Abstracts

RATIONALE: To compare the sensitivity and specificity of two methods of identifying children at risk for intractable epilepsy. METHODS: The study included 267 children with epilepsy (2 or more unprovoked seizures, age of onset 1 month - 18 years, treated with at least 1 antiepileptic drug and followed at least 1 year). From retrospective chart review, 73 were identified who were intractable at the last followup visit (? 1 seizure/month for at least 6 months, treatment with at least 2 antiepileptic drugs). RESULTS: Univariate analysis identified 10 variables associated with intractability: younger age of onset, remote symptomatic or progressive etiology, mental retardation (MR), seizure type (infantile spasms and myoclonic, generalized tonic or atonic seizures), abnormal electroencephalography (background abnormalities and epileptiform features), and abnormal neuroimaging. Multiple logistic regression (SPSS for Windows, version 6.1) identified 4 independent variables: age, MR (odds ratio (OR)= 2.98, 95% confidence interval (CI)=1.01,8.83), myoclonus (OR=1.88, CI=1.21,2.91), and generalized tonic seizures (OR=1.75, CI=1.21,2.53). These correctly classified 23/68 intractable and 184/189 non-intractable patients (positive and negative predictive values (PPV and NPV), 82 % and 80%, respectively). A classification tree algorithm (CART for windows, version 3.21) independently selected the same 4 variables, and correctly classified 38/73 intractable and 182/194 nonintractable patients (PPV=76%, NPV=84%). Correctly classified intractable cases include 22 with MR and myoclonus, 14 with MR, neuroimaging abnormalities, and onset before 21 months of age, and 2 with MR, generalized tonic seizures, and onset after 20 months of age. Missed cases included 17 without MR, and 18 with MR (11 without myoclonus or neuroimaging abnormalities and 8 with onset after 20 months of age). CONCLUSIONS: Classification tree algorithms may be useful in identifying children at risk for intractable epilepsy.