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
NALA was a randomized, open-label, parallel phase 3 trial comparing neratinib plus capecitabine with lapatinib plus capecitabine in patients with HER2+ metastatic breast cancer who had received two or more prior HER2-directed regimens in the metastatic setting.
| Feature | NALA |
|---|---|
| Phase | Phase 3 |
| Condition | HER2+ metastatic breast cancer |
| Population | Patients who had received two or more prior HER2-directed regimens in the metastatic setting |
| Design | Randomized, parallel, open-label |
| Allocation | Randomized |
| Primary endpoints | Centrally assessed progression-free survival and overall survival |
| Primary endpoint type | Time-to-event |
| Enrollment | 621 |
| Lead sponsor | Puma Biotechnology, Inc. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT01808573 |
2. Clinical Question
The central statistical question was whether neratinib plus capecitabine produced superior time-to-event outcomes compared with lapatinib plus capecitabine in patients with HER2+ metastatic breast cancer who had received two or more prior HER2-directed regimens in the metastatic setting.
Population
Patients with HER2+ metastatic breast cancer who had received two or more prior HER2-directed regimens in the metastatic setting.
Intervention
Neratinib plus capecitabine.
Comparator
Lapatinib plus capecitabine.
Primary question
Does the neratinib-plus-capecitabine regimen provide superior centrally assessed PFS and overall survival compared with lapatinib plus capecitabine?
3. Trial Design
Neratinib Plus Capecitabine
- Neratinib
- Capecitabine
- Randomized comparison arm
Lapatinib Plus Capecitabine
- Lapatinib
- Capecitabine
- Reference treatment group for the reported hazard ratios
4. Trial Timeline and Registry Status
| Milestone | Registry information |
|---|---|
| Trial start | March 29, 2013 |
| Primary completion | September 28, 2018 |
| Trial status | Completed |
| Enrollment | 621 |
| Phase | 3 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
The registry reports statistical results for seven posted outcome measures, including six formal statistical analyses. Two of those analyses correspond to the registered primary endpoints.
5. Primary Endpoints
| Endpoint | Registered definition and time frame | Analysis |
|---|---|---|
| Centrally Assessed Progression Free Survival | Progression Free Survival (PFS), measured in months, for randomized subjects of the central assessment. The time interval from the date of randomization until the first date on which recurrence, progression, or death due to any cause is documented. The registry time frame is from randomization to recurrence, progression or death, assessed up to 38 months. | Stratified log-rank test; hazard ratio |
| Overall Survival | Time from randomization to death due to any cause, censored at the last date known alive on or prior to the data cutoff employed for the analysis. The registry reports the result as restricted mean survival time, defined as the area under the survival function up to 48 months. The registry time frame is from randomization to death, assessed up to 59 months. | Stratified log-rank test; hazard ratio |
Both primary endpoints are time-to-event endpoints. That means the analysis incorporates not only whether an event occurred but also the amount of follow-up available for each randomized participant.
6. Statistical Methodology
Intention-to-treat analysis
The primary PFS analysis was performed in the Randomized ITT Population. An intention-to-treat framework analyzes randomized participants according to the treatment assignment generated by randomization. This preserves the central comparison created by the randomized design rather than redefining treatment groups according to subsequent treatment exposure.
Log-rank testing
The registry reports a log-rank test for both primary endpoints. The log-rank test is designed for comparing time-to-event distributions between treatment groups while accounting for censored observations.
The log-rank framework compares the observed pattern of events with the pattern expected under the null hypothesis across the follow-up period. It is therefore different from a simple comparison of proportions at one fixed time point.
Stratified analysis
The primary analyses were stratified by hormone receptor status, number of prior HER2-directed regimens in the metastatic setting, and visceral disease versus non-visceral disease.
Stratification allows the time-to-event comparison to account for these prespecified factors rather than treating the randomized population as if every participant had the same baseline risk. The same stratification factors are reported for the primary PFS and OS analyses and for the secondary CNS-disease analysis.
Hazard ratio
The registry reports hazard ratios as the effect measure for the primary PFS and OS analyses. With lapatinib plus capecitabine as the reference group, a hazard ratio below 1 indicates a lower estimated instantaneous event rate in the neratinib-plus-capecitabine group under the fitted time-to-event comparison.
The hazard ratio is a relative time-to-event measure. It is not a probability, not a percentage of patients who benefit, and not a statement that every participant experiences the same proportional change in risk.
Confidence intervals
Each primary endpoint has a two-sided 95% confidence interval around its hazard ratio. The interval describes statistical uncertainty around the estimated relative treatment effect under the analysis framework. A narrower interval generally indicates greater precision than a wider interval, although precision and clinical importance are separate questions.
P-values
The P-value addresses the statistical evidence against the null hypothesis under the specified testing framework. It does not measure the size of the treatment effect. For that reason, the P-value should be interpreted together with the hazard ratio and its confidence interval.
7. Primary Results: Centrally Assessed Progression-Free Survival
Hazard ratio for progression, recurrence, or death
95% CI: 0.626–0.926 · P = 0.0059
Randomized ITT population · Stratified log-rank analysis
| Element | Reported result |
|---|---|
| Endpoint | Centrally Assessed Progression Free Survival |
| Population | Randomized ITT Population |
| Comparison | Neratinib Plus Capecitabine vs Lapatinib Plus Capecitabine |
| Analysis | Log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.762 |
| 95% CI | 0.626–0.926 |
| P-value | 0.0059 |
| Hypothesis | Superiority |
| Time frame | From randomization date to recurrence, progression or death, assessed up to 38 months |
The reported HR of 0.762 means that, under the reported time-to-event analysis, the estimated instantaneous rate of the PFS event was about 76.2% of the corresponding rate in the lapatinib-plus-capecitabine reference group. Equivalently, 1 − 0.762 = 0.238, so the estimate corresponds to an approximately 23.8% lower estimated hazard for the PFS event in the neratinib group.
This does not mean that 23.8% of patients avoided progression, that 23.8% of patients were cured, or that every patient experienced exactly a 23.8% reduction in risk. It is a relative hazard estimate from a time-to-event analysis.
The 95% CI of 0.626–0.926 describes uncertainty around the estimated hazard ratio. It does not describe the range of outcomes for individual patients. Because the interval is below 1 throughout, the reported estimate is consistent with a lower event hazard for the neratinib-plus-capecitabine group under this analysis.
The P-value of 0.0059 is evidence against the null hypothesis used for the reported superiority comparison. It does not tell us that the treatment effect is "0.59%" or that there is a 0.59% probability that the treatment has no effect. The magnitude and precision of the effect are better conveyed by the HR and its confidence interval.
As with other hazard-ratio analyses, interpretation also depends on the time-to-event model and censoring structure. A single HR summarizes a relative event-rate comparison; it is not an absolute difference in survival probabilities at a particular time.
What the PFS endpoint actually measures
The registered endpoint begins at randomization and ends at the first documented recurrence, progression according to RECIST v1.1, or death from any cause. Participants without recurrence, progression, or death are censored according to the registry definition.
This makes PFS different from a binary response endpoint. A patient who remains progression-free contributes information for as long as that patient remains under observation without the event, rather than simply being classified as "progressed" or "not progressed" at a single time point.
8. Primary Results: Overall Survival
Hazard ratio for death
95% CI: 0.723–1.073 · P = 0.2086
Stratified log-rank analysis · Lapatinib Plus Capecitabine as reference
| Element | Reported result |
|---|---|
| Endpoint | Overall Survival |
| Comparison | Neratinib Plus Capecitabine vs Lapatinib Plus Capecitabine |
| Analysis | Log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.881 |
| 95% CI | 0.723–1.073 |
| P-value | 0.2086 |
| Hypothesis | Superiority |
| Time frame | From randomization date to death, assessed up to 59 months |
| Additional registry definition | Time to death due to any cause, censored at the last date known alive on or prior to the data cutoff. The result was reported as restricted mean survival time, defined as the area under the survival function up to 48 months. |
The reported OS HR of 0.881 corresponds to an estimated instantaneous rate of death approximately 88.1% of that in the lapatinib-plus-capecitabine reference group under the reported analysis. Expressed as a relative hazard difference, 1 − 0.881 = 0.119, corresponding to an approximately 11.9% lower estimated hazard of death.
That estimate should not be translated into an 11.9% increase in survival probability or an 11.9% absolute improvement in survival. The hazard ratio is a relative time-to-event measure, not an absolute survival difference.
The 95% CI is 0.723–1.073. Unlike the PFS interval, this interval crosses 1.00. Consequently, the reported estimate is compatible with both a lower and a higher instantaneous mortality hazard for the neratinib group relative to the reference group within the uncertainty represented by this interval.
The P-value of 0.2086 does not measure the size of the observed effect. It indicates that the reported superiority test did not provide strong statistical evidence against its null hypothesis under the specified analysis framework. A non-small P-value should not be interpreted as proof that the two treatments are identical.
The registry also describes the OS result using restricted mean survival time, with the restricted mean defined as the area under the survival curve up to 48 months. That is conceptually different from a hazard ratio because it summarizes survival time accumulated through a specified time horizon rather than expressing an instantaneous relative event rate.
9. Understanding the Two Primary Endpoints Together
Progression-Free Survival
HR 0.762 with a 95% CI of 0.626–0.926 and P = 0.0059. The reported analysis provides evidence against the null hypothesis for the superiority comparison.
Overall Survival
HR 0.881 with a 95% CI of 0.723–1.073 and P = 0.2086. The confidence interval includes 1.00 and the reported superiority comparison has a larger P-value.
The two endpoints answer different questions. PFS evaluates the time from randomization to recurrence, progression, or death, whereas OS evaluates the time from randomization to death from any cause. A treatment effect on one endpoint does not mathematically guarantee the same magnitude of effect on another.
The numerical contrast between the two reported hazard ratios is therefore itself statistically informative: the estimated relative effect is smaller for PFS than for OS, and the uncertainty around the OS estimate is sufficiently broad to include 1.00.
10. Secondary Endpoint Results
Duration of Response
Hazard ratio for duration of response
95% CI: 0.332–0.736 · P = 0.0004
Central assessment among the population that had a response with measurable disease at screening
The registry defines this endpoint as the time from the start date of response after randomization to first progressive disease, with assessment up to 33 months. The analysis used a log-rank test and reported lapatinib plus capecitabine as the reference group.
The HR of 0.495 indicates an estimated instantaneous rate of the duration-of-response event of about 49.5% of that in the reference group. Equivalently, the estimated hazard is approximately 50.5% lower under the reported model.
The confidence interval of 0.332–0.736 quantifies uncertainty around that estimate. This endpoint is conditional on having a response with measurable disease at screening, so it is not a substitute for the ITT PFS analysis.
The P-value of 0.0004 provides evidence against the reported superiority null hypothesis. It does not indicate the probability that a particular patient will have a durable response.
Intervention for Symptomatic Metastatic Central Nervous System Disease
| Element | Reported result |
|---|---|
| Endpoint | Intervention for Symptomatic Metastatic Central Nervous System Disease |
| Time frame | From randomization date to first intervention for symptomatic metastatic CNS disease, assessed up to 59 months |
| Analysis | Gray's test |
| Effect measure | No hazard ratio reported in the registry statistical analysis |
| P-value | 0.043 |
| Hypothesis | Superiority |
| Stratification | Hormone receptor status, number of prior HER2-directed regimens in the metastatic setting, and visceral disease vs. non-visceral disease |
The registry reports a Gray's test P-value of 0.043 for this endpoint but does not provide a hazard ratio or corresponding confidence interval in the registry-reported statistical analysis. Therefore, the registry result supports a statistical comparison of the cumulative incidence framework but does not provide a numerical relative effect estimate that can be interpreted like the PFS or OS hazard ratios.
Gray's test is used for comparing cumulative incidence functions when competing risks are relevant. The key distinction is that a competing event can prevent the event of interest from subsequently occurring, so treating every competing event as ordinary censoring can produce an inappropriate risk comparison.
Objective Response Rate
| Element | Reported result |
|---|---|
| Endpoint | Objective Response Rate (ORR) — Central Assessment |
| Population | ITT Population With Measurable Disease at Screening |
| Time frame | From randomization date to first confirmed Complete or Partial Response, whichever came earlier, up to 42 months |
| Analysis | Cochran-Mantel-Haenszel test |
| Effect measure | No effect estimate reported in the registry statistical analysis |
| P-value | 0.1201 |
| Hypothesis | Superiority |
| Stratification | Hormone receptor status, number of prior HER2-directed regimens in the metastatic setting, and visceral disease vs. non-visceral disease |
The registry reports a Cochran-Mantel-Haenszel P-value of 0.1201, but it does not provide the response percentages, an effect estimate, or a confidence interval in the registry-reported statistical analysis. The formal method is therefore reported without adding unprovided response rates.
Clinical Benefit Rate
| Element | Reported result |
|---|---|
| Endpoint | Clinical Benefit Rate (CBR) — Central Assessment |
| Population | ITT Population With Measurable Disease at Screening |
| Time frame | From randomization date to either first confirmed CR or PR or Stable Disease, whichever came earlier, up to 42 months |
| Analysis | Cochran-Mantel-Haenszel test |
| Effect measure | No effect estimate reported in the registry statistical analysis |
| P-value | 0.0328 |
| Hypothesis | Superiority |
| Stratification | Hormone receptor status, number of prior HER2-directed regimens in the metastatic setting, and visceral disease vs. non-visceral disease |
The reported P = 0.0328 indicates evidence against the null hypothesis for the specified superiority comparison under the Cochran-Mantel-Haenszel framework. However, because the ClinicalTrials.gov record does not report the underlying percentages or an effect estimate, the size of the difference cannot be quantified from this result alone.
This illustrates why a P-value should not be presented as a stand-alone measure of clinical effect. Without the corresponding group estimates and confidence interval, the numerical magnitude and precision of the CBR difference remain unspecified in the ClinicalTrials.gov record.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. These figures are presented as affected participants over participants at risk.
| Safety measure | Neratinib Plus Capecitabine | Lapatinib Plus Capecitabine |
|---|---|---|
| Serious adverse events | 103/303 | 93/311 |
Neratinib Plus Capecitabine
103/303 participants were affected by serious adverse events among the reported number at risk.
Lapatinib Plus Capecitabine
93/311 participants were affected by serious adverse events among the reported number at risk.
The ClinicalTrials.gov record does not provide a formal statistical comparison or confidence interval for these serious-adverse-event counts. Accordingly, the page reports the registry values without constructing an unreported hypothesis test.
12. Statistical Methods Explained
Why was a log-rank test used for PFS and OS?
PFS and OS are time-to-event outcomes. Some participants may not experience the event during the observation period, creating right-censored observations. The log-rank test is designed for comparing survival distributions while incorporating that censoring rather than treating all participants as if they had identical follow-up.
What does a PFS hazard ratio of 0.762 mean?
With lapatinib plus capecitabine as the reference, the reported HR of 0.762 means that the estimated instantaneous event rate for the PFS endpoint was approximately 76.2% of the reference rate under the fitted analysis. It corresponds to an approximately 23.8% lower estimated hazard, but it is not a 23.8-percentage-point improvement in the probability of remaining progression-free.
Why does the confidence interval matter?
The point estimate is only one possible estimate supported by the data. The 95% CI of 0.626–0.926 for PFS communicates uncertainty around the HR. For OS, the interval is wider relative to the estimate and spans 1.00, which means the reported data are compatible with a range of relative effects that includes no difference on the hazard-ratio scale.
Why is the OS P-value not a measure of treatment effect?
The OS P-value of 0.2086 summarizes evidence against the specified null hypothesis under the reported testing framework. It does not describe how large the treatment effect is. Effect magnitude is communicated by the HR, while uncertainty is communicated by its confidence interval.
Why was Gray's test used for the CNS endpoint?
The endpoint concerns the first intervention for symptomatic metastatic CNS disease, and the registry identifies Gray's test as the analysis method. Gray's test is used when cumulative incidence must be compared in the presence of competing risks. This is different from a standard Kaplan-Meier comparison that treats competing events as ordinary censoring.
Why was the Cochran-Mantel-Haenszel test used for ORR and CBR?
ORR and CBR are categorical endpoints rather than continuous time-to-event measurements. The Cochran-Mantel-Haenszel framework allows a treatment comparison while accounting for the reported stratification factors. In this trial those factors included hormone receptor status, number of prior HER2-directed regimens in the metastatic setting, and visceral disease versus non-visceral disease.
Why does the analysis population matter?
The primary PFS analysis is explicitly identified as using the randomized ITT population. By contrast, the duration-of-response analysis is restricted to the population that had a response with measurable disease at screening. These populations answer different questions. A response-duration analysis cannot be interpreted as though it represented every randomized participant.
13. Stratification and the Cochran-Mantel-Haenszel Framework
The registry identifies three stratification factors for the reported stratified analyses:
| Stratification factor | Categories as reported |
|---|---|
| Hormone receptor status | Hormone receptor status |
| Prior HER2-directed treatment | Number of prior HER2-directed regimens in the metastatic setting |
| Visceral disease | Visceral disease vs. non-visceral disease |
Stratification is useful when important baseline factors are expected to influence event rates. Instead of ignoring those factors, the analysis compares treatment groups within the relevant strata and combines information across strata using the prespecified statistical framework.
The same general principle applies to the Cochran-Mantel-Haenszel analyses of ORR and CBR: the categorical treatment comparison is adjusted for the specified stratification factors rather than being reduced to an unstratified comparison.
Stratified comparison → comparisons within prespecified strata, combined across strata
The purpose is not to manufacture a larger treatment effect. It is to account for the trial's prespecified structure and the prognostic factors used in the analysis.
14. Time-to-Event Endpoints and Censoring
NALA provides several examples of why time-to-event analysis requires more information than a simple event percentage.
| Endpoint | Starting point | Event | Reported statistical framework |
|---|---|---|---|
| Centrally Assessed PFS | Randomization | Recurrence, progression, or death | Log-rank test; HR |
| Overall Survival | Randomization | Death due to any cause | Log-rank test; HR |
| Duration of Response | Start of response after randomization | First progressive disease | Log-rank test; HR |
| Symptomatic metastatic CNS intervention | Randomization | First intervention for symptomatic metastatic CNS disease | Gray's test |
For PFS, the registry definition states that participants without recurrence, progression, or death are censored at the relevant end of follow-up. For OS, participants are censored at the last date known alive on or prior to the analysis data cutoff.
Censoring allows participants with incomplete event information to contribute the follow-up that is actually observed. The validity of the resulting survival analysis depends on the censoring and analysis assumptions, so a hazard ratio should always be understood as an estimate generated from a particular time-to-event framework rather than as a simple observed percentage.
15. Restricted Mean Survival Time in the OS Definition
The registry includes an additional description of the OS result: time to event was reported as the restricted mean survival time, with the restricted mean defined as the area under the survival function up to 48 months.
Here, the registry defines the relevant restriction at 48 months. Conceptually, the area under the survival curve represents average survival time accumulated through the specified horizon.
RMST is useful because it expresses treatment differences on a time scale rather than only as a relative hazard. It does not require the reader to interpret a hazard ratio as an instantaneous event-rate ratio. In this registry record, however, the registry-reported statistical analysis reports the primary OS comparison as a hazard ratio and does not provide a numerical RMST difference.
16. Primary Endpoint Interpretation: Relative vs Absolute Evidence
The NALA registry results illustrate why statistical interpretation should not rely on a single number.
Relative effect
The PFS HR is 0.762 and the OS HR is 0.881. These summarize relative event-rate differences under the reported time-to-event analyses.
Precision
The PFS 95% CI is 0.626–0.926, while the OS 95% CI is 0.723–1.073. The intervals communicate uncertainty around the point estimates.
Statistical evidence
The PFS P-value is 0.0059 and the OS P-value is 0.2086. These address evidence against the respective null hypotheses rather than clinical magnitude.
Endpoint definition
PFS and OS measure different events. A difference in one endpoint cannot simply be substituted for the other.
17. Multiplicity and the Two Primary Endpoints
The registry identifies two registered primary endpoints: centrally assessed PFS and overall survival. Both are superiority hypotheses and both have formal statistical analyses posted.
| Endpoint | Role | Hypothesis | Formal analysis |
|---|---|---|---|
| Centrally Assessed PFS | Primary | Superiority | Log-rank test |
| Overall Survival | Primary | Superiority | Log-rank test |
The ClinicalTrials.gov record does not provide an alpha-allocation scheme, multiplicity-adjustment procedure, or hierarchical testing sequence for the two primary endpoints. Accordingly, this page does not infer one.
18. Interim Analysis and Other Design Features
The ClinicalTrials.gov record does not report an interim-analysis plan, alpha-spending procedure, Bayesian method, non-inferiority margin, crossover rule, or missing-data/imputation method. Those design topics are therefore not characterized here.
This distinction matters because the absence of a field in the ClinicalTrials.gov record is not evidence that a particular method was or was not used in the full protocol or statistical analysis plan. The appropriate conclusion from the available data is simply that the ClinicalTrials.gov record does not specify the method.
19. Statistical Interpretation of the Secondary Analyses
| Secondary endpoint | Method | Reported estimate | 95% CI | P-value |
|---|---|---|---|---|
| Duration of Response | Log-rank | HR 0.495 | 0.332–0.736 | 0.0004 |
| Intervention for Symptomatic Metastatic CNS Disease | Gray's test | Not reported | Not reported | 0.043 |
| Objective Response Rate | Cochran-Mantel-Haenszel | Not reported | Not reported | 0.1201 |
| Clinical Benefit Rate | Cochran-Mantel-Haenszel | Not reported | Not reported | 0.0328 |
The four secondary analyses demonstrate four distinct reporting situations. Duration of response has a hazard ratio and confidence interval; the CNS endpoint has a competing-risks test without a reported effect estimate; ORR and CBR have stratified categorical tests without reported effect estimates in the registry-reported statistical analysis.
That difference in reporting is important. A P-value without an effect estimate cannot tell the reader how large the between-group difference was. Conversely, an effect estimate without its uncertainty would provide an incomplete picture of precision.
20. Important Limitations and Interpretation Issues
- Open-label design: masking was listed as none. Knowledge of treatment assignment can affect aspects of trial conduct even when the statistical endpoint is defined independently.
- Incomplete effect reporting for secondary endpoints: the ClinicalTrials.gov record does not provide effect estimates and confidence intervals for the CNS, ORR, or CBR analyses.
- Restricted response population: duration of response was evaluated in the population that had a response with measurable disease at screening, so it should not be interpreted as an ITT endpoint.
- Competing risks: the CNS intervention endpoint was analyzed with Gray's test, highlighting that the event structure differs from an ordinary survival endpoint.
- Hazard-ratio interpretation: an HR is a relative time-to-event measure, not an absolute risk difference or a guarantee of proportional benefit for every participant.
- Confidence intervals: the OS interval of 0.723–1.073 spans 1.00, so the point estimate alone should not be treated as establishing a definitive reduction in the hazard of death.
- Multiplicity: the ClinicalTrials.gov record identifies two primary endpoints but does not specify the multiplicity-control strategy. No unreported adjustment is assumed here.
- Unreported design details: interim-analysis, Bayesian, crossover, non-inferiority, and missing-data procedures are not specified in the ClinicalTrials.gov record.
- Registry reporting limits: the ClinicalTrials.gov record does not include median PFS, median OS, subgroup estimates, baseline characteristics, or arm-specific ORR and CBR values. Those quantities are therefore not added.
21. Why This Trial Matters Statistically
NALA is a useful teaching case because it combines several major clinical-trial statistical concepts in one randomized phase 3 study: two primary time-to-event endpoints, stratified log-rank testing, hazard ratios, confidence intervals, an ITT efficacy population, a response-duration analysis restricted to responders, a competing-risks analysis, and stratified categorical analyses.
| Concept | How it appears in NALA |
|---|---|
| Randomization | Randomized phase 3 parallel-group comparison |
| Intention-to-treat analysis | Primary PFS analysis uses the Randomized ITT Population |
| Time-to-event endpoints | PFS, OS, duration of response, and symptomatic CNS intervention |
| Log-rank test | Used for the two primary endpoints and duration of response |
| Hazard ratio | Reported for PFS, OS, and duration of response |
| Confidence interval | 95% two-sided intervals reported for the primary HR estimates and duration of response |
| Stratified analysis | Hormone receptor status, prior HER2-directed regimens, and visceral disease status |
| Gray's test | Used for the symptomatic metastatic CNS intervention endpoint |
| Cochran-Mantel-Haenszel test | Used for ORR and CBR |
| Competing risks | Relevant to the Gray's-test analysis of symptomatic CNS intervention |
| Restricted mean survival time | OS definition describes RMST as area under the survival function through 48 months |
22. What the Primary Hazard Ratios Do — and Do Not — Mean
A PFS HR of 0.762 means that the estimated instantaneous event rate was approximately 76.2% of the reference-group rate under the reported survival analysis. It corresponds to an approximately 23.8% lower estimated hazard.
It does not mean that 23.8% more patients remained progression-free, that 23.8% of patients were cured, or that each patient had exactly the same proportional reduction in risk.
An OS HR of 0.881 means that the estimated instantaneous mortality rate was approximately 88.1% of the reference-group rate under the reported analysis. It corresponds to an approximately 11.9% lower estimated hazard.
The 95% CI of 0.723–1.073 includes 1.00, so the uncertainty interval encompasses the possibility of no relative difference on the hazard-ratio scale.
A duration-of-response HR of 0.495 corresponds to an estimated event hazard approximately 49.5% of the reference-group hazard, or an approximately 50.5% lower estimated hazard under the reported model.
This estimate applies to the response-defined analysis population, not to all randomized participants.
23. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The randomized comparison produced a PFS HR of 0.762 and an OS HR of 0.881. The PFS confidence interval lies below 1.00, whereas the OS confidence interval spans 1.00.
Endpoint interpretation
PFS, OS, duration of response, CNS intervention, ORR, and CBR are distinct outcomes. Their analyses should not be collapsed into a single numerical measure of treatment effect.
Uncertainty interpretation
Confidence intervals describe uncertainty around estimates. The OS interval is particularly important because it extends above 1.00.
Reporting interpretation
For secondary endpoints without reported effect estimates, the P-values provide evidence from the specified tests but do not quantify the size of the treatment difference.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: NCT01808573 — NALA.
- Linked publication: PubMed record: PMID 34553296.
- Linked publication: PubMed record: PMID 32678716.
Continue through Clinical Biostats
Connect this trial's endpoints and statistical methods to deeper biostatistics tutorials, statistical calculators, and other clinical trial analyses.
27. Record Summary
NALA provides a useful example of how a randomized phase 3 oncology trial can combine several statistical frameworks according to endpoint type. The primary PFS and OS analyses use stratified log-rank testing and hazard ratios, while duration of response uses another time-to-event comparison, symptomatic CNS intervention uses Gray's test for competing risks, and ORR and CBR use Cochran-Mantel-Haenszel analyses.
The primary PFS result is reported as HR 0.762 (95% CI 0.626–0.926; P = 0.0059), while the primary OS result is HR 0.881 (95% CI 0.723–1.073; P = 0.2086). The statistical interpretation of these results depends on the distinction between relative effect, absolute outcome, uncertainty, and the specific endpoint being analyzed.
The secondary analyses further demonstrate why the method must match the outcome. Duration of response is analyzed as a time-to-event endpoint, the CNS intervention endpoint uses a competing-risks method, and ORR and CBR are categorical outcomes analyzed with the Cochran-Mantel-Haenszel test. For several secondary endpoints, the ClinicalTrials.gov record reports P-values without effect estimates, so the size of those differences cannot be inferred from the P-values alone.