A contextual error occurs when a plan of care is clinically correct in the abstract and wrong for the individual, because something about their circumstances was not asked about or not taken into account. It is distinct from a biomedical error, which is wrong for anyone.
The term comes from Saul Weiner and colleagues, who sent unannounced actors trained to present as patients into consultations with 111 internal medicine physicians. Each presented a clinical red flag and a contextual one. The physicians probed the contextual flags about half the time, and in the encounters where context complicated the picture, only 22% produced an error-free plan of care.
The finding transfers to AI directly. A system optimised to be correct in general will produce a great deal of care that is wrong in particular, and it will look accurate the whole time it is doing so. Transport, cost, caring responsibilities, shift work, literacy and trust all decide whether a correct recommendation is a usable one.
Where this is explored in depth