Industries / Health / Healthcare / Medical labs / Genetic testing and diagnostics
Newborn screening
Testing a few drops of blood from every baby, in the first days of life, for rare conditions that can be treated if they are found before symptoms start.
7
Milestones achieved, from 1963 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1963
Blood spots on filter paper
Robert Guthrie, a microbiologist at Buffalo Children's Hospital, found phenylketonuria (PKU), an inherited disorder that harms the brain unless diet is controlled, by the way high phenylalanine in a drop of blood dried on filter paper lets bacteria grow. With US Children's Bureau funds he packed test kits for state laboratories; Massachusetts began the first universal newborn program in the fall of 1962, more than 100,000 newborns had been tested by late 1962, and Guthrie and A. Susi published the method in 1963. Massachusetts then made the test compulsory, and in 1964 the American Medical Association's House of Delegates voted to oppose laws requiring it, but by 1965 most newborns were being screened.
Credit
Robert Guthrie and his technician A. Susi; Robert MacCready, Massachusetts Public Health Laboratory
1968
Ten principles for screening
The World Health Organization published Principles and Practice of Screening for Disease by J. M. G. Wilson and Gunnar Jungner. Its ten principles are still used to judge screening programs: among them, the condition should be an important health problem, there should be an accepted treatment, its natural history should be adequately understood, and there should be a suitable test. A 2025 history by P. C. J. I. Schielen found that six of the ten trace back to earlier US publications, among them a 1955 list of ten criteria, and that Wilson wrote nearly all of the monograph.
Credit
J. M. G. Wilson and Gunnar Jungner; WHO
2003
One test, many disorders
Tandem mass spectrometry (MS/MS) weighs and counts many small molecules in one run. D. S. Millington and colleagues proposed it in 1990 for profiling acylcarnitines, markers of fat-burning disorders, in newborns, and in the 1990s it let one 3 mm punch from a blood spot screen for an array of metabolic conditions. B. Wilcken and colleagues compared 362,000 babies screened this way in New South Wales and the Australian Capital Territory, Australia, from April 1998 to March 2002 with six earlier four-year groups: diagnosed metabolic disorders other than PKU rose to 15.7 per 100,000 births, from 8.6 to 9.5.
LessonScreening finds more cases than doctors do, and the authors noted that it was not yet clear which of those babies would ever have become ill.
Credit
D. S. Millington and colleagues; B. Wilcken, V. Wiley and the New South Wales Newborn Screening Programme
2008
A DNA test for SCID
Severe combined immunodeficiency (SCID) leaves a baby without working T cells, the immune cells that fight infection. Wisconsin began screening in 2008 with a DNA test for T-cell receptor excision circles (TRECs), small loops of DNA cut out as T cells form, and SCID joined the national US panel in 2010. Across 11 US programs and 3,030,083 newborns, screening found 52 cases, 1 in 58,000 births against an earlier estimate of 1 in 100,000, and 87% of the affected infants survived.
Credit
Wisconsin's newborn screening program; A. Kwan, J. M. Puck and colleagues, University of California, San Francisco
2016
Krabbe screening's hard lesson
Krabbe disease is a fatal nerve disorder whose only treatment is a stem cell transplant. New York began screening every newborn for it in August 2006; of more than 1.9 million babies screened, 348 were sent for diagnostic tests and five had the infantile form. Four had transplants: two died of complications and two survived with moderate to severe disabilities; 46 more babies were found at risk of a later-onset form, and only 1.4% of positive screens were confirmed infantile cases. In 2024 the US health secretary added infantile Krabbe to the national panel, with a second test on the same spot, for psychosine, to cut false positives.
LessonAdding a condition before its treatment is proven leaves families facing hard choices on uncertain results.
Credit
J. J. Orsini, M. Caggana and the New York State Krabbe Disease Consortium, Wadsworth Center
2025
Genomes beside the heel prick
The GUARDIAN study added genome sequencing to standard screening in six New York City hospitals, reading genes for 156 early-onset treatable conditions, with 99 seizure-linked disorders as an option. Of 5,555 families approached, 4,000 (72.0%) joined. 3.7% of babies screened positive, 92 of the 147 for G6PD deficiency, an enzyme shortage that can cause anemia; leaving G6PD aside, 0.6% screened positive against 0.3% on New York's standard screening. The positive rate was 6.9% for babies of African ancestry and 1% for European ancestry. Sanofi, Illumina and GeneDx funded it, and Illumina and GeneDx scientists helped collect and interpret the data.
Credit
A. Ziegler, W. K. Chung and colleagues, Columbia University Irving Medical Center; M. Caggana, New York State Department of Health
2025
The panel's review body ends
The Recommended Uniform Screening Panel (RUSP) is the US list of conditions advised for every state's program. It began with the 29 core conditions an American College of Medical Genetics panel proposed in 2005, was adopted by the health secretary in 2010, and held 38 core conditions by July 2024, each added after the Advisory Committee on Heritable Disorders in Newborns and Children (ACHDNC) reviewed the evidence. The committee was terminated on 1 April 2025. In August 2025 the Health Resources and Services Administration (HRSA) asked for public comment directly on adding Duchenne muscular dystrophy and metachromatic leukodystrophy, which the committee had sent to evidence review but never voted on.
LessonA national list that rests on one advisory committee stops growing when the committee is ended.
Credit
US Department of Health and Human Services; account by the Congressional Research Service
Watch for
Genomes for 100,000 babies
Genomics England's Generation Study, run with NHS England, aims to sequence the genomes of 100,000 newborns and looks for more than 200 rare conditions for which NHS treatment is available. At its June 2026 research summit Genomics England said more than 50,000 participants had joined.
Achieved whenresults for the cohort, with screen-positive, confirmed and false-positive counts and follow-up of the babies, appear in a peer-reviewed journal.
Claimed by
Genomics England, June 2026
A restart for the panel
HRSA announced a National Newborn Screening Workgroup in August 2026 and funded the Association of Public Health Laboratories (APHL) to re-establish a way for conditions to be considered and recommended for the RUSP. Duchenne muscular dystrophy and metachromatic leukodystrophy have waited since HRSA's August 2025 comment notices.
Achieved whena condition is added to the RUSP through the new pathway.
Claimed by
HRSA, August 2026
Open
01
AI evidence reviews, scored
An AI agent that drafts the evidence review a national body needs before it adds a condition to newborn screening: how common the condition is, how well the screening test performs, what early treatment changes and what harms follow, in the format of the published US reviews. It is scored on past reviews with the literature frozen at each review's date, so it cannot copy the answer.
The briefCredit · unclaimed02
A synthetic screening test bed
An open generator of made-up MS/MS screening results (amino acids and acylcarnitines, with birth weight, gestational age and age at sampling), with disorder cases mixed in at realistic birth rates and an answer key. Labs and software makers use it to compare fixed cutoffs, ratio rules and machine-learning classifiers on equal terms, with no real baby's data.
The briefCredit · unclaimed
Open 01 · the brief
AI evidence reviews, scored
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An AI agent that drafts the evidence review a national body needs before it adds a condition to newborn screening: how common the condition is, how well the screening test performs, what early treatment changes and what harms follow, in the format of the published US reviews. It is scored on past reviews with the literature frozen at each review's date, so it cannot copy the answer.
Why it is still open
The US committee that commissioned these reviews was terminated in April 2025, HRSA's 2026 workgroup is rebuilding the route, and stakeholders had already called the review process too burdensome to keep pace with new tests. Published reviews with recommendations exist for SCID (2010), X-linked adrenoleukodystrophy (2017), MPS II (2023), GAMT deficiency (2023) and infantile Krabbe disease (2025). DAISY, an open decision model published in 2026, speeds the modeling step but not the literature work. Europe PMC and web searches found AI tools tested on systematic reviews in general, and AI used to draft newborn screening lab reports, but none tested on newborn screening evidence reviews.
Built with what exists
the Europe PMC and Crossref APIs, which can filter by publication date; current AI models (Claude, Gemini or an open-weights model) in an agent loop for searching, screening papers and pulling numbers; the published reviews as answer keys; the DAISY model for the projection step.
First step
Take the 2025 Krabbe review, list its cited studies and key numbers, and run a first draft with the literature cut off at the year before publication.
Done looks like
drafts for five past reviews. For each, at least 90% of the agent's key numbers (birth prevalence, screening sensitivity and false-positive rate, treatment effect) fall within the published review's ranges, and the agent finds at least 90% of the studies the review cited that fall inside its date window. Two newborn screening specialists rate the agent's drafts and the published reviews without knowing which is which.
Open 02 · the brief
A synthetic screening test bed
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open generator of made-up MS/MS screening results (amino acids and acylcarnitines, with birth weight, gestational age and age at sampling), with disorder cases mixed in at realistic birth rates and an answer key. Labs and software makers use it to compare fixed cutoffs, ratio rules and machine-learning classifiers on equal terms, with no real baby's data.
Why it is still open
Each method for cutting false positives has been tested on one program's own data: a Random Forest model on 2,777 California screen positives (Peng et al., 2020), CLIR's multivariate tools in Sweden (2020) and the Netherlands (2026). A 2026 lab workflow that drafts reports with an AI model was checked on only 20 synthetic cases. Searches of Europe PMC, the web and GitHub found cutoff scripts and a general mass-spectrometry simulator (SMITER) but no shared synthetic newborn set with an answer key.
Built with what exists
the published percentiles and cutoff target ranges of the Region 4 Stork project (McHugh et al., 2011), built from about 25 to 30 million normal newborns and 10,742 true-positive cases across 114 markers; Python with the open-source Synthetic Data Vault (SDV) library to keep analytes correlated; Peng et al.'s Random Forest approach as one baseline; an AI coding agent to build, test and document the generator.
First step
Rebuild the normal-newborn distribution of five acylcarnitines from the published percentiles, add cases of MCAD deficiency from its published disorder range, and score one fixed-cutoff rule against the answer key.
Done looks like
1 million synthetic profiles with at least 500 cases across 20 disorders, published with a DOI. Three methods are scored on the same set and the scores published. Two newborn screening lab specialists sort 100 synthetic disorder profiles by disorder and agree with the answer key on at least 90 of them.