Abstracts

Rationale: Generalized tonic-clonic seizures (GTCs) are associated with vocalization, asymmetric clonic jerking, diffuse tonicity, and ending with a rhythmic clonic phase. GTCs can be easily identified and diagnosed by video EEG monitoring. We explore feasibility of audio recognition of GTCs by epilepsy experts, fellows and residents and provide education to identify specific features. Methods: Retrospective review of patients from Epilepsy Monitoring Unit was done and events diagnosed as GTCs or Psychogenic non-epileptic seizures (PNES) were selected for audio review. Two- to 3- minute audio segments of 18 events were selected for the study. Clinical information, the video clips and the diagnosis corresponding to the events were blinded. Audio samples were introduced to four "experts" consisting of epilepsy faculty and fellows with variable years of experience and they were asked to score the event as a GTC or PNES. The same procedure was repeated to 9 to 11 neurology residents in training twice. The second resident trial occurred after education (post-intervention) was provided about the identification of vocalization (ictal cry and gurgling sounds) and patterns of audio changes unique to GTC. Audio scores were compared to clinical diagnosis and analysis was done. Sensitivity, specificity, and inter-rater agreement was identified at each level. Results: The sensitivity for identifying GTC was 72.50% (CI=56.11% to 85.40%) in the expert group with a specificity of 75%. Positive likelihood ratio (PLR) was 2.90 (CI= 1.6-5.27) and Negative likelihood ration (NLR) was 0.37 (0.21-0.63). Inter rater reliability (IRR) using Davies and Fleiss method showed K value of 0.56 (moderate agreement) for evaluation of PNES and K value of 0.53 (moderate agreement) for evaluation of GTC seizures. The pre-intervention sensitivity and specificity for audio identification of GTCs by resident participants was 54.21% and 68.18%, respectively, with a PLR of 1.7 and NLR of 0.67. Post education, the sensitivity and specificity was 61.90% and 81.90% respectively with a PLR of 3.43 and NLR of 0.46. IRR showed K value of 0.17 pre-intervention (slight agreement) and 0.32 post-intervention (fair agreement). Conclusions: Audio analysis can be used to identify GTCs and can be used to differentiate from PNES. This is better identified by expert rather than non-expert participants and improves with identification of specific features that can aid in audio recognition of GTCs. Further digital analysis of specific features can potentially lead to the creation of at-home or in-hospital automated audio recognition tools. Funding: N/A