Determining the Causal Effect of Neurotrophic Tropomyosin Receptor Kinase Gene Fusions on Survival Prognosis for Adult Patients With Advanced Cancers

Introduction

Fusions in the neurotrophic tropomyosin receptor kinase (NTRK) gene represent one of the rarest actionable oncogenic drivers in adult oncology, yet they have broad implications across multiple tumor types.¹ Fusions involving NTRK1, NTRK2, and NTRK3 occur in fewer than 1% of most common adult cancers, including lung (0.2%), breast (0.4%) and colorectal (0.2%) malignancies, although they are enriched in a small number of rare tumor subtypes (eg, ≥90% in secretory carcinoma).^1,2 These rearrangements result in constitutive TRK signaling and can be detected in routine clinical practice using next-generation sequencing (NGS).

The tumor-agnostic approvals of first-generation TRK inhibitors, including larotrectinib and entrectinib (the latter currently approved with conditions in the United States, Europe, and Canada),^3,4 have expanded the clinical role of NGS for NTRK fusions. The rarity of these fusions and their occurrence across a broad range of tumor types limit the feasibility of traditional trial designs^5-7 and constrain understanding of their natural history under standard clinical care.⁸ As a result, uncertainty persists regarding their independent prognostic significance in adults with advanced cancers,⁹ complicating clinical decisions and treatment recommendations. In this setting, generating direct comparative evidence among NTRK fusion–positive (NTRK+) patients is inherently challenging. Quantifying the intrinsic prognostic effect of NTRK fusions supports the use of appropriately constructed external controls that reflect tumor-specific heterogeneity among NTRK fusion–negative (NTRK−) patients, thereby reducing uncertainty in the interpretation of comparative outcomes in routine oncology.

A recent meta-analysis reported lower overall survival among patients with NTRK+ tumors based on three previously published studies.⁹ Interpretation of these findings is constrained by methodological challenges common to the included studies, including heterogeneous index and censoring dates, unsatisfactory confounder control, overall and by tumor type, and small sample sizes.^10-13 Notably, the largest study included only 28 NTRK+ patients.¹³ Furthermore, evidence that the proportional hazards (PH) assumption was not met in some of the included studies suggests caution in interpreting Cox model estimates included in meta-analysis.^9,14 These limitations leave substantial uncertainty regarding the prognostic role of NTRK fusions⁹ and the attribution of survival benefits to TRK inhibitors in the absence of appropriate comparative evidence.

Using large real-world clinicogenomic data with a target trial emulation, we determined whether NTRK gene fusions independently influence overall survival in adults with advanced cancers. By comparing NTRK+ and NTRK– patients with similar tumor-specific clinical characteristics across multiple tumor types, we estimated the causal, tumor-agnostic prognostic effect of NTRK fusions as a directly interpretable difference in average survival. To contextualize and assess the robustness of our results, we replicated the three studies included in the recent meta-analysis of the prognostic effect of NTRK fusions by mirroring study design, inclusion criteria, and matching approaches.

Methods

Results

During our study period, there were 91,314 adult patients who had an advanced or metastatic solid tumor cancer diagnosis, received NGS, and were alive at the time of the testing report (Fig 1). We included 68,145 NTRK– patients for matching to 73 NTRK+ patients within seven tumor groups (Table 1), including colorectal, non–small cell lung, sarcoma, breast, head and neck, neuroendocrine, and prostate, and 14 NTRK+ patients representing eight other cancers. Among the 87 NTRK+ patients, 58 (66.7%) had a NTRK1 fusion, 5 (5.7%) had a NTRK2 fusion, and 24 (27.6%) had a NTRK3 fusion. Before matching, the most frequent tumor types among NTRK– patients for the same seven tumor groups (n = 47,480; 69.7%) included NSCLC (n = 15,473; 32.6%), colorectal cancer (n = 11,032; 23.2%), and breast cancer (n = 10,346; 21.8%).

We achieved balance overall (Fig 2 and Table 1) and within each tumor group (Data Supplement, Fig S6). Given our stratified matching approach by tumor type, our matched cohort of patients had the same distribution across tumor types for the most common tumor groups, with the largest groups consisting of colorectal cancer (n = 22; 25.3%), NSCLC (n = 13; 14.9%) sarcoma (n = 13; 14.9%), and breast cancer (n = 10; 11.5%).

Discussion

Drawing on a cohort of NTRK+ patients more than three times larger than prior studies, our study provides a comprehensive estimate of the prognostic effect of NTRK fusions for patients with advanced cancer. We estimated mortality risk to be more than two-fold higher within the first year after receiving NGS (HR, 2.1 [95% CI, 1.2 to 3.7]), suggesting a larger short-term prognostic effect than previously reported. Survival was, on average, 50 to 123 days shorter over 1- and 2-year periods, providing clinically meaningful estimates of the prognostic value of NTRK fusions among patients not receiving TRK inhibitors. Building on a literature characterized by small, heterogeneous cohorts, our study examines clinically comparable patients with and without NTRK fusions across several common cancers. Although stratified analyses were not statistically significant, tumor-specific prognostic estimates are novel in this setting and may provide a hypothesis-generating starting point for understanding prognosis across tumor types.

Our findings enhance the clinical interpretability and validity of evidence indicating poorer overall survival among NTRK+ adult patients. The 64% and 59% increases in mortality risk we observed over 2 and 3 years, respectively, are of larger magnitude compared with the 51% increased mortality risk previously reported in a recent meta-analysis.⁹ Our estimates are derived from short- to medium-term observation periods, while the meta-analysis summarized survival over a 10-year horizon from diagnosis or treatment initiation, drawing on studies with median follow-up of <3 years.^9,11-13 Replication of prior prognostic studies showed that reported effects were sensitive to model assumptions and selection of study index date, with analyses aligned to our design yielding consistently attenuated estimates. Our short- to medium-term findings of survival differences are therefore more closely aligned with the period over which outcomes are directly observed and most relevant to clinical decision making. Given that few patients remained under observation beyond 3 years, the attenuated survival difference at 5 years should be interpreted cautiously, as statistical precision decreases with the number of patients remaining at risk and does not necessarily indicate a diminution of the underlying prognostic effect. Our RMST findings contribute to the literature by providing estimates of absolute survival benefit that are directly expressible in units of time, facilitating interpretation over time horizons that are meaningful for patients. This study's clear and causal estimates of the tumor-agnostic prognostic effect of NTRK fusions thus improve interpretability for clinicians and patients while enhancing our understanding of important considerations for the validity of real-world external comparators.^32,33

Although uncertainty remains, by providing tumor-specific estimates of overall survival in adults with NTRK+ solid tumors, our study advances understanding of prognosis across cancer types where outcomes have previously been poorly characterized.⁹ These findings contextualize observed heterogeneity in survival by suggesting differences in the prognostic value of oncogenic drivers,^34,35 which may inform treatment considerations in routine oncology practice.^36,37 The strong overall balance in patient characteristics achieved across tumor types supports causal interpretation under assumptions commonly accepted in randomized trials,^16,38 while highlighting the need for tumor-specific inference within tumor-agnostic frameworks to manage persistent evidentiary uncertainty and support informed decision making in precision oncology.³⁷

By replicating prior studies using our large clinicogenomic data, our findings offer additional insights into the transportability of real-world findings of the prognostic effect of NTRK fusions. Our findings were attenuated but consistent with Hibar et al¹³ using a larger, more recent cohort from the same US database, whereas differences from studies conducted in the United Kingdom and the Netherlands likely reflect limited comparability due to fundamental differences in study design, including index date definitions and confounder adjustment, as well as underlying population characteristics across countries.^11,12

To manage evidentiary uncertainty in precision oncology, regulators and payers are increasingly signaling the acceptability of real-world evidence (RWE).^39,40 Realizing the promise of decision-grade RWE will require harmonized guidance on fit-for-purpose analytic standards that are grounded in causal study design.⁴¹ Across tumor types, heterogeneity in comparative effectiveness and cost-effectiveness can reflect differences in prognosis and available treatment options,^36,37 suggesting that tumor-agnostic market approval and reimbursement decisions may need to be tailored to clinically relevant subgroups.⁴² Persistent uncertainty driven by limited comparative evidence and gaps between identifying actionable alterations and delivering matched therapies complicates clinical decision making for patients in routine care.^37,43 A lifecycle assessment approach embedded within learning health systems can leverage real-world evidence to support patient-centered treatment decisions while informing regulatory and reimbursement processes that balance timely access with responsible evidence development.⁴⁴

This study had limitations. First, the overall sample of NTRK+ patients remains small, hindering inference regarding heterogeneity of effects. Second, NGS testing was performed selectively in clinical practice in patients with advanced disease or refractory to prior therapies, limiting generalizability to nonadvanced cancer settings. Nonetheless, our findings are based on the broadest cohort of NTRK+ patients with advanced cancers examined to date, with a study design that addressed key methodological limitations of prior real-world analyses. Third, although our intention-to-treat estimand did not disentangle intrinsic prognostic differences from effects mediated by subsequent treatment initiation, it remains appropriate for estimating prognosis as it reflects survival outcomes as observed in routine clinical practice. Finally, the inclusion of laboratory markers data was limited. Although matching achieved excellent balance on measured sociodemographic and clinical covariates, the potential for residual confounding on other genomic markers cannot be fully ignored. Good covariate balance overall and within tumor types reduced concerns about bias and supports the generalizability of our findings. More complete genomic data collection will be essential for unbiased outcome evaluation in future studies.

In conclusion, in this large real-world target trial emulation, NTRK gene fusions were associated with reduced short-term overall survival among adults with advanced cancers not treated with TRK inhibitors. By contributing novel tumor-specific prognostic evidence that accounts for heterogeneity of key clinical and sociodemographic characteristics across tumor types, our findings provide additional context for reliable tumor-agnostic inference. Collectively, these findings clarify the prognostic role of NTRK fusions and inform clinical interpretation of outcomes in routine oncology.

Data Sharing Statement

Patient-level data used in this study are confidential and are not available in a public repository, in accordance with institutional policies and license agreements. The data that support the findings of this study were originated by and are the property of Flatiron Health, Inc. and Foundation Medicine, Inc. Requests for data sharing by license or by permission for the specific purpose of replicating results in this manuscript can be submitted to PublicationsDataAccess@flatiron.com and cgdb-fmi@flatiron.com.

Author Contributions

Conception and design: Emanuel Krebs, Deirdre Weymann, Dean A. Regier

Administrative support: Emanuel Krebs, Deirdre Weymann

Provision of study materials or patients: Deirdre Weymann, Dean A. Regier

Collection and assembly of data: Emanuel Krebs, Nirupama Tamvada, Deirdre Weymann

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

Authors' Disclosures of Potential Conflicts of Interest

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).