This page separates reported registry information from statistical interpretation. The ClinicalTrials.gov record identifies a posted primary endpoint but do not provide a formal statistical analysis record for that endpoint.
1. Trial at a Glance
GEOMETRY mono-1 was a phase 2, non-randomized, parallel study of capmatinib in patients with EGFR wild-type, advanced non-small cell lung cancer. The study enrolled 373 participants across 10 arms and has a registered binary primary endpoint based on overall response rate assessed by a blinded independent review committee.
| Feature | GEOMETRY mono-1 |
|---|---|
| Trial name | GEOMETRY mono-1 |
| NCT ID | NCT02414139 |
| Phase | Phase 2 |
| Condition | Carcinoma, Non-Small-Cell Lung |
| Intervention | Capmatinib (drug) |
| Allocation | Non-randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 373 |
| Number of arms | 10 |
| Status | Completed |
| Lead sponsor | Novartis Pharmaceuticals |
| Sponsor type | Industry |
2. Clinical Question
The study evaluated capmatinib in patients with EGFR wild-type, advanced non-small cell lung cancer. Because the registry describes the allocation as non-randomized, the central statistical question is different from the question posed by a randomized comparative trial: the registered primary endpoint asks how frequently participants achieved a prespecified objective tumor response under blinded independent review.
Population
Patients with EGFR wild-type, advanced non-small cell lung cancer.
Intervention
Capmatinib, an oral cMET inhibitor.
Comparator
The ClinicalTrials.gov record does not identify a randomized comparator. The allocation is registered as non-randomized.
Primary question
What percentage of participants achieved a confirmed complete or partial response according to the registered BIRC assessment framework?
3. Trial Design
The combination of a non-randomized allocation and multiple arms is statistically important. A multi-arm structure can permit evaluation of different cohorts or treatment contexts, but without randomization, observed differences between cohorts cannot automatically be interpreted as causal treatment effects. The statistical unit of interpretation therefore matters: an observed response proportion is not equivalent to a randomized treatment contrast.
4. Study Population and Cohort Structure
The registry profile reports 373 participants and 10 arms. The serious-adverse-event information reported with the record identifies multiple cohorts, including Cohorts 1a, 1b, 2, 3, 4, 5a, 5b, 6.1, and 6.2.
| Registered cohort information reported | Serious adverse events, on-treatment |
|---|---|
| Cohort 1a | 42/69 |
| Cohort 1b | 21/42 |
| Cohort 2 | 30/54 |
| Cohort 3 | 15/30 |
| Cohort 4 | 38/69 |
| Cohort 5a | 9/15 |
| Cohort 5b | 14/28 |
| Cohort 6.1 (Expansion of Cohort 1a) | 2/3 |
5. Primary Endpoint
| Endpoint | Registered definition / assessment | Time frame |
|---|---|---|
| Overall Response Rate (ORR) by Blinded Independent Review Committee (BIRC) Assessment | Percentage of participants with a best overall response defined as confirmed complete response (CR) or partial response (PR) by BIRC assessment per RECIST 1.1. CR: disappearance of all non-nodal target and non-target lesions. In addition, any pathological lymph nodes assigned as target and non-target lesions must have a reduction in short axis to < 10 mm. PR: at least a 30% decrease in the sum of target-lesion measurements. | Up to approximately 5 years |
The primary endpoint is binary at the participant level: each participant is classified according to whether the best overall response meets the registered CR or PR definition. The resulting ORR is a proportion, not a time-to-event measure.
6. Why the Primary Endpoint Is Statistically Different From Overall Survival
ORR and survival endpoints require different statistical thinking. For ORR, the basic quantity is the proportion of participants meeting a response definition. There is no need to estimate a hazard function simply to calculate the response rate. By contrast, overall survival requires accounting for the timing of deaths and for participants who are censored before the event occurs.
The numerator is determined by the registered BIRC/RECIST 1.1 response classification. The denominator must be defined by the applicable analysis population; it should not be silently assumed to be the full enrollment count.
This distinction is especially important for a non-randomized, multi-cohort study. A response rate can describe the observed antitumor activity in an analyzed population, but by itself it does not establish that capmatinib caused the observed responses relative to an untreated or alternative-treatment counterfactual.
7. Planned Analysis
The ClinicalTrials.gov record indicates that results have been posted, but no formal statistical analyses were posted. Consequently, the ClinicalTrials.gov record does not provide a formal effect estimate, confidence interval, or p-value for the primary ORR endpoint.
For a binary ORR endpoint, a conventional analysis would begin with an estimate of the response proportion in the prespecified analysis population. An appropriate confidence interval for a binomial proportion would then quantify sampling uncertainty. If the statistical objective were to compare a response rate with a prespecified benchmark, a formal binomial hypothesis test or an equivalent model-based procedure could be used, provided that such a benchmark and testing framework had been prespecified.
What can be measured
The registered endpoint measures the percentage of participants whose best overall response is a confirmed CR or PR under BIRC assessment per RECIST 1.1.
What is not reported here
The ClinicalTrials.gov record does not report a formal ORR estimate with a confidence interval or p-value.
Typical uncertainty measure
A confidence interval for a binomial proportion would describe statistical uncertainty around the observed response-rate estimate.
Comparative inference
Because allocation is non-randomized, a comparison across cohorts would require particular care about differences in patient composition and treatment context.
8. Statistical Methodology
Binary endpoint analysis
The registered primary endpoint is binary at the participant level. Each participant has a best overall response classification, and the primary outcome is the proportion meeting the confirmed CR or PR definition.
Binomial proportion
If x participants among n evaluable participants have a qualifying response, the observed response rate is
The estimate describes the observed proportion in the analysis population. It is not automatically a causal treatment effect.
Confidence intervals
A confidence interval around the response rate quantifies uncertainty associated with estimating a population proportion from the observed sample. For a clinical-trial response endpoint, an exact or score-based binomial interval is generally preferable to relying automatically on a simple normal approximation, particularly when the response count or non-response count is small.
Blinded Independent Review Committee assessment
The registered endpoint uses BIRC assessment. Independent review can reduce the influence of investigator expectations when classifying radiologic response. It does not, however, transform a non-randomized study into a randomized comparison.
RECIST 1.1 response classification
The primary endpoint is anchored to RECIST 1.1 and requires a confirmed CR or PR. This creates a standardized rule for translating tumor measurements into a binary response outcome. The statistical analysis therefore depends on the quality and timing of the underlying radiologic assessments as well as the prespecified rules for confirmation.
Multi-cohort interpretation
Ten registered arms and multiple named cohorts create a potential issue of differing denominators and patient characteristics. If response rates are summarized separately by cohort, each estimate has its own sample size and uncertainty. Pooling cohorts can also change the estimand and should be based on a prespecified rationale rather than performed simply to increase the denominator.
9. Statistical Methods Explained
What does a binary ORR endpoint measure?
It classifies each participant according to whether their best overall response is a confirmed CR or PR, then summarizes the percentage meeting that definition. It does not measure how long a response lasts unless a separate duration-of-response endpoint is analyzed.
Why is BIRC assessment important?
A blinded independent review committee provides an independent assessment of radiologic response. This can reduce assessment bias compared with an entirely investigator-determined endpoint, although the underlying study remains non-randomized.
Why is RECIST 1.1 used?
A standardized response framework makes the binary outcome reproducible. In this registry definition, CR requires disappearance of specified lesions and reduction of pathological lymph nodes to < 10 mm in short axis, while PR requires at least a 30% decrease in the sum of target-lesion measurements.
Why should the denominator be examined carefully?
An ORR is always a numerator divided by a defined analysis population. In a multi-cohort study, the denominator may differ across analyses depending on eligibility, evaluability, or the prespecified response population. Using total enrollment without checking the analysis definition can produce a misleading rate.
Why doesn't an observed response rate prove causality?
Because GEOMETRY mono-1 is registered as non-randomized. Without random allocation, patients receiving treatment can differ systematically in characteristics associated with prognosis or response. The observed response proportion therefore describes what happened in the analyzed population; it does not by itself identify what would have happened to the same participants under another treatment.
What would a confidence interval add?
The point estimate alone gives the observed response proportion. A confidence interval adds information about statistical precision. A narrow interval indicates greater precision than a wide interval under the same inferential framework, but neither interval width nor the point estimate alone establishes clinical causality.
10. Safety Results
The ClinicalTrials.gov record contains serious-adverse-event information for several named cohorts. These are reported as affected participants divided by participants at risk, with the registry description identifying them as on-treatment serious adverse events.
| Cohort | Serious adverse events, on-treatment | Observed fraction |
|---|---|---|
| Cohort 1a | 42/69 | 42 affected / 69 at risk |
| Cohort 1b | 21/42 | 21 affected / 42 at risk |
| Cohort 2 | 30/54 | 30 affected / 54 at risk |
| Cohort 3 | 15/30 | 15 affected / 30 at risk |
| Cohort 4 | 38/69 | 38 affected / 69 at risk |
| Cohort 5a | 9/15 | 9 affected / 15 at risk |
| Cohort 5b | 14/28 | 14 affected / 28 at risk |
| Cohort 6.1 | 2/3 | 2 affected / 3 at risk |
The denominators differ substantially among the registry-reported cohorts. This matters statistically because an adverse-event proportion based on a small cohort has greater sampling variability than an otherwise similar proportion based on a larger cohort.
11. Safety Denominators and Statistical Precision
Safety analyses require explicit attention to the population at risk. An adverse-event proportion is conditional on exposure and follow-up, and the denominator can differ from total enrollment. In this study, the ClinicalTrials.gov record explicitly uses cohort-specific affected and at-risk counts.
Small denominators
Cohort 6.1 has 2 affected participants among 3 at risk. Such a small denominator means that the observed fraction is highly sensitive to a single participant.
Different cohorts
The cohort-specific denominators range from 15 to 69 among the complete entries reported, so direct visual comparison requires attention to precision.
On-treatment definition
The registry-reported safety field identifies these serious adverse events as on-treatment, which distinguishes them from an unrestricted all-study-period adverse-event summary.
Pooling
Pooling safety counts across cohorts would require a clear definition of the target population and confirmation that the underlying exposure and reporting periods are compatible.
12. Non-Randomized Design and Causal Interpretation
The most important statistical feature of GEOMETRY mono-1 is its non-randomized allocation. Randomization is not simply a procedural detail: it is the mechanism that makes treatment groups comparable in expectation with respect to both measured and unmeasured baseline characteristics.
In a non-randomized study, an observed response rate can still be scientifically informative. However, the inferential question is narrower. The response rate describes the frequency of RECIST-defined responses among the studied participants. It does not, without additional assumptions and design features, provide the same causal contrast that would arise from randomizing participants between capmatinib and a comparator.
The first can be estimated from a single treated population. The second requires an appropriate comparator and a design or modeling strategy that supports the intended causal comparison.
This distinction should remain visible when interpreting a multi-cohort phase 2 oncology study. A high response proportion can coexist with uncertainty about the counterfactual response that the same population would have experienced under another therapy.
13. Multiplicity and Multiple Arms
The registry profile identifies 10 arms but does not provide a statistical-analysis record describing a multiplicity strategy. Therefore, no specific alpha allocation, hierarchical testing procedure, or multiplicity-adjusted p-value should be attributed to the study from the ClinicalTrials.gov record.
| Issue | Statistical relevance |
|---|---|
| Multiple arms | Creates multiple possible cohort-level summaries and comparisons. |
| Primary endpoint | One registered primary endpoint is identified: ORR by BIRC assessment. |
| Formal alpha allocation | Not reported in the ClinicalTrials.gov record. |
| Multiplicity adjustment | No formal adjustment is reported in the ClinicalTrials.gov record. |
| Interpretation | Individual exploratory comparisons should not automatically be treated as confirmatory evidence without a prespecified testing framework. |
The absence of a registry-reported multiplicity procedure does not establish that no such procedure existed elsewhere in the study documentation. It means only that the provided ClinicalTrials.gov statistical-analysis extract does not supply one, so this page does not infer it.
14. Interim Analysis, Crossover, and Bayesian Methods
The ClinicalTrials.gov record does not report an interim-analysis procedure, crossover design, factorial structure, missing-data or imputation strategy, stratification factors, non-inferiority margin, or Bayesian analysis method.
Interim analysis
No interim-analysis method is provided in the ClinicalTrials.gov record.
Crossover
No crossover procedure is identified in the ClinicalTrials.gov record.
Missing data
No imputation method or missing-data strategy is provided in the ClinicalTrials.gov record.
Bayesian methods
No Bayesian statistical method is identified in the ClinicalTrials.gov record.
These omissions are important because each method would materially change how a statistical analysis should be interpreted. A response endpoint with missing assessments, for example, requires a prespecified rule for defining the analysis population and handling unevaluable participants. Without that rule, simply assuming that all enrolled participants form the denominator can be inappropriate.
15. Confidence Intervals and the Meaning of Statistical Precision
No confidence interval for the primary ORR is reported in the ClinicalTrials.gov record. Nevertheless, the statistical role of a confidence interval is central to understanding how a binary endpoint should be reported.
The ORR point estimate answers: what proportion of the analyzed participants met the response definition? It is a descriptive summary of the observed sample.
A confidence interval answers a different question: how much sampling uncertainty surrounds the estimated response proportion under the chosen inferential framework? It does not describe the range of responses experienced by individual patients.
A p-value, if a formal hypothesis test had been specified, would quantify compatibility with a null hypothesis under the chosen testing framework. It would not measure the magnitude of the response rate and would not substitute for reporting the response estimate and its uncertainty.
16. Clinical Biostats Interpretation of the Primary Endpoint
The primary endpoint is particularly useful for teaching the difference between measurement and causal inference. BIRC assessment and RECIST 1.1 provide a structured measurement process: participants either meet the registered confirmed CR/PR criterion or they do not. The resulting ORR is therefore a clearly defined binary summary.
However, the study's non-randomized allocation changes the inferential context. The ORR can characterize antitumor activity in the enrolled population, but it should not automatically be expressed as a percentage reduction in risk, a treatment effect versus placebo, or a causal benefit relative to another therapy.
The confidence interval, when available, should be presented alongside the ORR rather than treating the point estimate as exact. The p-value, if a prespecified hypothesis test exists, should likewise be interpreted as evidence relative to a null hypothesis rather than as a measure of clinical importance.
17. Limitations
- Non-randomized allocation: treatment assignment was not randomized, limiting the causal interpretation of observed response differences across cohorts.
- Multiple arms: the study contains 10 arms, creating potential differences in cohort composition and precision.
- Incomplete statistical-analysis record: no formal statistical analyses were posted, so formal effect estimates, confidence intervals, and p-values for the primary endpoint are not available in the ClinicalTrials.gov record.
- Endpoint denominator: total enrollment of 373 should not automatically be assumed to be the denominator for ORR without the prespecified response analysis population.
- Safety denominators: the registry-reported serious-adverse-event data use cohort-specific at-risk populations, which vary substantially in size.
- Missing methodological details: the ClinicalTrials.gov record does not specify an interim-analysis method, crossover procedure, imputation strategy, stratification factors, Bayesian methods, or multiplicity adjustment.
- Response assessment: ORR captures whether a participant achieved a confirmed CR or PR; it does not by itself characterize the duration of response or overall survival.
18. Why This Trial Matters Statistically
GEOMETRY mono-1 is a useful teaching example because it illustrates how statistical interpretation changes when a phase 2 oncology study uses a non-randomized, multi-cohort design and a binary tumor-response endpoint.
| Concept | How it appears in GEOMETRY mono-1 |
|---|---|
| Binary endpoint | ORR is defined from confirmed CR or PR according to BIRC assessment. |
| RECIST 1.1 | Provides the response classification framework used for the primary endpoint. |
| Independent review | BIRC assessment is specified for the primary ORR endpoint. |
| Non-randomization | The registry identifies allocation as non-randomized. |
| Multi-arm design | The study contains 10 arms. |
| Binomial inference | ORR is naturally summarized as a response proportion with corresponding uncertainty. |
| Analysis population | The denominator for ORR must be defined by the applicable response analysis population. |
| Safety denominators | Serious adverse events are reported as affected/at-risk counts by cohort. |
| Statistical reporting | The registry-reported statistical-analysis field contains no formal primary-endpoint analysis records. |
| Causal inference | Observed response rates should not automatically be interpreted as randomized treatment effects. |
19. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
20. Related Statistical Calculators
21. Sources
- ClinicalTrials.gov: NCT02414139 — GEOMETRY mono-1.
- Linked publication: PubMed PMID 39362249.
- Linked publication: PubMed PMID 36822130.
- Linked publication: PubMed PMID 32877583.
Continue through the Clinical Biostats statistical library
Explore tutorials and calculators covering response endpoints, binomial inference, confidence intervals, and clinical-trial methodology.
22. Record Summary
GEOMETRY mono-1 provides a useful example of statistical analysis for a phase 2, non-randomized, multi-arm oncology study with a binary primary endpoint. The registered primary endpoint is Overall Response Rate (ORR) by Blinded Independent Review Committee (BIRC) Assessment, defined through confirmed CR or PR under RECIST 1.1 and measured over up to approximately 5 years.
The ClinicalTrials.gov record reports 373 enrolled participants, 10 arms, capmatinib as the intervention, and a completed study status. They also provide cohort-level serious-adverse-event counts for several cohorts.
The central statistical lesson is therefore methodological: a response proportion is a binary endpoint that can be estimated with binomial methods, but its interpretation depends on the analysis population, response-assessment rules, uncertainty interval, and study design. Because allocation was non-randomized, observed response rates should be distinguished from causal treatment effects.