Results
We recruited 85 participants (15 children and 70 adults), giving a total of 170 ears. One case was lost-to-follow up.
Completion of the questionnaire took an average of 6 minutes (range 1-14 minutes). In 23 ears wax precluded a diagnosis, and in a further 9, technical faults interrupted data capture. Therefore, we analysed questionnaire data for 138 ears.
For otoscopy, we obtained video data for 138 ears. However, experts considered three videos inadequate for tele-diagnosis (poor view of ear canal and/or tympanic membrane), leaving 135 cases for this method.
Gold standard diagnosis was achieved in all remaining cases, with 56 classified as normal external and middle ear, 22 CSOM, 21 otitis externa, 9 inactive tympanic membrane perforation, and 4 AOM. Performance of each diagnostic method compared to the gold standard is summarised in Table 1.
For prediction model diagnosis using the questionnaire, no particular symptom or combination of symptoms reliably differentiated diagnostic categories. For a symptom in isolation, sensitivity for a named diagnosis varied from 33-76%, except for presence of otorrhoea, which had 95% sensitivity for CSOM. Specificity varied between 67-80%.
Non-expert (nurse) diagnosis performed well for specificity (90-99% for all diagnoses) but poorly for sensitivity (<43% for all diagnoses).
Remote expert diagnosis also performed well for specificity (86-99%), but again less well for sensitivity (32-100%).