Neurosurgery

Systematic review: epilepsy-surgery prediction models show modest performance and high bias

This systematic review included 42 papers and 113 prediction models for epilepsy surgery outcomes. Median discrimination was modest with AUC 0.75 and median accuracy 0.76 across models. Only 20.4% of models had external validation and 81% were judged high risk of bias. Apply these models cautiously in clinical decision-making until externally validated and bias reduced. 1

LLMs accurately screened Hebrew clinical notes to identify epilepsy surgery candidates

In 110 tertiary-clinic records, six general-purpose LLMs identified surgical eligibility with sensitivity up to 1.00 and specificity up to 0.96. Majority voting produced near-perfect sensitivity (1.00) for identifying eligible patients in this cohort. Among patients meeting criteria, 45% had not previously been considered for surgery, suggesting missed referrals. Consider LLM-assisted screening to flag candidates, but pilot and validate performance on local notes and workflows. 2

IDH‑mutant astrocytoma prognosis linked to tumor volume, grade, and extent of resection

Cohort of 210 newly diagnosed IDH‑mutant astrocytomas had median PFS 106.8 months and 5‑year PFS 67.5%. Multivariable analysis found higher WHO grade (grade 4 HR 2.91), larger total tumor volume (HR 1.03 per unit), and lesser EOR predicted shorter PFS. Higher baseline KPS was protective (HR 0.76); age and MGMT methylation were not significant predictors. These data support aiming for maximal safe resection and incorporating tumor volume into prognostic discussions. 3