Abstracts

Rationale: Infantile spasms (IS) is characterized by epileptic spasms, hypsarrhythmia (including variants thereof), and developmental impairment. There are no established EEG-based predictors of response to treatment for IS, and the use of hypsarrhythmia resolution as a criteria for response is undermined by poor interrater reliability in the identification of hypsarrhythmia. Our previous efforts have identified associations between response to therapy and (1) long-range temporal correlations (LRTCs, derived from detrended fluctuation analysis), (2) a cross-correlation measure of functional connectivity, and (3) Shannon entropy. We set out to replicate these findings in a separate, larger, and more diverse cohort, and ultimately develop an unbiased multi-feature quantitative EEG-based metric of response to therapy for IS. Methods: We identified 50 patients with IS, with (1) baseline overnight EEG, (2) subsequent treatment with a first-line therapy (prednisolone, ACTH, or vigabatrin), and (3) follow-up overnight EEG within 1 month. Thereafter, each pre- and post-treatment EEG was sampled four times (2 awake, 2 sleep) with specific samples selected using a randomization algorithm to mitigate sample selection bias. In a blinded fashion, each EEG was then post-processed to exclude artifact, with subsequent calculation of LRTCs, connectivity, and entropy. We then compared quantitative EEG features of clinical responders (resolution of epileptic spasms without relapse over the next 28 days) with nonresponders, and lastly used multiple logistic regression to (1) evaluate the independent contribution of each EEG feature in the prediction of response to therapy (based on baseline EEG only) and (2) develop a metric to quantify response based on interval EEG change, with adjustment for baseline probability of response. Results: Among the 50 cases, there were 28 (56%) clinical responders and 22 (44%) nonresponders. The sole clinical predictor of response was lead-time to treatment. Baseline hypsarrhythmia was specifically not predictive of response. After adjustment for lead-time, we found statistically significant differences between responders and nonresponders for all quantitative metrics (all p < 0.05). In a multivariate fashion, we found that response is independently predicted by lead-time and a combined metric of baseline connectivity and LRTCs. In a model utilizing post-treatment data and specifically evaluating interval changes in quantitative metrics, response was independently predicted by lead-time and a combined metric of connectivity, LRTCs, and sleep entropy. Lastly, we derive a measure of response ('probability-weighted response index') which quantifies response to therapy (interval EEG change) with adjustment for baseline probability of response (refractoriness). Conclusions: This study demonstrates that clinical response can be predicted by baseline quantitative EEG features, and that response is manifested by predictable changes in these features. However, this method does not perfectly predict response and further refinement and validation is needed. Funding: This study was accomplished with support from UCB Biopharma, the Elsie and Isaac Fogelman Endowment, the Hughes Family Foundation, and the UCLA Children’s Discovery and Innovation Institute.