NAVIGATE: Complete Statistical Analysis of Larotrectinib in NTRK-Fusion Positive Solid Tumors

1. Trial at a Glance

NAVIGATE was a phase 2, nonrandomized, parallel-design study evaluating larotrectinib in adults and children with solid tumors harboring NTRK fusion. The registry records 215 participants enrolled across 11 arms and identifies best overall response as the primary endpoint.

2. Clinical Question

The central statistical question in NAVIGATE is whether treatment with larotrectinib produces confirmed tumor responses in adults and children whose solid tumors harbor an NTRK fusion. The registry defines the primary endpoint in terms of the patient's best overall response, with complete response or partial response determined by independent radiologic review.

3. Trial Design

Study timeline

4. Endpoints

The endpoint is a categorical response outcome rather than a time-to-event endpoint. The key distinction is that the registry defines the endpoint around the best overall response observed during the specified assessment period, rather than defining the primary outcome as time until progression, time until death, or another event time.

Independent radiology review

The endpoint is determined by an independent radiology review committee. This is statistically important because response classification is not simply a clinician's unblinded impression; the registry specifies an independent review process for determining whether the tumor response meets the relevant criteria.

RECIST v1.1 and RANO

The registry allows RECIST v1.1 or RANO criteria as appropriate to tumor type. These criteria provide structured rules for classifying tumor response. The use of criteria appropriate to the tumor type recognizes that radiologic response assessment is not necessarily identical across all solid tumors and relevant imaging contexts.

5. Statistical Methodology

The registry identifies the primary endpoint and its assessment framework but does not report a posted statistical analysis for that endpoint. For this type of endpoint, the natural statistical summary is the proportion of evaluable participants who achieve a confirmed CR or PR, accompanied by an appropriate confidence interval.

Response proportion

If \(R\) participants have a confirmed CR or PR among \(N\) participants included in the response analysis, the observed response proportion is

The distinction between enrollment and the denominator used for a response analysis is important. A registry enrollment number describes the number enrolled in the study; it does not, by itself, establish the number of participants who would necessarily enter a particular response-evaluable population.

Confidence intervals for a response proportion

A binomial confidence interval can be used to quantify uncertainty around an observed response proportion. Exact binomial intervals or other prespecified binomial methods are commonly used, particularly when the number of responses is small or the proportion is near 0 or 1.

Why the analysis population matters

For a nonrandomized response study, the denominator must be defined carefully. A response proportion among all enrolled participants answers a different question from a response proportion among participants who were evaluable for radiologic assessment. Statistical interpretation therefore depends on the prespecified definition of the response analysis population.

Independent review and measurement error

Radiologic response is a classification derived from imaging assessments. Independent review is intended to provide a consistent assessment framework. Nevertheless, response classification can still involve measurement variability, image availability, timing of assessments, and the criteria used to classify changes in tumor burden.

6. Planned Analysis

The registry's primary endpoint is best overall response defined as confirmed CR or PR by an independent radiology review committee using RECIST v1.1 or RANO criteria, as appropriate to tumor type, with a time frame of up to 120 months.

For an endpoint of this type, the primary descriptive analysis would ordinarily report the number and proportion of participants achieving the defined response, together with a confidence interval for that proportion. If the statistical analysis plan specified a formal hypothesis test against a prespecified response threshold, that test would provide an additional inferential framework.

Because NAVIGATE is registered as nonrandomized, a response proportion is not inherently a randomized estimate of the difference between larotrectinib and a concurrent control treatment. Interpretation instead centers on the magnitude and precision of the observed response proportion and on the population to which it applies.

7. Statistical Methods Explained

What does "best overall response" mean?

Best overall response records the best response category achieved during the relevant assessment period. For NAVIGATE, the primary endpoint is specifically the occurrence of a confirmed complete response or partial response. A participant therefore contributes to the primary response outcome if the participant meets the registry's confirmation requirements for CR or PR.

Why is a response proportion appropriate?

CR and PR are categorical outcomes. Unlike overall survival, there is no requirement to estimate a survival curve simply to describe whether a participant achieved one of these response categories. The basic estimand is a proportion: the fraction of the defined analysis population achieving the response outcome.

Why is a confidence interval important?

An observed response proportion is calculated from a finite sample. A confidence interval communicates sampling uncertainty around that estimate. A narrow interval indicates greater statistical precision under the specified model and confidence procedure; a wide interval indicates greater uncertainty.

Why doesn't a response rate establish treatment superiority?

A response proportion by itself describes the observed response experience in the studied population. Without randomized allocation to a concurrent comparator, it does not automatically provide an unbiased estimate of a treatment difference. Differences from historical response rates can be affected by differences in patient selection, disease characteristics, assessment procedures, follow-up, and other factors.

Why does independent radiology review matter?

Response classification depends on imaging. An independent review committee provides an assessment separate from the treating investigators, which can reduce the influence of investigator expectations on the classification process. The registry specifically identifies this independent radiology review committee as the basis for determining the primary response endpoint.

Why are RECIST v1.1 and RANO both mentioned?

The registry states that RECIST v1.1 or RANO criteria are used as appropriate to tumor type. These frameworks provide structured response-assessment rules, but they are not interchangeable labels for a single universal measurement process. The criterion appropriate to the tumor type forms part of the definition of the endpoint itself.

8. Nonrandomized Design and Statistical Interpretation

The nonrandomized design is one of the most important statistical features of NAVIGATE. Randomization is the mechanism that, in a randomized controlled trial, helps balance both measured and unmeasured prognostic factors between treatment groups. NAVIGATE does not use that mechanism.

This distinction does not make a nonrandomized response endpoint statistically uninformative. Rather, it changes the inferential question. The analysis should emphasize the estimated response probability, its precision, the definition of the response population, and the assessment process instead of treating the response rate as though it came from a randomized head-to-head comparison.

Multiple arms

The registry identifies 11 arms within the nonrandomized parallel design. Multiple arms can be useful for studying a treatment across different study-defined groups, but the presence of multiple arms also means that the statistical analysis needs to distinguish the overall study population from any arm-specific or subgroup-specific analysis.

Without a posted statistical analysis, the registry does not specify an inferential hierarchy across the 11 arms. Consequently, a reader should not assume that every arm-level comparison was a prespecified confirmatory hypothesis test.

9. Analysis of the Primary Endpoint

Confidence interval versus p-value

A confidence interval and a hypothesis test answer related but different questions. The confidence interval focuses on the plausible statistical uncertainty around an estimated response proportion. A p-value, when a formal null hypothesis is specified, measures the compatibility of the observed data with that null hypothesis under the test's assumptions. Neither quantity is itself a measure of clinical importance.

Response duration is a different endpoint

Best overall response should not be confused with duration of response. A participant can achieve a CR or PR, but the best overall response endpoint alone does not quantify how long that response persists. Duration of response requires a separate time-to-event definition and corresponding censoring rules.

10. Time Frame and Censoring Considerations

The primary endpoint has a time frame of up to 120 months. Because the endpoint is defined over an observation period, the timing and completeness of response assessments can affect which response category is ultimately observed.

For a binary response endpoint, incomplete observation is not handled in exactly the same way as right censoring in a Kaplan-Meier analysis. The statistical analysis must instead define how participants without sufficient assessment information are classified or excluded from the response denominator.

The registry record summarized here specifies the endpoint definition and its time frame but does not provide a posted statistical analysis describing missing-response handling or imputation. Those rules are important because different denominator conventions can change the estimated response proportion.

11. Statistical Interpretation of Response

12. Multiplicity and the 11-Arm Design

NAVIGATE is registered with 11 arms. Multiple treatment arms can create several distinct statistical questions, particularly if investigators compare response rates across arms or perform separate subgroup analyses.

Multiplicity is particularly important when a study contains several possible comparisons. A collection of nominal p-values can create an exaggerated impression of statistical certainty if the testing strategy is not prespecified and adjusted appropriately. The registry information summarized here does not report a posted multiplicity strategy.

13. Safety and Other Outcomes

The registry information summarized here identifies the intervention, study design, enrollment, and primary response endpoint. It does not provide posted numerical serious adverse-event results by arm in the statistical analyses available for this record.

Safety is analytically distinct from the primary response endpoint. A response analysis asks whether the tumor met a defined radiologic response category, whereas a safety analysis summarizes adverse events, serious adverse events, laboratory findings, treatment discontinuation, or other safety outcomes under their respective definitions.

14. What Is Known About the Registry Record

15. Important Limitations and Interpretation Issues

• Nonrandomized allocation: without randomization, observed response differences between study groups or arms cannot automatically be interpreted as causal treatment effects.
• No concurrent randomized comparator: the registry identifies a nonrandomized design rather than a randomized comparison with a control treatment.
• Multiple arms: the presence of 11 arms creates potential for multiple descriptive or inferential comparisons, depending on the analysis plan.
• Response denominator: enrollment of 215 does not by itself establish the denominator for a response analysis. The definition of the response-evaluable population matters.
• Assessment criteria: response depends on independent radiologic review using RECIST v1.1 or RANO as appropriate to tumor type.
• Observation period: the endpoint is defined over a period of up to 120 months, so assessment opportunities and follow-up can influence response classification.
• Missing assessments: the registry information summarized here does not specify posted rules for missing or unevaluable response assessments.
• Endpoint scope: best overall response does not by itself describe duration of response, progression-free survival, or overall survival.
• Generalizability: the response estimate applies to the population enrolled under the study's eligibility and assessment framework and should not automatically be generalized to every patient with an NTRK-fusion positive solid tumor.

16. Why This Trial Matters Statistically

NAVIGATE is a useful teaching example because it illustrates a different statistical structure from a conventional randomized phase 3 trial. Its central endpoint is a categorical radiologic response outcome, while its design is nonrandomized and includes 11 parallel arms.

17. A Deeper Statistical View of the Primary Endpoint

The primary endpoint can be viewed as a binary classification applied after radiologic assessment: a participant either achieves the defined confirmed CR/PR response category or does not. This makes the endpoint fundamentally different from continuous outcomes such as tumor measurements and from time-to-event outcomes such as survival.

Estimand

The practical estimand is the probability of achieving a confirmed CR or PR under the study's treatment and response-assessment framework, in the population represented by the specified analysis denominator and over the stated time frame of up to 120 months.

Estimator

The natural estimator is the observed fraction of participants meeting the response definition. Its numerical value depends on both the number of responders and the population included in the denominator.

Uncertainty

Because the response outcome is binary, binomial sampling provides a natural framework for uncertainty estimation. The confidence interval should be selected in accordance with the prespecified statistical analysis plan and the characteristics of the observed response data.

Inference

Formal hypothesis testing requires a null hypothesis. For example, a one-sample response test could compare an observed response probability with a prespecified benchmark. In a nonrandomized multi-arm study, however, the interpretation of such a test depends critically on how the benchmark was chosen and whether it was prespecified.

Causal interpretation

The statistical distinction between estimation and causal inference is especially important here. Estimating a response probability is not equivalent to estimating the causal effect of larotrectinib relative to another treatment. The latter requires additional assumptions or a suitable comparative design.

18. Statistical Concepts in Context

19. Clinical Interpretation vs Statistical Interpretation

Keeping these two levels of interpretation separate prevents a response classification from being treated as a complete summary of treatment benefit. Statistical analysis describes the frequency and precision of the endpoint; clinical interpretation considers what that endpoint represents in the broader course of disease and treatment.

20. Record Summary

NAVIGATE is a phase 2, nonrandomized, parallel study of larotrectinib in adults and children with solid tumors harboring NTRK fusion. The registry records 215 enrolled participants across 11 arms and identifies best overall response of confirmed CR or PR, determined by an independent radiology review committee using RECIST v1.1 or RANO as appropriate to tumor type, as the primary endpoint over a time frame of up to 120 months.

Statistically, the trial is best understood as a response-estimation study rather than as a randomized comparative efficacy trial. The central analysis is naturally expressed through the proportion of participants achieving the defined response, with an appropriate confidence interval and a clearly specified response-analysis population. The nonrandomized allocation and multi-arm structure are essential to interpreting any arm-level estimates or formal comparisons.

21. Sources

• ClinicalTrials.gov: NAVIGATE, NCT02576431.