A partial least squares analysis of seizure outcomes following resective surgery for tuberous sclerosis complex in children with intractable epilepsy.

INTRODUCTION: Previous studies aimed at identifying predictors of seizure outcomes following resective surgery for tuberous sclerosis complex (TSC) are limited by multicollinearity among predictors, whereby the high degree of correlation between covariates precludes detection of potentially significant findings.
PURPOSE: Here, we apply a data-driven method, partial least squares (PLS) to model multidimensional variance and study significant patterns in data that are associated with seizure outcomes.
METHODS: Post hoc analysis of 186 children with TSC who underwent resective epilepsy surgery derived from an individual participant data meta-analysis was performed. PLS was used to derive a latent variable (component) that relates clinical covariates with Engel classification. Permutation testing was performed to evaluate the significance of the component, and bootstrapping was used to identify significant contributors to the component.
RESULTS: A significant component was identified, which represents the pattern of covariates related to Engel class. The strongest and significant factors contributing to this component were focal ictal electroencephalogram and concordance of electroencephalography (EEG)-magnetic resonance imaging (MRI) abnormality. Interestingly, covariates contributing the least to the seizure-free patient phenotype were continent of treatment and age at the time of surgery.
CONCLUSION: Using a data-driven, multivariate method, PLS, we describe patient phenotypes that are associated with seizure-freedom following resective surgery for TSC.