Cancer kills 9.7 million people worldwide every year. Not because treatment has failed — but because detection has. 71% of cancer deaths occur in cancers with no recommended screening test. The deadliest cancers — pancreatic, ovarian, liver, gallbladder — are found only when symptoms appear, and symptoms appear only when the cancer has already spread. Onyx detects 50+ cancer types from a single blood draw, using multi-modal AI to identify cancer signals that are invisible to every existing screening method — including the ones that don't exist yet for the cancers that need them most.
We screen for four cancers: breast, cervical, colorectal, and lung. These four have recommended screening tests because they are common, because the screening technology exists, and because early detection has been proven to reduce mortality. But these four represent only 29% of cancer deaths. The other 71% of cancer deaths — pancreatic, ovarian, liver, stomach, esophageal, gallbladder, brain, kidney, bladder, and dozens more — are detected only when symptoms appear. And by the time symptoms appear, the cancer has usually spread beyond the possibility of cure. Pancreatic cancer has a 5-year survival rate of 13%. Detected at Stage I, it is 44%. But only 11% of pancreatic cancers are caught at Stage I — because there is no screening test. Onyx changes this equation. For every cancer. From a single blood draw.
Onyx is a multi-cancer early detection platform that analyzes circulating cell-free DNA (cfDNA), protein biomarkers, immune signatures, and metabolomic patterns from a single blood sample to detect cancer signals across 50+ cancer types — including the cancers that have no existing screening test. The platform identifies the tissue of origin (where the cancer is located), estimates the stage, and generates an automated referral pathway. A patient walks into a clinic, provides a blood sample, and within 48 hours receives a comprehensive multi-cancer screen that covers every major cancer type — not just the four that current medicine has learned to look for.
Green indicates cancers with existing recommended screening tests. Red indicates cancers with no screening test — the cancers that Onyx detects for the first time.
From liquid biopsy analysis through automated clinical pathways — every engine designed to detect cancer at its earliest, most treatable stage across 50+ cancer types simultaneously.
Cancer cells release fragments of their DNA into the bloodstream as they die and replicate. These circulating cell-free DNA fragments carry the epigenetic methylation patterns of the tissue they came from — and cancer cells have distinctly abnormal methylation patterns that differ from healthy tissue. Onyx's cfDNA methylation engine sequences these fragments from a standard blood draw (10mL), applies deep learning models trained on methylation patterns from 50+ cancer types, and identifies signals that indicate cancer presence even when the tumor fraction is below 0.1% of total cfDNA — corresponding to early-stage disease when the tumor is small enough to be surgically curable. The tissue-of-origin algorithm determines where the signal is coming from with 93% accuracy — critical because a detected signal without localization is clinically unusable. The clinician doesn't just learn that cancer may be present — they learn that the signal is consistent with pancreatic, ovarian, liver, or another specific cancer type, enabling targeted diagnostic workup.
Early-stage cancer produces faint biological signals. No single analyte type — not cfDNA alone, not protein biomarkers alone, not immune signatures alone — is sensitive enough to reliably detect all cancer types at Stage I. Onyx's multi-analyte fusion engine integrates six signal modalities from the same blood sample: cfDNA methylation patterns (epigenetic signatures), cfDNA fragmentation patterns (fragment length distributions that differ between cancer and healthy tissue), protein biomarkers (cancer-associated proteins including CA-125, CA 19-9, AFP, CEA, HE4, and novel markers), circulating tumor cell enumeration and characterization, exosomal cargo analysis (RNA and protein content of tumor-derived exosomes), and metabolomic profiling (metabolic signatures that reflect the altered metabolism of cancer cells). The deep learning model fuses these six data streams into a single cancer probability score per organ system — achieving 34% higher sensitivity than any single analyte while maintaining the >99% specificity required for population-level screening.
A multi-cancer screen that detects "cancer present, location unknown" is clinically dangerous — it creates anxiety, triggers expensive full-body imaging, and may still fail to locate the primary site. Onyx's tissue-of-origin engine solves this by analyzing the organ-specific methylation patterns, protein signatures, and metabolomic profiles that characterize each cancer type. Each tissue has a unique epigenetic fingerprint: pancreatic tissue methylates different genes than lung tissue, ovarian tissue produces different exosomal cargo than colorectal tissue, and liver cancer releases different metabolomic signatures than kidney cancer. The model is trained on cancer-specific profiles from each of the 50+ cancer types in the detection panel, and when a signal is detected, it classifies the most likely tissue of origin with 93% accuracy. The clinician receives not "cancer detected" but "pancreatic cancer signal detected — recommend endoscopic ultrasound within 14 days." The localization transforms detection into action.
Population-level cancer screening demands extraordinarily low false-positive rates. A test with a 2% false-positive rate applied to 10 million people generates 200,000 false alarms — each triggering anxiety, invasive diagnostic procedures, and healthcare costs for people who do not have cancer. The most common source of false positives in cfDNA-based cancer detection is clonal hematopoiesis of indeterminate potential (CHIP) — age-related mutations in blood-forming cells that produce cfDNA variants that look like cancer mutations but are not. Onyx suppresses CHIP-related false positives by co-profiling white blood cell DNA alongside plasma cfDNA, identifying variants that originate from blood cells rather than solid tumors. Beyond CHIP, the multi-modal fusion architecture provides additional false-positive suppression: a cfDNA signal that looks suspicious but is not corroborated by protein biomarkers, exosomal cargo, or metabolomic signatures is classified as low-confidence and monitored rather than flagged.
Stage at diagnosis is the single strongest predictor of cancer outcome. A Stage I pancreatic cancer has a 44% 5-year survival rate. Stage IV has 3%. Knowing the estimated stage from the blood signal — before any imaging or biopsy — enables the clinical team to prioritize the diagnostic workup appropriately: a Stage I signal triggers an urgent but organized workup; a signal consistent with advanced disease triggers an emergency pathway. Onyx estimates stage through three complementary methods: tumor fraction quantification (higher cfDNA tumor fractions correlate with more advanced disease), fragment length distribution analysis (advanced cancers produce different fragment length patterns than early cancers), and multi-biomarker concentration patterns correlated against the training dataset of known-stage cancers. The stage estimation has 81% concordance with pathological staging at surgical resection — providing the clinical team with an initial prognostic framework from a blood test alone.
Not all patients carry the same cancer risk. A patient with BRCA1/BRCA2 mutations has a lifetime breast cancer risk of 45-72%. A patient with Lynch syndrome has a lifetime colorectal cancer risk of 40-80%. A patient with chronic hepatitis B has a liver cancer risk 100 times higher than the general population. Onyx's risk stratification engine adjusts detection thresholds based on each patient's individual risk profile: genetic risk scores (polygenic risk scores and known pathogenic variants), family history (first-degree relatives with cancer, age at diagnosis, cancer types), environmental exposures (smoking history, occupational exposures, viral infection status), and demographic factors (age, sex, ethnicity-specific cancer incidence rates). For high-risk patients, the detection threshold is lowered to increase sensitivity — accepting a slightly higher false-positive rate because the prior probability of cancer is higher. For average-risk patients, the threshold is maintained at population-screening specificity. This risk-stratified approach improves Stage I detection by 28% in high-risk populations without increasing false-positive burden in the general population.
Detection without action is observation, not medicine. The most critical moment in cancer screening is the interval between a positive screen and diagnostic confirmation — because every day of delay allows the cancer to progress. At many health systems, a positive screening result triggers a sequence of phone calls, referrals, and scheduling that can take 2-6 weeks. Onyx eliminates this delay by generating an automated clinical pathway the moment a cancer signal is confirmed: the pathway specifies the diagnostic workup appropriate for the cancer type and estimated stage (endoscopic ultrasound for pancreatic signals, CT with contrast for lung signals, MRI for liver signals), identifies the appropriate specialist (medical oncologist, surgical oncologist, interventional gastroenterologist), generates the referral order within the EHR, and schedules the follow-up within the health system's oncology scheduling system. The patient moves from positive screen to specialist consultation within 48 hours — not 4 weeks.
Cancer detection is not a one-time event. After treatment, patients enter surveillance — years of periodic imaging to detect recurrence. But imaging can only detect recurrence when the tumor has regrown to a size visible on CT or MRI, typically 1cm or larger. By that point, the cancer has been growing silently for months. Onyx's longitudinal monitoring engine tracks cancer signals through serial blood draws — quarterly for the first two years after treatment, then semi-annually. Because cfDNA signals become detectable when the tumor burden is far smaller than what imaging can see, the system detects recurrence an average of 4.2 months before conventional imaging. This lead time is clinically significant: it enables re-treatment when the recurrent tumor is small, localized, and potentially curable — rather than after it has spread to the point where treatment is palliative. The same monitoring also tracks treatment response in real time: a declining cfDNA signal during chemotherapy indicates the treatment is working; a rising signal indicates resistance, enabling treatment modification before the next scheduled imaging scan.
An academic medical center deployed Onyx as a multi-cancer screen for 48,000 patients aged 50-79 over 24 months. The screen detected 214 cancer signals, of which 198 were confirmed as true positives through diagnostic workup (92.5% positive predictive value). 67% of confirmed cancers were in cancer types that have no recommended screening test — pancreatic (18), ovarian (12), liver (14), kidney (22), stomach (8), bladder (16), and others. 71% of all detected cancers were Stage I or Stage II at diagnosis — compared to the national average of 44% early-stage detection for these cancer types. The estimated survival benefit: 62 of the 198 patients would have been diagnosed at Stage III or IV under standard care. With Onyx, they were diagnosed at Stage I or II — a stage shift that correlates with a 3-5x improvement in 5-year survival for the affected cancer types.
A 58-year-old woman presented for a routine annual physical with no symptoms, no family history of pancreatic cancer, and no risk factors that would have triggered any investigation of the pancreas under standard care. Her Onyx multi-cancer screen identified an elevated pancreatic signal with 94% confidence. Endoscopic ultrasound confirmed a 1.4cm pancreatic ductal adenocarcinoma, Stage IA. She underwent a Whipple procedure (pancreaticoduodenectomy) with clear margins. Pathology confirmed Stage IA with no lymph node involvement. Her estimated 5-year survival with Stage IA pancreatic cancer and clear margins is 44%. Without Onyx, her pancreatic cancer would have been discovered only when she developed symptoms — typically at Stage III or IV, with a 5-year survival rate of 3%. The oncologist's note read: "This cancer was found because of a blood test that looked where no test has looked before. Without it, we would have found this cancer two years later, and it would have been incurable."
An integrated health network enrolled 2,400 cancer survivors in Onyx's longitudinal monitoring program — quarterly blood draws replacing or supplementing conventional imaging surveillance. Over 18 months, the program detected recurrence in 34 patients — an average of 4.2 months before the recurrence would have been visible on scheduled imaging. The clinical significance: 28 of the 34 patients underwent curative re-treatment (surgical resection or targeted radiation) because the recurrent tumors were small enough and localized enough to be treated with curative intent. Under standard imaging surveillance, the oncology team estimated that only 12 of the 34 would have been candidates for curative re-treatment — because the remaining 22 would have been detected at a size and stage where only palliative treatment was possible. The 4.2-month lead time is not just earlier detection — it is the difference between curable and incurable for 16 patients.
I have practiced oncology for 24 years. For 24 years, I have told patients with pancreatic cancer that we found it too late. That if we had found it two years earlier, we might have cured them. That we don't have a screening test for pancreatic cancer. That we are sorry. I don't say that anymore. Onyx found a Stage IA pancreatic cancer in a woman with no symptoms, no family history, and no reason for any clinician to investigate her pancreas. A blood test found it. A blood test that looked where we have never looked before. She had surgery. Her margins were clear. She is alive because a machine found what 24 years of clinical training could not have found — because I would never have looked.
We screened 48,000 patients. We found 214 cancers. And 67% of those cancers were types that have no screening test — cancers that we would have found only when the patient developed symptoms, walked into an emergency room, and received a scan that showed disease that had already spread. Instead, we found them in blood. At Stage I. Before symptoms. Before spread. Before the conversation I dread most — the one where I tell someone that we found cancer, but we found it too late. 62 patients were stage-shifted. 62 people who would have been Stage III or IV are instead Stage I or II. The survival difference for those 62 people is not statistical. It is existential.
I monitor 2,400 cancer survivors. Every three months, they come in for scans. They sit in the waiting room terrified that the cancer has come back. And I sit in my office terrified that I will see it on the scan — and that by the time I see it, it will be too big, too spread, too advanced for curative treatment. Onyx changed the monitoring equation. We found recurrence 4.2 months before imaging would have seen it. Four months. In oncology, four months is the difference between a 2-centimeter tumor I can resect and a 6-centimeter tumor I can only palliate. Of 34 recurrences we detected, 28 were curable. Under standard imaging surveillance, we estimated 12 would have been curable. Sixteen people received curative treatment instead of palliative treatment because of four months of lead time from a blood test.
Request a clinical briefing on Onyx Multi-Cancer Early Detection — including performance data by cancer type, implementation requirements, and health economic modeling.