Accuracy of clinical neuropsychological versus statistical prediction in diagnostic classification of seizure

Neuropsychological differences among various seizure groups (e.g., left [LTLE] and right temporal lobe epilepsy [RTLE], frontal lobe epilepsy [FLE], psychogenic non-epileptic seizures [PNES]) have been widely reported. However, only two studies have evaluated the ability of the clinical neuropsychologist to accurately predict LTLE or RTLE from the results of neuropsychological assessment. Prior research has also shown that accuracy of clinical neuropsychological prediction is equal or superior to statistical methods of prediction of neurologic status. The purpose of this study was to evaluate the accuracy of clinical neuropsychological versus statistical prediction in three seizure classification scenarios thought to be representative of actual clinical practice (organicity of seizures, laterality of epilepsy, and differentiation among high base rate seizure diagnoses). A final sample of 352 individuals with confirmed seizures was administered a battery of neuropsychological tests. A clinical neuropsychologist formulated diagnostic impressions of seizure disorders from the results of these tests, blinded to both the confirmed seizure diagnosis and medical history. Clinical accuracy was determined by comparing the neurologically confirmed seizure diagnoses to the classifications made by the neuropsychologist. Statistical accuracy was determined by discriminant function analyses that incorporated the same neuropsychological and demographic variables available to the clinician. When differentiating between organic and non-organic seizures, hit rates for clinical and statistical prediction were respectively 84% and 71%. When differentiating among left, right, bilateral hemisphere and mixed epilepsies hit rates for clinical and statistical prediction were respectively 69% and 43%. When differentiating among LTLE, RTLE, FLE, PNES, and mixed epilepsies, hit rates for clinical and statistical prediction were respectively 76% and 52%. Accuracy of clinical methods were 38%, 41%, and 49% and statistical methods were 24%, 15%, and 24% better than chance prediction in the organicity, laterality, and most common seizure diagnoses scenarios, respectively. In all seizure classification scenarios, clinical neuropsychological prediction was more accurate than statistical prediction, while both methods outperformed chance classification. Results highlight the importance and validity of clinical neuropsychological judgment in the seizure treatment setting. Statistical prediction may be more useful when the diagnostic question is simple and future research should explore more sophisticated methods of statistical classification.