Cost-effectiveness analysis of resective epilepsy surgery in drug-resistant patients: an artificial intelligence data...

Resective epilepsy surgery has been proven to reduce the number of seizures and improve quality of life in patients with drug-resistant epilepsy (DRE) but implies high direct costs. Cost-effectiveness analyses have shown that surgery is cost-effective. We aimed to evaluate whether we can determine the cost-effectiveness of surgery for DRE using artificial intelligence-based extraction of surgical outcomes from electronic health records (EHR) to assess health outcomes for epilepsy in a semi-automatic manner.

A retrospective, pre-post, observational within-subject study was conducted at Hospital del Mar, Barcelona, Spain in patients operated for DRE with resective surgery. Clinical outcomes were retrieved from the EHR using large language models (LLM) based on Mistral 24B and compared with the Epilepsy Reference Center prospective database. The incremental Cost-Effectiveness Ratio (ICER) was calculated using direct hospital costs and quality-adjusted life-years (QALYs) as the outcome measure under a provider perspective. Different scenarios were calculated, both at the cost level, using costs from the two and three years before and after surgery, and at the quality-of-life level. Both a probabilistic sensitivity analysis (PSA) and a projection of possible ICERs up to 15 years post-surgery were carried out.

The ICERs for seizure free status (Engel I) were €28,630.32/QALY and €29,002.15/QALY two and three years after surgery, respectively. PSA showed that surgery had a 76.4% probability of being cost-effective under a €30,000/QALY willingness to pay threshold for Spain. Direct costs when considering anti-seizure medications would be lower in the surgery group after 12-14 years, becoming a dominant strategy with lower costs and higher quality of life. The LLM extraction strategy enabled a subsequent rule-based binary classification of outcomes (i.e., "good" vs. "bad"), achieving 91.2% accuracy at this level of aggregation, which resulted in an ICER of €57,858.61/QALY two years after surgery.

AI (LLM) based data extraction of epilepsy surgery outcomes enables accurate cost-effectiveness analysis to assess health outcomes in a clinical setting. Resective epilepsy surgery proved to be cost-effective in the mid-term, achieving ICER values below the WTP at three years after surgery.