Industries / Health / Pharmaceuticals / Biotech
Diagnostics
This work builds tests from biological molecules, such as antibodies, enzymes and DNA, that find and measure signs of disease in blood, swabs and other samples.
7
Milestones achieved, from 1959 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1959
Antibodies measure hormones
Rosalyn Yalow and Solomon Berson published a way to measure insulin in human blood using antibodies and a radioactive label, a method called radioimmunoassay. Their paper appeared in Nature in 1959, and Yalow received the 1977 Nobel Prize in Physiology or Medicine for it, according to the University of Illinois Grainger College of Engineering.
Lessontheir 1956 finding of insulin antibodies was first doubted, Grainger notes, and within a few years became a practical test.
Credit
Rosalyn Yalow and Solomon Berson
1971
Enzymes replace radioactivity
Eva Engvall and Peter Perlmann of Stockholm University published the enzyme-linked immunosorbent assay (ELISA). It works like radioimmunoassay, but an enzyme that makes a colour change reports the result instead of a radioactive tag, so ordinary labs could run antibody tests.
Credit
Eva Engvall and Peter Perlmann
1985
Copying DNA to diagnose
A team including Randall Saiki and Kary Mullis described a way to copy a chosen stretch of human DNA over and over, later named the polymerase chain reaction (PCR), and used it to diagnose sickle-cell anaemia. Their Science paper reports a 220,000-fold increase in copies of the target and a result in under a day from well under 1 microgram of DNA.
Credit
Saiki, Scharf, Faloona, Mullis, Horn, Erlich and Arnheim
2016
A blood-testing lab shut
Theranos said a few drops of blood from a finger prick could run a wide range of tests. In July 2016 the US Centers for Medicare & Medicaid Services (CMS), which oversees clinical labs, revoked the federal certificate (CLIA) of its Newark, California lab after a 2015 inspection found deficient practices, and barred Elizabeth Holmes from owning or running a lab for at least two years. Holmes was sentenced on 18 November 2022 to 135 months in prison for defrauding investors.
Lessonthe claims fell apart once an outside inspector looked at how the lab worked.
Credit
CMS and US federal prosecutors
2020
A failed kit, then cheap tests
Some of the first COVID-19 test kits that the US Centers for Disease Control and Prevention (CDC) sent to public health labs in early 2020 "could not be verified" by those labs. The inspector general of the US health department found weaknesses in how CDC built the kits and in its lab quality processes. In August 2020 the FDA gave emergency authorisation to Abbott's BinaxNOW, a 15-minute paper-strip antigen test, which detects a virus protein, priced at USD 5 per test.
Lessonthe inspector general recommended written procedures for developing test kits and quality systems for CDC's labs, and recorded all six recommendations as put in place by 2025.
Credit
HHS Office of Inspector General (review); Abbott (BinaxNOW)
2024
A blood test for colon cancer
On 26 July 2024 the FDA approved Guardant Health's Shield blood test to screen adults aged 45 and over at average risk of colorectal cancer. The test reads fragments of tumour DNA that circulate in blood. The FDA reports that in a study of 7,861 people, Shield found colorectal cancer 83% of the time, and about 1 in 10 people without advanced disease got a false positive.
Credit
Guardant Health; FDA
2025
Court strikes down test rule
In 2024 the FDA issued a rule to regulate laboratory-developed tests, tests designed and run inside a single lab, as medical devices. On 31 March 2025 a US federal court in the Eastern District of Texas vacated the whole rule, holding that these tests are services, not devices under US food and drug law.
Lessonin the US, who oversees a lab-made test depends on whether the law treats it as a product or a service.
Credit
American Clinical Laboratory Association (plaintiff); US District Court
Watch for
A multi-cancer blood test approval
GRAIL's Galleri test looks for signals of many cancer types in one blood draw. GRAIL filed the last part of its FDA premarket approval application on 29 January 2026, using data from 25,490 people in its PATHFINDER 2 study and the first round of the NHS-Galleri trial. On 23 September 2026 an FDA advisory panel voted 7 to 2, with 1 abstention, that benefits outweigh risks, and 6 to 4 on effectiveness.
Achieved whenthe FDA approves the application.
Claimed by
GRAIL, January and September 2026
Fewer late cancers from screening
The NHS-Galleri trial in England enrolled 142,250 people aged 50 to 77, each giving three yearly blood samples. GRAIL reported at the ASCO meeting on 30 May 2026 that the trial missed its main goal: combined stage III and IV diagnoses in 12 chosen cancers did not fall. A secondary measure, stage IV diagnoses, fell 14% across three rounds.
Achieved whenthe results appear in a peer-reviewed journal and longer follow-up confirms fewer late-stage cancers or cancer deaths in the screened group.
Claimed by
GRAIL, May 2026
Open
01
A table of test accuracy
A free, machine-readable table of the accuracy figures for FDA-cleared or authorised diagnostic tests, taken from FDA decision summaries (510(k), De Novo and emergency use documents). It records sensitivity (the share of sick people the test catches), specificity (the share of healthy people it correctly clears), sample size and the comparison method, and every number links to the page it came from.
The briefCredit · unclaimed02
One yardstick for cancer blood tests
An open re-analysis of published studies of multi-cancer early detection tests. Pull the raw counts by cancer type and stage from each paper, recompute sensitivity and specificity under shared definitions (the same cancer list, stage groups and follow-up window), and publish the extraction sheet and code.
The briefCredit · unclaimed
Open 01 · the brief
A table of test accuracy
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, machine-readable table of the accuracy figures for FDA-cleared or authorised diagnostic tests, taken from FDA decision summaries (510(k), De Novo and emergency use documents). It records sensitivity (the share of sick people the test catches), specificity (the share of healthy people it correctly clears), sample size and the comparison method, and every number links to the page it came from.
Why it is still open
In September 2023 the lawyer Bradley Merrill Thompson wrote that "there is no central database" of these figures; they sit in unstructured PDF summaries, and his own pass needed text processing plus hand review. Bigio and colleagues (Lancet Microbe, 2023) compared the packaging claims of 22 rapid COVID-19 antigen tests with independent Cochrane estimates and found that 12 claimed significantly higher sensitivity, by 12.0 percentage points on average, but that work covers one test type.
Built with what exists
The FDA's public 510(k) and De Novo databases and decision-summary PDFs; the openFDA device API; a PDF table extractor such as pdfplumber; a current large language model to pull figures into a fixed schema; Python and SQLite. Models that read tables in PDFs reliably are the advance that makes this a small-team job.
First step
Pull the diagnostic product codes for one FDA review panel, such as microbiology, from openFDA, download 50 decision summaries, hand-label 20, and measure how well the model's extraction matches.
Done looks like
At least 500 test summaries extracted, a second person's hand check of 100 random summaries finding 95% or more of fields correct, and the table, code and check sheet published under an open licence.
Open 02 · the brief
One yardstick for cancer blood tests
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open re-analysis of published studies of multi-cancer early detection tests. Pull the raw counts by cancer type and stage from each paper, recompute sensitivity and specificity under shared definitions (the same cancer list, stage groups and follow-up window), and publish the extraction sheet and code.
Why it is still open
A 2025 review for the UK National Institute for Health and Care Research covered 36 studies of 13 tests. It found that tests detect anywhere from 3 to more than 50 cancer types, so their overall figures "are not directly comparable", and that pooling the results "was not considered appropriate".
Built with what exists
The published papers and supplements, found through the Europe PMC API; Python with pandas; a large language model for a first-draft extraction; two human checkers; R or Python for confidence intervals.
First step
List the studies in the review, record which report counts by cancer type and stage, and draft the shared definitions.
Done looks like
Sensitivity by stage recomputed under the shared definitions for every study in the review that reports enough counts, with two independent extractors agreeing on 95% or more of cells and every number traced to a table in its source.