How Neurologists Combine Clinical Signs and Subjective Factors to Diagnose Epileptic and Functional Seizures: Insights From Seizure Video Analysis

Evaluating seizure semiology is crucial for distinguishing functional seizures (FS) from epileptic seizures (ES). This study examines how neurologists integrate clinical signs, observations, and subjective factors in making this distinction. Neurologists at Monash Health assessed unlabeled seizure videos on a web-based platform, providing free-text descriptions, selecting observed clinical signs, offering diagnoses, and rating their certainty (0-100). Non-negative matrix factorization (NMF) clustered clinical signs, and thematic analysis was performed on free-text responses. Logistic regression evaluated how single signs, sign combinations, and rater consensus influenced diagnostic alignment. Model performance was assessed using area under the receiver operating characteristic curve (AUC). NMF identified two sign clusters: one associated with FS (eye closure, asynchronous limb jerking, headshaking, pelvic thrusting) and another with ES (eyes open, head version, tonic/dystonic posturing, automatisms, synchronous limb jerking). Diagnostic alignment was influenced by clinical sign combinations (OR = 1.37, 95% CI = 1.15-1.64, p < 0.001) and rater consensus (OR = 1.40, 95% CI = 1.15-1.77, p = 0.002). Models incorporating these factors incrementally improved performance, with the final model achieving an AUC of 0.88 (95% CI: 0.84-0.92, McFadden's R2 = 0.37). Neurologists rely on distinct clinical sign patterns and subjective factors to differentiate FS from ES. Understanding these decision-making processes can improve diagnostic workflows and training programs.