Abstracts

Rationale: Study of semiology can help clinicians differentiate psychogenic nonepileptic seizures (PNES) from epileptic seizures (ES). Although the published literature has mostly focussed on clinical signs in isolation, clinicians rely on patterns of clinical signs when diagnosing PNES and ES. Literature focussing on combinations of clinical signs is sparse. In this study, we sought to understand how combinations of clinical signs can help clinicians diagnose PNES and ES.

Methods: We invited neurologists at our institution to participate in a web-based survey. Each participant independently reviewed 11 ES and 11 PNES videos (unaccompanied by clinical data) and were required to: 1) describe the video, 2) select clinical signs present in each video, 3) offer a diagnosis, and 4) state their degree of certainty. Non-negative matrix factorisation (NMF) was performed on categorical data (‘yes’ or ‘no’) from the survey. The NMF method was used to cluster patients into groups based on their similarity with the co-efficient matrix representing the ‘weights’ attributed to each clinical sign. Descriptive free text responses were analysed using Latent Dirichlet Allocation (LDA) algorithm to identify themes within the response. The LDA algorithm was used to scan participants’ free text responses and cluster words into themes based on multinomial probabilistic distribution.

Results: Seven neurologists completed the survey (median years of experience: 15, range: 6-33). Median patient age at time of video EEG was 33 (range: 18-76). Seizure videos comprised of nine focal onset motor seizures, two generalised onset motor seizures, seven complex motor PNES, three rhythmic motor PNES and one mixed PNES. Six different themes or ‘topics’ were identified through analysis of free text responses. The ten most common and ten least common terms in each topic and their corresponding probabilities of occurrence in those topics are summarized in Table 1. Terms such as ‘eye closure’ ( 0.063), ‘asynchronous’ (0.034), ‘pelvic thrusting’ (0.020) and ‘waxing and waning’ (0.013) had higher probabilities of occurrence in PNES topics, whilst terms such as ‘synchronous’ (0.0003) and ‘clonic’ (0.0003) had lower probabilities of occurrence in PNES topics. Terms such as ‘dystonic’ (0.050), ‘eyes open’ (0.042), ‘tonic’ (0.028) and ‘head version’ (0.013) had higher probabilities of occurrence in ES topics, whilst terms such as ‘voluntary’ (0.003) and ‘talks’ (0.0002) had lower probabilities of occurrence in ES topics. Analysis of categorical responses yielded the following combinations of clinical signs for diagnosing ES and PNES: a) head version and synchronous limb jerking (ES), b) eyes open during seizure and tonic or dystonic posturing (ES), c) ictal eyes closure, side-to-side head shaking and asynchronous limb jerking (PNES) (Figure 1). Individual clinical signs such as crying, pelvic thrusting, and waxing and waning course were associated with PNES, whilst automatisms and continuous time course were associated with ES.

Conclusions: Our study identified several combinations of clinical signs that can help clinicians differentiate PNES and ES in the absence of video EEG.

Funding: No funding was received in support of this abstract.