Photo: Hero Images via Getty Images
Researchers from Seoul National University Hospital have designed a machine learning model that predicts an individual's likely response to an anticonvulsant drug, aiming to reduce trials and errors in finding the right treatment.
WHAT IT'S ABOUT
Based on a media release, the AI model was trained on clinical data from a cohort of approximately 2,600 epilepsy patients treated at SNUH from 2008 – when the current epilepsy classification system was introduced – to 2017.
Data included antiepileptic drug use patterns, seizure types, brain MRI, EEG, blood tests, and treatment progress. Major antiepileptic drugs with high prescription frequency were analysed, including levetiracetam (LEV), oxcarbazepine (OXC), valproic acid (VPA), and lamotrigine (LMT).
A total of 84 variables were considered, and a tree-based ensemble analysis technique was utilised to design the machine learning model to predict antiepileptic drug treatment response. The team defined treatment response as a 50% or greater reduction in seizure frequency after drug use.
FINDINGS
Findings published in Scientific Reports showed that VPA, LMT, and OXC demonstrated the strongest predictive performance, respectively, among major anticonvulsant monotherapies. Among combination therapies, the CBM-LEV regimen showed the highest predictive performance.
Researchers also noted that patients with generalised seizures were more likely to respond to VPA, while older patients and those with a shorter disease duration showed a higher likelihood of response to LMT.
WHY IT MATTERS
According to SNUH researchers, identifying the most effective treatment for epilepsy, a neurological condition characterised by recurrent, unprovoked seizures, remains challenging because patients respond differently to anti-seizure medications. They noted that around three in ten patients are considered drug-resistant, failing to respond to two or more appropriately chosen medications.
More than 20 antiepileptic drugs are currently used in clinical practice, but repeated trial-and-error treatment carries risks for patients. The research team underscored the need for tools that can predict treatment response early in the course of epilepsy.
The SNUH researchers now plan to train their AI model with more data from other institutions, aiming to develop it into a clinical decision support tool.
THE LARGER TREND
An international study led by Monash University in Australia was possibly the first in the world to demonstrate the use of an AI model to predict the optimal anti-seizure medication for patients newly diagnosed with epilepsy in 2022. It used clinical information from around 1,800 patients in five health care centres in Australia, Malaysia, China and the United Kingdom, with the deep learning prediction model designed by the Monash Medical AI Group and trained using Monash's MASSIVE computing facility.
Meanwhile, AI has been utilised in research in Australia, India, and South Korea to detect the epileptogenic zone in the brain and for seizure monitoring.
ON THE RECORD
"The selection of antiepileptic drugs for epilepsy patients has traditionally relied on the empirical judgment of specialists, but this study presents a systematic approach to predicting antiepileptic drug treatment responses based on a large cohort and various clinical data accumulated over a long period of time," Dr Park Kyung-il, research co-lead and professor at the Department of Neurology of SNUH, said in a statement.
