Diagnostic accuracy of clinical signs and symptoms for psychogenic nonepileptic attacks versus epileptic seizures: A systematic review and meta-analysis.

BACKGROUND:Psychogenic nonepileptic attacks (PNEA) are events of altered behavior that resemble epileptic seizures (ES) but are not caused by abnormal electrical cortical activity. Understanding which clinical signs and symptoms are associated with PNEA may allow better triaging for video-electroencephalogram monitoring (VEM) and for a more accurate prediction when such testing is unavailable. METHODS:We performed a systematic review searching Medline, Embase, and Cochrane Central from inception to March 29, 2019. We included original research that reported at least one clinical sign or symptom, included distinct groups of adult ES and PNEA with no overlap, and used VEM for the reference standard. Two authors independently assessed quality of the studies using the Quality Assessment of Diagnostic Accuracy Studies tool. Pooled estimates of sensitivity and specificity of studies were evaluated using a bivariate random effects model. RESULTS:We identified 4028 articles, of which 33 were included. There was a female sex predominance in the PNEA population (n = 22). From our meta-analysis, pooled sensitivities (0.27-0.72) and specificities (0.51-0.89) for PNEA were modest for individual signs. History of sexual abuse had the highest pooled specificity (89%), while the most sensitive feature was female sex (72%). Individual studies (n = 4) reported high levels of accuracy for ictal eye closure (sensitivity 64-73.7% and specificity 76.9-100%) and post-traumatic stress disorder (no reported sensitivity or specificity). Assuming the pre-test probability for PNEA in a tertiary care epilepsy center is 14%, even the strongest meta-analyzed features only exert modest diagnostic value, increasing post-test probabilities to a maximum of 33%. CONCLUSIONS:This review reflects the limited certainty afforded by individual clinical features to distinguish between PNEA and ES. Specific demographic and comorbid features, even despite moderately high specificities, impart minimal impact on diagnostic decision making. This emphasizes the need for the development of multisource predictive tools to optimize diagnostic likelihood ratios.