By O Devinsky, C Dilley, M Ozery-Flato, R Aharonov, Y Goldschmidt, M Rosen-Zvi, C Clark, P Fritz. Epilepsy & Behavior 56(2016):32–37.


There is little data available to help health care providers decide on which seizure medication (also called AED) is best for an individual person. This study used data from insurance claims databases to develop a computer model to help choose appropriate treatments for individuals with epilepsy.

Description of Study

  • Data from a large private medical claims database were used.
  • First data from 40,000 patients were used to develop the computer model. Then data from 10,000 patients were randomly selected for testing the model.
  • A change in seizure medicines (either starting a new one, adding one, or switching medicines) was used in the prediction model.
  • The goal of the computer model was to pick an AED or seizure medication that had the lowest chance of being changed later (called the model-predicted AED regimen).
  • The researchers then used the model to look at the use of health care services (such as hospitalizations or physician visits).

Summary of Study Findings

  • The computer model had good predictive power.
  • Patients who were given the model-predicted seizure medications had a significantly longer time until a change in treatment was needed. Their use of health care services was lower than expected compared with those who received another treatment.
  • The actual seizure medications prescribed matched the ones predicted by the model only 13% of the time. There were large differences in how often certain AEDs or combinations of seizure medications were given.

What does this mean?

  • This study suggests that computer models might be able to help health care providers choose a seizure medication with a greater chance for good outcomes. This is one example of “personalized medicine.”
  • Seizure frequency data was not used in this study as an outcome. Only computer modeling was used. Also, the computer model was not tested in actual patients. Future research might use this model in a clinical trial to determine if the model works in everyday care of people with epilepsy.
  • With widespread use of computer-based medical records, this model could theoretically be part of an electronic medical record to suggest a seizure medication for an individual patient.

Article published in Epilepsy & Behavior, March 2016

Authored By: 
David M. Ficker MD, FAAN
Reviewed By: 
Nathan B. Fountain MD