Automatic Generation of Computer Interpretable Guidelines

College At Oswego, Oswego NY

Investigators

Abstract

PROJECT SUMMARY / ABSTRACT Compliance with clinical practice guidelines (CPGs) has been demonstrated to markedly improve patient care, but the tools and processes available for physicians to rapidly and meaningfully leverage these guidelines are currently sub-optimal. Compliance improves greatly with the introduction of clinical decision support systems (CDSS), which implement guideline recommendations and are integrated into electronic health record (EHR) systems. Unfortunately, CPGs, as commonly distributed, contain recommendations for care of only a single disorder (or class of disorders) and are not easily consumable by computers for integration with CDSS. Recent work has focused on methods to resolve conflicts between guidelines, but only once they are in a computer interpretable form - a Computer Interpretable Guideline (CIG). We aim: to (1) develop a system for computational understanding of CPGs from the unstructured text; (2) to ground the clinical terms in each guideline in its definition so that the produced CIG can better integrate with electronic health record systems; (3) to generate CIGs in a format already developed which allows guidelines to be mediated with each other and allows the creation of CDSS. The project will focus on four clinical guidelines, relating to diabetes, heart disease, non-small cell lung and prostate cancers. Samples of each of the guidelines will be annotated by clinicians with the appropriate output of systems accomplishing each of the three aims. These samples will be split into datasets for testing and for evaluation, with the overall goal to achieve human levels of competence for each aim.