative effect measure for comparing survival distributions with a Cox model. |
| Censoring | Relevant to survival analysis when patients remain alive at the end of observed follow-up. |
| Open-label design | The registry lists masking as none. |
| Large randomized population | Enrollment is 1713. |
15. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registered primary endpoint is a time-to-event measure. Its analysis would ordinarily focus on survival distributions, treatment-group comparisons, relative hazard, confidence intervals, and the handling of censoring.
Clinical interpretation
The clinical question concerns whether patients receiving the two treatment regimens experience different overall survival from baseline to death from any cause. The registry alone does not provide the numerical treatment effect.
Keeping these two levels separate prevents a common statistical error: treating a trial design or endpoint definition as if it were evidence of an observed treatment effect. The design establishes how the comparison is intended to be made; the observed data determine the resulting estimate.
16. Why the Absence of a Posted Estimate Matters
A hazard ratio is meaningful only in relation to the observed event history and the statistical model used to estimate it. The same randomized design can produce a hazard ratio below 1, close to 1, or above 1 depending on the observed outcomes.
Similarly, statistical significance cannot be determined from sample size alone. The number 1713 describes enrollment, but a p-value depends on the observed treatment-group outcomes, event information, censoring, and the prespecified statistical test.
Core statistical principle
A randomized phase 3 design defines a strong framework for inference, but the treatment effect must still be estimated from the observed survival data.
17. Clinical Biostats Interpretation
Overall survival measures time from baseline until death from any cause. It therefore incorporates both whether a patient experiences the event and when the event occurs.
If a Cox model produced an HR below 1 for pemetrexed plus cisplatin relative to gemcitabine plus cisplatin, that would indicate a lower estimated instantaneous rate of death in the pemetrexed-plus-cisplatin group under the model. It would not mean that the same percentage reduction applied to every individual patient.
A confidence interval would show the precision of the estimated treatment effect. A narrow interval would indicate greater precision than a wide interval, while the location of the interval relative to the null value of 1 would inform conventional hypothesis-test interpretation for a hazard ratio.
A p-value would address the compatibility of the observed evidence with a null hypothesis. It would not tell the reader how large the treatment effect is or whether an effect is clinically important. Those questions require the effect estimate and its uncertainty.
18. Sources
- ClinicalTrials.gov: NCT00087711 — JMDB.
Continue through Clinical Biostats
Explore additional clinical trial analyses, biostatistics tutorials, and statistical calculators.
19. Record Summary
JMDB is a randomized phase 3, parallel-group trial with 1713 enrolled patients comparing pemetrexed and cisplatin with gemcitabine and cisplatin in patients with locally advanced or metastatic non-small cell lung cancer. The registry identifies overall survival as the primary endpoint, defined from baseline to the date of death from any cause.
From a statistical perspective, this endpoint naturally calls for time-to-event methodology, with Kaplan-Meier estimation, treatment-group comparison, and a relative effect measure such as a Cox-model hazard ratio forming the core analytical framework. The ClinicalTrials.gov record represented here does not post formal statistical analyses, so the record establishes the trial's design and primary statistical question rather than a numerical estimate of the treatment effect.