Astemizole carried a 128-fold hERG margin and was pulled from the market for TdP. The margin reads it Low. The mechanism read reads it Intermediate, the class CiPA assigns. The market withdrew the drug.
The single-channel hERG margin is the cheapest cardiac-safety screen that exists, and it is wrong in a specific, preventable way. It reads a number and stops: how far is exposure from the concentration that blocks hERG? A wide margin says safe. Astemizole shows why the number is not enough.
The compound
Astemizole is an antihistamine that was withdrawn for QT prolongation and torsades de pointes. It blocks hERG very potently, with an IC50 of 33.3 nM, and it is given at a tiny exposure, a free Cmax of 0.26 nM. The therapeutic margin against hERG is therefore enormous: 128-fold. A margin-based screen reads that number and calls it safe.
The market judged it unsafe. The disconnect is a real signal. The margin asks only one question; the block's effect on a repolarizing cell is what the market answered.
What a mechanism read sees
Run the same two public inputs through a mechanism read on a published human iPSC-cardiomyocyte model, and the answer changes. Only one current moves at the doses that matter, hERG, and it moves by less than 8% even at four times free Cmax. The model prolongs APD90 by 1.3% at 1x and 3.3% at 4x.
The mechanism call on astemizole awaits a confirming assay. The class is right, and the ranked mechanism list under it stays unconfirmed; the model's mechanism ranking is unvalidated against wet data. Read a correct class as a class, and a mechanism ranking as a separate claim.
3% APD90 prolongation is near a plate's noise floor. Our own noise testing recovers small-magnitude causes correctly about a third of the time. Astemizole is a case where the right class survives, and the resolving power is thin.
Astemizole is the single cleanest case for mechanism. Most cases carry more ambiguity; on the full 28-drug panel, the margin is ahead on aggregate. That trade, and the loss, are published here.
Why the margin fails
The margin is a ratio of two concentrations. It carries no information about what blocking the channel does to a cell, because a block of 8% matters or stays silent depending on the other currents. Astemizole is the clean version of that: a potent blocker whose effect at therapeutic exposure is small but present, and the model simulates the effect where the margin simulates nothing.
The market is the final referee, and the market withdrew the drug. A screen that would have read it as Low at the time carries a label and no signal; the model simulates the effect the market then confirmed. Astemizole is the fourth of four worked reads on this site, run end to end with every miss stated.
your compound, read the same way
Send one compound with published ion-channel IC50 data and an approximate free Cmax. The first read is free and returns the same report format: ranked mechanisms, risk class, and the experiment that confirms the cause. What a free read covers.
Ion-channel IC50, Hill and free Cmax from the CiPA 28-drug reference dataset (FDA/CiPA public repository, branch Model-Validation-2018; Li et al. 2017); TdP class from Colatsky et al. 2016. Model runs on the Kernik-Clancy 2019 human iPSC-CM model (doi.org/10.1113/JP277724) integrated with Myokit/CVODE, frozen CiPA validation run T-151 (2026-07-03). Mechanism ranking unvalidated against wet data.