Early disease detection

What overdiagnosis measures in practice

• Lung CT screening: in a meta-analysis restricted to the two randomised trials at low risk of bias (8,156 participants), 49% of screen-detected cancers were estimated to be overdiagnosed — and restricting to higher-quality trials made the estimate larger, not smaller (PMID 32194774).
• PSA screening: modelled overdiagnosis of 22.7% of screen-detected cancers in white men and 34.4% in Black men (PMID 17501937).
• Overdiagnosis needs three ingredients, and multi-cancer blood tests and whole-body MRI supply all three: a reservoir of harmless indolent disease, a test sensitive enough to find it, and a screened population with meaningful chances of dying of something else first.

Early-detection tests by the strongest evidence that genuinely exists

Ranked on: the best human evidence for using the test in the population it is being sold to. "Marketed, unproven" here means the test is sold to healthy adults without an approval for that use and without randomised evidence that acting on the result improves any outcome — not that the laboratory is doing anything wrong. Analytical quality and clinical usefulness are different things, and the regulatory system checks the first far more carefully than the second.

1. ApoB

   Approved

   Counts atherogenic lipoprotein particles directly — LDL cholesterol measures the cargo, apoB measures the vehicles.

   The marker itself is not the intervention. What makes apoB strong is that it points to lipid lowering with statins, ezetimibe and PCSK9 inhibitors — among the most robust randomised mortality evidence in all of medicine. In the 20-year ATTICA cohort of 3,042 initially disease-free adults, apoB ≥100 mg/dL predicted major cardiovascular events regardless of hs-CRP, while LDL-C and non-HDL-C predicted only when hs-CRP was elevated.

   Standard clinical laboratory assay; ACC/AHA guidance supports its use to refine cardiovascular risk assessment.PMID 42435999

2. Blood pressure measurement

   Approved

   Arterial pressure, the most-studied modifiable risk factor there is.

   Included here as the benchmark. It is cheap, repeatable, and every step of the chain from measurement to treatment to fewer deaths has randomised evidence behind it. That is the standard every test below should be measured against.

   Universal standard of care; effectively free at home.

3. Low-dose CT lung screening

   Approved

   Lung nodules in people with a heavy smoking history.

   The strongest example of screening working. NLST: 20.0% lung-cancer mortality reduction and 6.7% all-cause mortality reduction. NELSON: lung-cancer death rate ratio 0.76 (0.61–0.94) in men at 10 years. The cost is real too — around 49% of screen-detected cancers may be overdiagnosed, and 96.4% of positive scans in NLST were false positives.

   USPSTF-recommended; performed on FDA-regulated CT devices.PMID 21714641 · NCT00047385

4. Coronary artery calcium (CAC) scoring

   Approved

   Calcified atherosclerotic plaque in the coronary arteries, reported as an Agatston score — a direct measurement of accumulated disease, not a statistical estimate.

   In 7,042 asymptomatic MESA and Dallas Heart Study participants followed 12.3 years, a score ≥100 predicted cardiovascular events across all sex and race groups with no interactions, and improved discrimination and reclassification for coronary heart disease — though not for stroke. What CAC does is reallocate a proven therapy toward people more likely to benefit. The mortality evidence lives in the statin trials, not in the scan, and it should be described in exactly those terms.

   FDA-regulated CT; established risk-stratification use. No randomised trial has ever tested whether CAC screening reduces mortality.PMID 32806939

5. Lipoprotein(a) measurement

   Approved

   An LDL-like particle carrying apolipoprotein(a); concentration is largely genetically set and essentially fixed for life.

   Causally linked to atherosclerotic disease by Mendelian randomisation. But no completed randomised outcomes trial has yet shown that lowering Lp(a) reduces cardiovascular events; the pivotal trials are pelacarsen (NCT04023552) and olpasiran (NCT05581303, NCT07136012). Knowing your Lp(a) changes how aggressively you treat everything that is modifiable, and identifies relatives who should be tested. It does not yet unlock an Lp(a)-specific treatment.

   Standard clinical laboratory assay. Because the level does not move, guidelines converge on one-time lifetime measurement in adults.NCT04023552

6. Screening colonoscopy

   Approved

   Adenomas and cancers, seen and removed directly.

   NordICC, the only randomised trial: 10-year colorectal cancer incidence 0.98% versus 1.20% (RR 0.82, 0.70–0.93), but colorectal-cancer death 0.28% versus 0.31% (RR 0.90, 0.64–1.16, not significant) and all-cause death 11.03% versus 11.04%. Number needed to invite to prevent one cancer: 455 (270–1,429). The incidence reduction is real and substantial; the mortality signal has not been cleanly demonstrated in ten years of follow-up.

   Standard of care worldwide.PMID 36214590 · NCT00883792

7. Cologuard Plus (stool DNA)

   Approved

   Methylated DNA markers plus faecal haemoglobin in stool.

   In BLUE-C (20,176 adults, all with reference colonoscopy): 93.9% sensitivity for colorectal cancer (87.1–97.7), 43.4% for advanced precancerous lesions, 90.6% specificity for advanced neoplasia. Better than FIT on both sensitivities and worse on specificity, meaning more people sent for a colonoscopy they did not need. Note that press coverage widely quotes 95%/94% — use the peer-reviewed figures.

   FDA-approved 4 October 2024 for average-risk adults 45+.PMID 38477986 · NCT04144738

8. Shield (blood-based colorectal screening)

   Approved

   Circulating tumour DNA alterations from colorectal cancer.

   83.1% sensitivity for colorectal cancer, but only 13.2% for advanced precancerous lesions. FDA labelling states the test has limited detection (55–65%) of stage I colorectal cancer and does not detect 87% of precancerous lesions. Since colonoscopy's main value is removing polyps before they become cancer, that gap is the whole story. The American Cancer Society positions blood tests as second-line, for people who will not do colonoscopy or stool testing.

   FDA-approved PMA P230009, July 2024 — the first blood test approved as a primary colorectal screening option, average risk, 45+.PMID 38477985 · NCT04136002 · FDA P230009

9. ctDNA companion diagnostics (Guardant360 CDx, FoundationOne Liquid CDx)

   Approved

   Actionable tumour mutations in people who already have a diagnosed cancer, to match them to targeted therapy.

   A legitimate, regulated, evidence-backed use of liquid biopsy — and one with nothing to do with screening healthy people. It is included here because the credibility of this category is routinely borrowed by tests that have not earned it.

   FDA-approved companion diagnostics, with indications added across lung, breast and colorectal cancer through 2026.

10. p-tau217 blood tests for Alzheimer's (Lumipulse, Elecsys)

    Approved

    Plasma markers of amyloid pathology.

    A genuine advance in the population it was validated in. In clinical validation, 91.7% of positive Lumipulse results had confirmed amyloid pathology and 97.3% of negative results were confirmed negative, with fewer than 20% indeterminate. "Cleared" is a lower bar than "approved": it means substantial equivalence to an existing device, not proof of clinical benefit.

    FDA 510(k)-cleared for patients with signs and symptoms of cognitive impairment — Lumipulse (K242706, 16 May 2025) and Elecsys pTau217 (24 August 2026).FDA K242706

11. p-tau217 ordered in cognitively normal adults

    Approved · off-label use

    The same assay, used as a screening test on people with no symptoms.

    Amyloid can be present 15–20 years before symptoms, and many amyloid-positive older adults never develop dementia. The ~92% PPV was measured where amyloid prevalence is high; in a symptom-free population it will fall sharply. Anti-amyloid therapies are approved for early symptomatic disease, not asymptomatic positivity — so a positive result today changes nothing about management, cannot be un-known, and carries insurance, employment and psychological consequences.

    Outside every current label and guideline. The Alzheimer's Association's first blood-biomarker practice guideline explicitly does not extend to cognitively unimpaired individuals, citing the lack of clinical relevance in that population.DOI 10.1002/alz.70535

12. Galleri (multi-cancer early detection)

    Phase 3 trials

    Methylation signatures in cell-free DNA, across more than 50 cancer types, plus a predicted cancer signal origin.

    The only multi-cancer blood test with a randomised trial, and it missed its primary endpoint: stage III/IV incidence rate ratio 1.03 (0.92–1.14), p=0.6324. Because the trial used a fixed-sequence testing strategy, everything after that failure is formally exploratory. Cancer mortality has not been reported; study completion is January 2031.

    A Laboratory Developed Test — not FDA cleared or approved. Breakthrough Device designation 2018; PMA submitted 29 January 2026; FDA Advisory Committee meeting scheduled 23 September 2026.NCT05611632 · PMID 34176681

13. ctDNA molecular residual disease monitoring (Signatera and similar)

    Phase 3 trials

    Residual or recurrent tumour DNA after curative-intent treatment, before it is visible on imaging.

    That ctDNA-detected recurrence predicts worse outcomes is well established. Whether acting earlier on that information improves survival is not, in most indications. Detecting recurrence sooner and starting the same therapy sooner is the textbook setting for lead-time bias producing impressive survival curves without changing when anyone dies.

    Laboratory Developed Test with Medicare coverage under MolDX; Natera submitted a PMA for a bladder-cancer companion-diagnostic indication on 2 February 2026, supported by the randomised phase 3 IMvigor011 dataset.

14. Whole-body MRI screening (Prenuvo, Ezra)

    Sold · unproven

    Anatomical abnormalities across the body in people with no symptoms.

    A meta-analysis of 12 studies and 5,373 asymptomatic subjects found pooled critical plus indeterminate incidental findings of 32.1% (18.3–50.1) and false-positive findings of 16.0% in the six studies reporting them. Crucially, no included study verified negative findings beyond five years, so the true false-negative rate is unknown: a reassuring whole-body MRI is reassurance of unmeasured value.

    MRI scanners are cleared devices; the screening use is not an approved indication. The American College of Radiology states there is no documented evidence that total body screening is cost-efficient or effective in prolonging life, and the Canadian Association of Radiologists formally opposed the practice in June 2025.PMID 30932247

15. Polygenic risk scores

    Sold · unproven

    Hundreds to millions of common variants summed into a single percentile of genetic liability for a trait.

    No randomised trial has shown that returning a score to a patient improves any outcome, and scores from different vendors for the same disease frequently disagree. Moving from the population median to the top few percent typically shifts lifetime risk two- to three-fold — usually less informative than a family history, a blood pressure reading or a smoking status.

    Effectively unregulated, sold as Laboratory Developed Tests or wellness products. There is no FDA-approved polygenic risk score for population risk stratification.PMID 30926966

16. Whole-genome or exome sequencing in healthy adults

    Sold · unproven

    Monogenic disease risk plus the ACMG secondary-findings list (SF v3.3, 84 genes).

    The one randomised trial, MedSeq, sequenced 100 generally healthy adults: 22% (12–36) received a new monogenic disease-risk result, and 4% (0.01–15) had any evidence of the predicted condition. The authors concluded that adding sequencing to primary care reveals findings of uncertain clinical utility and may prompt additional clinical actions of unclear value. A narrow, high-actionability gene panel in a younger adult is a different and more defensible proposition than an open-ended genome scan in middle age.

    Laboratory Developed Test. No randomised trial of mortality.PMID 28654958 · NCT01736566

The exploratory findings, stated with the caveat attached

• Stage IV cancers, overall: 14% reduction, IRR 0.86 (0.744–0.998) — nominally significant, with an upper confidence bound of 0.998.
• Stage IV by round: 0.91 (0.71–1.18) in round 1, 0.78 (0.57–1.06) in round 2, 0.74 (0.57–0.95) in round 3.
• Stage I–II diagnoses after three rounds: +16%. Emergency-presentation diagnoses: −25%. Cancers first found after symptoms: −21%.
• Test performance in the trial: detection rate 0.48% (937 cancers among 1,801 positive tests), PPV 52.0%, specificity 99.55%, cancer signal origin accuracy 92.5%, and no serious adverse events reported.
• Episode sensitivity was 54.7% for the 12 pre-specified cancers and 30.7% across all cancers — meaning roughly two of every three cancers arising in the screened population were missed.

Polygenic risk scores: the ancestry problem is not a footnote

• A polygenic risk score is a population-level statistical construct — it places you in a percentile of a genetic liability distribution. It is not a diagnosis and it is not a mechanism.
• Scores available today are several times more accurate in people of European ancestry than in anyone else, an inescapable consequence of Eurocentric bias in the genome-wide association studies they are built from. Unlike a drug that works somewhat better in some groups, polygenic scores perform better in European-descent populations essentially universally, so deploying them clinically today would widen health disparities rather than narrow them (PMID 30926966).
• The population-genetic mechanism is well characterised: across 37 UK Biobank traits, roughly 30% of heritability comes from European-specific variants, and under negative selection about half the heritability of some traits can come from variants present in Europe and absent in Africa. The people in the tails of the risk distribution — precisely the ones a score is meant to identify — may not be identified at all when the score was built elsewhere (PMID 33691092).
• No randomised trial has shown that returning a score to a patient improves any clinical outcome. Vendors frequently disagree with each other for the same disease and the same person. And a low score is the more dangerous error: it does nothing to protect against the behaviour and environment that cause most disease, while feeling like permission.
• For scale, in the most aggressive commercial application — polygenic embryo screening — a narrative review calculated absolute risk reductions of roughly 0.02% to 10.1%, implying between 10 and 5,000 IVF patients tested to prevent one affected offspring (PMID 40213363).

Sequencing healthy adults: read the 22% and the 4% together

• In MedSeq, the only randomised trial, 100 generally healthy adults aged 40–65 were randomised to a family-history report alone or family history plus interpreted whole-genome sequencing. 22% (12–36) of sequenced patients received a new monogenic disease-risk result. Only 4% (0.01–15) had any evidence of the predicted condition.
• Genotype is not phenotype. Penetrance measured in unselected populations is generally far lower than penetrance estimated from families who came to attention because they were affected.
• Primary-care physicians recommended new clinical actions for 34% of sequenced patients versus 16% in the family-history arm, and geneticists rated the management of 18% of monogenic results as inappropriate. The authors' own conclusion: sequencing in primary care reveals findings of uncertain clinical utility and may prompt additional clinical actions of unclear value.
• The current ACMG secondary-findings list, SF v3.3 (June 2025), covers 84 genes and adds ABCD1, CYP27A1 and PLN. Only pathogenic or likely pathogenic variants that are clinically actionable are reported — actionability is the inclusion criterion, which is the right principle.
• The realistic downstream picture: variants of uncertain significance vastly outnumber actionable ones, classifications get revised in both directions over time, and a positive monogenic result triggers cascade testing across a family, insurance and disclosure questions, and often lifelong surveillance for something that may never appear.

The four questions worth asking about any early-detection test

• Is there a randomised trial with death or a hard clinical event as the endpoint? If the only evidence is survival, stage distribution, or number of cancers found, the biases in chapter two can produce all of it.
• Is it FDA-approved, FDA-cleared, or a Laboratory Developed Test — and for which indication and which population? A performance figure measured in one population does not transfer to another.
• What treatment does a positive result unlock, and does that treatment have outcome evidence? If the answer is surveillance, the test has bought you monitoring, not health.
• What happens if it is wrong in each direction? Ask about the false-positive pathway and the cost of false reassurance before the blood is drawn, not after.

2025–2026: what actually changed

1. 31 Mar 2025

   Court vacates the FDA's LDT rule

   The US District Court for the Eastern District of Texas vacated and set aside the FDA's 6 May 2024 final rule (89 FR 37286), which had brought Laboratory Developed Tests into the in-vitro-diagnostic device definition, holding that the agency lacks statutory authority over laboratory-developed testing services.

   American Clinical Laboratory Association v. FDA, No. 4:24-CV-479-SDJ (E.D. Tex.)

2. Mar 2025 onward

   USPSTF suspended

   The US Preventive Services Task Force has not met since March 2025 and its chairs were terminated in May 2026. The practical effect is that no new US preventive-services recommendations are being issued while these tests reach the market.

3. 16 May 2025

   First Alzheimer's blood test cleared

   FDA cleared the Lumipulse G pTau 217/β-Amyloid 1-42 Plasma Ratio (K242706) to aid Alzheimer's diagnosis — in symptomatic adults aged 55 and over only.

   FDA 510(k) K242706

4. Jun 2025

   ACMG secondary findings list expands to 84 genes

   SF v3.3 published, adding ABCD1, CYP27A1 and PLN. Only pathogenic or likely pathogenic and clinically actionable variants are reported.

   Lee et al., Genet Med 2025 (PMID 40568962)

5. Jul 2025

   Alzheimer's Association guideline excludes the asymptomatic

   The association's first clinical practice guideline on blood-based biomarkers states explicitly that it does not extend to cognitively unimpaired individuals.

   Palmqvist et al., Alzheimers Dement 2025 (DOI 10.1002/alz.70535)

6. 19 Sep 2025

   FDA implements the vacatur and declines to appeal

   A final rule (Federal Register 2025-18239) removed the added language and reverted 21 CFR 809.3(a), leaving laboratory-developed tests under CLIA alone — outside any FDA premarket review of clinical validity for the foreseeable future.

   Federal Register 2025-18239

7. ~Nov 2025

   NHS-Galleri misses its primary endpoint

   The result becomes public: stage III/IV incidence rate ratio 1.03 (0.92–1.14), p=0.6324. GRAIL's share price falls sharply.

   NCT05611632

8. 29 Jan 2026

   GRAIL submits a PMA for Galleri

   The application draws on roughly 25,490 consented PATHFINDER 2 participants with one year of follow-up, plus more than 70,000 participants from the NHS-Galleri intervention arm.

9. 3 Feb 2026

   Medicare MCED coverage act signed

   The Nancy Gardner Sewell Medicare Multi-Cancer Early Detection Screening Coverage Act creates a Medicare benefit category for multi-cancer tests, letting CMS begin an evidence-based coverage process once a test is FDA-approved.

10. May 2026

    Full NHS-Galleri and PATHFINDER 2 results at ASCO

    The detailed figures on this page were presented at the 2026 ASCO Annual Meeting. They are conference-presented, not peer-reviewed.

    GRAIL / ASCO 2026

11. 24 Aug 2026

    Roche Elecsys pTau217 cleared

    The fourth FDA-cleared blood-based Alzheimer's biomarker, a single-biomarker assay supporting rule-in and rule-out of amyloid pathology in primary and specialty care — again, for patients with symptoms.

12. 23 Sep 2026

    FDA advisory committee on Galleri

    A meeting is scheduled to review the Galleri premarket approval application. It falls after this page's evidence cut-off and its outcome is unknown.