Present affiliation of Dr. Armon: Department of Neurology, Loma Linda University Medical School and Medical Center, and Jerry L. Pettis Memorial Department of Veterans'Affairs Medical Center, Loma Linda, California, U.S.A. Present affiliation of Dr. Dawson: Department of Epidemiology and Biostatistics, Case Western Reserve School of Medicine, Cleveland, Ohio, U.S.A.
Predictors of Outcome of Epilepsy Surgery: Multivariate Analysis with Validation
Article first published online: 3 AUG 2005
Volume 37, Issue 9, pages 814–821, September 1996
How to Cite
Armon, C., Radtke, R. A., Friedman, A. H. and Dawson, D. V. (1996), Predictors of Outcome of Epilepsy Surgery: Multivariate Analysis with Validation. Epilepsia, 37: 814–821. doi: 10.1111/j.1528-1157.1996.tb00033.x
Presented in part at the 46th Annual Meeting of the American Academy of Neurology, Washington, DC, May 1–7, 1994.
Based on a thesis submitted (by C.A.) in partial fulfillment of the requirements for the degree of Master of Health Sciences in Biometry, Duke University, Durham, North Carolina.
- Issue published online: 3 AUG 2005
- Article first published online: 3 AUG 2005
- Received October 26, 1995; revision accepted May 22, 1996.
- Logistic regression;
- Confirmatory and exploratory analysis;
- Bootstrapping techniques;
- Imaging and EEG localization
Summary: Purpose: To identify predictors of outcome of epilepsy surgery, using the Duke experience, applying multivariate analysis and validation techniques. To compare the results of different modeling algorithms. Few previous studies have reported multivariate analysis, or validated their results.
Methods: Records of 116 patients with focal resections for intractable epilepsy from January 1, 1980 through June 30, 1989 were analyzed. Primary outcome variable was patient's condition in second postoperative year: seizure free (except auras), or not. Three predictors of biologic interest were specified a priori for confirmatory analysis. Additional predictors were considered within exploratory analysis. Logistic regression techniques were applied to assess relations with pre-and postoperative predictors. Internal validity was assessed by repeated random selection of training and validation samples, used in conjunction with bootstrap techniques.
Results: By using multivariate analysis, percentage of epileptic EEG activity arising from the site of resection and either imaging localization or lack of use of invasive monitoring were the only statistically significant preoperative predictors for good outcome at 2 years. Presence of seizures'within 2. months of surgery was a significant postoperative predictor for a poor outcome. Adding more variables did not result in significantly improved models. Use of validation techniques reduced the degree of optimism in the predictive value of the models.
Conclusions: Pooling of data from multiple institutions is needed to attain the large sample sizes needed for multivariate analysis with validation.