Data Scientist Seer Medical Inc Melbourne, Australia
Rationale: Video-encephalography (vEEG) is the clinical standard for diagnosing epilepsy and is also an important component in surgical planning. Nevertheless, inpatient vEEG monitoring fails to capture seizures in up to one-third of patients across both diagnostic and pre-surgical monitoring (Ghougassian et al 2004). We hypothesized that personalized seizure forecasts could be used to optimise the timing of diagnostic or pre-surgical monitoring with the potential to improve the number of seizures captured during testing. We have recently shown that accurate forecasts can be developed from self-reported seizure diaries by tracking individual cycles of seizure likelihood (Karoly et al 2020). In this study, we conducted a retrospective analysis to assess whether diary-based forecasts correspond to the yield of diagnostic vEEG monitoring. Methods: We used a database of ambulatory vEEG studies to select a cohort with linked mobile seizure diaries of more than 20 reported seizures over at least eight weeks. The total cohort included 50 participants. vEEG monitoring duration ranged from one to ten days (mean 6.4 days). Diary duration ranged from eight weeks to two years (range of 21 to 754 seizures). Event detection during EEG monitoring was undertaken by neurophysiologist review and confirmed by a neurologist. Diary seizure times were used to detect an individual’s multiday cycle/s and forecast times of high seizure risk. vEEG monitoring outcomes were assessed retrospectively, projecting forecasts to determine what proportion of monitoring was high-risk. We then compared whether seizure risk forecasts could differentiate between participants with and without epileptic events captured during monitoring (‘seizure’ vs ‘no-seizure’ groups). Results: In total, 20 out of 50 people (40%) recorded epileptic activity during EEG monitoring. In this seizure group, vEEG monitoring windows showed significantly more time in high risk compared to the no-seizure group (Table 1). This trend was consistent across all cycle strengths, although was only significant (p < 0.05 using a one-sided t-test) when considering people with stronger cycles – where forecasts are expected to be more powerful.Individuals spend different amounts of time in high risk, so it is important to compare risk during vEEG monitoring to each person’s baseline risk. In the seizure group, risk during EEG monitoring was higher than baseline for 70% of people, compared to just 12% of people in the no-seizure group, with a similar trend observed across all cycle strengths. Possible confounding factors: vEEG duration, seizure frequency, or forecast accuracy were not significantly different between the seizure and no-seizure groups at any cycle strength (p > 0.05 using a two-sided t-test). Conclusions: People with seizures during monitoring showed significantly higher risk forecasts compared to people who did not have a seizure. These results show that forecasts developed from mobile seizure diaries correspond to the likelihood of recording an electrographic seizure during ambulatory EEG monitoring. This study provides a proof of principle that scheduling monitoring times based on personalized seizure forecasts can improve the yield of EEG monitoring. Importantly, forecasts can be developed at low cost from mobile seizure diaries. Even a small increase in EEG yield could provide meaningful time and cost savings for patients and providers. Funding: Please list any funding that was received in support of this abstract.: This project received funding from the Australian Government National Health and Medical Research Council (NHMRC Investigator Grant 1178220). This project was also supported by the Epilepsy Foundation of America’s Epilepsy Innovation Institute “My Seizure Gauge” grant. Click here to view image/table
