Abstracts

Rationale: The diagnosis of seizures and epilepsy requires electroencephalography (EEG). There is a need for a rapid diagnostic test that can identify individuals with seizures post-hoc. We devised a predictive "seizure score" using multiple algorithms based on plasma levels of 3 inflammatory cascade proteins, tumor necrosis factor alpha (TNFa, thymus and activation regulated chemokine (TARC) and soluble intercellular adhesion molecule 5 (sICAM5). Methods: Plasma was isolated from blood samples (n=28) in epilepsy patients admitted to the epilepsy monitoring unit (EMU) for pre-surgical evaluation. All patients had EEG confirmed complex partial or generalized tonic-clonic seizures. All patients exhibited inter-ictal epileptiform EEG discharges (EEGs assessed by 2 epileptologists). EMU patients gave a sample (up to 10ml) of blood each morning during their EMU stay and an additional blood sample within 24 hours of a clinical event captured by video-EEG. Normal controls (n=29, no history of seizures) enrolled gave a single blood sample (up to 10ml). Plasma was isolated from centrifuged whole blood. TNFa, TARC, and sICAM5 levels were assayed in a multiplex ELISA and electrochemiluminescent detection system (Meso Scale Discovery Platform with Sector Imager 2400). Results: The statistical analysis was designed to produce a predictive "seizure score" as a function of patient characteristics (age and sex) and plasma protein levels (pg/ml) within 24 hours of an EEG confirmed seizure (Table). Protein levels were highly reproducible in duplicate assays. The mean plasma levels of TARC and sICAM5 were increased in seizure patients compared with controls at 24 hours post-seizure (p < 0.05). The relative ratios of TNFa to TARC and TNFa to sICAM5 were decreased in seizure patients compared with controls at 24 hours (p < 0.01). Independent diagnostic models were developed through the use of logistic regression and classification trees with multivariate logistic regression to yield a "seizure score" (Figure) at 24 hours post-seizure. The "seizure score" algorithms resulted in Receiver Operating Characteristic (ROC) area-under-the-curves (AUC) in the range of 0.8707 (95% C.I. 0.7739 - 0.9675) to 0.927 (95% C.I. 0.862 ?" 0.993) within 24 hours of a confirmed seizure. The diagnostic performance ranges were: sensitivity 89.3% to 93.9%, specificity 75.9% to 83.3%, positive predictive value (PPV) 78.1% to 86.1%, negative predictive value 88% to 92.6% and overall "seizure score" accuracy of 82.5% to 88.9%. Conclusions: We demonstrate predictive algorithms using plasma levels of sICAM5, TARC and TNFa that can positively identify seizures post-hoc at 24 hours post-seizure with 82.5% to 88.9% accuracy and 89.3% to 93.9% sensitivity. These data provide the first serological, non-electrographic assay to consistently and accurately identify a seizure after it has occurred that could have wide applicability in clinical seizure diagnostics. Funding: 1R43NS079029-01A1