HOUSTON — Researchers have developed and validated a model to determine individual variability in seizure frequency, which could go a long way toward streamlining clinical trials and improving daily management of patients with drug-resistant epilepsy.
As it stands now in a drug study, researchers compare a drug's effectiveness against a placebo during a set period. If study patients get more than 50% better during the trial, they're considered to be responders.
But this doesn't consider that seizure frequency of some patients with epilepsy may already vary up to 50%, explained Daniel Goldenholz, MD, PhD, clinical fellow at the National Institutes of Health. Dr Goldenholz, who is also an engineer, developed the model.
He used the example of a patient who recently had 10 seizures a month but this month had 9. "Is that an improvement for him or is that the usual affairs? If that person can vary between 5 and 15, then 9 is no different."
The research was presented here at the American Epilepsy Society (AES) 2016 Annual Meeting.
Dr Goldenholz and colleagues studied three separate seizure frequency data sets: patient-reported data (SeizureTracker.com [ST]); physician-curated, patient-reported data (Human Epilepsy Project [HEP]); and data from a chronically implanted intracranial device (NeuroVista [NV]).
They fit a linear model to capture the relationship of log-transformed mean monthly seizure counts to log-transformed standard deviation of monthly seizure counts.
To test the ability of the model to predict variability prospectively, they segmented each dataset into trial-sized blocks of 6 months (2 for baseline, 1 for titration, 3 for test). Patients contributed as many blocks as available in their diaries. Two methods of predicting the variability were computed: the "linear" model and the "fixed" model.
After criteria for inclusion of complete data were applied, ST had 3016 patients, HEP had 107, and NV had 15.
Individual Predictions
The model was able to predict an individual patient's overall standard deviation by using overall mean, with high accuracy across each data set. The sequential predictions were correct 79% to 100% for the linear model and 42% to 80% for the fixed model, across data sets. In the ST validation set (n = 1820), the predictions were correct 94% for the linear model and 77% for the fixed model.
"We found using three different data sets, that you can actually predict how much variability someone is going to have, up and down, in their seizure frequencies," said Dr Goldenholz.
"That way, when I take a drug, if I have increased or decreased more than expected, then I am actually responding to the drug," he added.
"We can make a prediction that, whatever you do with this patient, they will go up and down a certain amount and if they are outside of that, something has changed."
This means that clinical trials can be carried out at a faster rate and require fewer patients, said Dr Goldenholz. "So from a clinical trial point of view, this is a very big deal; we can save money and accelerate drug discovery."
The model could also be used in clinical practice. If patients report having more seizures, the doctor has to decide whether this is actually outside of their normal pattern. "But they have no idea how to do that; there's no equation."
That's where this new model would come in.
"We can tell the doctor; if the patient normally has 10 seizures and now has 12, don't worry; that's within the expected range; but if they now have 20, for example, then they should be more concerned because that's way outside what is expected."
Every patient would get his or her own prediction. The information could easily be built into an app or posted on a website.
But that's down the road. This new research is "proving the concept at this point," said Dr Goldenholz.
"Terrific" Idea
Commenting on the research for Medscape Medical News, Michael R. Sperling, MD, Baldwin Keyes Professor of Neurology, and director, Comprehensive Epilepsy Center, Thomas Jefferson University, Philadelphia, Pennsylvania, said the idea is "terrific" and could really be effective.
"There's a natural variability in the rate that seizures occur, and if we can account for some of this, we can really design our trials much more intelligently, require fewer patients, and do it faster."
But Dr Sperling stressed that determining individual variation requires complicated mathematics. "The trick is going to be translating it into something that can be used relatively easily."
Although he understands the plan is to "take this forward" and develop more effective ways of treating patients with epilepsy, "this is very much a work in progress," said Dr Sperling.
American Epilepsy Society (AES) 2016 Annual Meeting. Poster 1.072. Presented December 3, 2016.
For more Medscape Neurology news, join us on Facebook and Twitter
Medscape Medical News © 2016 WebMD, LLC
Send comments and news tips to news@medscape.net.
Cite this: New Model Determines Individual Seizure Variability - Medscape - Dec 06, 2016.
Comments