Discussion
This is the first study to directly compare performance of different
methods for diagnosis of external and middle ear disorders. We report
variable results. Prediction model diagnosis using automated history
performed poorly, but an absence of otorrhoea (perhaps unsurprisingly)
was found to reliably exclude CSOM. This implies that individuals who
deny a history of ear discharge could confidently be excluded in
programs screening for CSOM.
Whereas both nurse-led and remote expert diagnosis showed high
specificity, they had low sensitivity. The reasons for poor sensitivity
can only be surmised, but could include poor performance of the
questionnaire, poor quality video capture on otoscopy, insufficient
video resolution, low experience or confidence of the assessor, or
diagnostic uncertainty or disagreement. Poor image capture and
diagnostic incongruity has been reported in other
studies.7
We recognise our study as a preliminary exploration of methods for
screening or diagnosis of middle or external ear disease (where on-site
specialist assessment is unavailable). Future work could look to
evaluate if and how validity of these approaches may be improved,
including through improving the quality or platform for image capture to
assist remote diagnosis; targeted training of non-experts to develop
diagnostic skills; or providing additional or enhanced information to
facilitate prediction model diagnosis (in particular the addition of
automated interpretation of otoscopic images).10