The clues to the heart attack that killed a Stanford researcher weren’t hiding in his anatomy — they were stored on his wrist.

The man, a 76-year-old visiting scholar in geneticist Michael Snyder’s lab, collapsed moments after a workout. When Snyder later analyzed the data from the man’s smartwatch, he found a subtle, sinister trajectory had been unfolding for months.

“You could see a step change 4 months earlier,” said Snyder, who has pioneered methods combining wearable sensors, molecular monitoring, and machine learning to decode genetic risk in real time. Six parameters had shifted: His heart rate variability dropped, while his heart rate went up. “But there was no mechanism to relay the information back to him.”

Snyder is highlighting a long-standing gap in medicine: the lethal disconnect between what the body reveals and what medicine acts on.

For more than half a century, across dozens of diseases, researchers have been striving to close that gap. Tools now span blood tests that screen for 50 cancers at once, plasma assays that flag the pathology of Alzheimer’s disease, AI systems that decode genetic risk, and wearables streaming data in real time. In some cases, these technologies can detect disease years or even decades before symptoms appear. But while the laboratory promise of “early detection” has never been brighter, the failure of a recent high-profile randomized trial to meet its endpoint shows how complex the transition to the real world can be.

“You don’t drive a car around without a dashboard. I would argue it’s just as crazy to go around without a health monitor,” Snyder said.

For now, we’re still driving blind.

A Dashboard for the Human Body

Scientists are now assembling that “dashboard.” Snyder’s contribution is a high-resolution genetic lens.

Most diseases aren’t caused by a single mutation like BRCA (linked to breast and ovarian cancers) but rather are influenced by hundreds or thousands of genetic variants, each contributing a small effect, Snyder said. Polygenic risk scores (PRSs) sum them all up into a single risk estimate. But Snyder argues the math is wrong.

“One plus one is not always two,” Snyder said. “If you have two mutations in the same pathway, one plus one is one — they won’t be additive. And sometimes you have two mutations in different pathways, and one plus one can be 15.”

To spot the synergy, you need to know which genes are active in which tissues. A new tool from Snyder’s team can reveal just that. Single-cell PRSs (scPRSs) are graph neural networks that analyze genetic variants at single-cell resolution, capturing interactions that linear math models miss. In a Nature Biotechnology study, the method outperformed traditional PRSs across type 2 diabetes, hypertrophic cardiomyopathy, Alzheimer’s disease, and severe COVID.

In diabetes, it exposed a major blind spot: Genetic risk isn’t driven by insulin-producing beta-cells alone, as experts have long assumed, but also by glucagon-producing alpha-cells — a unique mechanism that flat, linear models had missed.

You could think of scPRS as a high-resolution lens for drug discovery, revealing both the probability of disease and the cellular circuits that need fixing.

“Right now, the American Diabetes Association dietary guidelines do not work that well because they lump everyone together,” said Snyder, whose own longitudinal omics profiling, documented in a 2012 Cell study, revealed a predisposition to type 2 diabetes despite fitting none of the conventional risk factors.

The same logic applies to genetic risk: Different people may need different interventions based on which cellular pathways drive their disease.

But the scPRS alone isn’t enough to categorize patients for treatment. Truly knowing a person’s subtype requires real-time visibility into how their body responds to the environment. Combined with their personal scPRS map, these clues could help forecast medical problems, as with the Stanford researcher’s heart attack, giving clinicians more time to intervene.

Metabolites That Predict Diabetes

If Snyder is looking at the engine’s flaws, Qibin Qi, PhD, examines the exhaust fumes. His team at the Albert Einstein College of Medicine in Bronx, New York, looks for evidence of impending disease in metabolites, the chemical leftovers of our biology. In a January study published in Nature Medicine, they pooled metabolomic data from more than 23,500 people, following them for up to 26 years. They identified 235 circulating metabolites associated with incident diabetes.

Pathways spanned bile acids, lipids, amino acids, and gut-derived compounds — reflecting dysfunction across the liver, pancreas, adipose tissue, and digestive system. “Type 2 diabetes is not just a disease of one or two organs,” Qi said.

From there, the team built a 44-molecule panel. It improved diabetes risk prediction beyond conventional factors such as fasting glucose, BMI, and family history. “Our signature can predict type 2 diabetes decades in advance,” Qi said. It could someday help personalize prevention because different people with similar risk may have distinct metabolic profiles. “That means we may need a different intervention strategy,” Qi said.

Lifestyle factors — diet, physical activity, and body weight — explained substantial variation in these disease-linked metabolites, said Qi, who predicted clinical translation is likely 5-10 years away.

Cancer’s Shadow in the Blood

But perhaps the most high-stakes “dashboard” light of all is the one that signals cancer. Multi-cancer early detection (MCED) blood tests exploit a unique biology: cell-free DNA.

When cancer cells release DNA into the bloodstream, they carry distinctive “methylation” patterns — chemical tags that help cells remember their identity. Cancer hijacks this machinery to fuel growth, but the DNA’s original tissue signature remains. By analyzing these tags, a single blood test can reveal the presence of cancer and its organ of origin.

Urologist Eric A. Klein, MD, a distinguished scientist at GRAIL, has spent years developing and validating the company’s trailblazing MCED test, Galleri. The test analyzes methylation signatures from a single blood draw, scanning for signals across more than 50 cancer types simultaneously.

The need is stark. The US currently screens for just five cancers — breast, colorectal, cervical, lung, and prostate — and only 14% of all cancers are caught through screening. “More than 80% of [cancer] deaths come from cancers we don’t screen for,” Klein said.

Early results were promising: In GRAIL’s PATHFINDER 2 study, the test achieved 99.6% specificity, predicted the cancer’s tissue of origin 92% of the time, and found more than half of the cancers at stage I or II. In another study, some patients initially classified as false positives were later diagnosed with the cancer the test had predicted. Advocates marveled: The test was seeing tumors before imaging could.

Still, independent analyses hinted at a flaw. A 2025 systematic review by the UK’s National Institute for Health and Care Research (NIHR) across 13 MCED tests found they detect late-stage cancer far better than early-stage disease. For Galleri, sensitivity for stages I and II ranged from 27% to 37% but jumped to 84%-90% for stages III and IV. The “early detection” test was best at finding cancer that wasn’t early at all.

In February 2026, the field met its first real test — and the results were complicated.

The National Health Service (NHS)-Galleri trial — which screened 142,000 participants in England — failed to meet its primary endpoint: a statistically significant reduction in combined stage III and IV cancer diagnoses.

“It was a very high bar,” said Josh Ofman, MD, president of GRAIL. The endpoint was chosen to inform a policy decision: “basically a commercial agreement” with NHS, Ofman said. A trial designed purely to evaluate the test might have had a less ambitious threshold, such as stage IV reduction alone or a longer follow-up window, he said.

It was a huge gamble, said Anna Schuh, MD, PhD, a professor of genomic medicine at the University of Oxford in Oxford, England, who is developing an MCED test called TriOx. “Nobody really knows, in a multi-cancer setting, how long it will take to downstage — whether it’s 3 years or 5 years or 10 years.”

Schuh also highlighted a key constraint: Galleri’s bisulfite sequencing method destroys roughly 80% of input DNA. In early-stage cancers where tumor fragments are already scarce, that loss matters. “Retrospectively, one wouldn’t do that anymore,” she said.

The challenge runs deeper than any single assay. A pea-sized tumor — roughly 1 cm³, near the limit of what imaging can see — may shed DNA into the blood at a ratio of 1 in 100,000 fragments. Current technology can only reliably detect tumor DNA at a 1-in-10,000 scale.

Compounding the challenge is the noise floor. As humans age, noncancerous mutations accumulate in blood cells — a process called clonal hematopoiesis — that can trigger the same assays designed to catch tumors. Furthermore, most performance data came from case-control studies comparing blood from diagnosed patients against samples from healthy volunteers — a methodology that inflates accuracy because it’s confirming diagnosis rather than discovering one, Schuh said.

The Clinical Dilemma of Early Detection

These tests raise a clinical dilemma, too: the people it gets wrong.

Even at 99.6% specificity, screening millions of healthy adults generates thousands of false positives. The NIHR review found that time to diagnostic resolution was prolonged, particularly for those cases, with patients shuttled through imaging, biopsies, and specialist consultations, chasing a cancer that wasn’t there. Even more alarming: What if the test is right, but imaging can’t confirm it? It leaves patients in a medical purgatory: told they have cancer but unable to treat it.

The problem isn’t unique to cancer. In May 2025, the FDA cleared the first blood test for pathology of Alzheimer’s disease. The assay measures phosphorylated tau 217, a protein modification that rises in lockstep with amyloid plaque formation — the slow-building brain pathology that can precede memory loss by decades. A “clock” model built on the biomarker, published in Nature Medicine earlier this year, predicted the onset of symptoms of Alzheimer’s disease to within 3-4 years.

But detection is not destiny. Amyloid prevalence even in healthy adults can range from 10% at age 50 to 44% at age 90, and many will never develop dementia. As with MCEDs, the Alzheimer’s disease test may reveal results that medicine can’t act on yet.

Despite missing its endpoint, the NHS-Galleri trial was not a clean failure. Stage IV diagnoses in a prespecified group of 12 deadly cancers with no existing screening (including pancreatic, ovarian, lung, and liver) fell by more than 20%. And adding Galleri to standard-of-care screening produced a fourfold increase in the overall cancer detection rate. GRAIL has submitted its FDA premarket approval application and is extending the trial’s follow-up by 6-12 months, allowing more time for cancers to surface.

“This is far too important, and far too ambitious yet realizable a dream, for us to just say, ‘This failed, it’s time to move on,’” said Dan Landau, MD, PhD, a cancer geneticist at Weill Cornell Medicine in New York City.

Even imperfect tests can shift outcomes when applied at scale. The fecal occult blood test for colorectal cancer has a sensitivity as low as 13% — yet it has saved tens of thousands of lives. Modeling suggests that current MCED technology could still reduce cancer deaths by 17%-26% if widely adopted.

The question is whether current tests are good enough to start — or if we must wait for the next generation of technology to justify mass screening. That answer won’t be settled by lab performance. It will come from trials like NHS-Galleri as we wait to see if a flashing light on a dashboard actually keeps the car from crashing.

Ofman and Klein reported being employees of GRAIL; studies of the Galleri test were funded by GRAIL.

Schuh reported being the co-founder and chief medical officer of SerenOx, to which the TriOx multi-cancer detection test is licensed. TriOx research was conducted in collaboration with Exact Sciences.

Landau reported being the co-founder of and holding equity in C2i Genomics, a liquid biopsy company focused on residual disease detection. 

Snyder reported being the co-founder or scientific advisor of multiple precision medicine companies, including Personalis (cancer genomics), Qbio (health risk prediction), and January AI (glucose monitoring). 

Qibin Qi reported having no relevant financial disclosures.