Abstracts

Rationale:
Self-reported seizures are often used to adjust medications for outpatients with epilepsy but their accuracy is questionable. Previous studies have shown that self-report can miss many true seizures, raising doubts about their usefulness. However, no study has examined how self-report affects the long-term outcomes of medication management.

Methods:
We used CHOCOLATES, a realistic seizure diary simulator, to generate 100,000 simulated patient diaries lasting 10 years. Each patient had a clinic visit every three months and received medication changes based on a standardized algorithm. We assumed that medications reduced seizures by 20% on average and had typical chances of achieving seizure-freedom. We varied the sensitivity of self-report from 0.1 to 1.0 and fixed the false alarm rate at one seizure per month. We calculated the average monthly seizure burden and medication burden for each patient. We defined signal-to-noise ratio (SNR) as sensitivity/false-alarm-rate.

Results:
The median seizure rate across patients decreased from 1.4 seizures/month (SNR=0.1) to 1.0 seizures/month (SNR=1.0). The median number of drugs increased from 2.4 (SNR=0.1) to 4.1 (SNR=1.0). Thus, SNR values had a small effect on seizure rates but a larger effect on medication loads.

Conclusions:
Self-reported seizures can be useful for medication management in epilepsy, even with low accuracy. The SNR of self-report determines the trade-off between seizure control and medication burden. A device that can detect all seizures would not improve seizure burdens significantly but it might reduce medication use.

Funding: NIH K23NS124656.