Project Details - Ended
- Grant Number:R21 HS024977
- Funding Mechanism:
- AHRQ Funded Amount:$296,567
- Principal Investigator:
- Project Dates:9/30/2016 to 9/29/2018
- Care Setting:
- Medical Condition:
- Type of Care:
- Health Care Theme:
Epilepsy is one of the leading neurological disorders in the United States, affecting more than 479,000 children and over 2 million adults. Approximately 30 percent of those with epilepsy are poorly controlled with medications and are potential candidates for neurosurgical intervention. In general, providers are cautious about referring individuals for surgery due to its risks, the difficulty in distinguishing those in which surgery is indicated or contraindicated, and changing indications for surgery. The early identification of surgical candidates could significantly improve the quality of life for both those with epilepsy and their caregivers.
This research prospectively evaluated a machine learning algorithm that identified surgical candidates for epilepsy surgery, utilizing data from an electronic health record (EHR) and natural language processing (NLP) of written notes.
The specific aims of the research were as follows:
- Implement and prospectively evaluate the existing NLP system by integrating the system with the EHR for patients identified as potential surgical candidates.
- Perform a clinical pilot test to evaluate the effectiveness of electronic alerts, reminders, and no intervention (i.e., standard of care) for eligible patients.
The NLP algorithm identified patients with upcoming appointments as possible candidates for surgery. A sample of these underwent a chart review by two epileptologists to evaluate the accuracy of the algorithm. Patients were randomized such that their neurologists were notified about the possible need for surgical consult via an ‘honest broker’ email notification, an EHR alert, or no notification at all. Performance of the algorithm was determined by the number of missed surgical candidates it identified.
Over 12 months, the algorithm identified 200 of 6,395 patients with epilepsy as potential surgical candidates. Within the subsequent year, 6 percent were referred for surgical evaluation and 1 percent underwent surgery. The epileptologists indicated that 42 of the 200 found by the algorithm would not have otherwise been identified as surgical candidates, an increase in 43 percent.
The research showed that an EHR-integrated machine learning algorithm can aid clinicians in identifying patients with epilepsy who could benefit from surgery. This has the potential to decrease time from diagnosis to surgical evaluation, with a resulting improvement in quality of life for patients and caregivers, reduction in suffering, and a decrease in treatment cost.