Abstracts

Rationale: The distinction between epilepsy and psychogenic non-epileptic seizures (PNES) is a difficult. History-taking is the key diagnostic tool. Several previous studies have used Conversation Analysis to examine how patients with epilepsy or PNES describe their seizures rather than focusing on what symptoms they mention. These studies suggest that the observation of linguistically describable features may contribute to the differentiation of epileptic and non-epileptic seizures. This study aimed (1) to confirm that video-EEG proven diagnoses can be predicted by linguistic analysis of transcripts of clinical encounters between doctors and patients with seizures; and (2) to explore whether this analysis can be translated into a quantitative score with acceptable sensitivity, specificity and interrater-reliability on the basis of a Diagnostic Scoring Aid (DSA). Methods: Twenty consecutive patients with diagnostically unclear, refractory seizure disorders were recruited. Typical seizures were recorded by video-EEG in all cases. All patients were interviewed by one neurologist (unaware of the diagnosis) using a previously established interview schedule characterised by open questions. Two independent linguists blinded to all medical information analysed video-recordings and transcripts of the encounters using a methodology derived from Conversation Analysis. Guided by a “Diagnostic Scoring Aid” (DSA) both developed a final diagnostic hypothesis (epilepsy or PNES) on the basis of their qualitative observations of 17 different linguistic and interactional features. They also translated all 17 qualitative assessments for each patient into simple quantitative scores (+1 - in favor of epilepsy; 0 - not diagnostic or rateable; - 1 - in favor of PNES; table 2). The subscores were added up to generate a total DSA score. Results: Seizure recordings demonstrated that, prior to admission, 60% of patients had carried an incorrect working diagnosis. The qualitative assessment of both linguists predicted the correct diagnosis in 17/20 (85%) patients (Kappa 0.57, moderate to good). 229/340 (67.4%) quantitative ratings showed agreement between both linguists. There was non-agreement on 109/340 (32%) of ratings. Frank disagreement was only recorded in 12/340 (3.5%) of ratings. Mean total DSA scores were significantly higher in patients with epilepsy than those with PNES (rater one: p=0.017; rater two: p=0.047). Using the DSA and the optimal diagnostic cut off score suggested by a receiver operator characteristic (ROC) curve, rater one rated 80% of the cases correctly (sensitivity 85.7%, specificity 84.6%) and rater two 75% of the cases correctly (sensitivity 71.4%, specificity 92.3%). Conclusions: Two medically blinded linguists were able to predict the video-EEG proven diagnosis of epilepsy and PNES in 75-85% of cases although the admission diagnosis was only accurate in 40% of patients. Interactional and linguistic observations can be operationalized and scored with acceptable interrater reliability.