Estimating the effect of treatment when patients in the control arm later take the new treatment

What is it about?

This paper considers how to estimate the effect of a new treatment from a clinical trial when many patients in the control arm of the trial switch and start using the new treatment. The paper considers what to do when there is a binary outcome (e.g. recovered/ did-not recover) and also when there is a time to event outcome using the proportional hazards model. The new estimators are studies through computer simulation and are applied to a trial of breast cancer treatment.

Featured Image

Why is it important?

It permits causal inference in clinical trials when because the treatment seems to work lots of patients int he control arm switch tot he new treatment.