This page separates reported trial results from statistical interpretation. Numerical results are taken only from the ClinicalTrials.gov record. This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. View the ClinicalTrials.gov record.
1. Trial at a Glance
EGF104900 was a completed randomized phase 3 parallel-group trial evaluating lapatinib in combination with trastuzumab versus lapatinib monotherapy in subjects with HER2-positive metastatic breast cancer. The registered primary endpoint was progression-free survival (PFS), a time-to-event endpoint analyzed using a log-rank test with a hazard ratio as the effect measure.
| Feature | EGF104900 |
|---|---|
| Trial name | EGF104900 |
| ClinicalTrials.gov identifier | NCT00320385 |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Oncology |
| Condition | Neoplasms, Breast |
| Population | Subjects with HER2-positive metastatic breast cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 296 |
| Interventions | Lapatinib; trastuzumab |
| Lead sponsor | GlaxoSmithKline |
| Sponsor type | Industry |
| Start | 2005-11 |
| Primary completion | 2007-06 |
2. Clinical Question
The central statistical question was whether lapatinib in combination with trastuzumab produced a different time-to-event outcome from lapatinib monotherapy, with progression-free survival registered as the primary endpoint and superiority specified as the hypothesis type.
Population
Subjects with HER2-positive metastatic breast cancer, represented in the registry under the condition category “Neoplasms, Breast.”
Intervention
Lapatinib in combination with trastuzumab.
Comparator
Lapatinib monotherapy.
Primary question
Does the combination improve progression-free survival relative to lapatinib alone under the trial's superiority framework?
3. Trial Design
Trastuzumab + Lapatinib
- Lapatinib
- Trastuzumab
- Combination treatment
Lapatinib
- Lapatinib
- Monotherapy comparator
The enrollment count is 296, while the serious-adverse-event denominator reported in the ClinicalTrials.gov record is 149 for the combination group and 146 for the lapatinib group. Those denominators should not be silently treated as the randomized sample sizes because the ClinicalTrials.gov record identifies them specifically as the numbers at risk for the serious-adverse-event summaries.
4. Primary Endpoint
| Endpoint | Registered definition / time frame | Statistical analysis |
|---|---|---|
| Progression-Free Survival (PFS) | Baseline to disease progression or death due to any cause or 30 days after last dose (up to 216 weeks). PFS was defined as the time from randomization until the first documented sign of disease progression or death due to any cause. | Log-rank test; hazard ratio estimated using the Pike estimator based on the log-rank test. |
The primary endpoint is therefore not a simple binary outcome such as “progressed” versus “did not progress.” It incorporates when progression or death occurs. Participants who remain free of progression and death during their available observation period can contribute censored follow-up rather than being treated as if they experienced the event.
5. Primary Result: Progression-Free Survival
The registry reports a formal superiority analysis of PFS in the Intent-to-Treat population. The comparison was between trastuzumab plus lapatinib and lapatinib alone, using a two-sided log-rank test and the Pike estimator of the treatment hazard ratio.
Hazard ratio for progression or death
95% CI: 0.57–0.93 · P = 0.008
Analysis population: Intent-to-Treat population; all randomized participants irrespective of whether or not they actually received study treatment, with the registry noting that only participants with progesterone receptor status were considered.
| Primary PFS result | Reported value |
|---|---|
| Comparison | Trastuzumab + Lapatinib vs Lapatinib |
| Endpoint | Progression-Free Survival (PFS) |
| Method | Log-rank test |
| Effect measure | Hazard Ratio |
| Estimate | 0.73 |
| 95% confidence interval | 0.57–0.93 |
| Confidence interval | Two-sided |
| P-value | 0.008 |
| Hypothesis type | Superiority |
The reported hazard ratio of 0.73 means that the estimated hazard of a PFS event was about 73% as high in the trastuzumab-plus-lapatinib group as in the lapatinib group under the reported time-to-event analysis. Expressed as a relative difference, an HR of 0.73 corresponds to an estimated 27% lower hazard in the combination group.
That interpretation is about the estimated event hazard, not about individual patients. An HR of 0.73 does not mean that 27% of participants avoided progression, that every participant experienced a 27% reduction in risk, or that the absolute probability of progression or death was 27% lower.
The 95% confidence interval of 0.57–0.93 describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It is narrower than a statement about the range of outcomes that individual patients might experience. The interval remains below 1, which is consistent with the reported two-sided P-value of 0.008 under the superiority analysis.
The P-value does not measure the size of the treatment effect. It addresses the statistical evidence against the relevant null hypothesis under the specified test. The HR of 0.73 provides the effect estimate, while the confidence interval communicates its precision.
Because PFS is time-to-event data, the analysis also depends on censoring and the structure of event times. The ClinicalTrials.gov record identifies the log-rank test and Pike hazard-ratio estimator, but do not provide enough information here to independently assess the proportional-hazards assumption or reconstruct a Kaplan-Meier curve. The HR should therefore be interpreted as the reported summary of the survival comparison rather than as a claim that the relative hazard was identical at every time point.
6. Secondary Endpoint Results
Overall Survival
Overall survival was a secondary time-to-event endpoint. The registry defines its time frame as baseline to death or 30 days after last dose for the last participant, up to 216 weeks. The reported analysis used a log-rank test and a hazard ratio.
Hazard ratio for overall survival
95% CI: 0.53–1.07 · P = 0.106
| Overall survival result | Reported value |
|---|---|
| Comparison | Trastuzumab + Lapatinib vs Lapatinib |
| Time frame | Baseline to death or 30 days after last dose for the last participant (up to 216 weeks) |
| Analysis population | ITT Population; only participants with progesterone receptor status were considered for evaluation |
| Method | Log-rank test |
| Effect measure | Hazard Ratio |
| Estimate | 0.75 |
| 95% confidence interval | 0.53–1.07 |
| Confidence interval | Two-sided |
| P-value | 0.106 |
| Hypothesis type | Superiority |
The reported OS hazard ratio of 0.75 estimates a lower hazard of death in the trastuzumab-plus-lapatinib group relative to lapatinib alone: approximately 75% of the estimated hazard in the comparator group under the reported model, or approximately a 25% lower estimated hazard.
The confidence interval of 0.53–1.07 is important because it crosses 1.00. It therefore includes values compatible with a lower hazard as well as values slightly above the comparator hazard. The interval is also substantially wider than the PFS interval, reflecting greater uncertainty in this reported estimate.
The P-value of 0.106 should not be translated into a probability that the treatment is ineffective, nor should it be used to conclude that the two treatments have exactly equal survival. A P-value is a measure of evidence under a statistical testing framework, not a measure of treatment magnitude or clinical importance.
The distinction between the PFS and OS results is particularly useful statistically. The PFS analysis has an HR below 1 with a 95% CI of 0.57–0.93 and P = 0.008, whereas the OS analysis has an HR of 0.75 with a 95% CI of 0.53–1.07 and P = 0.106. These are different endpoints with different event definitions and different amounts of statistical information. One result should not be substituted for the other.
As with PFS, the HR is a relative time-to-event measure rather than an absolute survival difference. The ClinicalTrials.gov record does not provide median OS, survival probabilities at specified time points, or event counts, so those quantities should not be inferred from the reported HR.
Overall Tumor Response
Overall Tumor Response (OR) was evaluated as a binary endpoint. Its registered time frame was baseline to disease progression or death or discontinuation from study or 30 days after last dose, up to 216 weeks. The reported comparison used stratified Fisher's exact tests and an odds ratio.
Odds ratio for overall tumor response
95% CI: 0.6–3.9 · P = 0.460
| Overall tumor response result | Reported value |
|---|---|
| Comparison | Trastuzumab + Lapatinib vs Lapatinib |
| Endpoint type | Binary |
| Outcome unit | Percentage of participants |
| Analysis population | ITT Population; only participants with progesterone receptor status were considered for evaluation |
| Method | Stratified Fisher exact test |
| Effect measure | Odds Ratio |
| Estimate | 1.5 |
| 95% confidence interval | 0.6–3.9 |
| Confidence interval | Two-sided |
| P-value | 0.460 |
| Hypothesis type | Superiority |
An odds ratio of 1.5 means that the estimated odds of the reported overall tumor response were 1.5 times the odds in the lapatinib group, based on the reported stratified analysis. It is an odds comparison, not a risk ratio and not a difference in response percentages.
The distinction matters because odds and probabilities are related but are not interchangeable. Without the underlying response counts or arm-specific response percentages in the ClinicalTrials.gov record, the OR cannot be converted here into an absolute response difference without introducing information that was not provided.
The 95% CI of 0.6–3.9 is relatively broad and includes 1.00. Thus, the estimate of 1.5 should be understood as uncertain rather than as a precise description of the magnitude of the response difference.
The P-value of 0.460 does not mean that there is a 46.0% probability that the treatment effect is zero. It is the reported result of the specified statistical test. The effect estimate and its confidence interval are needed to describe the direction and precision of the observed comparison.
The use of a stratified Fisher exact test is also informative: the registry explicitly identifies stratified analysis as part of the analysis text. Fisher's exact testing is particularly suited to categorical comparisons when exact inference is desired, while stratification allows the comparison to account for specified grouping factors rather than collapsing all observations into a single unstratified table.
Clinical Benefit Response
Clinical Benefit Response (CBR) was another binary secondary endpoint. Its time frame was baseline to disease progression or death or discontinuation from study or 30 days after last dose, up to 216 weeks. The analysis population was the ITT population, and the registry reports a stratified Fisher exact test with an odds ratio.
Odds ratio for clinical benefit response
95% CI: 1.2–4.5 · P = 0.010
| Clinical benefit response result | Reported value |
|---|---|
| Comparison | Trastuzumab + Lapatinib vs Lapatinib |
| Endpoint type | Binary |
| Outcome unit | Percentage of participants |
| Analysis population | ITT Population |
| Method | Stratified Fisher exact test |
| Effect measure | Odds Ratio |
| Estimate | 2.2 |
| 95% confidence interval | 1.2–4.5 |
| Confidence interval | Two-sided |
| P-value | 0.010 |
| Hypothesis type | Superiority |
The CBR odds ratio of 2.2 means that the estimated odds of clinical benefit response were 2.2 times those in the lapatinib group in the reported stratified comparison. This is an odds-based relative measure; it does not mean that 2.2 times as many participants necessarily experienced clinical benefit.
The 95% CI of 1.2–4.5 quantifies uncertainty around the OR and does not include 1.00. The lower endpoint of 1.2 indicates that the estimated effect remains above the null value within the reported confidence interval, although the interval is broad enough that the precise magnitude of the effect remains uncertain.
The P-value of 0.010 indicates the result of the reported statistical test under its superiority framework. It should not be interpreted as a 1.0% probability that the null hypothesis is true, and it does not establish the size of the clinical benefit. The OR and confidence interval provide the effect-size information.
CBR is also conceptually distinct from PFS. A binary clinical-benefit endpoint asks whether a prespecified benefit classification was achieved, whereas PFS uses the timing of progression or death. Consequently, the two endpoints can provide complementary information without being expected to produce identical statistical estimates.
7. Secondary Results in Context
The four posted statistical analyses illustrate two different branches of clinical-trial statistics. PFS and OS are time-to-event endpoints, for which the registry reports log-rank tests and hazard ratios. Overall tumor response and clinical benefit response are binary endpoints, for which the registry reports stratified Fisher exact tests and odds ratios.
| Endpoint | Type | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|---|
| PFS | Time-to-event | Log-rank | Hazard ratio | 0.73 | 0.57–0.93 | 0.008 |
| OS | Time-to-event | Log-rank | Hazard ratio | 0.75 | 0.53–1.07 | 0.106 |
| Overall Tumor Response | Binary | Stratified Fisher exact | Odds ratio | 1.5 | 0.6–3.9 | 0.460 |
| Clinical Benefit Response | Binary | Stratified Fisher exact | Odds ratio | 2.2 | 1.2–4.5 | 0.010 |
Reading this table correctly requires resisting the temptation to compare the numerical estimates directly. An HR of 0.73 and an OR of 2.2 are not quantities on the same scale. The HR describes a relative event hazard over time, while the OR compares the odds of a binary outcome. Their numerical values cannot be ranked as if they represented the same quantity.
Time-to-event evidence
PFS and OS incorporate the timing of events and censoring. The log-rank test compares survival experience across the follow-up period, while the HR summarizes the relative event hazard.
Binary-outcome evidence
Overall tumor response and clinical benefit response reduce the outcome to a binary classification. Fisher's exact test and an odds ratio address that categorical comparison.
Precision differs by endpoint
The confidence intervals communicate how much uncertainty surrounds each estimate. The registry-reported intervals vary considerably across endpoints.
Statistical tests are endpoint-specific
A P-value from a log-rank test and a P-value from a Fisher exact test answer questions within different statistical frameworks and should not be interpreted as interchangeable quantities.
8. Statistical Methodology
Intention-to-Treat Analysis
The primary PFS analysis was reported in the Intent-to-Treat (ITT) Population. The registry defines this population as all randomized participants irrespective of whether or not they actually received study treatment, with the additional statement that only participants with progesterone receptor status were considered.
Using randomized assignment as the basis of the efficacy comparison preserves the treatment groups created by randomization. In a randomized trial, that principle is important because treatment assignment is established before subsequent treatment exposure, discontinuation, progression, and other post-randomization events occur.
The ITT framework preserves the comparison generated by randomization. It does not mean that every participant necessarily receives or completes the assigned treatment.
Log-Rank Test
The registry reports the log-rank test for PFS and OS. This is a standard method for comparing time-to-event distributions between randomized groups. Rather than comparing only one selected time point, the log-rank framework uses information from event times across follow-up, subject to the treatment of censored observations.
For this trial, the primary PFS analysis compared trastuzumab plus lapatinib with lapatinib monotherapy. The OS analysis used the same broad time-to-event comparison framework as a secondary endpoint.
The log-rank test evaluates the evidence against the null hypothesis using the observed pattern of events and individuals at risk over time. It is not a direct estimate of the magnitude of treatment effect; that role is filled here by the reported hazard ratio.
Pike Hazard-Ratio Estimator
For both PFS and OS, the registry states that the Pike estimator of the treatment hazard ratio based on the log-rank test was provided together with a 95% confidence interval.
A hazard ratio is a relative time-to-event measure. An HR below 1 indicates a lower estimated event hazard in the first-listed treatment group relative to the comparator, whereas an HR above 1 indicates a higher estimated hazard.
The HR is not a probability, risk ratio, absolute risk difference, or percentage of patients benefiting. It summarizes relative event rates within a time-to-event framework.
Fisher Exact Test
Overall tumor response and clinical benefit response were analyzed using stratified Fisher exact tests. Fisher's exact test is designed for categorical data and calculates exact probabilities based on the observed contingency-table configuration rather than relying on a large-sample approximation.
The registry specifically states that responses were compared between treatment arms using stratified Fisher exact tests and identifies both intention-to-treat analysis and stratified analysis as concepts in the analysis text. That means the categorical analyses should be understood as stratified comparisons rather than as simple unadjusted two-group proportions.
Odds Ratio
The effect measure for both binary endpoints was the odds ratio. For two groups, an OR compares the odds of the specified binary outcome rather than the probability of that outcome itself.
OR = 1 is the conventional null value. OR > 1 indicates higher estimated odds in the first-listed group, while OR < 1 indicates lower estimated odds.
Confidence Intervals
Every posted statistical analysis includes a 95% two-sided confidence interval. Confidence intervals are especially valuable because they place the point estimate in the context of statistical uncertainty.
For PFS, the reported HR of 0.73 is accompanied by a 95% CI of 0.57–0.93. For OS, the HR of 0.75 has a 95% CI of 0.53–1.07. For overall tumor response, the OR is 1.5 with a 95% CI of 0.6–3.9, while CBR has an OR of 2.2 with a 95% CI of 1.2–4.5.
The intervals should not be read as a range containing 95% of individual treatment effects. They describe uncertainty in the estimated parameter under the statistical model and sampling framework used to construct the interval.
9. Statistical Methods Explained
Why was a log-rank test used for PFS?
PFS is explicitly registered as a time-to-event endpoint. The important information is not only whether a participant eventually progresses or dies, but also when that event occurs and whether a participant remains event-free through the available follow-up. The log-rank test is designed for comparing time-to-event distributions while incorporating the available follow-up from censored observations.
What does an HR of 0.73 mean?
It means the estimated hazard of a PFS event in the trastuzumab-plus-lapatinib group was approximately 73% of the estimated hazard in the lapatinib group under the reported analysis. Equivalently, it corresponds to an estimated 27% lower hazard. It does not mean that 27% of participants were protected from progression or death, nor does it describe an absolute probability difference.
Why is the OS HR of 0.75 not interpreted as a 25% survival improvement?
Because a hazard ratio is not a survival probability. It is a relative measure of event hazard over time. A 25% lower estimated hazard does not translate automatically into a 25-percentage-point improvement in survival, a 25% increase in median survival, or any other particular absolute benefit.
Why use an odds ratio for overall tumor response?
Overall tumor response is registered as a binary endpoint, so the analysis asks whether the odds of response differ between treatment groups. An odds ratio provides a relative comparison of those odds. It should not be substituted for a risk ratio or absolute response-rate difference unless the necessary underlying event counts are available.
Why does the CBR confidence interval matter as much as the point estimate?
The OR of 2.2 is only one estimate. Its 95% CI of 1.2–4.5 shows that the statistical uncertainty around the estimate is substantial. A point estimate alone can give a false impression of precision; the interval shows the range of parameter values compatible with the stated confidence procedure.
What does a P-value of 0.008 actually tell us?
For the primary PFS comparison, 0.008 is the reported two-sided P-value from the log-rank analysis. It quantifies the statistical evidence under the specified hypothesis-testing framework. It does not give the probability that the treatment works, the probability that the null hypothesis is true, or the magnitude of the treatment effect. The HR and its confidence interval answer the effect-size and precision questions.
Why are PFS and response not interchangeable?
PFS is a time-to-event endpoint that incorporates the timing of progression or death and censoring. Response is binary. A participant can contribute to a response analysis through a yes/no classification, whereas PFS retains information about when the event occurs. The endpoints therefore measure different aspects of treatment experience.
10. Primary and Secondary Endpoint Architecture
| Endpoint | Role | Type | Time frame | Analysis |
|---|---|---|---|---|
| Progression-Free Survival (PFS) | Primary | Time-to-event | Baseline to disease progression or death due to any cause or 30 days after last dose (up to 216 weeks) | Log-rank; Pike hazard ratio |
| Overall Survival (OS) | Secondary | Time-to-event | Baseline to death or 30 days after last dose for the last participant (up to 216 weeks) | Log-rank; Pike hazard ratio |
| Overall Tumor Response (OR) | Secondary | Binary | Baseline to disease progression or death or discontinuation from study or 30 days after last dose (up to 216 weeks) | Stratified Fisher exact; odds ratio |
| Clinical Benefit Response (CBR) | Secondary | Binary | Baseline to disease progression or death or discontinuation from study or 30 days after last dose (up to 216 weeks) | Stratified Fisher exact; odds ratio |
This architecture illustrates why clinical-trial statistical analysis is usually endpoint-specific. There is no single universal test that is appropriate for every outcome. The outcome's structure determines the statistical framework: time-to-event endpoints call for survival methods, while binary endpoints can be handled with categorical-data methods.
11. Randomization and Stratified Analysis
Randomization is the central design feature that creates the treatment-group comparison. In EGF104900, allocation was randomized and the design was parallel, meaning participants were assigned to one of the two treatment groups rather than being assigned to a sequence of treatments.
The registry also identifies stratified analysis for the overall tumor response and clinical benefit response analyses. The ClinicalTrials.gov record does not specify the individual stratification factors, so no particular clinical variable should be inserted into the analysis description.
Randomization
Random allocation creates the basis for comparing outcomes between treatment assignments without choosing treatment based on observed baseline characteristics.
Stratification
A stratified categorical analysis evaluates treatment differences while accounting for specified strata. The ClinicalTrials.gov record does not identify those strata individually.
It is important not to overstate what stratification accomplishes. Stratification does not transform an observational comparison into a randomized one; rather, here it is an analysis feature applied within an already randomized trial. Likewise, because the individual strata are not reported, the statistical interpretation should remain at the level supported by the registry record.
12. Kaplan-Meier Estimation and Time-to-Event Interpretation
Although the ClinicalTrials.gov record identifies the log-rank test and hazard ratio rather than providing a Kaplan-Meier estimate table, PFS and OS are inherently time-to-event outcomes. Kaplan-Meier estimation is the standard descriptive framework for displaying survival probabilities over follow-up.
Here, di represents events at a particular event time and ni represents participants at risk immediately before that time.
The advantage of the Kaplan-Meier approach is that participants do not have to experience the event during the observation window to contribute information. A participant who remains event-free through the last available observation can be censored and still contributes follow-up before censoring.
The ClinicalTrials.gov record does not provide the individual event and censoring times needed to reconstruct a Kaplan-Meier curve. It would therefore be inappropriate to fabricate survival curves from the reported hazard ratios alone.
13. Confidence Intervals, Null Values, and Effect Size
The four reported analyses provide a useful comparison of how confidence intervals work across different effect measures.
| Endpoint | Effect measure | Estimate | 95% CI | Null value |
|---|---|---|---|---|
| PFS | Hazard ratio | 0.73 | 0.57–0.93 | 1.00 |
| OS | Hazard ratio | 0.75 | 0.53–1.07 | 1.00 |
| Overall Tumor Response | Odds ratio | 1.5 | 0.6–3.9 | 1.00 |
| Clinical Benefit Response | Odds ratio | 2.2 | 1.2–4.5 | 1.00 |
For both hazard ratios and odds ratios, the null value is 1.00 because the comparison is multiplicative. An HR of 1 indicates equal estimated hazards, while an OR of 1 indicates equal estimated odds.
The confidence intervals also demonstrate why an effect estimate should never be presented without its uncertainty. The OS HR of 0.75 and the OR for overall tumor response of 1.5 are point estimates that sit in directions suggesting a difference between groups, but their confidence intervals communicate substantially different levels of uncertainty. The CBR interval of 1.2–4.5 and the PFS interval of 0.57–0.93 do not cross 1.00, whereas the OS and overall tumor response intervals do.
That distinction is statistical, not a statement about whether an individual patient benefited. A confidence interval is a property of the estimated parameter and sampling process; it is not an individual prognosis.
14. Understanding the Four Reported Estimates
The estimated hazard of progression or death was lower in the combination group under the reported analysis. The 95% CI of 0.57–0.93 and P = 0.008 accompany that estimate. No median PFS or absolute PFS rate is reported here, so those measures should not be inferred.
The estimated hazard of death was lower in the combination group, with a 95% CI of 0.53–1.07 and P = 0.106. The interval crosses 1.00, indicating uncertainty that includes the null value under the reported confidence procedure.
The estimated odds of overall tumor response were higher in the combination group, with an OR of 1.5. The 95% CI of 0.6–3.9 is broad and includes the null value of 1.00.
The estimated odds of clinical benefit response were 2.2 times those in the lapatinib group. The 95% CI of 1.2–4.5 lies above 1.00, while the reported P-value is 0.010.
15. Safety Results
The ClinicalTrials.gov record reports serious adverse events by arm as the number affected divided by the number at risk. These data should be presented as reported rather than converted into an inferred comparative safety statistic.
| Safety measure | Trastuzumab + Lapatinib | Lapatinib |
|---|---|---|
| Serious adverse events | 40/149 | 24/146 |
The denominators are part of the reported safety information and matter for interpretation. A count such as 40 cannot be compared directly with a count such as 24 without considering the corresponding numbers at risk. At the same time, the ClinicalTrials.gov record does not provide a formal statistical comparison, confidence interval, or P-value for serious adverse events, so this page does not construct one.
Safety and efficacy are also different statistical questions. PFS and OS evaluate time-to-event efficacy outcomes, whereas serious adverse events are adverse outcomes summarized through safety surveillance. A treatment comparison therefore should not be reduced to a single numerical balance between efficacy and safety without a prespecified framework for doing so.
16. Analysis Populations
| Endpoint | Analysis population specified in the registry data |
|---|---|
| Progression-Free Survival | Intent-to-Treat (ITT) Population: all randomized participants irrespective of whether or not they actually received study treatment. Only participants with progesterone receptor status were considered. |
| Overall Survival | ITT Population. Only participants with progesterone receptor status were considered for evaluation. |
| Overall Tumor Response | ITT Population. Only participants with progesterone receptor status were considered for evaluation. |
| Clinical Benefit Response | ITT Population. |
The difference in the registry wording for CBR is important. The ClinicalTrials.gov record explicitly specify the ITT population for CBR without the additional progesterone-receptor-status qualification that appears in the PFS, OS, and overall tumor response records. Those distinctions should be preserved rather than harmonized by assumption.
17. What the P-Values Do — and Do Not — Tell Us
The P-values in EGF104900 range from 0.008 to 0.460 across the four reported analyses. They should be interpreted in the context of their corresponding endpoint, statistical method, effect measure, and hypothesis framework.
| Endpoint | P-value | What it accompanies |
|---|---|---|
| PFS | 0.008 | Two-sided log-rank test with HR 0.73 |
| OS | 0.106 | Two-sided log-rank test with HR 0.75 |
| Overall Tumor Response | 0.460 | Two-sided stratified Fisher exact test with OR 1.5 |
| Clinical Benefit Response | 0.010 | Two-sided stratified Fisher exact test with OR 2.2 |
A common statistical error is to treat the P-value as if it were an effect-size measure. It is not. For example, the PFS P-value of 0.008 does not mean the combination reduces risk by 99.2%, and the CBR P-value of 0.010 does not mean there is a 1.0% chance that the observed clinical-benefit effect is due to chance.
Likewise, the P-value of 0.106 for OS does not establish that there is no treatment difference. It indicates the result of the specified test under its statistical assumptions and hypothesis. The confidence interval, 0.53–1.07, adds the more informative description of the uncertainty surrounding the HR estimate.
18. Limitations and Interpretation Issues
- Registry-level reporting: The analysis is based exclusively on the ClinicalTrials.gov record. Detailed protocol and statistical-analysis-plan information beyond the reported methods is not introduced here.
- Hazard-ratio interpretation: A hazard ratio is a relative time-to-event measure and should not be translated directly into an absolute risk difference or a percentage of patients benefiting.
- Proportional-hazards considerations: The ClinicalTrials.gov record reports hazard ratios but do not provide sufficient information here to independently assess whether proportional hazards held over the complete follow-up period.
- Censoring: PFS and OS analyses depend on how follow-up and censoring are handled. The ClinicalTrials.gov record provides the endpoint time frames but do not provide individual event and censoring histories.
- Analysis populations: The registry uses ITT populations for the posted efficacy analyses, but the exact wording differs across endpoints, including the additional progesterone-receptor-status qualification for PFS, OS, and overall tumor response.
- Binary versus time-to-event outcomes: OR and CBR cannot be interpreted on the same scale as PFS and OS. Odds ratios and hazard ratios answer different statistical questions.
- Stratification details: The ClinicalTrials.gov record identifies stratified analysis for the two binary response endpoints but do not identify the individual strata. No specific stratification variables should therefore be inferred.
- Multiplicity: Four statistical analyses are posted, including one primary endpoint analysis and three secondary endpoint analyses. The ClinicalTrials.gov record does not specify an alpha-allocation or multiplicity-adjustment strategy across these endpoints, so the reported P-values should not be assigned an unstated multiplicity framework.
- Safety comparison: Serious adverse events are reported by affected participants and denominators, but no formal comparative safety test is reported.
- Incomplete absolute outcome information: The statistical analyses posted on ClinicalTrials.gov contain effect estimates, confidence intervals, and P-values but do not provide median PFS, median OS, event counts, or arm-specific response percentages.
19. Multiplicity and Endpoint Hierarchy
The registry data identify one primary endpoint and three posted secondary statistical analyses. The primary endpoint is PFS; OS, overall tumor response, and clinical benefit response are secondary endpoints.
| Endpoint | Role | Posted analysis | Reported effect |
|---|---|---|---|
| Progression-Free Survival | Primary | Log-rank | HR 0.73 (95% CI 0.57–0.93), P = 0.008 |
| Overall Survival | Secondary | Log-rank | HR 0.75 (95% CI 0.53–1.07), P = 0.106 |
| Overall Tumor Response | Secondary | Stratified Fisher exact | OR 1.5 (95% CI 0.6–3.9), P = 0.460 |
| Clinical Benefit Response | Secondary | Stratified Fisher exact | OR 2.2 (95% CI 1.2–4.5), P = 0.010 |
The distinction between primary and secondary endpoints is essential. A primary endpoint is the outcome around which the main confirmatory question is structured. Secondary endpoints provide additional information but do not automatically carry the same inferential role.
The ClinicalTrials.gov record does not report a multiplicity-adjustment procedure or an alpha hierarchy among the four posted analyses. Consequently, this page does not impose one after the fact. In particular, the P-values should be read as the values reported for the individual analyses rather than as if a particular familywise-error procedure had been documented in the ClinicalTrials.gov record.
20. Non-Inferiority, Crossover, Factorial, and Bayesian Methods
The trial data specify a superiority hypothesis type. No non-inferiority margin is reported, so non-inferiority logic is not applied to the reported results.
The design model is parallel. The ClinicalTrials.gov record does not identify a crossover or factorial design, so neither is used in the analysis framework presented here.
The statistical methods are Fisher exact test and log-rank test. No Bayesian method is identified in the ClinicalTrials.gov record, so the reported results are presented using the frequentist methods documented by the registry.
Superiority
The trial's stated hypothesis type is superiority. The PFS analysis therefore evaluates evidence for a treatment difference rather than testing whether one treatment is merely not unacceptably worse.
No non-inferiority margin
No non-inferiority margin is provided in the ClinicalTrials.gov record, so no margin-based conclusion can be constructed.
Parallel design
Participants were assigned to one of two parallel treatment groups. The ClinicalTrials.gov record does not describe a crossover or factorial structure.
Frequentist methods
The posted analyses identify log-rank and Fisher exact methods, with hazard ratios and odds ratios as effect measures. No Bayesian analysis is reported in the ClinicalTrials.gov record.
21. Why This Trial Matters Statistically
EGF104900 is a useful teaching case because its four reported analyses span two of the most common statistical structures in clinical research: time-to-event outcomes and binary outcomes. The same randomized treatment comparison can therefore be examined using fundamentally different statistical methods depending on how the endpoint is defined.
| Concept | How it appears in EGF104900 |
|---|---|
| Randomization | The trial uses randomized allocation to two parallel treatment arms. |
| Intention-to-treat analysis | PFS, OS, overall tumor response, and CBR analyses are reported in ITT populations, with endpoint-specific wording. |
| Time-to-event endpoint | PFS is the registered primary endpoint; OS is a secondary time-to-event endpoint. |
| Log-rank test | Used for the PFS and OS treatment comparisons. |
| Hazard ratio | Used to quantify the relative treatment effect for PFS and OS. |
| Confidence interval | 95% two-sided intervals accompany all four posted effect estimates. |
| Fisher exact test | Used for overall tumor response and clinical benefit response. |
| Odds ratio | Used as the effect measure for the two binary response endpoints. |
| Stratified analysis | Identified in the analysis text for overall tumor response and clinical benefit response. |
| Superiority | The registered hypothesis type for all four reported analyses. |
| Endpoint hierarchy | One primary PFS endpoint is distinguished from three secondary analyzed outcomes. |
The trial is particularly useful for understanding why statistical interpretation must follow the endpoint rather than precede it. A time-to-event endpoint retains information about follow-up duration and event timing, so a hazard ratio and log-rank test are appropriate tools. A binary response endpoint instead classifies participants according to whether the event occurred, making categorical methods such as Fisher's exact test and odds ratios appropriate.
It also illustrates why the phrase “statistically significant” is not enough to describe a result. The PFS estimate, its confidence interval, its P-value, the endpoint definition, the analysis population, and the time-to-event structure all contribute to the statistical story. The same principle applies to the CBR result: the OR of 2.2, CI of 1.2–4.5, and P-value of 0.010 together provide much more information than the P-value alone.
22. A Practical Reading of the Primary PFS Analysis
A disciplined reading of the primary result can be done in several steps.
- Identify the endpoint. PFS is defined from randomization to the first documented sign of disease progression or death due to any cause.
- Identify the population. The analysis is reported in the ITT population, with the registry's progesterone-receptor-status qualification.
- Identify the comparison. Trastuzumab plus lapatinib is compared with lapatinib.
- Identify the method. The registry reports a log-rank test.
- Identify the effect measure. The reported measure is a hazard ratio, specifically the Pike estimator based on the log-rank test.
- Read the estimate. HR 0.73 indicates a lower estimated hazard of progression or death in the combination group.
- Read the confidence interval. The 95% CI is 0.57–0.93, communicating uncertainty around the HR.
- Read the P-value. P = 0.008 is the reported two-sided test result under the superiority framework.
- No median PFS or absolute PFS percentages are reported in the ClinicalTrials.gov record, so those quantities cannot be derived from the HR alone.
This sequence prevents several common errors. It avoids treating an HR as a risk ratio, prevents the P-value from being mistaken for an effect size, and keeps the reported analysis population attached to the result rather than silently replacing it with another population.
23. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The reported primary PFS analysis produced an HR of 0.73 with a 95% CI of 0.57–0.93 and P = 0.008 using a log-rank test. The secondary OS analysis produced an HR of 0.75 with a 95% CI of 0.53–1.07 and P = 0.106.
Endpoint interpretation
The PFS and OS findings describe time-to-event outcomes, while overall tumor response and clinical benefit response describe binary outcomes. Their effect measures therefore have different meanings and should not be combined numerically.
Response interpretation
Overall tumor response had an OR of 1.5 with a 95% CI of 0.6–3.9 and P = 0.460. Clinical benefit response had an OR of 2.2 with a 95% CI of 1.2–4.5 and P = 0.010.
Safety interpretation
Serious adverse events were reported as 40/149 in the combination group and 24/146 in the lapatinib group. The ClinicalTrials.gov record does not include a formal comparative safety test.
A statistical result and a clinical judgment are not the same thing. The statistical analysis describes the estimated treatment difference and uncertainty under the trial's analysis framework. Whether a particular magnitude of effect is clinically meaningful requires additional context that is not reported by a P-value or hazard ratio alone.
24. Important Questions When Interpreting EGF104900
Does HR 0.73 mean 27% fewer patients progressed?
No. The 27% figure is a transformation of the hazard ratio into a relative hazard reduction. It does not represent a percentage-point reduction in the number of participants who progressed, and it does not identify how many individual participants benefited.
Does the OS P-value of 0.106 prove that there was no OS difference?
No. It reports the outcome of the specified two-sided log-rank test. The corresponding HR is 0.75 and the 95% CI is 0.53–1.07. The confidence interval includes the null value, so the estimate remains uncertain under the reported analysis.
Can the OR of 1.5 be called a 50% higher response rate?
No. An OR of 1.5 means 1.5 times the odds, not 1.5 times the probability and not a 50-percentage-point increase in response. Converting odds ratios into probability differences requires the underlying event rates.
Can the CBR OR of 2.2 be compared directly with the PFS HR of 0.73?
No. The two measures operate on different scales and summarize different types of endpoints. The HR concerns event hazards over time; the OR concerns odds of a binary outcome.
Does P = 0.010 mean there is a 99% probability that treatment is beneficial?
No. A P-value is not a posterior probability that a hypothesis is true. It is the reported probability measure generated by the specified frequentist testing framework under the null model.
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Statistical Calculators
27. Sources
- ClinicalTrials.gov: NCT00320385 — EGF104900.
Continue exploring Clinical Biostats
Use the related tutorials and statistical calculators to examine the survival-analysis and categorical-data methods illustrated by EGF104900.
28. Record Summary
EGF104900 provides a compact example of how a randomized phase 3 trial can use different statistical frameworks for different endpoint structures. Its registered primary endpoint, PFS, is a time-to-event outcome analyzed with a log-rank test and reported hazard ratio of 0.73 with a 95% CI of 0.57–0.93 and P = 0.008. The secondary OS analysis likewise used a log-rank test and reported an HR of 0.75 with a 95% CI of 0.53–1.07 and P = 0.106.
The two binary secondary endpoints demonstrate a different analytical pathway. Overall tumor response was analyzed with a stratified Fisher exact test and had an OR of 1.5 with a 95% CI of 0.6–3.9 and P = 0.460. Clinical benefit response had an OR of 2.2 with a 95% CI of 1.2–4.5 and P = 0.010.
The central statistical lesson is that the endpoint determines the appropriate measure of effect. Hazard ratios describe relative event hazards over time; odds ratios describe relative odds for binary outcomes. Confidence intervals communicate precision, while P-values quantify statistical evidence under the specified tests. None of these quantities, considered alone, describes the complete clinical experience of every participant.