The US drug regulator stopped new patients joining an epilepsy drug trial over a chemical found in rats. The drug was licensed for $795m nine days earlier.

The stories

25 stories on this beat today.

Every story told the same way: what happened, who is involved, why it matters, how it works, the truth and the good news.

1. 01

   The US drug regulator stopped new patients joining an epilepsy drug trial over a chemical found in rats. The drug was licensed for $795m nine days earlier.

   $350m of $795m

   paid straight away by SK Biopharmaceuticals for Biohaven's epilepsy drug

   the licence was signed on 26 August, nine days before the trial hold [1][2]

   15 of 20

   reviewed Coventry brain cancer cases judged unsatisfactory overall

   none of the 20 was judged good practice [3]

   468

   people given Boehringer Ingelheim's satoprodil across two depression trials

   no dose beat a dummy pill at six weeks [4]

   80,000-140,000

   people killed by snakebite each year, on the World Health Organization's estimate

   a mixture of rattlesnake blood proteins was ten times more potent than the current antivenom in lab tests [8]

   What happened

   • The US drug regulator, the Food and Drug Administration, paused new patients joining trials of Biohaven's experimental epilepsy drug opakalim on 4 September.
   • The company disclosed it on Thursday. [1][2]
   • The regulator said it had too little information to judge the human risk from a metabolite, a substance the body makes as it breaks a drug down, spotted during testing in rodents. [1]
   • Biohaven said the finding may be specific to rodents and that what it means for people is uncertain.
   • It extended the enrolment pause to sites outside the United States on its own. [1]
   • More than 1,200 people have already taken the drug, and the more than 600 patients already assigned to treatment can carry on. [1]
   • One fully enrolled late-stage trial is still due to report in the second half of 2026.
   • Enrolment in a second one, RISE-2, is paused. [1][2]
   • Minter, an analyst at the investment bank William Blair, called the earlier licensing deal prudent but necessary for a company whose cash had been dwindling. [2]
   • Nine days before the hold, on 26 August, Biohaven sold worldwide rights to the drug to South Korea's SK Biopharmaceuticals for up to $795 million, with $350 million paid straight away. [1][2]
   • Biohaven said in its filing that it gave SK all its clinical and non-clinical data before signing.
   • Minter said that suggests SK was comfortable with the metabolite issue. [2]
   • The deal has not closed and still needs a US antitrust filing.
   • Biohaven's shares fell more than 13% before the market opened. [1]
   • Opakalim, also called BHV-7000, is aimed at focal epilepsy, where seizures start in one area of the brain and keep coming despite at least two suitable anti-seizure medicines. [1]
   • Biohaven is the second company of that name, spun out by the same managers after the first was sold to Pfizer in 2022.
   • Its drugs have since failed in spinal muscular atrophy and in depression, and one was rejected by regulators. [2]
   • More animal data is due in the coming weeks, to test whether the rodent finding matters in people. [1]

   Who is involved

   • Biohaven a US drugmaker rebuilt by the managers who sold the first company of that name to Pfizer in 2022; its epilepsy drug is now partly on hold
   • The Food and Drug Administration the US drug regulator; it imposed the partial hold on 4 September over a chemical found in rodents
   • SK Biopharmaceuticals a South Korean drugmaker; it agreed on 26 August to pay up to $795 million for worldwide rights, and the deal has not closed
   • The Royal College of Physicians the British body that sets standards for hospital doctors; it is reviewing 20 brain cancer cases in Coventry, England
   • Robert F. Kennedy Jr the US health secretary; he accepted one Pennsylvania measles death this week and still disputes a second

   How it unfolded

   1. 26 Aug SK Biopharmaceuticals agrees to pay up to $795 million for worldwide rights to opakalim [1][2]
   2. 4 Sep the US drug regulator imposes a partial hold, citing a chemical found in rodents [1]
   3. Thu 10 Sep Biohaven discloses the hold and its shares fall more than 13% before the market opens [1]
   4. Coming weeks more animal data is due, to test whether the finding matters in people [1]
   5. Late 2026 one fully enrolled late-stage trial still reports [1]

   Why it matters

   Watch the animal data Biohaven expects in the coming weeks. [1]

   RBC Capital Markets analyst Leonid Timashev says the regulator's demand for more data itself adds risk, because a concerning signal might turn up in it. [2]

   The truth

   • What is really going on The US drug regulator stopped new patients joining Biohaven's epilepsy trial because a chemical rats made while breaking the drug down could not be judged safe for people.
   • Who gains Biohaven's balance sheet — The $350 million SK Biopharmaceuticals paid straight away landed nine days before the hold, at a company whose cash reserves had been dwindling. [2]
   • Who pays People with focal epilepsy that no medicine controls — Enrolment in the RISE-2 trial is paused, which delays a result an approval would probably need. [2]
   • Who pays Biohaven shareholders — The shares fell more than 13% before the market opened on the day the hold was disclosed. [1]
   • Nobody knows yet

     Whether the chemical found in Biohaven's rats does anything to a person.

     The regulator said it has too little information to judge the risk, Biohaven says the finding may be specific to rodents, and more animal data is due in the coming weeks. [1]

   • Nobody knows yet

     What SK Biopharmaceuticals knew, and when.

     Biohaven says it handed over all clinical and non-clinical data before the 26 August deal, the hold came on 4 September, and neither side has said what that data showed. [1][2]

1. 02

   Brain cancer patients kept on chemotherapy for years

   What happened

   • The Royal College of Physicians, the British body that sets standards for hospital doctors, is reviewing 20 cases at University Hospitals Coventry and Warwickshire in England.
   • Brain cancer patients there were prescribed the chemotherapy drug temozolomide for long stretches with little or no evidence of benefit. [3]
   • Guidelines say the drug is usually taken for six cycles after radiotherapy, about six months, but patients told the BBC they received it for years, one of them for 16. [3]
   • Fifteen of the 20 reviewed cases were judged unsatisfactory overall, and none was judged good practice. [3]

   Why it matters

   More than 40 patients are represented by lawyers, and one father says he believes eight years of the drug left his son with irreversible neurological damage, liver problems and muscle pain. [3]

   The hospital trust says it has put new checks on who prescribes the drug and for how long. [3]

   The truth

   • Who pays Brain cancer patients in Coventry — Fifteen of 20 reviewed cases were care well below standard, and one father says he believes eight years of the drug left his son with irreversible neurological damage, liver problems and muscle pain. [3]
   • Nobody knows yet

     How many Coventry patients were harmed.

     The Royal College of Physicians has reviewed 20 cases and the review is not finished, while lawyers say more than 40 patients are affected. [3]

2. 03

   A depression drug matched the dummy pill

   What happened

   • Boehringer Ingelheim's experimental pill satoprodil failed to beat a dummy pill in two mid-stage trials in major depression, published in Neuropsychopharmacology. [4]
   • The drug damps down one part of the NMDA receptor, a switch on brain cells that researchers have linked to depression for years. [4]
   • One trial gave it to 243 patients alongside their existing antidepressant; the other gave it alone to 225. [4]
   • At six weeks, none of the three doses moved depression scores meaningfully further than the dummy pill did, although the drug was well tolerated. [4]

   Why it matters

   Together the two trials dosed 468 people, so this is a clear answer rather than a small one that could go either way. [4]

   The authors say it is the most thorough human evidence yet on this target and should steer future work on the brain's glutamate system. [4]

   The truth

   • Who pays Boehringer Ingelheim — Two completed trials in 468 people produced no benefit over a dummy pill, closing the route it had been taking on the NMDA receptor. [4]
   • Nobody knows yet

     Why satoprodil did nothing.

     Both trials ran six weeks at three doses and neither beat a dummy pill, and the authors do not say whether the target was wrong, the doses too low, or the patients too varied. [4]

1. 04

   AI firm blocked possible bioweapon research

   What happened

   • Anthropic, the American company behind the Claude artificial intelligence models, said it had disrupted several efforts this year by scientists using its models for research that could help build biological weapons. [5]
   • In a report on misuse it said it could not tell whether the work was legitimate or harmful, because the same biology that produces vaccines can also engineer dangerous germs. [5]
   • It shut the work down anyway. [5]
   • Jacob Klein, who runs threat intelligence at the company, said nobody announces they want to build a weapon, and the situations are nuanced. [5]

   Why it matters

   Andrew Weber of the Council on Strategic Risks, who read the report before release, called the findings chilling examples of state-backed weapons developers using fast-improving AI. [5]

   Biological misuse has drawn less attention than AI-enabled cyberattacks, partly because earlier cases turned up only in research settings. [5]

   How it works

   Someone has to decide before the evidence is in

   Evidence runs out before the decision is due, and the person who has to act then picks the mistake they would rather make.

   The twist

   The same gap pushes different people opposite ways. The US drug regulator stopped a trial because it could not rule harm out. The US health secretary left a measles death out of the national count because he could not rule it in.

   Step by step

   1. 1 A test throws up a signal: a chemical in rats, an alarm on a ward, a death certificate
   2. 2 Nobody can say yet what the signal means
   3. 3 A decision falls due anyway: enrol the next patient or not, count the death or not
   4. 4 Whoever holds that decision picks the mistake they would rather make
   5. 5 What everyone else sees is the action, not the doubt behind it

   The same force, elsewhere today

   Where this chain is also running, in today's other stories.

   • Anthropic blocking possible bioweapon research

     The same step: it could not tell legitimate biology from work towards a weapon, so it shut the work down rather than wait for an answer it was not going to get.

   • The US health secretary and the Pennsylvania measles death

     A coroner and a state health department disagree about what killed an infant, and the person who publishes the national number chose zero while the state's number stands at two.

   • The sepsis alarm tested in Amsterdam

     The model warns a median 47 hours ahead and is wrong most times it fires, so the doctor on the ward has to act on a signal nobody can confirm.

   • The Coventry chemotherapy review

     The same gap, filled the other way: there was little or no evidence that more months of temozolomide helped, and the prescriptions carried on for years.

   Where you've seen this

   Airport security

   a screener who cannot see what is inside a bag opens it, because opening one wrongly costs less than missing one

   Food recalls

   a factory pulls a whole batch on one unclear test, long before anyone knows whether a single jar is contaminated

   A school closing for a storm

   the head teacher decides on Sunday night, on a forecast that will not be settled until Monday morning

   A bank freezing an account

   a payment it cannot explain is stopped, and the customer is the one who has to show nothing is wrong

   The catch

   Caution is not automatically the safe side. Pausing Biohaven's trial protects patients from a harm nobody has measured, and it also delays a drug for people whose seizures nothing currently stops.

   And the whole of it

   Everyone deciding here holds one piece of it. The US drug regulator has the rat results and not the human ones. The Lancaster County coroner has one infant and not the outbreak. Most of us decide the same way every day, and we rarely notice how little we are working from.

   The truth

   • Nobody knows yet

     Whether Anthropic blocked legitimate scientists.

     The company says it could not determine whether the research was legitimate or harmful and shut it down anyway, and it has not said how many accounts were involved. [5]

1. 05

   US health secretary accepts one measles death

   What happened

   • Robert F.
   • Kennedy Jr, the US health secretary, said on Friday that measles killed an infant in Lancaster County, Pennsylvania, citing the county coroner. [6]
   • Last month he had asked for two Pennsylvania measles deaths to be taken off the national count kept by the CDC, the US disease agency. [6]
   • He now disputes only the second, saying the coroner found that infant had a fatal inherited condition and was about two months old. [6]
   • Pennsylvania's health department says it has confirmed two measles deaths this year, and the national data still shows none for 2026. [6]

   Why it matters

   The national tally is the number the rest of the country reads during a fast-growing outbreak, and the state and the US federal government are now publishing different answers. [6]

   Public health experts called the earlier removal request extraordinary political interference in a state health department's authority. [6]

   How it works

   A coroner and a state health department disagree about what killed an infant, and the person who publishes the national number chose zero while the state's number stands at two.

   The idea, in story 04: Someone has to decide before the evidence is in →

   The truth

   • What is really going on The same week, the US health secretary kept a measles death out of the national count because a county coroner and Pennsylvania's health department disagree about what killed the infant.
   • Who gains Robert F. Kennedy Jr — While the national count shows no measles deaths in 2026, the outbreak's worst outcome does not appear in the figure his department publishes. [6]
   • Who pays Unvaccinated families in Lancaster County, Pennsylvania — An infant died of measles, a second death is contested, and the national figure for 2026 still stands at zero. [6]
   • Nobody knows yet

     How many people have died of measles in the United States this year.

     Pennsylvania's health department says two, the national data shows none for 2026, and the health secretary now accepts one of the two. [6]

2. 06

   Two ways antipsychotics disturb young bodies

   What happened

   • Researchers followed 510 young people aged 4 to 17 for a year after they started second-generation antipsychotics, medicines used for severe mental illness, and most of them had never taken one before. [7]
   • Blood tests confirmed who was actually taking them. [7]
   • Total cholesterol, fasting blood sugar and blood fats rose in a way that weight and fat gain did not explain, most clearly with olanzapine and partly with risperidone, and less with quetiapine and aripiprazole. [7]
   • Insulin resistance, by contrast, tracked closely with body composition. [7]

   Why it matters

   The authors say watching a young patient's weight alone therefore misses part of the damage, and that the risk differs by which of these drugs is chosen. [7]

   The truth

   • Who pays Young people taking olanzapine and risperidone — Cholesterol, fasting blood sugar and blood fats rose over 12 months in ways their weight did not explain. [7]
3. 07

   Snake blood outperforms a made antivenom

   What happened

   • University of Maryland researchers combined toxin-blocking proteins that western diamondback rattlesnakes carry in their own blood to avoid poisoning themselves. [8]
   • Alone, each protein countered only part of a venom's effect and none prevented death. [8]
   • Mixed, the best combinations were about ten times more potent in lab tests than the sheep-derived rattlesnake antivenom now used, and they neutralised venom from several viper species. [8]
   • Today's antivenoms are made by injecting large animals with venom and harvesting their antibodies, which is costly, variable and can trigger severe immune reactions. [8]

   Why it matters

   Snakebite kills an estimated 80,000 to 140,000 people a year and leaves hundreds of thousands disabled, mostly in rural places where antivenom is hard to reach. [8]

   The work is in the laboratory, and the team expects veterinary use first, with human treatments later. [8]

   Good news

   • Proteins that western diamondback rattlesnakes carry in their own blood, mixed in the right combinations, neutralised venom about ten times more powerfully in the laboratory than the antivenom now in use. Snakebite kills between 80,000 and 140,000 people a year.
4. 08

   A glue drug destroys a lung cancer protein

   What happened

   • Researchers reported TRI-611, a drug that does not block its target but sticks it to the cell's own disposal machinery so the cell destroys it. [9]
   • The target is ALK, a stuck-together protein that drives one form of non-small-cell lung cancer. [9]
   • TRI-611 degraded every version they tested, including ones that had already stopped responding to the approved ALK blockers, and it reaches the brain. [9]
   • Tumours shrank in cell lines and in mice carrying human tumours, both under the skin and inside the skull. [9]

   Why it matters

   Patients whose ALK-driven lung cancer stops responding to the current pills have few options left, and this cancer often spreads to the brain. [9]

   The authors say it is the first drug of this kind to reach the clinical stage against a cancer-causing gene fusion. [9]

5. 09

   A lab bladder shows why antibiotics miss

   What happened

   • University College London researchers built a small artificial bladder lined with human cells and flushed with flowing urine, then infected it with the strain of E. coli behind most urinary tract infections. [10]
   • In that setting the bacteria stuck harder to the lining, invaded it, and built protected pockets inside bladder cells that antibiotics could not reach. [10]
   • A cocktail of phages, viruses that infect and kill bacteria, wiped out those hidden bacteria. [10]
   • Hospitals normally test antibiotics in still, nutrient-rich liquid, which is nothing like a bladder. [10]

   Why it matters

   Urinary tract infections cause around 400 million cases a year, and repeated courses of antibiotics are one of the routes to drug-resistant bacteria. [10]

   Jennifer Rohn, who led the work, said the model shows why antibiotics that look powerful in standard tests often fall short in patients. [10]

   Good news

   • A cocktail of viruses that eat bacteria cleared E. coli out of pockets inside the wall of a laboratory bladder, where antibiotics cannot reach them.
6. 10

   Two drugs against shape-shifting prostate cancer

   What happened

   • Prostate tumours that stop responding to hormone-blocking treatment often survive by changing what kind of cell they are, dropping their gland-like genes and switching on stem-cell programmes. [11]
   • University of Michigan researchers, writing in JCI Insight, paired two drug classes at that switch. [11]
   • BET bromodomain inhibitors interfere with the new identity; DNMT inhibitors switch the lost gland genes back on. [11]
   • Either alone slowed growth, but together they suppressed it further in cell lines and in tumours grown in mice, at doses well below the recommended ones. [11]

   Why it matters

   Prostate cancer is the second-leading cause of cancer death among men in the United States, and nearly everyone on hormone-blocking drugs eventually becomes resistant. [11]

   DNMT inhibitors are already approved for blood cancers, which shortens the route to a human trial. [11]

   The truth

   • Who gains Makers of DNMT inhibitors — Drugs already approved for blood cancers are half of the prostate cancer pairing, so an existing product gets a new disease to chase. [11]
   • Nobody knows yet

     Whether any of today's dish and mouse results survive a human trial.

     The rattlesnake antivenom, the magnetic nanoantennas, the prostate drug pairing and the bone drug were each tested only in laboratory dishes or in mice. [8][11][12][13]

7. 11

   Magnets drive nanoantennas into brain tumours

   What happened

   • A team writing in Science Advances described HITMAN, tiny antennas that turn a weak magnetic field applied from outside the head into a sharp electric field right at a cell. [12]
   • That electric field unfolds proteins and tears cell membranes. [12]
   • In dishes it cut the survival of drug-resistant glioblastoma cells taken from patients by 52.2%, against 10% for temozolomide, the standard chemotherapy, while sparing neurons and support cells. [12]
   • In mice with tumours grown in the brain it slowed growth and extended median survival by more than half, with no whole-body toxicity. [12]

   Why it matters

   Glioblastoma resists both radiation and chemotherapy, and surgery cannot remove a tumour that grows into healthy tissue. [12]

   Temozolomide is the same drug at the centre of the Coventry review above, which is a measure of how little has changed in this cancer. [3][12]

8. 12

   A cancer drug protected bone in mice

   What happened

   • CADD522, an experimental cancer drug that blocks a protein called RUNX2, protected the bones of mice whose ovaries had been removed to imitate the menopause. [13]
   • The same mice also lost body fat and reversed some of the changes in fat handling that follow the menopause. [13]
   • The University of East Anglia team published the work in npj Drug Discovery. [13]
   • Osteoporosis, a thinning of bone that makes fractures easy, affects about one in three women and one in five men over 50, and existing drugs are held back by side effects and awkward dosing. [13]

   Why it matters

   A molecule already in development for cancer has a shorter path to a bone trial than a new one does. [13]

   This was mice, and the researchers say the drug needs further development before anything reaches women. [13]

9. 13

   Private biotechs buy their way onto Nasdaq

   What happened

   • About two dozen biotech reverse mergers have been announced in 2026, more than double the 10 across all of 2025, on a count by the advisory firm JB Strategy Partners. [14]
   • In a reverse merger a private drug developer takes over a listed company that has little left except its stock market listing, and inherits the listing. [14]
   • Bankers call those companies shells. [14]
   • Tim Opler of the investment bank Stifel said there is no question the move has taken off, and dealmakers are now hunting for shells worth buying. [14]

   Why it matters

   Testing a single molecule can cost hundreds of millions of dollars, so how a young drugmaker reaches public investors decides which medicines get funded at all. [14]

   Branden Berns, a partner at the law firm Gibson Dunn, said the danger is running out of high-quality shells. [14]

   The truth

   • Who gains Owners of listed biotech shells — Private drug developers are bidding for companies whose main asset is a stock market listing, and dealmakers say demand may outstrip supply. [14]
10. 14

    One flu shot stacks three older tricks

    What happened

    • CSL Seqirus published phase 3 results in The Lancet for aQIVc, a flu vaccine for older adults that combines three approaches already used separately. [15]
    • Those are an additive called MF59 that strengthens the immune response, manufacture in cell cultures rather than hens' eggs, and a higher dose. [15]
    • The trial enrolled 7,699 adults aged 50 and over across eight countries. [15]
    • It beat an adjuvanted egg-based vaccine on all four flu strains and every age group tested, and matched a recombinant vaccine on three of four strains in the over-50s. [15]

    Why it matters

    Older people respond less well to flu shots because the immune system weakens with age, which is why enhanced versions exist at all. [15]

    Britain's medicines regulator has already approved a three-strain version, and these figures come from the company's own announcement of the paper. [15]

    The truth

    • Nobody knows yet

      Whether the flu vaccine result holds outside the company's own summary.

      The figures for aQIVc come from CSL Seqirus's announcement of its Lancet paper, and the announcement does not carry the published detail. [15]

11. 15

    Turning RNA folding into a dose dial

    What happened

    • ARPA-H, the US government's health research funding agency, awarded up to $4.4 million for a one-year pilot called PROPEL. [16]
    • It is led by Silvi Rouskin's laboratory at Harvard Medical School with Jonathan Weissman's laboratory at MIT and the Whitehead Institute, alongside a company called RNAV8 Bio. [16]
    • The idea uses the shapes an RNA message folds into: those folds set how much protein the message makes, and a small molecule that binds a fold can change the amount. [16]
    • Bacteria already work this way, through elements called riboswitches. [16]

    Why it matters

    Most medicines act wherever their chemistry carries them, and most gene therapies change DNA permanently, so a dial that turns protein output up and down would be a different kind of control. [16]

    Devan Shah, who founded RNAV8 Bio, said the hard part has always been predicting what an RNA sequence actually does. [16]

12. 16

    Japan tries to fix cell therapy manufacturing

    What happened

    • Teijin, its subsidiary Teijin Regenet and Shinshu University Hospital signed a joint research agreement on manufacturing methods for cell and gene therapies in Japan. [17]
    • They will work through the hospital's cell processing centre on automation, efficiency, quality testing and quality control. [17]
    • The partners said Japan is short of both trained manufacturing staff and the facilities to make these treatments at scale. [17]
    • CAR-T therapies, which re-engineer a patient's own immune cells to hunt cancer, are built one patient at a time, which is what makes the factory the bottleneck. [17]

    Why it matters

    A cell therapy that works in a trial is not a treatment anyone can get until somebody can make it repeatedly and to standard. [17]

    The truth

    • Who gains Teijin Regenet — Japan's shortage of cell therapy staff and facilities is exactly the gap its contract manufacturing arm is being positioned to fill. [17]
13. 17

    Brain rhythms sorted who developed psychosis

    What happened

    • Researchers followed 204 people aged 13 to 38 judged at clinical high risk of psychosis, testing them at two months, one year and two years. [18]
    • At the start their symptom scores and their best functioning over the past year looked alike, but the groups then diverged. [18]
    • Visuospatial learning and working memory measured at the start went on shaping symptoms and functioning for two years. [18]
    • A brain-wave pattern called microstate D tracked outcomes too, and a model combining it with subtype picked out who developed psychosis. [18]

    Why it matters

    Everyone in this group carried the same label at the start and ended up in very different places, so a marker that separates them early decides who is watched and who is treated. [18]

    This is one group of 204 people and the participant data are not public. [18]

14. 18

    A sharper gene editor aimed at ataxia

    What happened

    • A team writing in Molecular Therapy Nucleic Acids reworked twin prime editing, a way of rewriting DNA that nicks the strands rather than cutting both, to make it more efficient and more accurate. [19]
    • They then pointed it at spinocerebellar ataxia type 3, an inherited disease that wrecks balance and coordination and has no approved treatment. [19]
    • The fault is a stretch of repeated DNA inside a single gene called ATXN3. [19]
    • The edit makes the cell skip past the repeat and stop the protein early, producing a shortened version already known to be less toxic. [19]

    Why it matters

    This was done in human cell lines engineered to carry the fault, not in animals or people. [19]

    Type 3 is the most common of the spinocerebellar ataxias, and a single faulty copy inherited from one parent is enough to cause it. [19]

15. 19

    A sepsis alarm tested on another continent

    What happened

    • A neural network called LiSep LSTM, trained to predict septic shock on a US intensive care database, was tested on an independent European one from Amsterdam UMC. [25]
    • It scored 0.8851 on a standard accuracy measure, slightly better than at home, and flagged patients a median of 47 hours before onset. [25]
    • But its precision ranged from 0.01 to 0.17, which means the large majority of its alarms were false. [25]
    • Training on both datasets together improved the scores. [25]

    Why it matters

    Sepsis kills when it is caught late, so a 47-hour warning is worth a great deal, and a ward that stops believing the alarm has gained nothing. [25]

    The authors say local recalibration and a way to handle false alarms are needed before anyone uses it. [25]

    How it works

    The model warns a median 47 hours ahead and is wrong most times it fires, so the doctor on the ward has to act on a signal nobody can confirm.

    The idea, in story 04: Someone has to decide before the evidence is in →

    The truth

    • Nobody knows yet

      What the sepsis alarm would do on a real ward.

      Its precision ran from 0.01 to 0.17 across the settings tested, so most alarms are false, and the authors say it needs local recalibration first. [25]

16. 20

    A hospital fall predictor explains itself

    What happened

    • Using electronic records from a large German university hospital covering 2016 to 2022, researchers trained a model to predict which inpatients would fall, then grouped the flagged patients by which factors had driven each alert. [24]
    • That produced nine clusters, each matching a recognisable reason people fall and each pointing at a different prevention step. [24]
    • The model scored 0.95 on one standard accuracy measure and 0.36 on a second one that is harsher when the event is rare. [24]
    • The authors say it needs testing at other hospitals first. [24]

    Why it matters

    Falls cause around 684,000 deaths a year worldwide and account for up to 1.5% of health spending, and one-size-fits-all prevention does not work because people fall for different reasons. [24]

17. 21

    Priority review for a disease with nothing

    What happened

    • The US drug regulator granted priority review to Genentech's Enspryng, known generically as satralizumab, for MOGAD, an autoimmune disease in which the body attacks the optic nerves, spinal cord or brain. [23]
    • There is no approved treatment for it. [23]
    • The filing rests on a phase 3 trial called METEOROID, which the company says cut the risk of relapse by 68% against a dummy injection in patients aged 12 and over. [23]
    • The drug is already approved for a related condition, neuromyelitis optica spectrum disorder. [23]

    Why it matters

    Each MOGAD attack can leave permanent damage, including vision loss, weakness and confusion, so preventing relapses is most of what a treatment can offer. [23]

    The figures come from Genentech's own announcement rather than from a published paper. [23]

    The truth

    • Who gains Genentech — A faster review on a disease with no approved treatment means there is no rival product to be priced against when it arrives. [23]

    Good news

    • MOGAD, a disease in which the immune system attacks the optic nerves, spinal cord and brain, has no approved treatment anywhere. A phase 3 trial of Genentech's satralizumab cut the risk of relapse by 68%, and the US drug regulator has put the application on its faster track.
18. 22

    Six ageing clocks read one trial

    What happened

    • Researchers writing in Nature Biotechnology ran six proteomic ageing clocks, models that estimate biological age from the proteins in blood, over serum from a published 12-week trial of rentosertib, an experimental drug for scarring of the lungs. [22]
    • All six read the treated patients as biologically younger. [22]
    • The authors say the clocks on their own cannot separate an anti-ageing effect from the drug simply treating the disease. [22]
    • They used pathway analysis instead, and found shifts in cell-ageing and metabolic processes alongside the drug's anti-scarring action. [22]

    Why it matters

    Standard trial designs cannot detect whether a drug for an age-related disease also slows ageing itself, and the authors argue for measuring both inside one study. [22]

19. 23

    An NHS trust joins a prevention trial

    What happened

    • Surrey and Borders Partnership NHS Foundation Trust in England joined a trial of a drug meant to prevent Alzheimer's disease rather than treat it. [21]
    • The drug attacks beta amyloid, a sticky substance that builds up in the spaces between brain cells in the disease. [21]
    • Blood tests will be used to find the people most likely to benefit, before symptoms appear. [21]
    • Ramin Nilforooshan, the consultant psychiatrist leading the trial, said research has focused for decades on treating Alzheimer's after symptoms start. [21]

    Why it matters

    About 982,000 people in the UK are living with dementia, on an Alzheimer's Society estimate, and this trial asks whether the disease can be headed off rather than slowed once it has begun. [21]

20. 24

    Yeast engineered to burst on cue

    What happened

    • Manus and Hal Alper's laboratory at the University of Texas at Austin finished a programme funded by BioMADE, a US bio-manufacturing body, in which industrial yeast were engineered to break their own cell walls at the end of fermentation. [26]
    • Products made inside yeast normally have to be freed by grinding the cells or dissolving them in hazardous solvents. [26]
    • At 300-litre pilot scale, engineered Yarrowia lipolytica cut the energy needed for mechanical separation by more than half. [26]
    • The team also achieved self-digestion in a production strain of brewer's yeast. [26]

    Why it matters

    A growing share of proteins, oils, vitamins and pigments is made by fermentation, and getting the product back out of the cells is a large part of the bill. [26]

21. 25

    A 16-patient probiotic trial in pneumonia

    What happened

    • Sixteen hospital patients aged 60 and over with pneumonia completed a randomised, blinded pilot in which some took a probiotic capsule called resB Lung Support for 14 days alongside normal care and the rest took a dummy. [20]
    • Stool samples were taken at the start, at day 15, and at day 30 or discharge. [20]
    • The probiotic group showed more of several gut bacteria thought to be helpful, and less Corynebacterium and Streptococcus. [20]
    • The authors call the trial exploratory, say the findings should be read cautiously, and ask for larger studies. [20]

    Why it matters

    Sixteen people is far too few to show that anyone got better, and the study measured bacteria rather than recovery. [20]

    It is a marker of how far the gut-and-lung idea has travelled from theory towards hospital wards. [20]