Outputs
Captured data were used to ascertain three outputs for non-expert diagnosis (figure 2).
The first output was data captured by the questionnaire, which was used for prediction model diagnosis. We also recorded time taken to administer the questionnaire.
The second output was diagnosis reached by the non-expert (nurse), using data they captured with the questionnaire and video otoscopy on the Cupris device. They were asked to select a diagnosis from a list: normal / acute otitis media (AOM) / CSOM / inactive tympanic membrane perforation / otitis externa / otitis media with effusion / cholesteatoma / other.
The third output was tele-diagnosis by an expert, using data captured by the non-expert (nurse). Questionnaire results were translated back into English, and these and the otoscopic videos were anonymised and sent to specialist otologists in the UK (authors [removed for blind peer review] and [removed for blind peer review]). Owing to logistical constraints, data were reviewed 9-11 months after capture. Experts were asked to select a diagnosis from the same list as the nurse.
All assessments were blind to diagnosis by other methods.