Medical Terminologies That Work: The Example of MeSH

One description of medical informatics is as a disciplined search for methods to answer a question that a physician might pose: "What is the experience with patients like mine?" The difficulty in practicing high quality medicine has resulted in a strong professional emphasis in clinical practice of evidence-based medicine. This emphasis on evidence-based medicine relies primarily on literature review, and has largely ignored the fundamental problems of identifying patients similar to the patient at hand and exploring their experience. Standard terminologies are an important tool for humans and machines to recognize and to describe similar patients and to measure outcomes. Other tools, such as suitable outcome and similarity measures, are not within the scope of this paper, but clearly are other major concerns. Despite efforts over many years, there are few examples of standard terminologies that have been instituted into working systems. Lessons from the success of the Medical Subject Headings, used to index the world's biomedical literature in MEDLINE, have led to the formulation of several principles in building terminologies that work. Essentials include starting with a simple user and information model, evolving as user needs indicate, and understanding the interaction between systems, information models, and terminology. Update models and managing changing terminology pose significant challenges. Overall, making terminologies that work is a complex engineering process.