#213 ‒ Liquid biopsies and cancer detection | Max Diehn, M.D. Ph.D.

Episode Summary

Executive Summary: Peter Atiyah and Stanford radiation oncologist Max Dean trace the evolution of liquid biopsies from protein markers and circulating tumor cells to ctDNA, explaining why mutation-based assays are far more specific for detecting minimal residual disease and guiding adjuvant therapy. They also stress the limits of current early-detection blood tests, the importance of prevalence in interpreting results, and why randomized trials are still needed before broad screening use.

Main Topics: Max Dean’s path from MD-PhD to radiation oncology and liquid biopsy research: Dean describes his Stanford training, work in Pat Brown’s lab using DNA microarrays, his move into radiation oncology, and how clinical frustration with lung cancer recurrence led him to liquid biopsy research. Diagnostic test fundamentals: sensitivity, specificity, PPV, and NPV: The conversation carefully defines sensitivity and specificity, then connects them to prevalence and pretest probability to show why even seemingly strong tests can perform poorly in low-risk screening populations. Lung cancer epidemiology and screening context: They review lung cancer as the leading cause of cancer death, its relationship to smoking and non-smoking risk factors, and the role and limitations of low-dose CT screening. Why protein biomarkers and circulating tumor cells fell short: Dean explains why PSA/CEA/CA19-9-like protein markers lack specificity and why circulating tumor cell assays are technically difficult, low-yield in early-stage disease, and sometimes positive in healthy controls. ctDNA as the most promising liquid biopsy modality for MRD: The discussion centers on cell-free DNA and circulating tumor DNA, how tumor-specific mutations can be detected with next-generation sequencing, and why this approach is highly specific for residual disease after treatment. Liquid biopsy for early detection: promise, limitations, and regulation: They examine methylation, fragmentomics, and machine-learning approaches for pan-cancer screening, but emphasize that current sensitivity in early-stage disease is still too low and that FDA-approved tests are mainly for advanced-disease genotyping. Clinical trials, reimbursement, and the need for outcome proof: Dean argues that liquid biopsies should be validated by randomized trials showing cancer-specific survival benefit before widespread adoption, despite commercial and CLIA-based availability of some tests.

Key Arguments: Liquid biopsy research emerged from a real clinical need: after curative-intent treatment, clinicians often cannot tell who is truly cured and who has microscopic residual disease. Sensitivity and specificity must always be interpreted together; a test can look impressive in isolation but still be clinically weak when disease prevalence is low. Protein biomarkers are limited because normal tissues can also produce them, reducing specificity and making low-level positives hard to interpret. Circulating tumor cells are conceptually attractive but practically limited by rarity, processing constraints, and false positives in healthy individuals. ctDNA is more specific because it tracks tumor mutations, which are the molecular cause of cancer and are generally absent from normal cells. Using multiple known tumor mutations in parallel increases sensitivity compared with searching for a single mutation. Early-stage screening is much harder than MRD detection because tumor burden is tiny and pretest probability is low, which depresses positive predictive value. Current liquid biopsy screening tests are not yet as good as established screening modalities like low-dose CT or colonoscopy, so they should not replace them without stronger evidence. Randomized trials demonstrating reduced cancer-specific mortality are necessary before broad population screening can be justified. A future best-in-class assay will likely combine mutations, methylation, fragment size, and other features rather than rely on one biomarker alone.

Data Points: Lung cancer mortality rank: #1 cause of cancer death - Dean notes lung cancer remains the leading cause of cancer-related death in both men and women. Never-smokers among lung cancer deaths: 15% - At the start of the episode, Atiyah notes that 15% of people who die of lung cancer have never smoked. Early-stage lung cancer recurrence rate: ~20-25% - Dean describes a subset of stage 1-2 lung cancer patients who recur after curative-intent radiation. CT detectability threshold: ~1 cm diameter tumor - They discuss that conventional imaging generally cannot see tumors much smaller than about 1 cm. Cell count at 1 cm tumor size: ~1 billion cells - At about 1 cm, a tumor is already roughly a billion cells, illustrating the blind spot of imaging. Low-dose CT mortality reduction: ~20% relative risk reduction - The National Lung Screening Trial showed fewer lung cancer deaths with low-dose CT versus chest x-ray. Low-dose CT absolute risk reduction: single-digit percent, likely ~0.5-2% - Atiyah and Dean discuss that the absolute benefit is much smaller than the relative reduction. Healthy plasma cell-free DNA concentration: ~1-5 ng/mL - Dean gives the typical cfDNA concentration in healthy individuals. Typical blood draw for liquid biopsy research: 10-20 mL blood - He says most assays use roughly this amount of blood. CTC assay yield in early-stage disease: often zero - In stage 1-2 non-small cell lung cancer, circulating tumor cell assays often found no cells. ctDNA fraction in advanced lung cancer: ~1% of circulating DNA - Dean notes advanced disease can have around 1% tumor-derived cfDNA. ctDNA fraction in early-stage disease: <0.01% to 0.001% - Early-stage tumors shed far less ctDNA, making detection much harder. First-generation mutation-based assay sensitivity: ~0.01% (1 in 10,000) - Dean contrasts older ctDNA methods with newer ones. Third-generation assay sensitivity: ~1 in 1,000,000 - He says his newer method improved sensitivity by about two logs. Methylation-based assay sensitivity: ~0.1% (1 in 1,000) - Dean estimates methylation approaches are less sensitive than mutation-based methods. Stage 1 lung cancer sensitivity in GRAIL presentations: ~5% or less - Atiyah cites that early-stage sensitivity remains very low for pan-cancer screening. Specificity example for early screening: 99-99.5% - They discuss that even high specificity may still yield poor PPV when prevalence is ~1%. Pretest probability example: ~1% - Atiyah uses a low-risk patient example to show how prevalence affects predictive value. CLIP study stage 1 sensitivity: ~20% - Dean says his mutation-based lung screening method reached about 20% sensitivity in stage 1 in a validation cohort. CLIP study specificity: ~98% - He reports specificity around 98% in that analysis. FDA-approved liquid biopsy tests: 4 - Dean states there are four approved liquid biopsy tests, mainly for advanced-disease genotyping.

Pivotal Quotes: "The way that we and others in the field do that is to focus on a unique molecular property of cancer cells that normal cells generally don't have, and that is the mutations that cause the cancer." — Max Dean: Explaining why ctDNA is more specific than protein biomarkers or circulating tumor cells. "If you already basically have no risk of having cancer, you're less than 1% chance of having cancer, the test isn't really moving the needle for you significantly." — Peter Atiyah: Illustrating why prevalence/pretest probability can make a seemingly good screening test clinically weak. "We absolutely should not be doing imaging if there's no point. Like if no matter what the scan shows, I'm not going to change what I'm doing, then I should not be ordering that scan." — Max Dean: Discussing the clinical logic behind ordering imaging and avoiding unnecessary tests.

Implications: Liquid biopsies are most compelling today for MRD and treatment guidance, not broad screening. The field needs better sensitivity, better specificity, and randomized outcome trials before blood tests can safely replace or augment established cancer screening.