argenx's Own Scientists Found Antibody Levels Alone Don't Explain Who Its $225,000 Myasthenia Drug Helps

For years I filed myasthenia gravis under solved enough. Your immune system builds antibodies against the receptor your nerves use to tell your muscles to move, the muscles stop listening, your eyelids droop and your grip goes, so a drug that hauls those antibodies out of your blood should hand you your body back. Clean, mechanical, satisfying. I believed that about efgartigimod, the drug argenx sells as Vyvgart, and it is the exact story the sales pitch rests on: pull the immunoglobulin down, the weakness lifts.

So a set of results out of argenx’s own labs stopped me cold. When the company’s scientists went back into the stored blood of 50 patients from its pivotal trial and measured what the drug had actually done to the antibodies, the tidy version came apart. Efgartigimod drove down every detectable class of these antibodies. How far it drove them down told you almost nothing about who got stronger.

Here is the drug’s résumé, because it decides what the rest means. Efgartigimod is an FcRn blocker: it jams the recycling receptor that normally rescues your antibodies from the cell’s disposal machinery, so all your IgG, the harmful ones included, gets swept toward the incinerator instead. In the phase 3 ADAPT trial behind its December 2021 FDA approval, 67.7% of treated patients improved on the standard daily-living score versus 29.7% on placebo. It works. It also carries a net price around $225,000 a year, and independent analysts put its per-patient cost far above therapeutic plasma exchange. At that number, who this actually helps is not an academic question.

The new substudy, published this month in the Journal of Neurology, Neurosurgery & Psychiatry, took 150 serum samples from 50 antibody-positive ADAPT patients (40 on efgartigimod, 10 on placebo) at baseline and again at days 29 and 57, and asked a sharper question than how much antibody is left. It asked what the antibodies could still do. That is where the biology got strange enough that I read it twice.

Because not all of these antibodies misbehave the same way. Some, once they latch onto the acetylcholine receptor, flag down the complement cascade, a chain of blood proteins that drills holes straight through the muscle-cell membrane. Others grab two receptors at once and haul them down inside the cell, so the nerve’s signal arrives and finds nothing to dock onto. A smaller group just plugs the acetylcholine binding site directly. At baseline, complement-triggering antibodies turned up in 84% of these patients, receptor-internalizing ones in 82%, and direct binding-site blockers in only 20%. Three different ways to break the same synapse, riding around in the same bloodstream.

Wait, so why would a drug that removes all of them equally produce such uneven results? That is the question I kept circling, and the answer the data gives is quietly radical for a six-figure therapy. Efgartigimod drove down every detectable subclass of these antibodies. What varied, patient to patient, was how far it dialed back what those antibodies could still do. The people who improved on both clinical scores, the double responders, were the ones in whom the drug cut the complement activity and the receptor-dragging the most. Same drop in numbers, different outcome, because the antibodies left behind were not equally dangerous. The company’s own scientists put it in careful language: clinical benefit is more closely associated with functional modulation of the autoantibodies than with their absolute quantity alone.

I want to sit with what that admission costs, and who is making it. This paper was not written by outside skeptics. Six of its authors are argenx employees who hold stock options in the company, and the samples came from argenx’s own trial. When a manufacturer’s team publishes that its own we-lower-your-bad-antibodies pitch does not explain who the blockbuster helps, that carries more weight than the same words from a critic. It is also worth watching where the reframe points. “We need to measure antibody function, not just level” is honest science. It is also a clean runway toward companion tests, responder-selection assays, and a story in which uneven results become a reason to sell more diagnostics rather than a reason to question a $225,000 price.

And the evidence is thinner than the confidence of that conclusion. Forty patients on the drug, with responders and non-responders split across two different scores, is a small base to carry a mechanistic rule. The assays run in a dish, and they read the antibodies floating in serum, not the ones locked onto the muscle where the weakness actually happens. So this is a clue about how the drug works, not a test you can ask your neurologist to run on Monday.

What it changes, if it holds up, matters anyway. It means two patients with identical antibody titers are not the same patient, and that the usual way of tracking this disease, and of justifying who stays on the most expensive option, may be measuring the wrong thing. For a family staring down a $225,000-a-year decision, that is not a footnote.

My conclusion follows. A falling antibody titer is not proof the money is working, and I would stop treating it as if it were, because the people who profit from the drug just published that it is not the thing that tracks who gets better. I would ask whether anyone can yet measure how my antibodies behave, not just how many I have, and I would want that measurement run by someone who does not hold the stock. Until that test exists and holds up in a bigger, independent sample, I would not let a dropping number talk me into calling a $225,000 answer solved.

Sources

1. JNNP – Khani-Habibabadi et al., pathogenic AChR autoantibody properties and efgartigimod response, ADAPT substudy (2026)
2. Lancet Neurology – Howard et al., ADAPT phase 3 efgartigimod trial (2021)
3. Pharmacy Times – efgartigimod cost far higher than therapeutic plasma exchange (2026)
4. CADTH – pharmacoeconomic review of efgartigimod alfa (Vyvgart), net price and cost