Human-centered artificial intelligence framework for overdose mortality surveillance

Overdose mortality surveillance relies heavily on Medicolegal Death Investigation (MLI) reports, which contain rich but unstructured clinical narratives. We propose a human-centered AI framework that combines evidence extraction with overdose prediction in a two-stage pipeline. In Stage 1, key evidence spans are extracted from MLI reports. In Stage 2, the extracted evidence is used to predict overdose mortality. Our framework is designed with clinical stakeholders in mind, prioritizing interpretability and transparency throughout.