🧠 One Blood Test Could Flag Dementia 25 Years Before It Hits

What if a blood test could see dementia coming decades before the first symptom? This week: a biomarker with a 25-year track record, cancer immunotherapy that builds itself inside the body, a tech founder who ChatGPT'd his way to a dog cancer vaccine, NIH funding in freefall, and Ozempic copies hitting India at sandwich prices. Let's get into it.

RESEARCH

If you've been reading Cereal Bio lately, p-tau217 is becoming a familiar face. This blood biomarker for Alzheimer’s-related brain changes keeps building its resume, and now it has its longest study yet.

A team led by Aladdin Shadyab at UC San Diego measured p-tau217 in frozen blood from 2,766 cognitively healthy postmenopausal women enrolled in the Women's Health Initiative Memory Study between 1996 and 1999. For each standard deviation increase in the biomarker, the risk of developing dementia rose 3.17-fold over up to 25 years of follow-up. Women on estrogen-plus-progestin hormone therapy showed an even stronger association (HR 4.18), while estrogen-alone therapy showed no such effect.

Now, about that "25 years" headline (yes, we used it too). That's the maximum follow-up window. Median follow-up was 14.1 years. As Dr. Ivan Koychev at Imperial College London pointed out, that number reflects the observation window, not the test's ability to predict dementia decades out for any individual. Big difference between population-level association and individual prediction.

Still, a simple blood draw outcompeting PET scans and spinal taps in terms of accessibility is meaningful, especially given that women face roughly twice the dementia risk as men for reasons we still don't fully understand. The study was women-only, so whether these findings extend to men remains an open question.

RESEARCH

CAR-T cell therapy is one of oncology's biggest wins, but it comes with a logistical headache: extract the patient's T cells, ship them to a specialized facility, genetically modify them over weeks, run quality checks, then infuse them back after lymphodepleting chemo. The whole process runs $400,000 to $500,000 per treatment. Apparently, there might be a simpler way.

A team led by Justin Eyquem at UCSF built a two-part system that engineers CAR-T cells directly inside the body with a single intravenous injection, no cell harvesting required. One particle delivers CRISPR editing machinery to cut open a precise spot in T cell DNA; the other carries the cancer-targeting gene that slots into that gap. Both are engineered to home in on T cells, and the gene only activates if it lands in exactly the right location (basically a molecular lock-and-key, except the lock is also a key and both have to agree).

In mice, a single injection cleared aggressive leukemia in nearly all animals within two weeks and achieved complete responses across all eight mice with multiple myeloma. With one of two cell donors, it even worked against solid tumors (CAR-T's traditional weak spot), with five of six mice showing complete responses.

These are mice, not humans, and that gap has humbled many a promising therapy. Eyquem's startup, Azalea Therapeutics, is advancing the approach toward clinical studies, though no specific timeline has been announced. But if this holds up, making cancer immunotherapy as simple as an IV drip would be a genuine game-changer.

NEWS

When Sydney tech entrepreneur Paul Conyngham's rescue dog Rosie was diagnosed with advanced mast cell cancer in 2024, vets gave her one to six months to live. Conventional treatments slowed but didn't shrink her tumors. So Conyngham, a machine learning engineer with no biology background, did what any reasonable person would do: he opened ChatGPT.

He spent $3,000 AUD on DNA sequencing at UNSW's Ramaciotti Centre for Genomics, used AlphaFold to model Rosie's mutated c-KIT protein, and worked with UNSW researchers to design a personalized mRNA vaccine targeting her tumor-specific mutations. After a pharmaceutical company declined compassionate use of an immunotherapy drug for a dog (because, of course, they did), UNSW's RNA Institute manufactured the vaccine in under two months.

Rosie received her first injection in December 2025. Her tennis ball-sized tumor has since shrunk substantially, though the exact figure is disputed: Conyngham claims ~75% on television, while early media reports put it closer to 50%. She's energetic enough to be jumping fences again, which is either a great sign or a whole new problem.

Before you start prompting ChatGPT for medical advice: this is one dog, one vaccine, no peer review, and no control group. Rosie also received a checkpoint inhibitor alongside the vaccine, making it impossible to tell what actually did the work. UNSW has been explicit: Rosie still has cancer, and it's still incurable. But a computer scientist building a custom mRNA therapy from his laptop is either the future of medicine or the most chaotic good energy we've seen all year.

NEWS

We've covered the squeeze on US science funding before, but the latest numbers are something else entirely. Elizabeth Ginexi, a former NIH program director who spent 22 years writing research funding announcements, tracked the agency's output of Notices of Funding Opportunity through grants.gov and found the pipeline has essentially frozen.

NIH published 756 NOFOs in 2024. In 2025, that dropped to roughly 120, an 84% decline. Through mid-March 2026, just 14 have been posted, a pace that Ginexi projects would yield a 98% collapse by year's end. Hundreds of completed announcements sit in a political approval queue that now routes every funding call through NIH leadership, HHS, and OMB before it can go live.

Important distinction: these are funding announcements (essentially calls for proposals telling researchers what NIH wants funded), not grants awarded or dollars spent. But the downstream effects are already real. Competitive grant awards are down 66% compared to recent averages, and the FY25 success rate hit approximately 17%, a nearly 30-year low. Early-career researchers are being hit hardest.

NIH Director Jay Bhattacharya has called the reported slowdown "hype," saying the agency plans to rely more on broad "parent announcements" instead of targeted funding calls. Researchers can still submit through those channels. But as Ginexi put it, these weren't ideas rejected on scientific merit: they were fully developed, internally reviewed, budget-allocated announcements waiting for political sign-off that never came.

NEWS

Semaglutide, the molecule behind Ozempic and Wegovy, costs roughly $970 per month in the US. In India, as of March 2026, you can get it for $14.

That's the price from Natco Pharma for a monthly supply in vial format at the lowest starting dose, launched on or immediately after Novo Nordisk's formulations patent expired on March 20. They weren't alone: at least 12 major drugmakers launched around the same date, with industry estimates pointing to over 40 companies entering the market this year.

A note on that $14 figure: it's the absolute floor, for a vial format at the lowest starting dose. Prefilled pens (which most patients prefer) start at about ₹1,800/month (~$19) from Alkem Laboratories and run up to $45 depending on the manufacturer. Still, even the most expensive Indian generic costs less than a month of streaming in the US.

Novo Nordisk responded by slashing prices up to 36% on Ozempic and 48% on Wegovy in India. The company's stock has dropped roughly 75% from its June 2024 peak, and its 2026 guidance projects the first sales decline in nearly a decade. Meanwhile, India has over 100 million people with diabetes and 250 million who are overweight or obese, but only about 200,000 currently use GLP-1 therapy. That gap is about to close, fast.

This week's theme, if you squint, is about who gets access to what. A blood test that could flag dementia risk years out, but only studied in women. CAR-T cells that might skip the half-million-dollar manufacturing process, but have only been tested in mice. An AI-designed cancer vaccine that worked (maybe, partially, in one dog) without peer review. Research funding announcements that aren't being announced. And a blockbuster drug that went from $970 to $14 by crossing an ocean.

The science keeps moving. The systems built to deliver it are struggling to keep up. If anything this week surprised you, hit reply and tell us which story. And if you know someone who'd enjoy reading about vibes-based veterinary oncology, forward this along.

Keep questioning everything (especially $970 price tags)

Prateek & Jere

P.S. Somewhere in Sydney, a machine learning engineer is explaining to a veterinarian that ChatGPT told him this would work. We're entering the vibes-based medicine era, and honestly, it's going better than expected.