Interpreting procedures from descriptive guidelines

Errors in clinical practice guidelines may translate into errors in real-world clinical practice. The best way to eliminate these errors is to understand how they are generated, thus enabling the future development of methods to catch errors made in creating the guideline before publication. We examined the process by which a medical expert from the American College of Physicians (ACP) created clinical algorithms from narrative guidelines, as a case study. We studied this process by looking at intermediate versions produced during the algorithm creation. We identified and analyzed errors that were generated at each stage, categorized them using Knuth's classification scheme, and studied patterns of errors that were made over the set of algorithm versions that were created. We then assessed possible explanations for the sources of these errors and provided recommendations for reducing the number of errors, based on cognitive theory and on experience drawn from software engineering methodologies.