Cancer Biomarkers: Three Real Studies, Zero Actionable Insight

The Setup: What Got Bundled Under "Early Cancer Detection"

Three peer-reviewed studies landed in the same news cycle, and someone decided they belonged together. One from Cell on urine DNA liquid biopsy in bladder cancer. One from Molecular & Cellular Proteomics on stool proteomics and machine learning for colorectal cancer screening. One large Swedish trial published in Nature Medicine confirming that stool-based screening beats doing nothing. The implied narrative: the era of biomarker-driven, AI-powered cancer detection is arriving, and you should care.

Here's what's actually true: all three studies are real, competently done, and published in serious journals. Here's what's also true: they cover two completely different cancer types, address questions at wildly different stages of clinical development, and none of them represent anything a healthy adult can act on today.

Study 1: The Cell Paper — A Clever Bladder Cancer Tool That Isn't for You

The urine liquid biopsy study in Cell is the most technically interesting of the three. Researchers developed a method using cell-free DNA extracted from urine to predict whether bladder cancer patients would respond to BCG therapy — a century-old tuberculosis vaccine repurposed to reduce tumor recurrence after surgical removal.

The genuinely clever part: bladder tissue accumulates age-related somatic mutations in normal cells, a phenomenon called the field effect. This previously contaminated liquid biopsy signals by making benign age-related mutations look like cancer. The team developed a correction algorithm that accounts for background mutation rates in normal urothelium — a real methodological advance that makes the assay more accurate, particularly in older patients.

What this is not: a consumer screening test. Every patient in this study had already been diagnosed with bladder cancer and had already undergone surgery. The question being asked is which of these post-surgical patients will benefit from BCG maintenance therapy — a narrow clinical question about treatment selection, not early population-level detection. You cannot biohack your way into or out of a BCG response prediction algorithm, and this test is not going to appear on any panel your primary care physician orders.

Study 2: The Stool Proteomics ML Paper — Where Methodology Gets Murky

The Molecular & Cellular Proteomics paper addresses a legitimate clinical problem: the standard FIT (fecal immunochemical test) generates false positives, sending patients to colonoscopy when they don't actually have cancer. Could proteomics analysis of leftover FIT samples, processed through a machine learning classifier, reduce those unnecessary procedures?

The study shows it might be possible — at least in this dataset. But several caveats make this premature for any real-world application:

• Pre-selected samples: the researchers used leftover material from patients who had already tested FIT-positive. This enriches the population for pathology and makes generalizability to the full general screening population unknown.
• Black-box classification: the paper reports model performance but does not clearly identify which specific proteins are driving the result. In medicine, black-box ML models have a documented history of collapsing when deployed outside their training data — different hospitals, different countries, different dietary contexts.
• Funding and replication: probable industry involvement is flagged in the paper. That doesn't automatically invalidate the results, but it raises the replication standard. Independent validation in a demographically and geographically distinct cohort does not yet exist.

The underlying goal — reducing unnecessary colonoscopies — is genuinely worthwhile. Colonoscopy carries real procedural risk and significant cost. If proteomics can reliably filter false positives, that matters clinically. But this is an early proof-of-concept, not a validated diagnostic tool, and treating it as one would be a mistake.

Study 3: The SCREESCO Trial — Confirming What We Already Knew

The SCREESCO trial in Nature Medicine is the most credible of the three by volume and design: a large Swedish prospective trial comparing colonoscopy-based screening, FIT-based screening, and no screening for colorectal cancer in a general population cohort.

The finding — that FIT-based stool screening reduces colorectal cancer incidence compared to no screening — is almost certainly correct. It is also what every major clinical guideline in every high-income country has recommended for years.

Read the study carefully before concluding this is news. The "no screening" control group is messy: people in the real world don't live in a screening-free bubble. They receive incidental diagnoses through other clinical pathways, which compresses the apparent benefit of the intervention. The comparison between colonoscopy and FIT is murkier still — colonoscopy detects more polyps but carries significantly higher procedural burden, and the trial doesn't definitively settle which approach is superior for cancer-specific mortality at the population level.

Most importantly: this trial validates existing guidelines, it does not generate new ones. If you're over 50 and not getting FIT screening, this is your reminder. If you already are, nothing has changed. This is useful data for healthcare policy. It is not a reason for any individual reading this to do something different tomorrow.

The Core Problem: "Biomarkers That Save Lives" Implies You Can Use Them

The framing of these three studies as a coherent "early detection" narrative illustrates how science journalism manufactures momentum from incremental research. Bladder cancer and colorectal cancer share almost nothing biologically, anatomically, or clinically. Grouping them under "cancer biomarkers" is an editorial convenience, not a scientific throughline.

More importantly: the framing implies actionability. You cannot order a field-effect urine liquid biopsy. Stool proteomics ML screening does not exist as a clinical product. The FIT test — the one real, accessible takeaway — has been recommended for everyone over 50 for years. The message embedded here is "talk to your doctor about getting screened," which is correct and useful advice, but it does not require three new studies to justify, and it is not the kind of insight that earns a headline.

What Would Actually Be Worth Covering

For multi-cancer early detection to generate a post worth reading, you'd need something like:

• A prospectively validated multi-cancer blood test with published sensitivity and specificity data in the general population — not post-diagnosis or high-risk cohorts only
• A biomarker study showing that early detection via the new test actually changes outcomes — survival, treatment burden, quality of life — not just detection rates
• A head-to-head comparative effectiveness trial showing a new test is superior to existing standard-of-care screening, not just better than doing nothing

The Galleri multi-cancer early detection trial is the closest thing to a study that clears that bar, and its outcome data remain mixed. When something credible and actionable emerges, it will earn a full post. These three did not.

Verdict

Three solid studies doing real work in narrow corners of oncology. The urine liquid biopsy is a post-surgical treatment selection tool for bladder cancer patients who are already in the system. The stool proteomics classifier is a pre-validation ML experiment that needs independent replication before it means anything clinically. The SCREESCO trial confirms that existing colorectal cancer screening guidelines are correct — which they were before this trial ran. The science is fine. The packaging is the problem.