Long-range prediction of epileptic seizures with nonlinear dynamics.

Patients with uncontrolled epilepsy have some significant problems with planning life routines, and thus one goal of the present study was to explore the viability of predicting seizures in time intervals of one week. The second goal was to utilize the principle of dynamic diseases and to assess the viability of a cusp catastrophe model for seizure onset that was proposed by Cerf (2006). A seizure history of 124 weeks from one adult male patient fit both the cusp and fold catastrophe models (R2 = .92 and .88 respectively) reasonably well using the pdf method and more accurately than counterpart linear models. Prediction of future states was possible, but somewhat compromised because of the nonstationary nature of the data and uncertainties regarding the control variables in the catastrophe models. Analyses of lag functions, however, revealed some surprising elements, suggesting that the precursory conditions for a seizure could be building up over a period of several weeks and that a self-correcting effect within the nervous system could have been occurring.