Abstracts

This abstract has been invited to present during the Pediatric Epilepsy Highlights platform session

Rationale: Identification of infants at risk of cerebral palsy is essential to implement early interventions to optimize outcomes. Sleep spindles, prominent bursts of 9-16 Hz oscillations during non-rapid eye movement (NREM) sleep, are typically present by 6 weeks of age and reflect thalamocortical circuit function. We hypothesized that abnormal early Rolandic sleep spindle activity would predict long-term contralateral cerebral palsy and improve prediction after controlling for clinical exam and MRI findings.

Methods: EEGs capturing sleep from all at-risk infants with a history of neonatal seizures between 1/2011-1/2017 (n=35; 13F; 1.0-7.3 months) and all available age-matched control infants who had normal neurodevelopment, non-epileptic events, and normal EEG between 2/2002-4/2021 (n=127; 63F; 1.0-7.4 months) were included. EEGs were manually reviewed for NREM sleep. To quantify spindles, we validated an automated infant spindle detector using a latent state model that measured sigma power, theta power, and oscillation regularity. To do so, two reviewers hand-marked 3362 sleep spindles from EEGs in 47 healthy infants ages 0-24 months, by consensus. The detector was then trained and validated using leave-one-out cross-validation. Spindle features were computed in the C3 and C4 channels. Clinical exam was obtained from all available neurology notes at the time closest to the EEG recording (1.6-7.4 months) and any noted abnormality included. Available neonatal MRIs in at-risk infants (n=32; median age 3.5 days, IQR 3 days) were evaluated for any abnormalities in the thalamus, basal ganglia, or posterior limb of the internal capsule (PLIC), or lactate peaks. Long-term unilateral motor outcome was obtained from chart review (median follow-up 4.5 years, IQR 3 years). Any noted weakness or altered tone in the upper or lower extremities was included as cerebral palsy. Differences in spindle features between groups were evaluated using mixed-effects linear regression models. Clinical exam, individual MRI findings, and any abnormal MRI finding were each tested as predictors of contralateral cerebral palsy among at-risk infants using mixed-effects logistic regression models. Significant features were then tested together.

Results: The infant spindle detector had excellent performance compared to manual markings (F1=0.50). Spindle rate, duration, and percentage were lower in hemispheres with subsequent cerebral palsy (p≤0.02, all tests; Figure 1). Clinical exam at the time of EEG (p=0.6) and individual MRI abnormalities did not predict contralateral cerebral outcome (PLIC sign: p=0.3; thalamus: p=0.8; basal ganglia: p=0.5; lactate peak in thalamus or basal ganglia: p=0.3). The presence of any MRI abnormality did predict contralateral motor outcome (p=0.02). After controlling for the presence of any MRI abnormalities, spindle rate, duration, and percentage remained significant predictors of contralateral cerebral palsy (p=0.001, all tests).

Conclusions: Abnormal Rolandic spindle activity reflects disrupted thalamocortical motor circuit function and provides an early prognostic indicator of cerebral palsy in at-risk infants.

Funding: NIH NINDS R01NS115868