Early clinical prediction of neurodevelopmental outcome in KCNQ2-related disorders

<jats:p>Objective: In KCNQ2-related disorders (KCNQ2-RD), neurodevelopmental outcome remains variable despite established genotype-phenotype correlations. Our aim is to improve counselling, by developing and internally validating models predicting neurodevelopmental outcomes based on early clinical and genetic features, universally available to clinicians. Methods: We conducted a multicentric retrospective cohort study including 277 individuals carrying a (likely) pathogenic variant in the KCNQ2 gene, with a minimum follow-up age of three years. Mosaic variants were excluded. The cohort was randomly split into training (70%) and validation (30%) sets. Ten expert selected parameters with minimal missing data were used to train random forest models to predict (i) dichotomous outcomes and (ii) three-category outcomes for cognition, language, and gross motor milestones. Results: Models incorporated seven clinical (neonatal hypotonia, EEG characteristics, age at seizure onset, seizure type, and seizure frequency at onset, prematurity, and sex) and three genetic variables (de novo status, exon localisation, and position within known KCNQ2-developmental and epileptic encephalopathy (DEE) hotspot regions). Dichotomous models showed the highest predictive performance, with accuracies of 0.83 for normal vs. mild-profound intellectual disability (ID), 0.83 for achievement of first words, and 0.86 for achievement of independent walking. Three category models remained clinically informative: accuracies were 0.79 for normal vs. mild vs. moderate-profound ID, 0.70 for first words ≤16 months vs. &gt;16 months vs. never, and 0.71 for independent walking ≤18 months vs. &gt;18 months vs. never. The strongest predictors for adverse neurodevelopmental outcomes were presence of hypotonia at birth, seizure onset within the first day of life, multiple seizures per day at onset, tonic seizures at onset, a burst-suppression pattern on EEG at onset, the presence of a de novo variant, and variant location within exons 6-7. Significance: These prediction models demonstrate the feasibility of early prognostication in KCNQ2-RD and support future prospective external validation. They enable more accurate individualised counselling by integrating clinical and genetic information readily available at time of genetic diagnosis and provide an objective foundation for early intervention planning and future precision medicine trial stratification.</jats:p>