General Surgery

Prediction models for epilepsy surgery are promising but under-validated

Systematic review of 42 papers and 113 models found median AUC 0.75 and median accuracy 0.76. Only 54.0% of models had internal validation and 20.4% had external validation. Overall risk of bias was high in 81% of models despite low applicability concerns in most. Use these models only as adjuncts; external validation and better methods are required before routine use. 1

Protocol: individualized BP targets for older hip fracture surgery

Single-center RCT will randomize 180 patients aged 65–85 to SBP within ±10% of baseline versus standard reactive care. All patients will maintain MAP ≥65 mmHg; primary outcome is organ dysfunction within 7 days. The trial tests whether proactive individualized BP reduces early postoperative organ dysfunction versus usual care. Results will be directly applicable to older hip fracture patients under general anesthesia once completed. 2

Large language models accurately screen notes for epilepsy surgery candidates

Retrospective analysis of 110 Hebrew-language records showed LLMs achieved near-perfect sensitivity for surgical eligibility. Majority voting produced sensitivity 1.00 and high specificity, with accurate Seizure Freedom Scale extraction. The models identified that 45% (13/29) of eligible patients had not previously been considered for surgery. LLM-based screening could scale candidate identification but requires local validation and integration into workflow. 3