This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
CRYSTAL was a randomized, parallel-group, open-label phase 3 trial evaluating cetuximab combined with FOLFIRI versus FOLFIRI alone in patients with epidermal growth factor receptor (EGFR)-expressing metastatic colorectal cancer.
| Feature | CRYSTAL |
|---|---|
| Trial name | CRYSTAL |
| Phase | Phase 3 |
| Status | COMPLETED |
| Condition | Epidermal Growth Factor Receptor (EGFR) Expressing Metastatic Colorectal Cancer |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 1221 |
| Interventions | Cetuximab (drug); FOLFIRI (5-Fluorouracil, Folinic acid, Irinotecan) (drug) |
| Primary endpoint type | Time-to-event |
| Results posted | Yes |
| Outcome measures posted | 15 |
| Statistical analyses posted | 11 |
2. Clinical Question
The central statistical question was whether adding cetuximab to FOLFIRI changed progression-free survival in the randomized population, including the prespecified analyses of the overall ITT population and tumor KRAS subgroups.
Population
Patients with epidermal growth factor receptor (EGFR)-expressing metastatic colorectal cancer.
Intervention
Cetuximab combined with FOLFIRI, where FOLFIRI consists of 5-Fluorouracil, Folinic acid, and Irinotecan.
Comparator
FOLFIRI alone.
Primary question
Does cetuximab plus FOLFIRI differ from FOLFIRI alone for the registered progression-free survival endpoints under a superiority framework?
3. Trial Design
Cetuximab combination
- Cetuximab
- FOLFIRI
- FOLFIRI consists of 5-Fluorouracil, Folinic acid, and Irinotecan
FOLFIRI comparator
- FOLFIRI alone
- FOLFIRI consists of 5-Fluorouracil, Folinic acid, and Irinotecan
4. Primary Endpoints
Three primary endpoints were registered. All were progression-free survival measures based on Independent Review Committee (IRC) assessments and analyzed as time-to-event outcomes.
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Progression-free Survival (PFS) Time - Independent Review Committee (IRC) Assessments | Duration from randomization until radiological progression (based on modified World Health Organisation (WHO) criteria) or death due to any cause. Only deaths within 60 days of last tumor assessment are considered. Patients without event are censored on the date of last tumor assessment. | Time from randomisation to disease progression, death or last tumour assessment, reported between day of first patient randomised, 10 Aug 2004, until cut-off date, 27 July 2006 |
| Progression-free Survival Time (Chinese V-Ki-ras2 Kirsten Rat Sarcoma Viral Oncogene Homolog (KRAS) Wild-Type Population) - Independent Review Committee (IRC) Assessments | Duration from randomization until radiological progression (based on modified WHO criteria) or death due to any cause. Only deaths within 60 days of last tumor assessment are considered. Patients without event are censored on the date of last tumor assessment. | Time from randomisation to disease progression, death or last tumour assessment, reported between day of first patient randomised, 10 Aug 2004, until cut-off date, 27 July 2006 |
| Progression-free Survival Time (KRAS Mutant Population) - Independent Review Committee (IRC) Assessments | Duration from randomization until radiological progression (based on modified WHO criteria) or death due to any cause. Only deaths within 60 days of last tumor assessment are considered. Patients without event are censored on the date of last tumor assessment. | Time from randomisation to disease progression, death or last tumour assessment, reported between day of first patient randomised, 10 Aug 2004, until cut-off date, 27 July 2006 |
Why the censoring definition matters
The registry definition is more specific than simply saying that PFS is the time until progression or death. Deaths are counted as events only when they occur within 60 days of the last tumor assessment, while patients without an event are censored on the date of the last tumor assessment. Consequently, the analysis depends not only on whether a patient eventually experiences an event, but also on the timing of assessments and censoring.
5. Statistical Methodology
Stratified log-rank test
All three primary PFS analyses used a stratified log-rank test. The registry analysis notes specify a two-sided test at the 5% significance level for the KRAS wild-type and KRAS mutant analyses and describe consideration of the randomization strata.
The log-rank framework compares the observed and expected numbers of events across treatment groups over follow-up. Stratification allows the comparison to account for prespecified strata rather than treating every participant as though they came from one homogeneous risk set.
Kaplan-Meier estimation
For the KRAS wild-type and KRAS mutant primary analyses, the registry states that the Kaplan-Meier method was used to estimate median PFS time. Kaplan-Meier estimation is appropriate for time-to-event data because it incorporates right-censored observations without requiring every patient to experience the event during follow-up.
Here, di is the number of events at time ti and ni is the number at risk immediately before that time.
Stratified Cox proportional-hazards model
The registry states that hazard ratios were calculated using a Cox proportional-hazards model stratified by randomization strata for the KRAS wild-type and KRAS mutant PFS analyses. The hazard ratio therefore summarizes a relative event-rate comparison from a model that accounts for the specified strata.
An HR is not a direct probability, an absolute risk reduction, or the percentage of patients who benefit. It is a relative measure of event rates under the fitted time-to-event model.
Cochran-Mantel-Haenszel test
The response-related secondary endpoints used the Cochran-Mantel-Haenszel (CMH) test, including the Best Overall Response Rate analyses and the analysis of participants with no residual tumor after metastatic surgery. The registry specifies stratification by the randomization strata for the response analyses where those details are reported.
Odds ratio
The binary secondary endpoints report odds ratios rather than risk ratios. An odds ratio compares the odds of the specified binary outcome between the randomized treatment groups, with the registry estimates adjusted through the stratified CMH framework.
6. Analysis Populations and Stratification
The primary overall PFS analysis was conducted in the Intent to Treat (ITT) population, defined in the registry as all randomized subjects who have received at least one dose of randomized treatment, with allocation to treatment groups as randomized.
| Analysis | Population | Role |
|---|---|---|
| Overall primary PFS | Primary analysis on Intent to Treat (ITT) population i.e. all randomized subjects who have received at least one dose of randomized treatment (allocation to treatment groups as randomized). | Primary efficacy analysis |
| KRAS wild-type PFS | Intent to Treat (ITT) population with KRAS Wild Type tumor status as collected until 28 August 2009 | Primary efficacy analysis |
| KRAS mutant PFS | Intent to Treat (ITT) population with KRAS Mutant tumor status as collected until 28 August 2009 | Primary efficacy analysis |
| Overall survival | ITT population (allocation to treatment groups as randomized and treated) | Secondary efficacy analysis |
| KRAS-specific OS | Intent to Treat (ITT) population with KRAS Wild Type or KRAS Mutant tumor status as collected until 28 August 2009 | Secondary efficacy analyses |
The registry-reported analysis notes identify the randomization strata as region — Western Europe, Eastern Europe, outside Europe — and KPS — <80 versus ≥80. The stratified analyses therefore preserve information about the design strata rather than ignoring them in the treatment comparison.
7. Primary Results: Overall PFS
The overall primary PFS analysis compared cetuximab plus FOLFIRI with FOLFIRI alone in the ITT population using a stratified log-rank test. The reported effect measure was the hazard ratio.
Progression-free survival
95% CI: 0.728–1.000 · P = 0.0479
Two-sided superiority analysis.
| Feature | Reported result |
|---|---|
| Analysis population | Primary analysis on Intent to Treat (ITT) population i.e. all randomized subjects who have received at least one dose of randomized treatment (allocation to treatment groups as randomized). |
| Groups compared | Cetuximab Plus FOLFIRI vs FOLFIRI Alone |
| Method | Stratified log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.853 |
| 95% CI | 0.728–1.000 |
| P-value | 0.0479 |
An HR of 0.853 means the estimated instantaneous rate of progression or qualifying death was lower in the cetuximab-plus-FOLFIRI group than in the FOLFIRI-alone group under the fitted time-to-event model. Expressed as a simple relative interpretation, 0.853 corresponds to an estimated hazard approximately 14.7% lower than the comparator.
The HR does not mean that 14.7% of patients avoided progression, nor does it mean that every patient experienced a 14.7% reduction in individual risk. It is a relative time-to-event measure.
The 95% confidence interval of 0.728–1.000 describes uncertainty around the estimated hazard ratio. Its upper boundary reaches 1.000, so the reported estimate is close to the conventional no-difference value.
The P-value of 0.0479 addresses the evidence against the specified null hypothesis under the stated testing framework; it does not measure the size or clinical importance of the effect. The estimate and confidence interval provide the more direct description of magnitude and precision.
The analysis is also subject to the assumptions and censoring rules of time-to-event analysis. In particular, the Cox hazard-ratio interpretation relies on a proportional-hazards framework, while the registered PFS definition determines which deaths count as events and when patients are censored.
Planned information and power
The registry analysis notes state that the study was planned with 633 progression events to provide 80% power to test the null hypothesis of no difference in PFS time between treatment groups, assuming a hazard ratio of 0.8 for cetuximab plus chemotherapy over chemotherapy alone. The significance level was fixed at 5%.
The prespecified assumption of HR 0.8 is a design parameter, not the observed treatment effect. It should therefore be distinguished from the reported HR of 0.853.
8. Primary Results: KRAS Wild-Type PFS
The registry separately reports PFS for the ITT population with KRAS wild-type tumor status as collected until 28 August 2009.
KRAS wild-type progression-free survival
95% CI: 0.558–0.867 · P = 0.0012
Two-sided stratified log-rank analysis.
| Feature | Reported result |
|---|---|
| Analysis population | Intent to Treat (ITT) population with KRAS Wild Type tumor status as collected until 28 August 2009 |
| Groups compared | Cetuximab Plus FOLFIRI vs FOLFIRI Alone |
| Method | Stratified log-rank test |
| Kaplan-Meier method | Used to estimate median PFS time |
| Hazard-ratio model | Cox proportional hazards model stratified by randomization strata |
| Estimate | 0.696 |
| 95% CI | 0.558–0.867 |
| P-value | 0.0012 |
An HR of 0.696 indicates an estimated instantaneous event rate about 69.6% of the comparator under the stratified Cox model. As a simple relative interpretation, this corresponds to an estimated 30.4% lower hazard in the cetuximab-plus-FOLFIRI group.
The HR is not an absolute reduction in the probability of progression or death, and it does not imply that every KRAS wild-type patient experiences the same relative change.
The 95% CI of 0.558–0.867 remains below 1.000, providing a more informative description of the uncertainty around the estimated relative effect than the P-value alone.
The P-value of 0.0012 quantifies evidence against the null hypothesis under the specified two-sided stratified log-rank framework. It does not quantify the magnitude of benefit and should not be interpreted as a probability that the treatment hypothesis is true.
This is a biomarker-defined analysis, so its interpretation depends on the KRAS classification used for the analysis population and on the stated collection cutoff of 28 August 2009. The registry does not provide enough information in the ClinicalTrials.gov record to establish whether the KRAS subgroup effect formally differs from the effect in the KRAS mutant population through an interaction test.
9. Primary Results: KRAS Mutant PFS
The third primary endpoint evaluated PFS in the ITT population with KRAS mutant tumor status as collected until 28 August 2009.
KRAS mutant progression-free survival
95% CI: 0.887–1.544 · P = 0.2648
Two-sided stratified log-rank analysis.
| Feature | Reported result |
|---|---|
| Analysis population | Intent to Treat (ITT) population with KRAS Mutant tumor status as collected until 28 August 2009 |
| Groups compared | Cetuximab Plus FOLFIRI vs FOLFIRI Alone |
| Method | Stratified log-rank test |
| Kaplan-Meier method | Used to estimate median PFS time |
| Hazard-ratio model | Cox proportional hazards model stratified by randomization strata |
| Estimate | 1.171 |
| 95% CI | 0.887–1.544 |
| P-value | 0.2648 |
An HR of 1.171 is above 1, corresponding to an estimated instantaneous event rate about 17.1% higher in the cetuximab-plus-FOLFIRI group under the fitted model. This is a model-based relative estimate, not a statement that progression risk was 17.1% higher for every individual patient.
The 95% CI of 0.887–1.544 includes 1.000. Thus, the interval contains values compatible with lower, similar, and higher hazards for the cetuximab group.
The P-value of 0.2648 does not provide evidence against the null hypothesis at the conventional 5% level under the stated analysis. It does not prove that the treatment groups are identical, and it does not measure the size of the observed estimate.
Most importantly, comparing the KRAS wild-type P-value with the KRAS mutant P-value is not itself a formal test of treatment-effect heterogeneity. Establishing a difference in treatment effects between biomarker groups requires an appropriate interaction or treatment-by-biomarker comparison.
10. Secondary Results: Overall Survival
Overall survival was registered as a secondary time-to-event endpoint. The registry defines OS as time from randomisation to death or the last day known to be alive.
| Population | HR | 95% CI | P-value |
|---|---|---|---|
| ITT population | 0.878 | 0.774–0.995 | 0.0419 |
| KRAS wild-type population | 0.796 | 0.670–0.946 | 0.0093 |
| KRAS mutant population | 1.035 | 0.834–1.284 | 0.7549 |
Overall survival in the ITT population
Overall survival
95% CI: 0.774–0.995 · P = 0.0419
The registry reports a two-sided stratified log-rank test, with Kaplan-Meier estimation of median OS and a Cox proportional-hazards model stratified by the randomization strata.
The OS HR of 0.878 corresponds to an estimated instantaneous rate of death about 12.2% lower in the cetuximab-plus-FOLFIRI group than in the FOLFIRI-alone group under the fitted model.
The 95% CI of 0.774–0.995 is relatively close to 1.000 at its upper boundary. The confidence interval should be read together with the effect estimate rather than replaced by the P-value.
The P-value of 0.0419 is below the registry's stated 5% significance level for the two-sided stratified log-rank framework. It is evidence against the specified null hypothesis in that analysis, but it does not quantify treatment benefit.
KRAS wild-type overall survival
KRAS wild-type OS
95% CI: 0.670–0.946 · P = 0.0093
The KRAS wild-type OS analysis used the ITT population with KRAS wild-type tumor status as collected until 28 August 2009. The two-sided stratified log-rank test was used, with Kaplan-Meier estimation and a stratified Cox model.
KRAS mutant overall survival
KRAS mutant OS
95% CI: 0.834–1.284 · P = 0.7549
The KRAS mutant OS estimate is close to 1, while its confidence interval spans 1.000. The reported P-value does not provide evidence against the null hypothesis under the stated two-sided stratified analysis.
11. Secondary Results: Best Overall Response Rate
Best Overall Response Rate was evaluated by Independent Review Committee (IRC) assessments, with evaluations performed every 6 weeks until progression. The registry used stratified Cochran-Mantel-Haenszel methods and reported odds ratios.
| Population | Odds ratio | 95% CI | P-value |
|---|---|---|---|
| ITT population | 1.40 | 1.12–1.77 | 0.0038 |
| KRAS wild-type population | 2.069 | 1.515–2.826 | <0.0001 |
| KRAS mutant population | 0.822 | 0.544–1.242 | 0.3475 |
Why an odds ratio is used here
Response rate is binary: a participant either meets the registry's response criterion or does not. An odds ratio compares the odds of that binary outcome between treatment groups. Unlike a hazard ratio, it does not incorporate the timing of an event.
ITT response analysis
The reported OR was 1.40 with a 95% CI of 1.12–1.77 and P = 0.0038.
KRAS wild-type response
The reported OR was 2.069 with a 95% CI of 1.515–2.826 and P < 0.0001.
KRAS mutant response
The reported OR was 0.822 with a 95% CI of 0.544–1.242 and P = 0.3475.
Assessment schedule
Evaluations were performed every 6 weeks until progression.
An odds ratio above 1 for the ITT and KRAS wild-type analyses indicates higher estimated odds of the response outcome in the cetuximab-plus-FOLFIRI group. It should not be translated directly into a percentage-point difference in response rate without the underlying event counts or group-specific response proportions.
12. Secondary Results: Disease Control Rate
Disease Control Rate was another binary endpoint assessed by Independent Review Committee. The registry reports a stratified Cochran-Mantel-Haenszel analysis with an odds ratio of 0.92.
Disease control rate
95% CI: 0.67–1.26 · P = 0.6004
An odds ratio of 0.92 is close to 1. The 95% CI of 0.67–1.26 spans 1.000, and the reported P-value of 0.6004 does not provide evidence against the null hypothesis under the stated two-sided stratified CMH analysis.
The result also illustrates why response endpoints should be interpreted separately. Disease control rate and best overall response rate are distinct binary outcomes, and an estimate for one cannot be substituted for an estimate for the other.
13. Secondary Results: No Residual Tumor After Metastatic Surgery
The registry also reports the binary endpoint Participants With No Residual Tumor After Metastatic Surgery. The analysis used the Cochran-Mantel-Haenszel test in the ITT population.
No residual tumor after metastatic surgery
95% CI: 1.45–6.27 · P = 0.002
An OR of 3.02 means that the estimated odds of the specified binary outcome were approximately three times those in the comparator group within the analyzed population and CMH framework. It does not mean that three times as many participants achieved the outcome.
The 95% CI of 1.45–6.27 indicates uncertainty around the estimated odds ratio and excludes 1.000. The P-value of 0.002 provides evidence against the null hypothesis under the stated analysis, but it does not measure the size of the effect.
14. Secondary Endpoint Summary
| Endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Overall Survival Time (OS) | Hazard ratio | 0.878 | 0.774–0.995 | 0.0419 |
| Overall Survival Time (KRAS Wild-Type Population) | Hazard ratio | 0.796 | 0.670–0.946 | 0.0093 |
| Overall Survival Time (KRAS Mutant Population) | Hazard ratio | 1.035 | 0.834–1.284 | 0.7549 |
| Best Overall Response Rate | Odds ratio | 1.40 | 1.12–1.77 | 0.0038 |
| Best Overall Response Rate (KRAS Wild-Type) | Odds ratio | 2.069 | 1.515–2.826 | <0.0001 |
| Best Overall Response Rate (KRAS Mutant) | Odds ratio | 0.822 | 0.544–1.242 | 0.3475 |
| Disease Control Rate | Odds ratio | 0.92 | 0.67–1.26 | 0.6004 |
| Participants With No Residual Tumor After Metastatic Surgery | Odds ratio | 3.02 | 1.45–6.27 | 0.002 |
The ClinicalTrials.gov record contains 11 posted statistical analyses, of which 3 are primary-endpoint analyses. The table above presents the posted secondary analyses contained in the ClinicalTrials.gov recordset; no additional estimates have been inferred.
15. Statistical Methods Explained
Why was a stratified log-rank test used?
PFS and OS are time-to-event endpoints, so the treatment groups need to be compared over follow-up rather than simply comparing whether an event occurred by a fixed date. The stratified log-rank test compares event patterns over time while incorporating the randomization strata identified in the registry analysis notes.
What does a hazard ratio of 0.696 mean?
For the KRAS wild-type PFS analysis, an HR of 0.696 means the estimated instantaneous event rate in the cetuximab-plus-FOLFIRI group was 69.6% of the corresponding rate in the FOLFIRI-alone group under the stratified Cox model. The simple relative interpretation is approximately a 30.4% lower estimated hazard. It does not mean a 30.4% absolute reduction in progression.
Why does the confidence interval matter?
A point estimate such as 0.696 is only one estimate of the treatment effect. The 95% CI of 0.558–0.867 communicates the statistical uncertainty around that estimate. It is not a range in which individual patients' treatment effects are expected to fall.
Why is a response endpoint analyzed differently from PFS?
Best overall response rate is binary, whereas PFS incorporates event timing and censoring. CRYSTAL therefore used a stratified CMH test and odds ratio for response, while PFS used stratified log-rank testing and hazard ratios.
Why does an odds ratio of 2.069 not mean "twice the response rate"?
An odds ratio compares odds, not probabilities directly. When outcomes are common, the odds ratio can differ materially from the corresponding risk ratio. The ClinicalTrials.gov record does not provide the treatment-group response proportions needed to convert the reported OR into an absolute response-rate difference.
Does a significant KRAS wild-type result prove a treatment effect difference from KRAS mutant disease?
No. A statistically significant result in one subgroup and a nonsignificant result in another do not by themselves demonstrate heterogeneity. A formal interaction or treatment-by-KRAS analysis is needed to test whether the estimated treatment effects differ between subgroups.
Why is ITT important here?
The primary PFS analysis was based on the ITT population as defined in the registry. An analysis anchored to randomized assignment helps preserve the comparison created by randomization. At the same time, the registry definition specifies that subjects had received at least one dose of randomized treatment, so the exact registry definition should be retained rather than replacing it with a generic definition of ITT.
16. Understanding the PFS Endpoint
The CRYSTAL PFS definition combines several rules that determine whether and when an observation contributes an event or is censored.
Progression event
Radiological progression based on modified World Health Organisation (WHO) criteria is an event.
Death event
Death due to any cause is considered, but the registry specifies that only deaths within 60 days of the last tumor assessment are considered.
Censoring
Patients without an event are censored on the date of their last tumor assessment.
Assessment timing
The endpoint is therefore dependent on the timing and availability of tumor assessments as well as the occurrence of progression or death.
This structure is important because censoring is not simply a technical footnote. The statistical analysis relies on assumptions about the relationship between censoring and the event process. Violations of those assumptions can affect the interpretation of Kaplan-Meier estimates and Cox-model results.
17. Confidence Intervals and P-values
95% CI → How precise is that estimate?
P-value → How compatible are the data with the specified null hypothesis?
These quantities complement one another. A P-value is not an effect-size measure, and a confidence interval is not a prediction interval for individual patients.
The CRYSTAL results illustrate this distinction particularly clearly. The KRAS wild-type PFS analysis has HR 0.696 with 95% CI 0.558–0.867 and P = 0.0012, while the KRAS mutant PFS analysis has HR 1.171 with 95% CI 0.887–1.544 and P = 0.2648. These are different estimates with different levels of uncertainty, not simply a "significant" versus "nonsignificant" label.
18. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment group. The outcome is affected by the registry caveat that a nonspecific outcome measure called Safety was deleted from the entry in error and a replacement outcome was created; the replacement outcome refers to adverse events.
| Treatment group | Serious adverse events affected | At risk | Reported format |
|---|---|---|---|
| Cetuximab Plus FOLFIRI | 263 | 600 | 263/600 |
| FOLFIRI Alone | 204 | 602 | 204/602 |
These counts should not be converted into percentages here because the task is to report registry numbers exactly as reported in the registry. The denominators also should not be substituted for the overall enrollment of 1221; they are the at-risk denominators registry-reported specifically for the serious-adverse-event comparison.
19. What the Effect Measures Do — and Do Not — Mean
The HR is a relative time-to-event measure. For example, the KRAS wild-type PFS HR of 0.696 indicates a lower estimated instantaneous event rate in the cetuximab-plus-FOLFIRI group under the stratified Cox model. It does not state how many months an individual patient will remain progression-free.
The OR compares odds for a binary endpoint. The Best Overall Response Rate OR of 1.40 indicates higher estimated odds of the response outcome in the cetuximab-plus-FOLFIRI group in the overall analysis. It does not directly state the difference in response percentages.
A 95% CI describes uncertainty around the estimated parameter under the statistical model and sampling framework. For the OS analysis, the interval of 0.774–0.995 indicates an estimate close to the no-difference value of 1.000 at its upper boundary.
A P-value measures compatibility of the observed data with a specified null hypothesis under the stated statistical procedure. It is not the probability that the null hypothesis is true and it is not a measure of clinical importance.
20. Statistical Design Features Supported by the Registry
| Design topic | What the ClinicalTrials.gov record establishes |
|---|---|
| Randomization | Allocation was RANDOMIZED. |
| Parallel design | Design model was PARALLEL. |
| Masking | Masking was NONE. |
| Superiority | The posted primary and secondary statistical analyses use a superiority hypothesis type. |
| Stratification | Analyses considered randomization strata of region and KPS: <80 vs. ≥80. |
| Interim analysis | No interim-analysis method is reported in the ClinicalTrials.gov record. |
| Multiplicity | No multiplicity-adjustment procedure is reported in the ClinicalTrials.gov record. |
| Non-inferiority margin | No non-inferiority analysis or margin is reported; the posted hypothesis type is superiority. |
| Crossover | No crossover information is reported in the ClinicalTrials.gov record. |
| Factorial design | No factorial structure is reported; the registered design model is PARALLEL. |
| Bayesian methods | No Bayesian method is reported in the ClinicalTrials.gov record. |
| Missing-data/imputation method | No imputation method is reported in the ClinicalTrials.gov record. |
This distinction is deliberate. Absence of a registry-reported method is not evidence that the trial did not address the issue in its protocol or statistical analysis plan. It means only that the ClinicalTrials.gov record do not support a more specific description.
21. Longitudinal Trial History
Trial start
The registry records the start of CRYSTAL in May 2004.
Primary completion
The registry records primary completion in December 2006.
KRAS analysis collection cutoff
The statistical analyses posted on ClinicalTrials.gov specify this tumor-status collection date for the KRAS wild-type and KRAS mutant analyses.
Results posted
The registry status is COMPLETED, with 15 outcome measures and 11 statistical analyses posted in the ClinicalTrials.gov record.
22. Why This Trial Matters Statistically
CRYSTAL is a useful teaching case because its registry results connect two major families of clinical-trial methods: time-to-event analysis for PFS and OS, and stratified categorical analysis for response-related outcomes.
| Concept | How it appears in CRYSTAL |
|---|---|
| Randomization | Randomized allocation in a parallel phase 3 treatment trial. |
| ITT analysis | Primary PFS analysis uses the registry-defined ITT population. |
| Time-to-event endpoints | PFS is defined from randomization to progression, qualifying death, or censoring; OS is from randomization to death or last day known alive. |
| Kaplan-Meier estimation | Used to estimate median PFS and median OS in the analyses posted on ClinicalTrials.gov. |
| Stratified log-rank test | Used for primary PFS and secondary OS comparisons. |
| Cox proportional-hazards model | Used to calculate HRs with stratification by randomization strata in the registry-reported KRAS analyses and OS analyses. |
| Hazard ratio | Primary and secondary time-to-event effects are expressed as HRs. |
| Cochran-Mantel-Haenszel test | Used for binary response and disease-control outcomes. |
| Odds ratio | Used for Best Overall Response Rate, Disease Control Rate, and no residual tumor after metastatic surgery. |
| Stratified analysis | Region and KPS strata are incorporated into the reported analyses. |
| Biomarker-defined analysis | PFS, OS, and response are separately reported for KRAS wild-type and KRAS mutant populations. |
| Endpoint interpretation | Time-to-event and binary endpoints require different effect measures and statistical methods. |
23. Limitations and Interpretation Issues
- Registry-only numerical scope: this analysis is intentionally restricted to the numbers and methods in the ClinicalTrials.gov record. It does not substitute publication results that are not contained in that dataset.
- Subgroup interpretation: KRAS-specific estimates should not be treated as formal evidence of effect modification without an appropriate interaction analysis.
- Hazard-ratio assumptions: Cox-model HRs are model-based and rely on the proportional-hazards framework. A single HR can be difficult to summarize if hazards vary materially over time.
- Censoring rules: the registered PFS definition has a specific 60-day rule for deaths after the last tumor assessment, which directly affects event classification.
- Analysis-population distinction: the registry's ITT definition specifies randomized subjects who have received at least one dose of randomized treatment. It should not automatically be replaced with a different textbook definition.
- Multiplicity: the ClinicalTrials.gov record identifies three primary endpoints and several secondary analyses but do not provide a multiplicity-adjustment strategy. The individual P-values should therefore be interpreted in the context of the registered analysis structure rather than treated as a collection of independent confirmatory tests.
- Safety data provenance: the registry notes an error involving deletion and replacement of the nonspecific "Safety" outcome measure.
- No inferred missing methods: the ClinicalTrials.gov record does not establish crossover, interim monitoring, imputation, Bayesian methods, or a non-inferiority margin, so none are inferred.
24. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The overall PFS analysis reported HR 0.853, 95% CI 0.728–1.000, and P = 0.0479. The KRAS wild-type PFS analysis reported HR 0.696, while the KRAS mutant analysis reported HR 1.171. These estimates describe different populations and should not be reduced to P-values alone.
Clinical interpretation
The registry data provide evidence on progression-free survival, overall survival, response, disease control, and a surgical residual-tumor outcome. They also provide serious adverse-event counts by treatment group. The ClinicalTrials.gov record does not provide enough numerical detail to construct a complete benefit-risk assessment or to report median survival values.
25. A Practical Reading of the CRYSTAL Results
A disciplined reading starts with the endpoint before looking at the P-value. For PFS, the endpoint is time from randomization to progression, qualifying death, or censoring. The next question is the analysis population, followed by the treatment comparison and the effect measure.
| Step | Question to ask | CRYSTAL example |
|---|---|---|
| 1 | What outcome is being measured? | Progression-free survival, overall survival, response, disease control, or no residual tumor after metastatic surgery. |
| 2 | Who is being analyzed? | Overall ITT population or the specified KRAS wild-type / mutant population. |
| 3 | How is the endpoint analyzed? | Stratified log-rank and Cox methods for time-to-event outcomes; CMH for binary outcomes. |
| 4 | What is the effect measure? | Hazard ratio for time-to-event outcomes; odds ratio for binary outcomes. |
| 5 | How precise is the estimate? | Read the 95% confidence interval alongside the point estimate. |
| 6 | What does the P-value answer? | Compatibility with the specified null hypothesis under the stated test; not effect size. |
| 7 | Is the subgroup comparison formal? | Not unless an interaction or heterogeneity test is reported. |
26. Related Tutorials
Learn more about the methods used in this trial:
27. Related Statistical Calculators
28. Sources
- ClinicalTrials.gov: NCT00154102 — CRYSTAL.
- Linked PubMed record: PMID 19339720.
- Linked PubMed record: PMID 30657405.
- Linked PubMed record: PMID 27722750.
- Linked PubMed record: PMID 23265711.
Continue through the Clinical Biostats statistical library
Explore the underlying survival-analysis, categorical-data, and clinical-trial methods used to interpret randomized evidence.
29. Record Summary
CRYSTAL provides a compact example of how randomized clinical-trial evidence can require several statistical frameworks at once. The primary endpoints were three PFS analyses: the overall ITT population, the KRAS wild-type population, and the KRAS mutant population. Each used stratified log-rank testing, with hazard ratios derived through stratified Cox models in the registry-reported subgroup analyses. Secondary endpoints extended the same time-to-event framework to overall survival while using stratified Cochran-Mantel-Haenszel testing and odds ratios for binary outcomes such as Best Overall Response Rate and Disease Control Rate.
The most important statistical lesson is that the effect measure, analysis population, confidence interval, P-value, and endpoint definition must be interpreted together. The overall PFS HR of 0.853, KRAS wild-type PFS HR of 0.696, and KRAS mutant PFS HR of 1.171 are estimates from different populations and should not be converted into a simple subgroup ranking without a formal interaction analysis. Likewise, the response ORs cannot be interpreted as hazard ratios, and the safety denominators cannot be substituted for overall enrollment.