Freenome

The model is the engine of the result, and the signal starts in the laboratory. SimpleScreen CRC reads cell-free DNA methylation at single base resolution, alongside other biomarkers, and machine learning classifiers turn that data into a positive or negative screening result; the company describes the approved test as built on base level cfDNA methylation and AI-enhanced accuracy, and its prospectus describes versioned classifiers deployed under controlled workflows.

The assay chemistry carries as much of the result as the model does, and the company attributes part of its next generation gains to a 2.6 fold reduction in the assay's limit of detection, so the product is a laboratory test with a learned component at its core.

What the test may and may not do is published, with the next step for each result and the limits that matter most. SimpleScreen CRC is indicated for adults aged 45 and older at average risk who are referred by a healthcare provider, and is not a replacement for diagnostic colonoscopy or for surveillance colonoscopy in high risk people. A positive result is followed by colonoscopy; after a negative result screening continues on clinical judgment, and a repeat test is due in three years if the patient chooses it again. The FDA precaution on the product page tells the clinician what a negative result cannot rule out: detection of stage I cancer is limited and most advanced precancerous lesions are missed, so one in ten patients with a negative result may have a precancer a colonoscopy would have found.

The lung test sold alongside it states its own route (a positive result goes to low-dose CT) and that it has not been reviewed by FDA.

The approach and the versions are named; the training data for the approved classifier is not characterized in what the company publishes. The method is described: cell-free DNA methylation read at single base resolution, alongside protein and autoantibody signals from the wider platform, with machine learning classifiers producing the result. The prospectus describes versioned classifiers deployed under controlled workflows for traceability, performance monitoring and regulatory review, and a fragment level deep learning architecture in research, in which an attention mechanism over fragment embeddings predicts cancer status.

Versions are explicit: the first generation test is the one FDA approved, and a next generation version with assay and algorithm changes has completed its validation study. A significant design change to an approved test needs FDA approval of a supplement or new application, which leaves a public record of when the approved test changes. Neither the data that trained the approved classifier nor how its performance is monitored after launch is described.

A structurally short chain for the model: the classifiers are Freenome's own and run on data from its own laboratory. The named parties sit around it: Abbott as exclusive US commercial partner for the colorectal test, with technology transfer under way; commercial couriers for samples; and, in the privacy notice, unnamed IT vendors and cloud service providers.

The prospectus describes a planned data chain that is not yet named: tokenization vendors to follow consented patients in real world data, and vendors supporting bidirectional exchange and secure cloud environments with partner health systems, whose records would train future models. No subprocessor list is published, and no party that receives patient data for model development is named.

Prospective, blinded, multicenter validation in the intended use population, published in a major journal, with one prespecified endpoint missed. PREEMPT CRC (Shaukat et al., JAMA 2025) enrolled 48,995 average risk adults aged 45 to 85 at 201 centers across 49 states and the United Arab Emirates, with colonoscopy and central pathology as the reference and all parties blinded. In 27,010 evaluable participants, colorectal cancer sensitivity was 79.2 percent (57 of 72), specificity for advanced neoplasia 91.5 percent, negative predictive value 90.8 percent and positive predictive value 15.5 percent, meeting all primary endpoints. Sensitivity for advanced precancerous lesions was 12.5 percent (321 of 2,567) and did not meet its prespecified criterion. Adjusted to the US Census age and sex mix, the figures FDA approved on are 81.1 percent, 13.7 percent and 90.4 percent.

A next generation version met its endpoints in a July 2026 validation on previously unevaluated PREEMPT samples (80.4 percent cancer sensitivity, 18.2 percent for precancerous lesions, census adjusted, at 90 percent specificity); the prospectus calls those data preliminary and they are not yet peer reviewed. Blood-based testing is in the American Cancer Society's 2026 colorectal screening guideline.

Categorical commitments are published, retention is general, and the plan to learn from patient data is stated to investors rather than patients. Freenome's HIPAA notice of privacy practices (effective 13 July 2026) states that it never shares a patient's information for marketing or sale without written permission, and that it may use or share information for health research. Its privacy notice (last updated 13 July 2026) says it does not sell personal information, may share de-identified data with third parties for any lawful purpose, and keeps information as long as necessary or as law requires.

The company's prospectus describes the other half: a data flywheel in which molecular data from its tests, real world data tokenization that follows consented patients over time, and records from partner health systems feed the models that improve existing tests and create new ones. The patient facing notices do not say how long a blood sample or its sequencing data is kept, whether a commercial patient's data trains models by default, or how to withdraw. Safety engineering for the classifiers is not published.

Covered entity status is evident and a substantive privacy document reaches the patient; no contracting instrument is published. Freenome runs the laboratory that performs the test and publishes a HIPAA notice of privacy practices (effective 13 July 2026) covering patient rights, uses and disclosures, breach notification and a privacy officer contact, and its separate privacy notice states that it does not govern protected health information and points to the HIPAA notice.

No provider agreement, data protection addendum or business associate terms for health system customers are published, and nothing states how subcontractors handling PHI are bound, which matters more as the colorectal test moves to Abbott and the company plans data partnerships with health systems.

Controls are described and an outside reporting channel is published. The site links a Freenome Trust Center hosted on Vanta, and its contents may say more. The coordinated vulnerability disclosure policy invites reports from researchers, commits to acknowledging them within five business days, offers safe harbor for good faith reporting and states an intent to publish advisories; there is no bug bounty. The privacy notice names encryption, access and authentication controls, data integrity controls, continuous security monitoring and event logging.

Neither the website nor the prospectus names a SOC 2, HITRUST or ISO/IEC 27001 attestation, and the laboratory's CLIA certification covers testing quality rather than information security.

An approval identifiable in FDA's database, with its indication. SimpleScreen CRC was approved through premarket approval P250029 on 24 July 2026, indicated for colorectal cancer screening in adults aged 45 and older at average risk, with positive results followed by colonoscopy and no role as a replacement for diagnostic or surveillance colonoscopy in high risk people; supplement S001 (26 August 2026) covers a change in a supplier's manufacturing process for critical reagents. The product page states the approval and the labeled precaution.

The positions around it are stated too. SimpleScreen Lung is labeled a laboratory developed test not reviewed, cleared or approved by FDA. A next generation colorectal version has met its study endpoints; under the rules the prospectus sets out, a significant design change to an approved test needs FDA approval of a supplement or new application, so the version approved today is the first generation test.

Subgroup performance is published across clinical and demographic groups, including where the test is weak. The PREEMPT CRC validation (Shaukat et al., JAMA 2025; 27,010 evaluable participants aged 45 to 85, 11.2 percent Black, 11.8 percent Hispanic or Latino) reports colorectal cancer sensitivity by stage (57.1 percent at stage I against 82 to 100 percent at stages II to IV), advanced precancerous lesion sensitivity by lesion type and size, and results by demographic group in its supplement: the paper reports no substantial difference in cancer sensitivity by demographic characteristics, specificity falling with age, and precancerous lesion sensitivity rising with age. A prespecified follow up (Shaukat et al., Journal of the National Cancer Institute 2026) adjusts the results to the US Census age and sex mix and reports predictive values by age group.

The weak spots are stated on the product page as well as in the papers: limited detection of stage I cancer and most advanced precancerous lesions missed. The studies were sponsored by the company, with academic investigators leading them.

Error rates are published with method and denominators, on the product page as well as in the literature; nothing makes the company answerable for a result. The SimpleScreen page carries the FDA precaution: limited detection (57 percent) of stage I colorectal cancer, 87 percent of advanced precancerous lesions not detected, and one in ten patients with a negative result possibly having a precancer a colonoscopy would have found. The JAMA validation behind those figures gives the counts: 57 of 72 cancers detected, 321 of 2,567 advanced precancerous lesions detected, specificity for advanced neoplasia 91.5 percent (22,306 of 24,371), positive predictive value 15.5 percent. The paper also states that precancerous lesion sensitivity did not meet its prespecified acceptance criterion.

No guarantee, indemnity or remediation obligation attaches to a result, and no correction route for a patient or clinician is described.

Integration is claimed with no system named. The health systems page offers Founding Partners Freenome Digital Health, described as EHR-native digital tools that identify eligible patients, engage patients and care teams and fit screening into routine care, with clinical and EHR workflow optimization as part of the program. No record system, interface standard, order or result path, or live deployment is named, and the prospectus says the company does not yet have any health system data partnerships. Abbott's Nexus platform is named as part of how Abbott will connect patients to screening, which describes Abbott's channel rather than Freenome's integration.

For a screening test repeated every three years, with a positive result that must lead to colonoscopy, the order and result path inside the record is what makes follow up reliable.

A laboratory service at one named site, with the computing layer unnamed. Samples reach Freenome's CLIA certified laboratory in Brisbane, California by commercial courier, and the prospectus names the single facility and the couriers as dependencies. The classifiers run on cloud computing infrastructure from third party providers the company does not name, and the privacy notice lists cloud service providers among the parties that receive personal information.

No provider, region, residency commitment or tenancy model is published. The licensing agreement with Abbott includes technology transfer, which the company is already being paid for, so where the colorectal test is run once Abbott commercializes it at scale is an open question for a buyer.

No price is published and the posture is discoverable. Abbott holds exclusive US rights to commercialize SimpleScreen CRC and plans to launch it in the fall of 2026 through the channels it built for Cologuard, so for the colorectal test alone the buyer's counterparty is Abbott. Freenome sells its own configuration to health systems through a Founding Partner program: SimpleScreen Lung, a laboratory developed test, with SimpleScreen CRC and its digital tools. The company states that the approved test meets Medicare's coverage criteria.

The partnership economics are public (a $100 million milestone on this approval, further milestones and royalties under the August 2025 agreement), and they describe what Abbott pays Freenome rather than what a health system or patient pays for a test. No list price, rate or program fee is published by either company.

Coverage is named, and validation covers the colorectal half. SimpleScreen CRC is for adults aged 45 and older at average risk who have not completed colonoscopy or stool testing, which matches the population PREEMPT CRC enrolled (average risk adults aged 45 to 85) and the FDA indication.

The company also sells SimpleScreen Lung to health systems in its Founding Partner program, for adults aged 50 to 80 with a 20 pack-year smoking history who have not had low-dose CT screening. Its performance appears in a 2026 conference abstract, with no peer reviewed validation listed, so that part of the named coverage runs ahead of its evidence.