Abstracts

Rationale: One of the major contributions of neuropsychological assessment in the preoperative setting is to assist in the lateralization of the seizure focus. The Boston Naming Test (BNT) is sensitive in respect to the side of seizure focus with the worst scores indicating seizure origin in the speech dominant temporal lobe. Busch et al. (Epilepsia 2009; 34(6):1270-1273) provided a method that may help to lateralize seizure focus using the Boston Naming Test while controlling for duration of epilepsy, onset, and intelligence. This study aims to test the clinical utility of this method in predicting the ultimate side of surgery using a sample of 34 patients with temporal lobe epilepsy (TLE). Methods: Patients (n=34; left TLE=18; right TLE=16) ranged in age from 18-59 (M=39.12; SD=12.31) and in education from 8-17 years. The study group was equivalent to the Busch et al., (2009) combined group in terms of intelligence, age at onset, duration of epilepsy, and BNT score. The method provided by Busch et al., in addition to alternative prediction models based on logistic regressions, was used to predict ultimate side of surgery. Results: Applying the equation introduced by Busch et al., it was possible to correctly predict the side of surgery in 61.8% of patients (Chi-Square=3.031; p=.087). A more parsimonious model using BNT as the sole predictor correctly predicted 73.5% of patients (Chi-Square=8.476; p=.004). Conclusions: The usage of the regression equation provided by Busch et al. resulted in an unsatisfactory prediction of the ultimate side of surgery in patients with medically intractable TLE. Improved prediction was possible after specifying model coefficients based on our own sample. However, a less complex model with the Boston Naming Test as the only predictor resulted in an equally good prediction. The failure of the proposed prediction method may have resulted from overfitting, caused by an excessively complex model. According to the principle of parsimony models with the fewest number of parameters are most preferable.