Interobserver agreement in BPPV diagnosis using eye movement recordings

Benign Paroxysmal Positional Vertigo (BPPV) is one of the most prevalent peripheral vestibular disorders seen in specialized dizziness clinics. Despite being a well-understood condition with effective treatment options, BPPV remains associated with significant diagnostic delays and healthcare costs. If proven reliable, telemedicine approaches could help address these challenges by improving diagnostic accessibility and efficiency. To investigate the interobserver agreement in BPPV diagnosis, when using eye movement recordings. Six vestibular medicine specialists (Specialist 1, 2, 3, 4, 5, 6) were recruited to participate in this study. The specialists were asked to evaluate the recordings of 240 patient cases who underwent assessment for BPPV (first assessment). After viewing the recordings of each case, they were required to make a BPPV diagnosis. Five specialists (2, 3, 4, 5, 6) agreed to repeat the procedure twice, to additionally evaluate the intraobserver agreement (second assessment). The proportion of agreement and Cohen's kappa were calculated for both interobserver and intraobserver agreement. Furthermore, agreement with the original diagnoses was evaluated. The interobserver agreement between experts was fair to moderate with a Cohen's kappa value of 0.40 (CI 95% [0.35, 0.45]) and a proportion of agreement of 60% (CI 95% [54, 67]). Specialists 2, 4, and 5 exhibited substantial intraobserver agreement. In contrast, Specialist 3 demonstrated fair intraobserver agreement, while Specialist 6 showed almost perfect intraobserver agreement. Regarding the first assessment, agreement with the original diagnoses ranged from fair to substantial, with kappa values between 0.40 and 0.70, and corresponding percentages between 58 and 78%. Similar results were observed for the second assessment. The interobserver agreement between specialists diagnosing BPPV using eye movement recordings, was fair to moderate. The suboptimal agreement could be related to missing clinical information (e.g. patient history and symptoms during positional maneuvers). Future studies should incorporate this information and reassess interobserver agreement.