Three out of how many? That was the first thing I wanted to know when I read that three people in Novartis’s autoimmune CAR-T trials were dead, and six weeks later none of the public reporting has an answer. CAR-T for autoimmune disease was the treatment I’d let myself fall for: one infusion, an immune system knocked back toward something like factory settings, and people who’d spent years cycling through steroids and immunosuppressants walking away from all of it. I read the early case series closely. I read the safety sections a little too fast.
On August 24, Novartis stopped screening, randomizing, and dosing patients in eight trials of rapcabtagene autoleucel, or rap-cel, its CD19-targeted CAR-T therapy, in lupus, systemic sclerosis, vasculitis, myositis, rheumatoid arthritis, myasthenia gravis, multiple sclerosis, and more. Three participants had developed immune effector cell-associated hemophagocytic lymphohistiocytosis-like syndrome, IEC-HS for short, and complications led to “fatal outcomes,” a Novartis spokesperson told BioPharma Dive. Rap-cel’s two cancer trials, in lymphoma and leukemia, kept running. Even the death count has wobbled in public: Novartis attributed three deaths to the reaction in a statement reported by Reuters, while the Lupus Foundation of America’s write-up says two of the three patients died.
Bristol Myers Squibb then paused enrollment in its own autoimmune CAR-T, zolacabtagene autoleucel (zola-cel), “out of an abundance of caution,” after routine safety surveillance turned up “transient and reversible inflammatory events,” according to BioPharma Dive’s report on both pauses. A results paper from February had already described one IEC-HS case in a phase 1 zola-cel study. This week Nature Medicine ran the fallout under a headline that would have been unthinkable a year ago: “A reality check for autoimmune CAR T cells.”
What IEC-HS does in a body is brutal. T cells, natural killer cells, and macrophages switch on and don’t switch off, and the activated macrophages start engulfing the body’s own blood cells. Ferritin climbs, blood counts fall, fibrinogen drops so clotting goes haywire, and liver enzymes rise. It’s sneaky about timing, too, typically showing up after the fever of cytokine release syndrome has begun to settle, right when a care team might reasonably exhale. Treatment means high-dose steroids and targeted drugs such as anakinra, ruxolitinib, or tocilizumab, with emapalumab used off-label. Even in cancer, where CAR-T has years of track record, the exact incidence “remains uncertain.”
Novartis hasn’t given the number that would let anyone judge the risk. None of the reports on the halt say how many people received rap-cel across those eight trials, and European Biotechnology notes that the company hasn’t said when the deaths occurred or detailed the individual cases: which diseases, what ages, how far apart. Three deaths among a few dozen patients and three among several hundred describe very different drugs to someone deciding whether to enroll, and that denominator is the number a patient most needs. This is the same program whose early AUTOGRAPH data in myositis and systemic sclerosis Novartis presented at EULAR 2026 with a safety profile it called “manageable,” in a phase 2 lupus effort built around sustained remission at 24 and 52 weeks.
Then came the framing. In its September 1 statement, Novartis called this kind of severe immune reaction “a known, potentially life-threatening risk of CAR-T therapy”; the version quoted in PharmTech calls it “a known and potentially fatal risk associated with CAR T-cell therapy generally.” I get why a sponsor reaches for that word “generally,” and I don’t accept it as the whole picture, because spreading the risk across the entire field turns three deaths in your program into everyone’s problem. Analysts weren’t reading it as field-wide either. TD Cowen’s Phil Nadeau noted that Cabaletta Bio hasn’t seen any IEC-HS cases and wrote, “Given the difference in protocols, we see limited read-through,” per BioPharma Dive, though that’s an analyst’s note without patient counts, not a head-to-head comparison. Kyverna hurried to say its product “has a distinct construct design and is produced using a well-established, validated manufacturing process.” And the day Novartis put out that statement, its shares rose about 4%, carried by late-stage success for its oral multiple sclerosis drug, remibrutinib.
None of this would sting if the early results had been ordinary. In Erlangen, Germany, Georg Schett, Andreas Mackensen, and colleagues reported that all 8 people with treatment-resistant lupus they had treated were in remission and off glucocorticoids and every other immunosuppressant. Their CASTLE basket trial, published in Nature Medicine in January, gave 24 patients with lupus, systemic sclerosis, or myositis a single infusion, and 22 met the six-month efficacy endpoint; cytokine release syndrome hit 71% of them but never rose above grade 2. A pooled review of 145 lupus patients across 16 studies found 70% met the strict DORIS definition of remission and 84% reached drug-free remission. In the same Healio report, rheumatologist Philip Mease gave a tidy, physical reason the side effects ran so much milder than in cancer: lupus patients carry far fewer disease-driving B cells for the CAR-T cells to chew through, so the inflammatory blast is smaller.
| Measure | Part | Whole |
|---|---|---|
| patients met the six-month efficacy endpoint after a single infusion | 22 | 24 |
So I keep asking the obvious question. If the gentleness came from a small target, how does a syndrome defined by runaway inflammation end up killing people? The leading suspect sits in the factory rather than the disease, and it’s still only a suspect. Rap-cel and zola-cel are both made on rapid manufacturing platforms, and the working hypothesis, as Washington University lupus specialist Alfred Kim described it, is that “faster production yields fitter T cells that expand greater in the patient,” with IEC-HS at the severe end of that inflammation. William Blair analyst Sami Corwin wrote that the speedier technology “could be driving increased cell expansion and the reported toxicities,” while cautioning that other factors could be involved. Picture what that would mean in a body, if it holds: cells that spend less time growing in a bag do more of their multiplying in the bloodstream, and that bloodstream belongs to someone whose immune system is already primed. Jefferies analyst Roger Song flagged exactly that, noting that autoimmune diseases “have more activated immunity at baseline.” Nobody has proven the link, and Novartis hasn’t publicly offered a different explanation.
Look at who the press has leaned on to steady the field. Kim supplied the most quotable verdict, “Three deaths have to be taken seriously, and pausing was the only call,” and framed the halt as the system working, not the end of cell therapy in autoimmunity. Kim reports research funding from AstraZeneca, Bristol Myers Squibb, CRISPR Therapeutics, and Novartis. Mease, whose account of mild side effects and mild, partial relapses carries the optimistic case in lupus, reports consulting fees, grants, and speaking fees from Novartis, plus consulting fees and grants from Bristol Myers Squibb, alongside more than a dozen other companies. Neither disclosure makes either of them wrong. It does mean the two companies whose programs are paused also fund the clinicians brought in to put the pause in perspective, and readers deserve to have that on the table.
The sharpest thing Kim said cuts against the rush rather than for it: people with lupus, myositis, scleroderma, or other autoimmune diseases “are often younger, are not facing imminent death and increasingly have other options.” A lymphoma patient out of chemo options can rationally accept a known fatal risk. A refractory lupus patient may be badly sick too, but is weighing that risk against a different set of alternatives, and the bet shouldn’t hinge on how fast a company can turn cells around. The Lupus Foundation of America called for “continued research and vigilance.” Meanwhile Kyverna could file for approval in stiff person syndrome by year-end, so a regulator may be asked to rule on an autoimmune CAR-T before anyone has explained these deaths.
I still think the Erlangen remissions are some of the most remarkable biology I’ve read, and three deaths don’t erase one of them. But if I had refractory lupus and a trial slot, I’d ask the coordinator two questions before signing: how many patients had already been dosed on this exact product and process, and how many of them developed IEC-HS. If I couldn’t get both numbers, I wouldn’t sign.
Sources
- BioPharma Dive: Novartis and Bristol Myers pause autoimmune CAR-T trials (Aug. 31, 2026; updated Sept. 1)
- Healio: Three deaths halt Novartis CAR T-cell trials in autoimmune diseases, with comment from Alfred Kim (Sept. 2026)
- Reuters via Yahoo News: Novartis pauses trials of experimental cell therapy after three deaths (Sept. 1, 2026)
- Nature Medicine: A reality check for autoimmune CAR T cells (Oct. 16, 2026)
- BioPharma Dive: Analyst and competitor reactions to the autoimmune cell therapy safety events
- European Biotechnology: Three deaths put CAR-T expansion into autoimmune diseases on hold
- PharmTech: CAR T-cell trial pause signals new manufacturing risk review
- American Society of Hematology: IEC-HS as a post-CAR-T complication
- ASTCT: Summary of the phase 1/2 CASTLE basket trial (Müller et al., Nature Medicine, 2026)
- ACR Convergence 2023: Schett, Mackensen et al., long-term remission after CD19 CAR-T in lupus
- Healio: Emerging data bring CAR T-cell therapy benefits and risks in lupus into sharper focus (pooled 145-patient review, comment from Philip Mease)
- Lupus Foundation of America: Novartis and Bristol Myers Squibb pause CAR-T trials for lupus and other autoimmune diseases