Earlier Identification of Epilepsy Surgery Candidates Using Natural Language Processing

Abstract

This research analyzed the clinical notes of epilepsy patients using techniques from corpus linguistics and machine learning and predicted which patients are candidates for neurosurgery, i.e. have intractable epilepsy, and which are not. Information-theoretic and machine learning techniques are used to determine whether and how sets of clinic notes from patients with intractable and nonintractable epilepsy are different. The results show that it is possible to predict from an early stage of treatment which patients will fall into one of these two categories based only on text data. These results have broad implications for developing clinical decision support systems.