Predicting Effects of Brain Stimulation from Observational Data

Poster Session B - Sunday, April 14, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Riley DeHaan¹, David Halpern¹; ¹University of Pennsylvania

Improving memory through neuroscientific interventions requires understanding of the neural activity that causes behavior. While decades of work have shown associations between specific neural activity and subsequent memory performance, it remains unclear whether these conclusions from observational data reflect causes of successful memory encoding. In this project we analyze a dataset of intracranial electroencephalography recordings of 140 subjects performing a delayed free recall task. We account for confounders of causal effects left unaddressed in the majority of prior work on the neural correlates of recall performance, including item and serial position effects. We compare results from these models with more traditional unadjusted models to find that accounting for these variables results in different conclusions about the relevant neural activity. We then validate the inferences of our model using a separate dataset of 20 subjects who received randomized neural stimulation while performing the free recall task. If neural activity better predicts behavioral outcomes after adjusting for behavioral confounders, we hypothesize that those deconfounded model predictions may better reflect an endogenous state of memory performance rather than the features of the presented stimuli. Such a model should in turn better predict the behavioral effects of brain stimulation. We find that electrophysiological activity after controlling for experimental variables predicts cognitive performance significantly better than task features alone. These results suggest that intracranial EEG provides unique information for understanding and modulating cognition that cannot be trivially explained by exogenous factors.

Topic Area: LONG-TERM MEMORY: Episodic