CHRYSALIS: Complete Statistical Analysis of Amivantamab in Advanced Non-Small Cell Lung Cancer

1. Trial at a Glance

CHRYSALIS is a phase 1, non-randomized, parallel-design treatment study in participants with advanced non-small cell lung cancer. The registry lists 751 participants, 2 arms, no masking, and a primary focus on treatment evaluation.

2. Clinical Question

The statistical question in CHRYSALIS is primarily descriptive and exploratory rather than a randomized treatment-effect comparison. The registered endpoints evaluate dose-limiting toxicity, adverse events and serious adverse events, tumor response, duration of response, clinical benefit, and pharmacokinetic exposure to amivantamab.

3. Trial Design

4. Registered Primary Endpoints

The registry lists multiple primary endpoints across different study components. They span acute toxicity, broader safety, antitumor activity, durability of response, clinical benefit, and pharmacokinetics. This endpoint structure is characteristic of an early-phase study in which safety and characterization of treatment activity are central objectives.

5. Planned Analysis

The registry has not posted formal statistical analyses for these primary endpoints. The appropriate analysis therefore follows from the nature of each registered endpoint rather than from reported outcome estimates.

Dose-limiting toxicity

The Part 1 DLT endpoint is a participant-level safety outcome assessed through Day 28. A natural primary summary is the number and percentage of participants meeting the prespecified DLT definition. Because DLT is based on specific drug-related toxicities, the analysis should preserve the clinical definition rather than collapsing all adverse events into a generic toxicity measure.

Adverse events and serious adverse events

Part 2 safety outcomes are ordinarily summarized using participant counts and percentages, with adverse events and serious adverse events tabulated separately. The follow-up window extends from screening through 30 [+7] days after the last dose. This time definition matters because an event occurring outside the defined observation period would not necessarily belong to the same primary safety summary.

Overall response rate

ORR is a binary participant-level efficacy outcome based on RECIST v1.1 classification. The basic estimand is the proportion of participants achieving CR or PR within the specified assessment period:

Duration of response

DOR is a time-to-event endpoint defined only among participants who achieve CR or PR. The event is progressive disease or death due to underlying disease, whichever occurs first. Participants without the defined event by the end of observation require appropriate censoring under the prespecified analysis convention.

Clinical benefit

The clinical-benefit endpoint extends beyond CR and PR by also including durable stable disease under the registry definition. Its primary summary would ordinarily be the percentage of participants meeting one of those response or disease-control criteria during the registered time frame.

Pharmacokinetic endpoints

Ctrough and AUCtau are continuous pharmacokinetic measures. Ctrough represents the observed serum concentration immediately before the next administration, whereas AUCtau represents exposure across a dosing interval. These measures are generally summarized descriptively, often with distributional summaries appropriate to concentration and exposure data. The registry does not specify a particular statistical model for these endpoints.

6. Statistical Methodology

Descriptive analysis for an early-phase study

The non-randomized phase 1 structure changes the role of statistics compared with a randomized confirmatory trial. Counts, percentages, response proportions, toxicity frequencies, and pharmacokinetic summaries can characterize what was observed. They do not, by themselves, establish that an observed difference between non-randomized groups was caused by treatment.

Binary endpoints

DLT, AE/SAE occurrence, ORR, and clinical benefit can all be represented at the participant level as binary outcomes. Their descriptive analysis focuses on the number of participants meeting the endpoint and the corresponding proportion. Confidence intervals are useful because a proportion calculated from a finite sample is an estimate rather than a known population quantity.

RECIST-based response assessment

ORR depends on categorical tumor-response assessments under RECIST v1.1. The statistical analysis therefore begins with consistent classification of complete response, partial response, stable disease, and progressive disease. The clinical definition is part of the estimand: changing the response criteria or assessment window would change the endpoint being measured.

Time-to-event analysis for DOR

DOR differs from ORR because it measures how long a response persists. A participant can contribute substantial information to a DOR analysis even when the response has not ended by the last assessment, provided the censoring rule is prespecified. Kaplan-Meier estimation is a standard way to describe such a distribution when sufficient event and censoring information are available.

Pharmacokinetic exposure

AUCtau and Ctrough answer different exposure questions. Ctrough samples the concentration immediately before another dose, whereas AUCtau summarizes exposure throughout a complete dosing interval. Because these are continuous measurements, their distributions and the presence of skewness are important when selecting descriptive summaries or downstream models.

7. Statistical Methods Explained

Why is DLT assessed through Day 28?

The registry defines the Part 1 DLT time frame as up to Day 28. This creates a fixed early observation window for identifying dose-limiting toxicity. Statistically, a fixed window makes the denominator and observation period explicit and helps distinguish early dose-limiting toxicity from the broader safety experience collected during later follow-up.

Why is ORR a percentage rather than a mean?

ORR is based on a categorical response classification: a participant either achieves CR or PR or does not. The natural summary is therefore a proportion. A mean is not the appropriate descriptive measure for a binary response endpoint.

Why does DOR use only participants who respond?

The registry defines DOR from initial CR or PR to progressive disease or death due to underlying disease and specifies that the endpoint applies only to participants who achieve CR or PR. A participant who never responds cannot have a duration of response under that definition.

Why is censoring important for DOR?

A responder may remain in response at the last available assessment. That participant has not necessarily experienced the event defining the end of response. Treating such a participant as if progression occurred at the last assessment would bias the distribution. Time-to-event methods instead distinguish an observed event from right censoring.

Why should non-randomized groups not be interpreted like randomized treatment arms?

With non-randomized allocation, treatment groups can differ systematically before treatment begins. A difference in response or toxicity rates may therefore reflect treatment, baseline prognosis, treatment selection, or other factors. Statistical adjustment can address measured covariates in some settings, but it cannot automatically reproduce the protection against unmeasured confounding provided by randomization.

What do Ctrough and AUCtau measure?

Ctrough measures the observed serum concentration immediately before the next administration. AUCtau measures the area under the serum concentration-time curve during a dosing interval. Together, they describe different aspects of pharmacokinetic exposure: a concentration at a specific point in the dosing cycle versus cumulative exposure over the interval.

8. Interpreting the Endpoint Structure

This endpoint architecture illustrates why a single summary statistic would be inadequate for an early-phase study. Safety, tumor response, response durability, disease control, and pharmacokinetic exposure describe different dimensions of treatment experience and should remain analytically distinct.

9. Limitations and Interpretation Issues

• Non-randomized allocation: the registry classifies allocation as non-randomized. Observed differences between groups therefore cannot automatically be interpreted as randomized causal treatment effects.
• Multiple endpoint families: the registry lists safety, response, clinical-benefit, duration-of-response, and pharmacokinetic endpoints. These address different questions and should not be collapsed into a single efficacy or safety conclusion.
• Part-specific endpoints: DLT is defined for Part 1, while several safety and efficacy endpoints are defined for Part 2. The populations and observation structures should therefore be kept distinct when analyzing the study.
• Response assessment: ORR depends on RECIST v1.1 classifications. Differences in assessment timing or evaluability can affect the observed response proportion.
• Responder conditioning: DOR is defined only among participants who achieve CR or PR. It therefore describes durability among responders rather than the probability of responding in the full enrolled population.
• Censoring: DOR requires appropriate handling of participants who remain in response at the end of observation or otherwise do not experience the defined event during follow-up.
• Pharmacokinetic distributions: Ctrough and AUCtau are continuous measurements and may have distributions for which means alone provide an incomplete description.
• Generalizability: phase 1 eligibility criteria and the non-randomized study structure determine which participants are represented and limit direct extrapolation to populations not represented in the study.
• No posted formal analyses: the ClinicalTrials.gov record does not report formal statistical analyses for the registered primary endpoints.

10. Why This Trial Matters Statistically

CHRYSALIS is a useful teaching case because it demonstrates how statistical methodology changes when the purpose of a clinical trial is early treatment characterization rather than randomized confirmation of a treatment effect. The study combines categorical safety endpoints, RECIST-based response assessment, time-to-event response durability, clinical-benefit classification, and pharmacokinetic exposure measures.

11. Statistical Interpretation: What Can Be Learned From the Design?

12. Statistical Interpretation vs Clinical Interpretation

13. Record Summary

CHRYSALIS is a phase 1, non-randomized, parallel study enrolling 751 participants with advanced non-small cell lung cancer. Its registered primary endpoints span seven distinct measurements: Part 1 DLT; Part 2 AEs and serious AEs; ORR; DOR; clinical benefit; amivantamab Ctrough; and amivantamab AUCtau. Statistically, the study is best understood as an early-phase characterization program in which safety, response, durability, clinical benefit, and pharmacokinetic exposure require different analytical frameworks.

14. Sources

• ClinicalTrials.gov: NCT02609776 — CHRYSALIS.