Ontology-Agnostic AI for Clinical Information Extraction and Quality Improvement in Cancer Care
Leveraging electronic health records (EHRs) for quality improvement (QI) in cancer care has been limited by the fact that the nuanced clinical concepts required to contextualize a patient's case for treatment decision-making, such as cancer histology, treatment history, biomarker status, and burden of disease, largely reside in unstructured clinical text documents. This forces essential quality assurance tasks, such as verifying that newly ordered systemic therapies align with a patient's specific indication, to rely on inefficient and error-prone manual review, contributing to preventable patient harm. To address this need, we propose an ontology-agnostic AI pipeline that stores "living," continuously updated, free-text summaries of a patient's history, which can be queried to rapidly generate application-specific structured data on demand. The plan is divided into two aims: first, developing this novel AI pipeline for clinical concept extraction and creating an open-source agentic query tool for researchers to analyze the data using natural language; and second, deploying this system at Dana-Farber Cancer Institute (DFCI) by integrating it into the daily workflow via a user-friendly dashboard for prior authorization staff and pharmacists. This deployment will culminate in a QI pilot to evaluate the system's effectiveness in increasing the rate of treatment-indication concordance in cancer care delivery.