Comparing normative and ipsative methods of IC-CoDE phenotype classification in frontal lobe epilepsy

To compare the relative utility of normative (i.e., measuring cognitive performance relative to peers) and ipsative (i.e., measuring cognitive performance relative to the patient's own baseline functioning) approaches for classifying cognitive phenotypes in adults with frontal lobe epilepsy (FLE) using the International Classification of Cognitive Disorders in Epilepsy (IC-CoDE).

One hundred adults with FLE (89% drug-resistant, 12% with premorbid IQ ≤ 70) who completed neuropsychological evaluations were classified into IC-CoDE phenotypes (intact, single-domain impairment, bi-domain impairment, generalized impairment) using two approaches: (1) a normative method, with impairment defined relative to normative means, and (2) an ipsative method, with impairment defined relative to premorbid functioning estimated from word-reading ability. Agreement between approaches was quantified using quadratic weighted kappa. Descriptive statistics and exploratory binary logistic regressions were used to compare associations between demographic/clinical variables and cognitive phenotypes (intact vs. any impairment) across the two classification methods.

Cognitive phenotype distributions differed by method, with the ipsative method classifying more patients as intact. Agreement between approaches was poor (weighted κ = 0.085). In exploratory regression analyses, greater anti-seizure medication burden (OR = 1.62; p = 0.042), younger age (OR = 0.96; p = 0.016), female sex (OR = 0.20; p < 0.001), and lower education (OR = 0.80; p = 0.029) were associated with greater odds of impairment when using the normative approach. In contrast, greater years of education was associated with impairment when using the ipsative approach (OR = 1.27; p = 0.016).

Normative IC-CoDE classifications in FLE appear more sensitive to clinical/demographic risk factors and to impairment among individuals with lower premorbid functioning, whereas ipsative approaches may reduce misclassification in highly educated individuals. These findings highlight the need to carefully consider method selection for deriving cognitive phenotypes given the specific research question, clinical scenario, or population of interest.