ADVERTISEMENT
Systematic Review Finds Problems With Epilepsy Outcome Prediction Models
Treatment outcome prediction models for patients with newly diagnosed epilepsy have a universally high risk of bias, researchers reported in a systematic review of nearly 50 such models published in the journal Epilepsia.
“A subset of the included studies also exhibited applicability concerns, indicative of a lack of consistency between study objectives and methods,” wrote corresponding author Corey Ratcliffe, a PhD student at the University of Liverpool, Liverpool, England, and study coauthors. “An adverse effect of low applicability is heterogeneity of study characteristics.”
>>QUIZ: Oxygen Saturation and Late-Onset Epilepsy
The systematic review included 32 studies with a total 48 multivariable treatment prediction models. Researchers had hoped to identify factors that help predict outcomes in patients with newly diagnosed epilepsy, but because of the heterogeneity of predictors and outcome selection across the studies, systematic comparison proved difficult.
Broadly, study outcomes fell into 4 categories: drug resistance, short-term treatment response, seizure remission, and mortality, researchers reported. Predictor categories most commonly linked with seizure remission were seizure characteristics and type, epilepsy history, and age at onset.
“Factors related to comorbidities, demographics, diagnosis, electroencephalography, neuroimaging, and neuropsychology were reported as significantly related to seizure remission either once or twice, whereas antiseizure medication-related factors were not,” the authors wrote. “Bias and applicability concern levels were largely influenced by several studies including response to treatment as a baseline variable.”
The use of antiseizure medication response as a baseline variable potentially obscured other baseline factor relationships in many studies, according to the review. The authors also reported an underrepresentation of electrophysiological and magnetic resonance imaging findings in multivariable prediction models.
“To improve the overall quality of prediction model development in newly diagnosed epilepsy, prospective authors are advised to adhere to TRIPOD [Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis] guidelines,” researchers wrote, “and to avoid including response to treatment as a baseline variable.”
Reference
