A seizure is a sudden burst of abnormal electrical activity, and after two medicines fail the next one works for about one person in 25

The rest of the day

38 more stories on this beat.

Each with its own sources. None of these is a link to the story above.

1. 02

   An implant recorded twice the seizures of a diary

   Most epilepsy care runs on what patients write down, and people cannot record a seizure they did not notice. A team led by Benjamin Brinkmann at the Mayo Clinic in the United States tested a small implant that sits under the skin behind the ear and records brain activity all day. Over 15 months it collected more than 72,000 hours of recordings and 754 seizures, nearly twice as many as the diaries held. About half the participants wore it more than 20 hours a day. [6]

   Recorded by the implant754 seizures

   Written in diaries377 seizures

   Why it matters — Every number in today's lead story, including the 4.4%, rests on counted seizures. If diaries miss half of them, the counts of who is doing well on which medicine are built on a shaky base.

2. 03

   A UK trial puts trackers in the brain

   Adam Atkinson, 22, from Whitehaven in north-west England, had a device implanted at the Freeman Hospital in Newcastle two months ago. He has a rare brain disease called seronegative autoimmune encephalitis, and with it drug-resistant epilepsy. The six-month UK-wide trial compares the implant against the usual method, which is a diary plus a hospital recording session that only catches a seizure if one happens. Rhys Thomas, the neurologist leading the north-east arm, says it lets doctors watch patients remotely. [7]

   Why it matters — It is the same gap as the Mayo work, with a named family in it: the people who most need their seizures counted are the ones whose seizures are hardest to catch.

3. 04

   New children's medicines cut seizures, rarely stopped them

   A review published in October 2025 looked at the newer medicines for severe childhood epilepsies, including Lennox-Gastaut syndrome and Dravet syndrome. Cannabidiol, everolimus, fenfluramine and ganaxolone all reduced how often seizures happened, and a share of children responded. But in the main trials that got them approved, almost no child stopped having seizures altogether. The authors say more options are still needed, and list adaptive stimulation devices among what is coming. [8]

   Why it matters — Reducing seizures by half and stopping them are different results, and the second is what changes whether a child can be left alone in a bath.

4. 05

   After valproate fails, which medicine to add

   Sodium valproate is the usual first medicine for several kinds of epilepsy, and nobody had compared what works best when it fails. Chinese researchers looked at 2,656 people and sorted them by which second medicine was added. For generalised seizures, which involve the whole brain at once, adding lamotrigine halved seizures in 89.6% of patients, against 75.9% for carbamazepine. For focal seizures, which start in one patch, oxcarbazepine came top at 88.9%. [9]

   Why it matters — It is a real answer to the question the lead story leaves open: if the choice of second medicine matters this much, the order people are given them in matters too.

5. 06

   Missing the odd dose did not set off seizures

   People with epilepsy are told that missing a dose is dangerous. Researchers gave 27 adults with drug-resistant epilepsy, each having at least three seizures a month, smartphone apps to record both their seizures and every dose taken, for ten months each. That produced 7,853 days of data. Only two people showed even a weak link between a missed dose and a seizure soon after. Across the group, seizures tracked a simple forecast based on recent seizure rate, not adherence. [10]

   Why it matters — The authors say clinicians can tell patients an occasional lapse is unlikely to trigger a seizure, while sustained non-adherence remains a separate worry. It removes one common source of guilt without removing the medicine.

6. 07

   Growing old with epilepsy is harder than getting it late

   Doctors at three outpatient clinics in Berlin, Germany, looked at 243 patients aged 65 and over, treated between 2010 and 2025. They split them by when the epilepsy started. Among those who had it since before 40, half were seizure free in the last year. Among those who developed it after 65, 69% were. The long-standing group also took a higher total dose of medicine, and were six times more likely to be on an older generation of drug. [11]

   Why it matters — The people who have had the most tries are the ones still having seizures, which is the lead story's arithmetic showing up in a clinic waiting room forty years later.

7. 08

   What kind of epilepsy it is predicts the outcome

   A study of children and young people with epilepsy, where 78% became seizure free overall, sorted them by the type of epilepsy and its cause. Among those with focal epilepsy, starting in one patch of brain, 84% became seizure free. Among those with both focal and generalised seizures, only 26% did. Where no cause was found, 89% became seizure free; where a structural or genetic cause was identified, 58% and 54% did. In Dravet syndrome and Lennox-Gastaut syndrome, none did. [12]

   Focal epilepsy84%

   Generalised75%

   Both kinds26%

   Why it matters — It says the one-third figure in the lead story is not spread evenly. Which group a child is in is largely settled before any medicine is tried.

8. 09

   A computer tried to pick the right medicine first

   Researchers took the records of 2,586 people who first saw an epilepsy specialist between 2008 and 2017 and were followed at least three years. They fed the first visit's tests, scans and brain recordings into machine learning models and asked them to predict which medicine that person would respond to. The best single-drug score was valproate, at 0.686 on a scale where 0.5 is a coin flip and 1.0 is perfect. A pair of drugs, levetiracetam with carbamazepine, reached 0.764. [13]

   Why it matters — This is the direct attack on the lead story's problem, and its current state is honest: better than chance, nowhere near good enough to replace trying.

9. 10

   Nobody agrees what seizure free means

   A review published in 2026 went through 147 papers, trial registries, regulator decisions and health technology reports, and kept 25 that reported seizure freedom as an outcome. The definitions varied by age group, by seizure type, by whether the study was a trial or real-world, and by how the data was analysed. Regulators and health technology bodies used different requirements again. The authors found no consensus and are now running an expert panel to build one. [14]

   Why it matters — Half of the numbers in today's lead story are shares of people who went a year without a seizure. If the twelve-month rule is not what everyone means, the shares are not comparable.

10. 11

    Rodent screens and the human ranking

    New anti-seizure medicines are screened in animals long before they reach people. Researchers asked whether those screens predict how the same drugs later rank in human add-on trials, where a new medicine is added on top of ones that have already failed. The oldest single screen, an electric-shock test in rodents, did not predict the human ranking, and nor did most individual models on their own. A panel of several focal-seizure models taken together did line up with the human results, including for seizure freedom. These were mice and rats, not patients. [15]

    Why it matters — Screening decides which molecules ever reach a trial at all. A screen that does not predict the human ranking quietly sets the list of medicines that people in today's lead story then work their way through.

11. 12

    Acting on a warning score within 36 hours

    Sepsis is the body's own reaction to infection turning against its organs, and the treatable window is short. Researchers re-analysed a hospital trial of the CONCERN early warning system, which reads nurses' recorded observations to flag patients getting worse. Of 100 patients later diagnosed with sepsis, 54 had a warning score rise before an unplanned transfer to intensive care. The shorter the gap between the rise and the transfer, the lower the odds of dying. Every patient moved more than 36 hours after the rise died in hospital. [16]

    Why it matters — The study is small and the 36-hour line is drawn from 54 people. What it shows is that the warning arrived in time in every one of those cases, and the time was spent elsewhere.

12. 13

    Starting the blood-pressure drug early in shock

    When sepsis drops blood pressure far enough it is called septic shock, and norepinephrine is the drug used to push it back up. Doctors argue about whether to start it in the first hour or wait. A review searched studies from 2010 to May 2025, found 28 and pooled nine. In randomised trials, starting early cut deaths by 10%, which was not statistically reliable. In studies that merely watched what doctors did, early starters had 25% fewer deaths. [17]

    Why it matters — The two kinds of study disagree in the way they usually do, because the patients doctors treat fastest are not a random group. It is an open question, not a settled practice.

13. 14

    Treating tuberculosis before the test comes back

    In east Africa, people with HIV who arrive at hospital with sepsis often have tuberculosis, and the tests are slow or unavailable. The ATLAS trial ran at four hospitals in Tanzania and Uganda and was published in April 2026. It randomly assigned 437 patients either to start tuberculosis treatment immediately or to wait for a diagnosis, and at either a conventional or a high dose. Across everyone, 25% died within 28 days either way. Among the 204 who did have tuberculosis, immediate conventional-dose treatment cut deaths from 34% to 12%. [18]

    Why it matters — Treating everyone made no difference overall; treating the right people early made a large one. The trial could not tell them apart in advance, which is exactly the problem it set out to solve.

14. 15

    England swapped its newborn sepsis rule

    Sixteen NHS hospital groups in England moved from the national guidance on newborn infection to an American tool, the Kaiser Permanente sepsis risk calculator, which gives each baby a number rather than a yes or no. The worry was that fewer babies on antibiotics would mean more infections caught late. Researchers tracked 222,122 live births from 2015 to 2023, 18 months either side of each switch. The share of confirmed cases found only after the baby went home and came back was 9.6%, against 10.8% before. [19]

    Why it matters — No overall rise in late diagnosis, but the authors flag that some individual hospitals did see increases, and the study was not big enough to measure deaths.

15. 16

    Computer sepsis alarms, tested mostly on old records

    A review pulled together 52 studies of computer models built to spot sepsis early in hospital patients from their records, vital signs and lab results. Their scores looked strong, between 0.79 and 0.96 on the measure where 0.5 is a coin flip. Almost all of the studies were retrospective: the model was run over records of patients whose outcome was already known. Very few were tested live on a ward, and the authors name generalisability and explainability as the blocking problems. [20]

    Why it matters — A model that finds sepsis in yesterday's notes is not the same as one that changes what happens to a patient tonight, and the review says that step has mostly not been taken.

16. 17

    A third of sepsis patients died despite the checklist

    The Sepsis Six is a list of six things to do within an hour of spotting sepsis, such as giving oxygen, taking blood cultures and starting antibiotics. Researchers at one tertiary hospital followed 660 patients diagnosed with sepsis in the emergency department. A third of them, 32%, died. Two-thirds had developed sepsis after 24 hours in hospital rather than arriving with it. When the bundle was created in 2007, an observational study reported 20% deaths with it against 44.1% without. [21]

    Why it matters — The authors conclude that adherence, not the list, is what needs work, and the two-thirds who got sick after admission suggests much of the delay happens inside the hospital.

17. 18

    Taking the fluid back off in intensive care

    Patients in shock are given large volumes of fluid, and the fluid then has to go somewhere. A single-centre trial in a medical intensive care unit randomly assigned 100 patients after their initial resuscitation. Half got usual care. Half got a strict plan aiming at near-zero fluid balance over three days, using diuretics or a machine to remove fluid. By day three, the strict group was 2,353 millilitres down where usual care was 793 millilitres up. [22]

    Why it matters — It was a feasibility trial, so it shows the plan can be carried out, not that it saves lives. The gap of more than three litres is the size of the thing a bigger trial would now be testing.

18. 19

    England's flu season started on a changed virus

    England's 2025/26 flu season began in late October, the earliest start since 2003/04. Almost all cases were influenza A, and most were an H3N2 strain called subclade K. When the virus was tested against blood from animals given the season's vaccine, the match was poor. Vaccine effectiveness against emergency department attendances and hospital admissions was still 72% to 75% in under-18s and 32% to 39% in adults. [24]

    Under 18s73%

    Adults35%

    Why it matters — A poor laboratory match and a working vaccine are not a contradiction. The antibodies a vaccine raises are only part of what the immune system brings.

19. 20

    Subclade K across Europe, reviewed

    A review published in 2026 traced how subclade K took over Europe in the 2025-26 season. It found the virus had picked up several changes in haemagglutinin, the protein on its surface that antibodies latch onto, at the very spots antibodies aim for. That gave it an advantage and an earlier, sharper peak. The review's reading of the vaccine data is moderate protection against catching flu at all, with protection against severe illness largely intact. [25]

    Why it matters — It is the mechanism under the English figures: a changed surface costs you the mild cases first, because stopping an infection needs a tighter match than stopping it from becoming serious.

20. 21

    The WHO changed next winter's flu recipe

    On 27 February 2026, after a four-day meeting over global surveillance data, the World Health Organization announced which virus strains should go into flu vaccines for the 2026-27 northern winter. Those recommendations are what national regulators and manufacturers build the season's vaccines from. The meeting happens twice a year, once for each hemisphere, and draws on samples collected by laboratories in the WHO's global influenza network. [26]

    Why it matters — The decision has to be made roughly six months before the season, because that is how long the vaccines take to make. That lead time is why a virus can change in between.

21. 22

    Fewer flu samples reaching the US agency

    In November 2025, NPR reported that the United States leaving the World Health Organization had cut the flow of virus samples and data reaching the US Centers for Disease Control and Prevention from other countries. The agency uses that global picture to judge which strains are spreading and to feed the twice-yearly vaccine decision. With fewer samples, American scientists have a dimmer view of what is circulating abroad. [27]

    Why it matters — The vaccine recipe is a forecast made from samples. Cutting the inputs does not show up as a failure this winter; it shows up as a worse guess in some later one.

22. 23

    An antibiotic matched a strict diet for gut pain

    Irritable bowel syndrome is pain and disturbed bowel habits with nothing visibly wrong in the gut. Two second-line treatments had never been compared directly: rifaximin, an antibiotic that stays in the gut, and the low FODMAP diet, which cuts out sugars that ferment. A trial published in October 2025 split 100 adults between them. At four weeks, 56% responded to rifaximin and 48% to the diet, a difference that could be chance. And 95.9% stuck with the antibiotic against 77.8% on the diet. [28]

    Why it matters — The result people will act on is the sticking-with-it gap, not the response rate: a treatment nobody can keep up is worth less than its trial result. A gastroenterologist writing in 2025 argues the field still hands patients measures that only ease symptoms; that is one doctor's argument, not a settled finding. [46]

23. 24

    Twenty-seven reviews of probiotics, almost all weak

    Probiotics are among the most sold products for gut symptoms. A review of reviews published in February 2026 gathered 27 systematic reviews of probiotics in adults with irritable bowel syndrome, published between 2009 and 2025. Probiotics were linked to modest gains: symptoms persisted in fewer people, with four to seven people needing treatment for one to benefit. But two-thirds of the reviews were rated critically low quality, 85% were at high risk of bias, and just 1% of the outcomes were rated high certainty. [29]

    Why it matters — The number of studies and the strength of the evidence are separate things, and here they point in opposite directions.

24. 25

    An anti-sickness drug for a bowel problem

    Ondansetron is best known for stopping nausea after chemotherapy. It also slows the gut, and it is used for the diarrhoea form of irritable bowel syndrome. A clinic in a specialist centre looked back at 92 patients seen between October 2016 and February 2024, most with that diagnosis and the rest with plain chronic diarrhoea. Just under three-quarters, 73.9%, responded, meaning at least one fewer bowel movement a day and firmer stools. [30]

    Why it matters — It is a record of what happened in a clinic rather than a trial, so it cannot separate the drug from time passing. It is also the kind of second try the lead story's arithmetic is about.

25. 26

    Sending current to the gut nerves

    Because medicines for irritable bowel syndrome work poorly, researchers have tried electrical stimulation of the nerves between gut and brain. A 2025 review went through the methods. They include electrodes implanted at the base of the spine, spinal cord stimulation, and non-invasive versions using pads on the skin or on the outer ear, where a branch of the vagus nerve runs close to the surface. The review sets out what each has shown, how it might work, and the side effects. [31]

    Why it matters — The gut and the brain send signals both ways, and the treatments that get tried follow that wiring rather than the gut wall itself. [32]

26. 27

    Three approved drugs for a pain nobody can see

    Fibromyalgia is widespread pain in muscles and joints, usually with tiredness, poor sleep and trouble thinking, and no damage that a scan can find. An update published in October 2025 reports that only three medicines have US approval for it: pregabalin, duloxetine and milnacipran. Each gives meaningful relief to a minority of patients, and side effects are frequent enough to stop many from continuing. Other drugs have been tried without approval following. [33]

    Why it matters — It is the lead story's problem in a condition with no visible target: a short list of options, each helping some, and no way to know in advance who.

27. 28

    Three telehealth sessions for fibromyalgia pain

    Pain reprocessing therapy tries to change what the brain concludes from a pain signal rather than the signal itself. In a single-arm pilot published in September 2025, 35 adults with fibromyalgia were offered three one-to-one sessions by video call, then followed for three months. Thirty-three finished. Average pain intensity, how much pain interfered with life, and fear of pain all fell, with the effects growing over the three months. At three months, 42.3% of completers called themselves much or very much improved. [34]

    Why it matters — There was no control group and no placebo, so the numbers cannot be read as proof. The authors say larger randomised trials are what comes next.

28. 29

    Sperm quality fell faster in heavier men

    Researchers looked back at 2,430 men with normal sperm results who had fertility checks at one centre between 2010 and 2024, and asked whether their results drifted over those 14 years. They did. In men of normal weight, total sperm count, movement, volume and shape all declined. In overweight men, movement and volume fell. In obese men, count and volume fell, with volume dropping fastest of all three groups. [35]

    Why it matters — These are men whose sperm results were in the normal range the whole time, so the decline is happening inside what a clinic calls normal.

29. 30

    Smoking, drink and heat showed up in sperm DNA

    A study at one tertiary centre in India tested 278 men aged 21 to 50, measuring semen by the World Health Organization's current standard and separately checking how broken the DNA inside the sperm was. Tobacco and alcohol were strongly linked to fewer sperm, poorer movement and more misshapen sperm. Alcohol and heat at work were both linked to more DNA damage. Low testosterone and raised prolactin, a hormone from the pituitary gland, went with abnormal results. Men over 40 had more DNA damage while their ordinary readings looked unchanged. [36]

    Why it matters — The ordinary test and the DNA test disagree in older men, which means a normal semen result is answering a narrower question than it appears to.

30. 31

    What work stress does to male fertility, unclear

    A scoping review searched four databases plus general web searches for studies on how the social side of work affects male fertility, published between January 1990 and January 2024. From 1,322 records, 18 peer-reviewed articles survived screening. On shift work, long hours and job strain, the findings were inconclusive and pointed in different directions. Job stress and mental exhaustion were linked to lower sperm quality, and support from colleagues appeared to blunt the effect of heavy demands. [37]

    Why it matters — Eighteen studies in 34 years is not much for something most working men are exposed to, and the review's main conclusion is that better studies are needed.

31. 32

    A new depression drug lost to the dummy pill

    Satoprodil is a pill aimed at the NMDA receptor, the brain switch that ketamine acts on, which is why drugs of this kind are tried in depression. Two phase two trials tested it against placebo for six weeks. In the trial where it was added to an existing antidepressant, 243 patients were split four ways. Depression scores fell 12.0 points on placebo, 7.8 points on the 5mg dose, 11.9 on 10mg and 12.3 on 20mg. The dummy pill did as well as the drug. [38]

    Why it matters — A placebo group improving 12 points is the reason depression trials are hard to read at all, and why a treatment can feel like it works to everyone in the room.

32. 33

    Aiming brain stimulation with a bargaining game

    Magnetic stimulation for depression is usually aimed at the same spot in everyone. In a trial in China, 70 patients with major depression played the Ultimatum Game, in which one person proposes a split of money and the other accepts or rejects, while their brain was scanned. The spot that lit up during the game became the target for two weeks of stimulation. Thirty-seven got real stimulation and 33 a sham. The real group improved more on depression, anxiety and social functioning. [39]

    Why it matters — The target was picked from a social task rather than an anatomical average, which is one way of answering the lead story's question of how to match a treatment to a person.

33. 34

    Online therapy for anxious children in Romania

    Anxiety and depression are the commonest mental health problems in young people, and most never reach a therapist. Romanian researchers tested a six-week internet-delivered programme covering both together, rather than one diagnosis at a time. They randomly assigned 106 children aged 11 to 17 with an anxiety or depressive diagnosis either to the programme or to a waiting list, and followed them for six months. More of the treated group recovered, and their symptoms fell by a moderate amount compared with waiting. [40]

    Why it matters — The comparison is against nothing at all, so it shows the programme beats a wait, not that it matches seeing a therapist.

34. 35

    Two medicines for eating disorders in thirty years

    A review covering 2010 to 2025 counted what has been approved for eating disorders. The answer is two medicines, both American approvals and neither recent: fluoxetine for bulimia nervosa in 1994, and lisdexamfetamine for binge eating disorder in 2015. Nothing at all is approved for anorexia nervosa. A few candidates are being studied, including solriamfetol and psilocybin, but the review found few drugs left in late-stage trials as of 2025. [41]

    Why it matters — Eating disorders carry some of the highest death rates in psychiatry, and this is what thirty years of drug development has left the clinic.

35. 36

    An AI assistant in Kenyan clinics changed nothing

    Sixteen primary care facilities in Kenya took part in a trial between April and July 2025. Clinical officers were randomly assigned to use their electronic records with or without help from a large language model. Across 9,691 patients seen by 103 clinical officers, an expert panel judged whether treatment failed within 14 days. It did for 2.2% of patients in the assisted group and 2.0% in the control group, a difference well within chance. No safety signal was found. [42]

    Why it matters — It is one of the few real-world tests of this technology in a working health system, and the honest answer it returned was no measurable difference either way.

36. 37

    Cancer immunotherapy given later in the day

    Immune checkpoint inhibitors release a brake the immune system puts on itself so it can attack a tumour. A large cancer centre looked back at 2,631 patients treated between 2018 and 2023, mostly for advanced lung cancer, melanoma and kidney cancer. It split them by whether most of their infusions came before or after the middle of the day. Median survival was 21.4 months for the earlier group and 13.1 months for the later one. [43]

    Why it matters — It is a look back at records, not a trial, and patients scheduled late in the day may differ from those scheduled early. Several earlier studies point the same way, which is why it is now being tested properly.

37. 38

    What actually stops catheter infections

    Urinary catheters cause one of the most common infections people pick up in hospital. A review searched seven databases up to August 2024 for trials and controlled studies of ways to prevent them. Antiseptic cleaning with chlorhexidine, catheters coated with silver alloy or other materials, and staff training sessions all cut infection rates, with some interventions reporting reductions as high as 94%. Reminder systems that prompt staff to take catheters out on time also worked. Side effects were minimal. [45]

    Why it matters — None of these is new or expensive. The review's conclusion is about implementation, which means the gap being reported is between what is known and what is done.

38. 39

    What raises the risk of dying from sepsis

    Researchers pooled 20 observational studies that had adjusted for other factors, looking for what predicts death within 30 days of sepsis. Septic shock, where blood pressure will not hold up, more than doubled the risk. Liver disease raised it by about a half and sudden kidney failure by about the same. Age, lactate in the blood and the standard intensive care severity scores all added smaller amounts. Diabetes and a urinary tract source of infection went the other way, with lower risk. [44]

    Why it matters — Most of the list is about how sick the person already was, not about what the hospital did. That is why the trials in today's other sepsis stories keep measuring the clock instead.