MULTIVARIABLE PREDICTION OF SEIZURE OUTCOME ONE-YEAR AFTER RESECTIVE EPILEPSY SURGERY - DEVELOPMENT OF A MODEL WITH INDEPENDENT VALIDATION

Purpose: To identify predictors of seizure-outcome after epilepsy surg ery and validate the findings in an independent series of patients. To use the results to develop a predictive model. Methods: Sequential pa tients undergoing resective surgery for medically intractable epilepsy were identified at Yale New Haven Hospital (1987-1990, group 1) and C olumbia Presbyterian Hospital (1991-1994, group 2). Information about seizure outcome and predictors of outcome was obtained from medical ch art review. Good seizure-outcome was defined as having been seizure-fr ee for one year beginning with discharge from the hospital. Multiple l ogistic regression was used to develop a model of predictors in group 1. It was then validated in group 2. Results: There were 133 patients in group 1 and 81 in group 2. In a multivariable analysis, independent predictors of outcome in group 1 were presence of mesial temporal scl erosis based on postsurgical pathological analysis (MTS) (relative ris k (RR) = 1.47), having a known underlying etiology (RR = 1.32), and pa rtial seizures only (RR = 1.17). In group 2, the findings for each fac tor were similar to those in group 1: MTS, RR = 1.49; etiology, RR = 1 .32; and partial seizures, RR = 1.24. Used in combination, these three factors can identify patients with nearly a 100% chance of being seiz ure-free (all three factors present) versus less than a 50% chance (no ne of the three factors present). Conclusions: With independent valida tion of the findings, we can be reasonably certain that the three fact ors identified in this analysis are meaningful and generalizable predi ctors of seizure outcome following epilepsy surgery. Use of predictive models should be considered in future studies to convert study result s into clinically relevant statements about a particular patient's lik elihood of surgical success. (C) 1998 Elsevier Science B.V. All rights reserved.