Abstracts

Rationale: While treatment with medications is often effective, 40% of the 300 000 Canadians suffering from epilepsy are refractory to medications while only 5% of patients with refractory epilepsy respond clinically to additional antiepileptic drug (AED) [1]. This group of patients is often referred to the referral centers such as the Calgary Epilepsy Program (CEP). To clarify diagnoses, optimize therapy or for pre-surgical evaluation, they are subsequently admitted to the Seizure Monitoring Unit (SMU) for on average 8 days to record and analyze seizures with video-EEG. However, over a fifth (21%) of SMU patients do not have seizure events while on the unit. The purpose of this study was to improve the efficiency of SMU admission by developing models to predict the likelihood of obtaining seizure events on the SMU and exploring associations between seizure events and clinical variables.Methods: Patient admission data (n=575) from 2008 to 2013 were analyzed to determine predictors. Preliminary descriptive analyses (means, standard deviation, and frequency distribution) were conducted to describe the data. Binary logistic regression was used to model the association between seizure event and 14 demographic and clinical variables. Two models were determined using the same constructs: patients with any events (n=533) and with epileptic seizures (ES) (n=516). The estimates of odds ratio (OR) of the regression coefficients were reported.Results: A higher seizure frequency would lead to a higher probability of having events of any type (OR=1.39). Events were more likely to occur among those who took more medications (OR=1.74). Patients with stopped AEDs were more likely to have events (OR=2.45) compared with those with no AEDs change. Activating procedures (hyperventilation, sleep deprivation) had no effect. Among those with ES, non-epileptic reason for referral was associated with lower risk of having events (OR=0.09) compared with epileptic reason for referral. Number of medications had the same effect as overall group with any events (OR=1.64). AEDs reduction had no significant effect on patients with ES events. The prediction accuracy for the above two models are 77.67% and 77.67%, respectively.Conclusions: The use of pre-admission variables to predict the likelihood of seizure events on the SMU can help improve hospitalization both for patients and clinicians. Further refinement of the model could result in a useful paradigm at admission. It is important to emphasize that the associations found are not necessarily causal and that more in depth analyses are required to explore the mechanisms of some of these associations.