Strand

A cell model, a strip of heart tissue and a statistical personalization step, wired together and tested on made-up patients. Move a slider and the whole chain reruns.

Every patient here is simulated. Nothing on this page is a clinical result.

Hidden repolarization reserveThe patient’s true value. The estimator never sees it.

Measurement noiseError on the one baseline QT reading, in milliseconds.

Drug doseShare of the outward repolarizing current the drug blocks.

Same patient, a different measurement error.

What the twin infers from one reading

The twin starts from a population prior and updates it with the baseline QT. The narrower the solid curve is compared with the dashed one, the more the reading actually told us.

Prediction against the truth

The question is how much the drug lengthens this patient’s QT. The truth comes from the full tissue model.

Many patients, two groups

Group B has a different true distribution and far fewer people in the reference data. The audit asks whether the twin is as accurate and as honest about its uncertainty for B as for A.

Give each group its own prior, partly pooled with the whole

Do the layers agree with each other?

Cells and tissue swap voltages every exchange interval. A longer interval is cheaper but less exact. The test measures the error that creates and checks it against the bound from Section 3.6.8 of the paper.