Abstracts

Rationale: While continuous electroencephalography (cEEG) constantly records data, interpretation is often periodic. Although cEEG returns higher rates of seizure detection, hardware and labour costs often render cEEG not feasible for smaller medical centres. In Manitoba, EEGs are recorded intermittently but interpreted live for the entire duration. This distinction is important because real time seizure detection affords more rapid initiation of treatment. However, the effectiveness of this model is not well studied, and we aim to determine whether seizure detection rates in Manitoba are comparable to those in centres where cEEG is the standard model of practice.

Methods: We calculated composite residual seizure risk percentages using the clinically validated 2HELPS2B score from 504 inpatient EEGs. This score determines the EEG duration required for a patient’s residual seizure risk to decay below an acceptable 5% threshold. EEGs were classified into three categories based on risk scores (low seizure risk ‘2HELPS2B’=0, moderate seizure risk ‘2HELPS2B’=1, high seizure risk ‘2HELPS2B’≥2) and followed for the full duration of recording before a residual risk was assigned. Specific patients had multiple EEG studies, but each recording was treated as an independent risk. As the 2HELPS2B protocol requires a 1-hr screening EEG before risk decay begins, and many of our recordings were under 1 hr in duration, we created one scenario where risks were allowed to decay immediately, and another scenario where risks could only decay after 1 hr. In the second scenario, recordings under 1 hr were assigned a residual risk based on the starting value of the appropriate 2HELPS2B decay curve. We repeated these analyses with seizure positive EEGs by converting each seizure into EEG features which could be used to assign a 2HELPS2B score.

Results: Of the 374 EEGs interpreted from October to December 2019, 57 displayed a seizure, and 164 were under 1hr in duration. 90 recordings were classified as low risk by 2HELPS2B, 83 were moderate risk, and 201 were high risk. In our seizure free EEG sample, the average residual risks with and without the screening hour were 15.08% and 13.24% respectively. Once seizure recordings were included, the residual risks with and without screening were 16.25% and 14.34% respectively. Among seizure recordings only, the residual risks with and without screening were 22.78% and 20.44% respectively.

Conclusions: These analyses sampled from three months of our EEG database suggest that the overall residual seizure risk for a non-cEEG model of practice at our site surpasses the 5% benchmark set by the 2HELPS2B validation study. Our findings suggest that we are failing to detect 13%-16% (around 3 times the acceptable benchmark) of subclinical seizures among inpatients. Future work includes examining EEG samples from other centres and determining the clinical impact of missed seizures. These results warrant further investigation into optimal EEG recording protocols in resource scarce healthcare systems and the determination of an acceptable residual seizure risk.

Funding: University of Manitoba BScMED medical student summer research program