Cancer of Unknown Primary with a HER2 Signal: Trusting a Biomarker Without a Matching Trial
A single patient whose cancer never told the team where it started, and a genomic report that offers an answer of a different kind. The disagreement is about how much confidence a biomarker found on a broad panel deserves when the specific population it would treat has never actually been studied.
R.T., a 71-year-old man, spent thirty-nine years as an accountant before retiring, work he says trained him to want a clear number before making any decision -- an instinct that has made the past three weeks genuinely difficult. He presented with a firm left supraclavicular lymph node, and biopsy showed a poorly differentiated adenocarcinoma; an exhaustive search for the primary site -- CT of the chest, abdomen, and pelvis, upper and lower endoscopy, PSA, and a whole-body PET scan -- found nothing else, leaving him with a diagnosis of cancer of unknown primary rather than a named disease with its own established treatment pathway.
Comprehensive genomic profiling, sent as part of his workup, identified a HER2 gene amplification in the tumor -- a finding that matters because CUPISCO, the trial testing molecularly-matched targeted therapy against standard empiric platinum-based chemotherapy in CUP patients outside a defined favorable clinical subset, found a real progression-free survival advantage for the molecularly-guided approach. What CUPISCO's result doesn't yet settle is whether that PFS advantage has translated into longer overall survival, and it doesn't erase the fact that HER2-directed antibody-drug conjugates like trastuzumab deruxtecan have never been tested in their own pivotal trials on a population of patients whose primary tumor site is unknown -- every prior approval rests on a defined cancer type his diagnosis, by definition, doesn't have.
The one place genuinely biomarker-driven CUP therapy already has a long track record is different from his situation in a way worth naming plainly: certain CUP presentations -- axillary-node-only disease in a woman, for instance, treated as though it were breast cancer -- are matched to a specific known cancer type by their whole clinicopathologic picture, not by a single mutation on a broad panel, and that favorable-subset paradigm has mature outcome data behind it that a HER2 amplification found in isolation does not yet have.
He has asked, in nearly the same words each visit, for someone to simply tell him which number is the right one to trust -- a request the team has had to answer honestly has no single correct figure behind it yet, only two real bodies of evidence that measure different things and haven't been reconciled by a trial built to compare them directly.
How much a biomarker can stand in for a known cancer type
CUPISCO is the one randomized trial actually built to test this question, and it found a real progression-free survival advantage for molecularly-matched therapy over empiric platinum-based chemotherapy in CUP patients outside a favorable subset. He has an actionable HER2 amplification. I don't think we should default to empiric treatment when a trial specifically designed for this scenario points the other way.
I'd want to be precise about what CUPISCO has and hasn't shown. Its progression-free survival benefit hasn't yet translated into a confirmed overall survival advantage. And no HER2-directed antibody-drug conjugate has ever been tested in a pivotal trial on a population defined by unknown primary site -- every existing approval rests on a named cancer type he doesn't have. Empiric platinum-based chemotherapy is the actual validated standard for CUP as its own disease category.
I take CUPISCO's result seriously as real evidence, not nothing. I just don't think a progression-free survival signal, without a matured survival benefit and without any trial data specific to HER2-directed therapy in an unknown-primary population, is strong enough yet to displace the treatment that's actually been validated for his exact diagnosis.
I'd add a distinction that I think gets lost when we call this 'biomarker-driven CUP therapy' as though it's all one thing. The form of that approach with genuinely mature outcome data -- axillary-node-only disease in a woman treated as breast cancer, for instance -- matches a patient's whole clinicopathologic picture to a known cancer type, not one mutation on a broad panel in isolation. His HER2 amplification is real information. It's not the same category of evidence as that whole-picture match.
Given that, and given the genuine uncertainty on both sides here, I'd start empiric carboplatin-paclitaxel now rather than treat the HER2 finding as though it already carries favorable-subset-level confidence -- and hold trastuzumab deruxtecan explicitly in reserve for next-line therapy if he progresses, at which point a real trial of molecularly-matched treatment, with much less to lose by that point, becomes a clearly reasonable next step.
Agreed: start empiric carboplatin-paclitaxel, with trastuzumab deruxtecan held explicitly in reserve as a next-line option given his HER2 amplification, contingent on progression rather than used as an immediate substitute for standard treatment.