A result can be real, reproducible — and still not tell you what to do.
A liquid biopsy comes back on a sixty-seven-year-old with metastatic lung cancer. The report flags DNMT3A and TP53. No targetable driver. A trainee asks: should this patient be worked up for a blood cancer? The oncologist pauses — because DNMT3A is one of the most common mutations in aging blood cells, not lung tumors. The report looks authoritative. Whether anything in it should change a clinical decision is a different question entirely.
The industry is mature. Analytic performance is rarely the limiting issue. The hard problems are downstream.
Analytic validity is a laboratory question about the instrument, the chemistry, and the quality of the specimen. Most commercially available molecular tests pass this bar for the mutations they are designed to detect — the clinical-grade assay industry is now mature enough that reproducibility and sensitivity are rarely the limiting issue. The DNMT3A signal in the opening case was real. The mutation was genuinely present in the plasma DNA. It cleared the first bar. That's not the same as being useful.
Clinical validity asks whether the measured result actually correlates with a clinical state. Does an EGFR exon 19 deletion predict response to an EGFR inhibitor? Does MRD positivity after colon surgery predict recurrence? This is answered by observational studies and trials that measure both the molecular result and the outcome in the same patients. DNMT3A in the plasma of a sixty-seven-year-old fails this test for lung cancer — its most likely source is aging white blood cells, not the tumor.
Utility is about decision strategies, not patients with or without a mutation. That distinction forces a different study design.
Clinical utility is the hardest bar because it requires a controlled study comparing outcomes between test-guided and non-test-guided decision-making. The comparison is between strategies, not between patients. That is why a correlation — the test predicts who will recur — is not enough. You still need to show that acting on the prediction changes what happens to the patient. Many molecular tests adopted into clinical practice have cleared only the first two bars.
Tumor-agnostic approvals are the structural shift: the biomarker, not the organ, is the selection criterion.
A companion diagnostic is a test whose results are necessary to safely and effectively use a specific drug. The FDA approves companion diagnostics alongside the drugs — which means the evidence for the test is the same evidence as the evidence for the drug. Changing the test changes what the trial evidence applies to. The recent tumor-agnostic approvals are a structural shift: microsatellite instability-high status predicts pembrolizumab response regardless of tumor type. The biomarker, not the organ, is the selection criterion.
TNM staging describes extent. T is the primary tumor's size and invasion. N is regional lymph node involvement. M is distant metastasis. These combine into stage groups from I to IV. The AJCC 8th edition made a structural change explicit: breast cancer staging now incorporates hormone receptor status, HER2, and the Oncotype DX recurrence score. A patient with small, node-negative, favorable biology may be assigned a lower prognostic stage than her anatomy alone would suggest. Biology overriding anatomy is now in the formal system.
ctDNA aggregates signal across all lesions — but it cannot tell you where the mutation lives or whether the tumor is even shedding.
Liquid biopsy samples ctDNA shed from all tumor sites. Genuine advantages: whole-tumor sampling, repeatability, and resistance detection.
Still open: multi-cancer early detection. The analytic validity is established. The clinical utility is not. Enthusiasm is not evidence.
Here is the chapter's discipline, stated plainly. Companion diagnostics work. MRD-guided adjuvant de-escalation in colon cancer works. Each succeeded because the three-validity chain was followed. A PIK3CA mutation on a comprehensive panel may be real, associated with a potential target, and still a hypothesis for a clinical trial — not a treatment directive. A multi-cancer early detection test has established analytic validity and some clinical validity. It has not established utility evidence that it reduces cancer mortality in a real population. Enthusiasm for the technology is not a substitute for evidence of benefit.
Cancer Research · Chapter 3 · Molecular Diagnostics, Staging, and the Liquid Biopsy
That is the skill this chapter builds. For every molecular result: does it measure what it claims, in this specimen, with this assay? Does it correlate with a clinical state in this cancer type? And is there evidence that acting on it improves outcomes in this specific decision? Most findings fail the third test in most contexts. The ones that pass it have earned the right to change what you do. Everything else is a hypothesis.