Many people around the world suffer from epilepsy and even though there have been efforts made to help those who have the condition, much is still unknown. Currently, there is no cure for epilepsy, but there are treatments that can help to better control the seizures. However, these treatments are not effective for everyone. In some cases, Sudden Unexpected Death in Epilepsy (SUDEP) occurs. Little is truly known about SUDEP, which often means there is a gap in counseling provided by doctors to their patients with epilepsy.
Fortunately, there may be a potential solution available due to automatic detection of risk factors from electronic medical records. If there is automated risk detection, it is possible to allow the automation to prompt the medical professionals to provide their patients with counseling. A recent study looked into how feasible this would be. Also, it looked at the “generalizability of using regular expressions to identify risk factors in EMRs and barriers to generalizability.”
The data that was used in the research included physician notes for 3,000 patients from a single medical center they termed the home center, as well as 1,000 from five additional medical centers they called away centers. During the review of the patient charts, the researchers were able to identify three SUDEP risk factors – generalized tonic-clonic seizures, refractory epilepsy and epilepsy surgery candidacy.
They manually created regular expressions of risk factors with home training data. The performance was evaluated with both home test and away test data. The researchers evaluated performance by “sensitivity, positive predictive value and F-measure.” They defined generalizability “as an absolute decrease in performance by <0.10 for away versus home test data.” In order to evaluate the underlying barriers to generalizability, they looked at causes of errors that were seen more often in away data than in the home data.
The researchers were able to find high performance in the home test data, while the away test data was low to high. Once they removed the aforementioned three boilerplate phrases, they were able to find improvements to the performance and improved generalizability in nearly all of the measures. Out of 171 errors, the boilerplate phrases that were used accounted for 104 of them, or 61%. Their removal was able to make quite a positive difference.
What does this mean for doctors and patients in terms of SUDEP? It means that regular expressions are a “feasible and probably a generalized method to identify variables related to SUDEP risk.” Therefore, it will be possible to implement the methods that were used in the study as a way to create larger patient cohorts for more research and to “generate electronic prompts for SUDEP counseling.”