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
START was a randomized, quadruple-masked, phase 3 parallel-group trial in oncology evaluating tecemotide (L-BLP25) with single low-dose cyclophosphamide against saline plus placebo in participants with non-small cell lung cancer. The registry reports overall survival as the primary endpoint and provides additional time-to-event analyses for symptom progression and disease progression.
| Feature | START |
|---|---|
| Phase | Phase 3 |
| Condition | Non-small Cell Lung Cancer |
| Brief title | Cancer Vaccine Study for Unresectable Stage III Non-small Cell Lung Cancer (START) |
| Design | Randomized, parallel-group |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 1513 |
| Lead sponsor | EMD Serono |
| Study period | January 2007 to April 2012 primary completion |
| ClinicalTrials.gov | NCT00409188 |
2. Clinical Question
The primary statistical question was whether treatment with tecemotide (L-BLP25) plus single low-dose cyclophosphamide changed overall survival compared with saline plus placebo over a follow-up period of up to 66 months.
Population
Participants enrolled in the phase 3 START study for non-small cell lung cancer, with the registry identifying the brief study title as a cancer vaccine study for unresectable stage III disease.
Intervention
Tecemotide (L-BLP25), classified in the registry as a biological intervention, together with single low-dose cyclophosphamide.
Comparator
Saline plus placebo.
Primary question
Does tecemotide plus cyclophosphamide alter overall survival relative to saline plus placebo?
3. Trial Design
Tecemotide + cyclophosphamide
- Tecemotide (L-BLP25)
- Single low-dose cyclophosphamide
- Biological and drug interventions as registered
Saline + placebo
- Saline
- Placebo
- Drug interventions as registered
The combination of randomization and quadruple masking is important statistically. Randomization establishes the basis for comparing outcomes according to assigned treatment, while masking is intended to reduce the influence of treatment knowledge on participant, investigator, outcome-assessor, and related trial processes. The registry identifies the trial as parallel rather than crossover or factorial.
4. Analysis Population and Clinical Hold
The primary analysis used a modified intention-to-treat population. The registry describes this population as being based on the ITT population but prospectively excluding 274 participants who were randomized during the 6 months from 22 September 2009 to 22 March 2010 before the effective date of the clinical hold.
The same primary analysis set is described for the secondary time-to-event analyses. For the time-to-symptom-progression outcome, the registry additionally notes that N signifies the number of participants evaluable for that outcome measure.
5. Endpoints
| Endpoint | Role | Time frame | Registry definition |
|---|---|---|---|
| Overall Survival | Primary | Up to 66 months | Overall survival time was defined as the time from randomization to death. Participants without events were censored at the last date they were known to be alive or the clinical cut-off date, whatever was earlier. |
| Time To Symptom Progression (TTSP) as Measured by the Lung Cancer Symptom Scale (LCSS) | Secondary | Up to 66 months | Time-to-event outcome measured using the Lung Cancer Symptom Scale. |
| Time To Progression (TTP) | Secondary | Up to 66 months | Time-to-event outcome measuring time to progression. |
All three posted analyses are time-to-event comparisons. This is important because a time-to-event endpoint uses both the occurrence and timing of events, while also allowing participants who have not experienced an event by the analysis cutoff to contribute censored follow-up.
6. Statistical Methodology
Cox proportional-hazards model
The registry identifies regression using a Cox model as the statistical method for the primary overall-survival analysis and for the two secondary time-to-event analyses. The statistical method is a Cox proportional-hazards model, with the hazard ratio as the effect measure.
For a two-group comparison, exponentiating the treatment coefficient gives a hazard ratio. A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the intervention group relative to the comparator under the fitted model.
Hazard ratio
The hazard ratio is a relative time-to-event measure. It compares the estimated event hazard between the two randomized groups over the analyzed follow-up. It is not a probability, and it is not equivalent to the percentage of participants who experience an event.
An HR below 1 indicates a lower estimated instantaneous event rate for the intervention group. For example, an HR of 0.893 corresponds to an estimated hazard that is approximately 89.3% of the comparator hazard, or approximately a 10.7% lower estimated hazard.
Censoring
For overall survival, participants without an event were censored at the last date they were known to be alive or the clinical cut-off date, whichever was earlier. Censoring allows incomplete follow-up to contribute information without treating an unknown future outcome as though it had already occurred.
Superiority framework
The registry identifies the hypothesis type for all three posted analyses as superiority. The intervention is therefore evaluated by estimating whether its time-to-event outcome differs favorably from the comparator under the prespecified superiority framework.
7. Primary Result: Overall Survival
The primary endpoint was overall survival over a time frame of up to 66 months. The primary analysis compared tecemotide plus cyclophosphamide with saline plus placebo using a Cox proportional-hazards model in the modified ITT analysis population.
Hazard ratio for overall survival
95% CI: 0.763–1.044 · P = 0.1566
Two-sided superiority analysis; primary analysis set was the modified ITT population.
| Primary endpoint | Tecemotide + cyclophosphamide vs saline + placebo |
|---|---|
| Outcome | Overall Survival |
| Time frame | Up to 66 months |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.893 |
| 95% CI | 0.763–1.044 |
| P-value | 0.1566 |
| Hypothesis | Superiority |
The estimated hazard ratio of 0.893 means that, under the fitted Cox model, the estimated instantaneous hazard of death in the tecemotide-plus-cyclophosphamide group was approximately 89.3% of the corresponding hazard in the saline-plus-placebo group. Expressed as a relative model-based quantity, this corresponds to an approximately 10.7% lower estimated hazard.
The hazard ratio does not mean that 10.7% fewer participants died, that survival increased by 10.7%, or that every participant experienced a 10.7% reduction in risk. A hazard ratio is a relative time-to-event measure rather than an absolute survival probability or risk difference.
The 95% confidence interval, 0.763 to 1.044, quantifies uncertainty around the estimated hazard ratio under the model and sampling framework. Because the interval extends across 1, the data are compatible with both a lower and a higher hazard relative to the comparator within the interval.
The P-value of 0.1566 is a measure of the evidence against the specified null hypothesis under the statistical model and testing framework. It is not a measure of the magnitude or clinical importance of the estimated effect. The estimate and confidence interval are therefore essential for understanding the size and precision of the treatment comparison.
The analysis also depends on the Cox model's proportional-hazards assumption. A single hazard ratio is most straightforward to interpret when the relative hazards are reasonably represented by a common ratio over time. The registry summary does not provide a separate assessment of that assumption.
8. Secondary Result: Time To Symptom Progression
The first secondary posted analysis evaluated Time To Symptom Progression (TTSP) as Measured by the Lung Cancer Symptom Scale (LCSS), also over up to 66 months. The same modified ITT population definition was used, with the registry noting that the number of evaluable participants is represented by N for this outcome.
Hazard ratio for time to symptom progression
95% CI: 0.732–0.977 · P = 0.0226
Two-sided superiority analysis using a Cox proportional-hazards model.
| Secondary endpoint | Result |
|---|---|
| Outcome | Time To Symptom Progression (TTSP) as Measured by the Lung Cancer Symptom Scale (LCSS) |
| Time frame | Up to 66 months |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.845 |
| 95% CI | 0.732–0.977 |
| P-value | 0.0226 |
| Hypothesis | Superiority |
The hazard ratio of 0.845 corresponds to an estimated instantaneous hazard of symptom progression approximately 84.5% of the comparator hazard under the Cox model, or approximately a 15.5% lower estimated hazard.
The 95% confidence interval of 0.732 to 0.977 is relatively close to 1 at its upper boundary. It therefore indicates a more limited range of compatible relative effects than would be conveyed by the point estimate alone. Because the entire interval is below 1, the interval is consistent with a lower estimated hazard for symptom progression under this analysis.
The P-value of 0.0226 indicates the strength of evidence against the null hypothesis in this superiority analysis; it does not quantify how large or clinically important the treatment effect is. The hazard ratio and confidence interval provide the effect-size information.
This result concerns time to symptom progression, not overall survival. A statistical finding for TTSP should therefore not be interpreted as evidence that the treatment necessarily changes mortality to the same degree. The two endpoints represent different clinical events.
9. Secondary Result: Time To Progression
The second secondary posted analysis evaluated Time To Progression (TTP) over up to 66 months using the same Cox proportional-hazards framework and modified ITT analysis population.
Hazard ratio for time to progression
95% CI: 0.752–1.002 · P = 0.0528
Two-sided superiority analysis using a Cox proportional-hazards model.
| Secondary endpoint | Result |
|---|---|
| Outcome | Time To Progression (TTP) |
| Time frame | Up to 66 months |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.868 |
| 95% CI | 0.752–1.002 |
| P-value | 0.0528 |
| Hypothesis | Superiority |
The estimated hazard ratio of 0.868 corresponds to an estimated instantaneous progression hazard approximately 86.8% of the comparator hazard, or approximately a 13.2% lower estimated hazard under the fitted model.
The confidence interval, 0.752 to 1.002, is particularly informative because its upper boundary is very close to 1 and extends slightly above it. The interval therefore includes the null value of 1, so the point estimate should not be interpreted without its uncertainty interval.
The P-value of 0.0528 is close to 0.05, but a P-value should not be converted into a graded measure of effect size. The estimated hazard ratio, its confidence interval, and the endpoint definition together provide the substantive statistical information.
TTP is also distinct from TTSP and overall survival. A participant can therefore have a progression-related event without that event being equivalent to symptom progression or death.
10. Summary of Posted Statistical Results
| Endpoint | Role | HR | 95% CI | P-value |
|---|---|---|---|---|
| Overall Survival | Primary | 0.893 | 0.763–1.044 | 0.1566 |
| Time To Symptom Progression | Secondary | 0.845 | 0.732–0.977 | 0.0226 |
| Time To Progression | Secondary | 0.868 | 0.752–1.002 | 0.0528 |
The pattern across endpoints illustrates why clinical-trial interpretation should not collapse several outcomes into a single P-value. The three estimates are all below 1, but their confidence intervals differ in width and in whether they include 1. Each result also concerns a different event definition.
11. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
All three posted analyses were time-to-event outcomes. A Cox model is well suited to this setting because it compares event hazards while allowing participants to contribute follow-up until an event or censoring. It produces a hazard ratio that summarizes the relative event hazard between the treatment groups under the proportional-hazards model.
What does an HR of 0.893 mean?
An HR of 0.893 means that the estimated instantaneous hazard in the intervention group is approximately 89.3% of the comparator hazard under the fitted model. It can therefore be described as an approximately 10.7% lower estimated hazard. It does not mean a 10.7 percentage-point improvement in survival and does not mean that every participant has a 10.7% lower individual risk.
Why is the confidence interval more informative than the point estimate alone?
A point estimate gives one value, but every estimated treatment effect is subject to statistical uncertainty. The 95% confidence interval communicates the range of values compatible with the model and sampling process at that confidence level. For overall survival, the interval is 0.763–1.044; for TTSP, it is 0.732–0.977; and for TTP, it is 0.752–1.002.
Why does the P-value not measure treatment effect size?
The P-value addresses evidence against a null hypothesis under the statistical framework. It depends on the magnitude of the observed effect, its precision, and the amount of information available. A small P-value does not tell us that an effect is large, while a larger P-value does not prove that the treatment effect is exactly zero.
What does modified intention-to-treat mean here?
The registry defines the primary analysis set as an ITT population from which 274 participants randomized during the 6 months from 22 September 2009 to 22 March 2010 before the effective clinical hold were prospectively excluded. The resulting population therefore differs from an analysis of all 1513 enrolled participants.
Why does censoring matter?
For overall survival, participants without an event were censored at the last date they were known to be alive or the clinical cut-off date, whichever was earlier. Censoring means that the exact event time is not observed within the available follow-up, but the participant still contributes information through the time at which follow-up ends.
12. Overall Survival: A Closer Statistical Reading
The primary endpoint is the most important place to distinguish effect size, uncertainty, and hypothesis testing.
Effect size
The estimated hazard ratio was 0.893. This is the model-based relative comparison between the two treatment groups.
Precision
The 95% confidence interval was 0.763–1.044. Its width describes uncertainty around the estimate.
Hypothesis test
The two-sided P-value was 0.1566 under the superiority analysis.
Analysis population
The analysis used the registry-defined modified ITT population rather than all 1513 enrolled participants.
These four elements answer different statistical questions. The hazard ratio asks how large the estimated relative difference is; the confidence interval addresses precision; the P-value describes evidence against the null hypothesis; and the analysis population defines the participants represented by the estimate.
13. Comparing the Three Time-to-Event Analyses
The three posted analyses provide an instructive example of how different event definitions can produce different statistical summaries within the same randomized comparison.
| Question | Endpoint | Interpretive focus |
|---|---|---|
| How long until death? | Overall Survival | Mortality over time |
| How long until symptom progression? | TTSP measured by LCSS | Patient symptom progression over time |
| How long until progression? | TTP | Disease progression over time |
The endpoint definitions should remain distinct even when the same statistical model is used. A Cox model does not make different endpoints interchangeable; it is the event definition that determines what the estimated hazard ratio describes.
14. Confidence Intervals and the Null Value
For hazard ratios, the null value is 1. An interval entirely below 1 is compatible with a lower event hazard for the intervention under the model. An interval crossing 1 is compatible with no difference as well as with effects in either direction within the interval.
The TTSP interval of 0.732–0.977 lies below 1. The overall-survival interval of 0.763–1.044 crosses 1, as does the TTP interval of 0.752–1.002. This distinction is statistical rather than a measure of clinical importance by itself.
15. Multiplicity and Interpretation Across Endpoints
The registry identifies one primary endpoint and two secondary endpoint analyses. Because several outcomes are evaluated, the collection of P-values should not automatically be treated as though each were an isolated hypothesis test with no relationship to the others.
| Endpoint | Role | Why the role matters |
|---|---|---|
| Overall Survival | Primary | The principal prespecified efficacy question represented in the registry. |
| Time To Symptom Progression | Secondary | Provides information on symptom progression in addition to the primary endpoint. |
| Time To Progression | Secondary | Provides disease-progression information distinct from overall survival and symptom progression. |
A secondary endpoint with a P-value below a conventional threshold should therefore be interpreted in the context of its secondary role, the overall endpoint structure, and any prespecified multiplicity strategy. The registry information summarized here identifies the hypothesis type and posted analysis but does not provide a multiplicity adjustment procedure in the statistical analysis fields.
16. Censoring and the Meaning of "Up to 66 Months"
The phrase "up to 66 months" defines the registry time frame for the three posted outcomes. It does not mean that every participant necessarily contributed 66 months of observed follow-up.
For overall survival, a participant who remained alive without an event was censored at the earlier of the last date known to be alive or the clinical cut-off date. Thus, the statistical analysis uses varying amounts of observed follow-up across participants rather than requiring every participant to have the same observation time.
Participants with observed events contribute event times. Participants without observed events contribute follow-up until their censoring time.
This structure is one reason ordinary methods based only on fixed-time proportions are not interchangeable with survival-analysis methods.
17. Safety
The registry reports serious adverse events by treatment arm. The reported counts are:
| Treatment group | Serious adverse events affected / at risk |
|---|---|
| Tecemotide (L-BLP25) + Cyclophosphamide | 303 / 1024 |
| Saline + Placebo | 151 / 477 |
The denominators in the registry's serious-adverse-event reporting are 1024 and 477, respectively. These are not the same as the total trial enrollment of 1513 and should therefore be preserved when interpreting the safety summary.
18. Trial Timeline
Study start
The START trial began in January 2007.
Clinical-hold analysis window
The modified ITT definition prospectively excluded 274 participants randomized during this 6-month period before the effective date of the clinical hold.
Primary completion
The registry lists April 2012 as the primary completion date.
Registered outcome follow-up
Overall survival, time to symptom progression, and time to progression were evaluated over time frames extending up to 66 months.
19. What the Hazard Ratio Does — and Does Not — Mean
The overall-survival HR of 0.893 means that the estimated instantaneous hazard of death under the Cox model was approximately 89.3% of the comparator hazard. In relative terms, this is approximately a 10.7% lower estimated hazard.
It does not mean that 10.7% of participants were saved, that survival increased by 10.7 percentage points, or that every participant experienced the same proportional reduction in risk.
The 95% CI of 0.763–1.044 describes uncertainty around the estimated hazard ratio. It is not a prediction interval for individual patients and does not mean that individual treatment effects must fall within that range.
The P-value of 0.1566 quantifies evidence against the null hypothesis within the specified statistical framework. It does not measure the magnitude, clinical relevance, or probability that the treatment is effective.
20. Interpreting the Secondary Endpoints Without Overreading Them
The TTSP and TTP results demonstrate why the interpretation of a clinical trial should preserve the distinction between endpoint, effect estimate, uncertainty, and statistical test.
TTSP
HR 0.845 with 95% CI 0.732–0.977 and P = 0.0226. The estimated hazard is lower in the intervention group under the model, with the confidence interval entirely below 1.
TTP
HR 0.868 with 95% CI 0.752–1.002 and P = 0.0528. The point estimate is below 1, but the confidence interval extends slightly above 1.
These two analyses cannot be treated as duplicates. TTSP is based on symptom progression measured by the LCSS, whereas TTP concerns progression. The clinical meaning of the event differs even though both analyses use Cox regression.
21. Important Limitations and Interpretation Issues
- Modified ITT population: the primary analysis excludes 274 participants who were randomized during the specified 6-month period before the effective date of the clinical hold. The estimate therefore applies to the defined modified ITT population.
- Endpoint distinctions: overall survival, symptom progression, and disease progression are different clinical events and should not be treated as interchangeable outcomes.
- Hazard-ratio assumptions: the Cox model relies on the proportional-hazards framework. The registry summary does not provide a separate assessment of that assumption.
- Confidence intervals: point estimates should not be interpreted independently of their uncertainty intervals, particularly for TTP where the upper confidence limit is 1.002.
- Multiplicity: multiple endpoint analyses create a broader inferential context than a single isolated hypothesis test. The registry data summarized here do not specify a multiplicity adjustment procedure for the posted analyses.
- Safety denominators: serious adverse-event counts are reported using arm-specific denominators of 1024 and 477, which differ from the overall enrollment of 1513.
- Follow-up: the registered time frame is up to 66 months, but participants can have different amounts of observed follow-up because of events and censoring.
22. Why This Trial Matters Statistically
START is a useful teaching case because its registry results bring together several core concepts in survival analysis: randomized treatment assignment, masking, a modified intention-to-treat population, right-censored time-to-event outcomes, Cox proportional-hazards modeling, hazard ratios, confidence intervals, P-values, and multiple clinically distinct endpoints.
| Concept | How it appears in START |
|---|---|
| Randomization | The trial used randomized allocation in a parallel-group design. |
| Blinding | The registry identifies quadruple masking. |
| Intention-to-treat analysis | The primary analysis set was based on ITT but prospectively excluded 274 participants under the modified ITT definition. |
| Time-to-event endpoints | Overall survival, TTSP, and TTP were all analyzed as time-to-event outcomes. |
| Cox model | All three posted statistical analyses used Cox regression. |
| Hazard ratio | The treatment effect was summarized with hazard ratios. |
| Confidence interval | 95% two-sided confidence intervals were reported for all three hazard ratios. |
| P-value | Two-sided P-values were posted for the three analyses. |
| Clinical endpoint hierarchy | Overall survival was primary; TTSP and TTP were secondary. |
| Censoring | Overall survival censored participants at the earlier of last known alive date or clinical cut-off date when no event had occurred. |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT00409188.
- PubMed: PMID 24331154.
- PubMed: PMID 24976972.
- PubMed: PMID 34870327.
- PubMed: PMID 25722382.
Continue through Clinical Biostats
Connect this trial's time-to-event endpoints and statistical methods with deeper biostatistics tutorials and statistical tools.
26. Record Summary
START provides a clear example of a randomized oncology trial in which the principal efficacy question was evaluated through time-to-event methodology. The primary endpoint, overall survival, was analyzed with a Cox proportional-hazards model in a modified ITT population, producing an HR of 0.893 with a 95% CI of 0.763–1.044 and a P-value of 0.1566. Two secondary time-to-event analyses used the same modeling framework: TTSP had an HR of 0.845 with a 95% CI of 0.732–0.977 and P = 0.0226, while TTP had an HR of 0.868 with a 95% CI of 0.752–1.002 and P = 0.0528.
The statistical interpretation is strongest when these estimates are considered together with their confidence intervals, endpoint definitions, analysis population, censoring rules, and the distinction between primary and secondary outcomes. The trial also illustrates an important practical lesson in clinical biostatistics: a hazard ratio is not a probability, a P-value is not an effect size, and different time-to-event endpoints answer different clinical questions even when they are analyzed with the same statistical model.