Automated Encoding of Clinical Guidelines into Computer-interpretable Format

Computer-interpretable guidelines (CIGs) are critical knowledge source for clinical decision support systems (CDSS). However, most of current CIGs are encoded by medical experts and knowledge engineers based on the clinical practice guidelines (CPGs). It is complex, time-consuming and error-prone. This paper proposes a model and a system framework that automates large part of the encoding process. The model employs a directed graph representing the knowledge of a guideline, and the framework consists of a pipeline of three steps: semi-structural guideline generation, graph reduction and validation, and CIG construction. Furthermore, we chose two CPGs issued by National Comprehensive Cancer Network (NCCN) to illustrate the use of this proposed framework. Automated encoding them into semi-products saves a tremendous amount of time, reducing 25 workdays for manual encoding work to 15 minutes of automated encoding plus 5 hours manual validation and correction. This indicates that automated encoding tools based on rigorous models is of practical value in a proper work framework.