For cancers of unknown primary (CUP) and other diagnostically ambiguous tumors, traditional pathology doesn’t always provide clear answers. In hopes of improving diagnostics, a new report from CAP explores how molecular testing could improve tumor origin detection, uncover actionable biomarkers, and guide patient care. We connected with lead researchers Jie-Fu Chen, FCAP and CAP Personalized Healthcare Committee member, and JinJuan Yao, FCAP, to learn more.
In routine sign-out, what typically makes a tumor “difficult to classify,” even after morphology and immunohistochemistry (IHC)?
A “difficult” case typically falls into a diagnostic gray zone, where morphology and IHC do not clearly support a single tumor type consistent with the clinical context.
Common causes include poor or divergent differentiation, in which tumors show limited lineage features, lose expected markers, or display equivocal or conflicting IHC staining. High-grade tumors may also undergo morphologic transformation and mimic other tumor types or primary sites, broadening the differential diagnosis.
Site-indifferent morphology can also complicate interpretation. Tumors, such as neuroendocrine neoplasms, may arise in multiple organs while maintaining similar histologic features, making it difficult to determine primary site or distinguish primary from metastatic disease using morphology and IHC alone.
Limited or suboptimal tissue – including small biopsies, crush and cautery artifact, low tumor cellularity, and processing-related artifacts – may further restrict evaluation. Necrosis, inflammation, and treatment-related changes can also obscure diagnostic features.
An unusual clinical context, in which morphology does not align with the expected tumor spectrum for a given site, or conflicts with imaging or other clinical findings, can add further uncertainty.
Finally, limitations of conventional tools contribute to ambiguity. Many IHC markers lack complete specificity, and staining must be interpreted in morphologic and clinical context. Unexpected or variable patterns can widen the differential diagnosis.
Together, these factors can leave routine diagnostic approaches inconclusive.
Your paper discusses key scenarios like cancers of unknown primary, narrow differentials, and multiple tumors in one patient. Which situations are most challenging in practice, and why?
These three scenarios represent significant diagnostic challenges that are not fully resolved by conventional pathology workups, each in a different way.
CUPs are often the most difficult. Morphology and IHC fail to identify tumor type or primary site, and the differential diagnosis is broad, making targeted workup challenging.
Difficult-to-classify tumors with a limited differential or prior oncologic history may initially seem narrower. However, conflicting morphologic findings or a history of prior malignancy can complicate interpretation and potentially misdirect the evaluation.
Determining clonal relationships between multiple tumors is critical for staging and management. Distinguishing separate primary tumors from metastatic disease often requires molecular analysis, as histology alone may be insufficient.
When a case is still unclear, how do you decide when molecular testing is likely to help – and when it may not change the diagnosis?
The role of molecular testing depends on clinical context and is most urgent when systemic therapy may be required. It is often prioritized in cases with multiple lesions suggestive of metastasis, histologic features concerning for spread, or a differential that includes aggressive tumor types requiring more than local treatment.
Difficult-to-classify tumors in patients with a prior cancer history and cases requiring comparison of multiple tumors frequently raise concern for recurrent or metastatic disease. Many CUPs are also high risk based on location or morphology. Because most systemic therapies remain tumor type– and site-specific, molecular testing can guide treatment selection and monitoring. Even when a primary site remains uncertain, actionable genomic alterations may support targeted therapy, particularly in CUP.
In contrast, testing may be deferred for localized tumors that can be managed with local excision or ablation, or when the differential includes tumor types without actionable molecular targets.
Decisions regarding molecular testing should be made through multidisciplinary discussion and shared decision-making with the patient and care team.
How often does molecular testing uncover actionable biomarkers in these difficult cases, and how can that information guide treatment decisions even when classification remains uncertain?
The frequency of actionable molecular alterations varies by patient population and over time. In a real-world analysis of nearly 50,000 patients who underwent clinical genomic profiling between 2017 and 2022, approximately one-third had predictive biomarkers classified as level 1 or 2 by the OncoKB knowledge base, indicating potential eligibility for targeted therapy.
Within that cohort, 21.4 percent of more than 1,400 patients with CUP became eligible for an FDA-approved targeted therapy based on genomic findings over the five-year study period. Other studies using different genomic platforms have reported similar results, with roughly 30 percent of CUP cases harboring actionable alterations.
As the number of available targeted therapies expands, molecular testing plays an increasing role in identifying patients who may benefit from these treatments. In addition, several tumor type–agnostic approvals now allow treatment of advanced solid tumors based on specific genomic alterations, regardless of primary site. Common biomarkers with tumor-agnostic approvals include high tumor mutational burden (≥10 mutations per megabase), microsatellite instability–high (MSI-H), BRAF V600E mutations, RET fusions, and NTRK rearrangements.
In cases with more than one tumor, how can next-generation sequencing (NGS) help determine whether they are separate primaries or metastases, and why does that matter clinically?
NGS, particularly large-panel targeted sequencing, can help distinguish separate primary lung carcinomas (SPLCs) from intrapulmonary metastases (IPMs). In general, SPLCs show completely non-overlapping genomic profiles or carry distinct driver hotspot mutations, supporting the interpretation that the tumors arose independently. In contrast, IPMs typically share multiple genomic alterations, indicating a clonal relationship and metastatic spread.
There are important caveats. A single shared oncogenic driver mutation does not necessarily confirm metastasis, as certain driver mutations are common in lung cancer and may arise independently in separate primary tumors. In addition, small NGS panels with limited genomic coverage may not provide enough discriminatory power, particularly in tumors with few or no detectable driver alterations.
This distinction has significant clinical implications. SPLCs are staged individually and often represent earlier-stage disease, typically managed with local treatment such as surgical resection, with or without adjuvant therapy. IPMs, however, indicate more advanced disease (T3, T4, or M1a) and generally require systemic therapy. Prognosis is typically more favorable for patients with SPLCs than for those with IPMs.
AI-based tumor classifiers are becoming more common. Where do you see these tools fitting into the workup?
AI-based tumor classifiers are currently most useful at the later stages of diagnostic workup or as decision-support tools. Their output must be interpreted in the context of the algorithm’s capabilities, the clinical presentation, and histologic findings. In some cases, AI-generated suggestions may guide additional confirmatory testing.
Ongoing development aims to integrate more comprehensive clinical and pathologic data, rather than relying solely on molecular findings, which may improve their utility as decision-support systems.
In situations where tissue is limited, AI-based classifiers may also serve as an early triage tool, helping prioritize testing and reduce unnecessary use of scarce tumor material.
What are the biggest risks of relying too heavily on AI outputs, especially for rare tumors or cases that do not match the model’s training data?
Over-reliance on AI-based tumor classifiers carries several risks. An algorithm cannot identify entities it was not trained or validated to recognize, including rare subtypes, newly described tumors, or evolving classifications. Probability scores may create false confidence if the correct diagnosis is not represented in the training set. In addition, specimen-related factors, such as low tumor purity or degraded material, can affect output accuracy. There is also the risk of automation bias, where teams may give more weight to AI results than to morphology or clinical and radiologic context.
AI can detect patterns in molecular data that may not be immediately apparent. However, its output is most effective when interpreted alongside clinical presentation, imaging findings, and histologic features. In practice, AI should be viewed as a complex ancillary tool that supports, but does not replace, integrated diagnosis.
What new data types do you think will be most important for future tumor classification?
Many tumor classifiers have shown strong performance using a single data type, such as genomic alterations (mutations, copy number changes, rearrangements), RNA expression, and DNA methylation. However, these approaches do not always capture the full biologic complexity of tumors.
Current development is moving toward multimodal or multi-omic models that integrate multiple types of molecular data. Increasingly, these systems also aim to incorporate clinical information and histologic features derived from whole-slide imaging to improve classification accuracy.
Looking ahead 5 to 10 years, what role do you predict molecular testing will play in routine diagnostics?
Based on current trends, molecular testing is expected to become more integrated into routine practice.
First, it is likely to be used earlier in the diagnostic workflow and incorporated directly into the pathology report, rather than ordered as a separate add-on test. For many tumor types, molecular findings are increasingly part of the core diagnostic definition.
Second, molecular testing will continue to play a central role in disease management. Reflex testing for actionable biomarkers is already standard for several cancers and is expected to expand. Additional markers may be evaluated at diagnosis to support risk stratification, prognosis, and disease monitoring.
Finally, advances in computational tools will support more integrated analysis of multimodal data. Artificial intelligence and machine learning are expected to become more embedded in molecular testing workflows, aiding data analysis and interpretation.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
