Development and Validation of Prediction Models for Developmental and Intellectual Outcome Following Pediatric Epilepsy Surgery

BACKGROUND AND OBJECTIVES: To identify predictors of postoperative intelligence and developmental quotients (IQ/DQ) and develop and validate clinically applicable IQ/DQ prediction models. METHODS: We retrospectively analyzed neuropsychological outcomes and their possible determinants for children treated in Bethel and Utrecht since 1990. We performed separate analyses for patients with IQ and those with only DQ available. We developed prediction models based on presurgical determinants to predict dichotomized levels of performance (IQ ≥85, IQ ≥70, DQ ≥50). RESULTS: IQ/DQ data before and 2 years after surgery were available for 492 patients (IQ n = 365, DQ n = 127). At a cutoff level ±10 points, the chance of improvement was considerably higher than the chance of deterioration (IQ 37.3% vs 6.6% and DQ 31.5% vs 15.0%, respectively). Presurgical IQ/DQ was the strongest predictor of postoperative cognition (IQ r = 0.85, p <0.001; DQ r = 0.57, p <0.001). Two IQ models were developed in the Bethel cohort (n = 258) and externally validated in the Utrecht cohort (n = 102). For DQ, we developed the model in the Bethel cohort and used 10-fold cross-validation. Models allowed good prediction at all 3 cutoff levels (correct classification for IQ ≥85 = 86%, IQ ≥70 = 91%, DQ ≥50 = 76%). External validation of the IQ models showed high accuracy (IQ ≥85: 0.82, confidence interval [CI] 0.75-0.91; IQ ≥70: 0.84, CI 0.77-0.92) and excellent discrimination (receiver operating characteristic curves: IQ ≥85: area under the curve [AUC] 0.90, CI 0.84-0.96; IQ ≥70: AUC 0.92, CI 0.87-0.97). DISCUSSION: After epilepsy surgery in children, the risk of cognitive deterioration is very low. Presurgical development has a strong effect on the postoperative trajectory. The presented models can improve presurgical counseling of patients and parents by reliably predicting cognitive outcomes. CLASSIFICATION OF EVIDENCE: This study provides Class II evidence that for children undergoing epilepsy surgery presurgical IQ/DQ was the strongest predictor of postoperative cognition.