Objective: Elderly onset epilepsy represents a distinct subpopulation that has received considerable attention due to the unique features of the disease in this age group. Research into this particular patient group has been limited by a lack of a standardized definition and understanding of the attributes associated with elderly onset epilepsy.
Methods: We used a prospective cohort database to examine differences in patients stratified according to age of onset. Linear support vector machine learning incorporating all significant variables was used to predict age of onset according to prespecified thresholds. Sensitivity and specificity were calculated and plotted in receiver-operating characteristic (ROC) space. Feature coefficients achieving an absolute value of 0.25 or greater were graphed by age of onset to define how they vary with time.
Results: We identified 2,449 patients, of whom 149 (6%) had an age of seizure onset of 65 or older. Fourteen clinical variables had an absolute predictive value of at least 0.25 at some point over the age of epilepsy-onset spectrum. Area under the curve in ROC space was maximized between ages of onset of 65 and 70. Features identified through machine learning were frequently threshold specific and were similar, but not identical, to those revealed through simple univariable and multivariable comparisons.
Significance: This study provides an empirical, clinically informed definition of "elderly onset epilepsy." If validated, an age threshold of 65-70 years can be used for future studies of elderly onset epilepsy and permits targeted interventions according to the patient's age of onset.
Keywords: Elderly; Epidemiology; Epilepsy; Machine learning.
Wiley Periodicals, Inc. © 2015 International League Against Epilepsy.