Neurosurgery is one of the most effective treatments for the ~30% of epilepsy patients who do not respond to medication. However, it is underutilized: it can take 10 years for patients to be referred to an epilepsy surgery program. Our lab researches methods for automatically detecting patients that are potential candidates for surgery. The goal is to increase surgery referrals and decrease the average lag time to referral for those patients that may benefit from the treatment.

Much of the information relevant to surgery candidacy can be found in clinician free-text notes in the electronic health record. Using notes from known candidates as a gold standard, we train natural language processing and machine learning software to find complex patterns of words and phrases in the clinician's notes that may be predictors of surgery candidacy. For example, there may be language indicating that the patient has focal epilepsy, has failed multiple medication trials, and has poor quality of life. Our retrospective studies analyzing past surgery patients show that these algorithms may be effective at identifying candidates up to one year before they were actually referred for surgery evaluation.

The Pestian lab is currently putting a system into production at Cincinnati Children's that uses the electronic health record to automatically identify potential surgery candidates from the population of epilepsy patients seen at Cincinnati Children's. The system alerts the physician that the patient shows similarities to known surgery candidates, and that a review of the patient's chart may be in order. Using feedback from the physician, the system then continues to learn and improve its detection algorithm.