Anyone who has implemented a commercial anesthesia information system in the past decade has been faced with the daunting task of creating a set of words, phrases and sentences that encapsulate their hospital’s anesthesia practice. Usually the vendor will supply a basic set or offer a copy from a reference institution. In either case, these lists are unlikely to match practice that is often highly individualized, forcing many centers to create their own terms.

We were no exception at Duke University Hospital with the installation of our Saturn system in 1999 (Draeger Medical, Telford, Pennsylvania), and we continue to struggle with the consequences in a database of more than 100,000 cases described using our highly customized lists of procedures, techniques and anesthesia outcomes. The current generation of computerized systems is superb at collecting data for anesthesia records at the point-of-care, displaying old records and managing administrative tasks such as billing. Unfortunately the promise of anesthesia information systems opening up new avenues of pooled data of the highest quality to answer more significant questions has so far been unfulfilled. The reason is no longer the scarcity of these systems; it is the inability to pool, share and query data in a meaningful way. When a nurse in the recovery area uses an information system to document the term “confirmed myocardial infarction,” what does this mean? Is she recording the act of confirming a suspected myocardial infarction (MI), or has she just received definitive diagnosis from creatine kinase activity (CKMB) results? When in the preceding perioperative time course did the event occur? Were there any identified causative factors? How is this vital piece of information to be assimilated and pooled with other cardiac outcomes occurring after surgery?

Researching cardiac events using such a database is a bit like running a Google search online. The results will be a mixture of relevant and near-relevant hits, but more importantly, it is hard to know how complete your analysis has been. You do not know what is missing. These issues are only compounded when trying to combine data from other centers that use different words to describe approximately the same thing. Can you really pool your “confirmed myocardial infarction” with someone else’s documented case of “postoperative myocardial infarction?” One problem lies in the ambiguity of natural language, another in the variable quality of data collected in routine clinical documentation not necessarily under the scrutiny of a research protocol.

Having spent years frustrated by these issues, it was a welcome challenge to be invited to create a data dictionary task force by the Anesthesia Patient Safety Foundation (APSF) to obviate the linguistic ambiguity component of this problem.* The initial expectation was to create and endorse a definitive list of all anesthesia terms and hope that it would be adopted by a user-base wide enough to allow the direct comparison of identical terms. Everyone would use the term “postoperative MI confirmed by rise in CKMB.” Of course there was no guarantee that we would be as successful as more eminent groups who had tried but whose efforts had fallen into obscurity. Endorsed lists suffer from the problem of needing to be adopted by a wider forum to become useful and the fact that they should not be separate from the broader context of medicine. Our “myocardial infarction” is also a medical diagnosis with considerable significance to cardiologists who might just be building their own lists. Our breakthrough came with the realization that an endorsed list of anesthesia terms was a gross oversimplification of a complex linguistic and semantic interplay of terms and their relationships and that we were not alone in our problems.

Industry and some other areas of medicine have already addressed such problems. The solution is to define a “terminology,” a semantic network of terms and their relationships to each other. You would say, for example, that “postoperative myocardial infarction” is a “postoperative cardiac event” and that “postoperative cardiac event” is a “postoperative complication of anesthesia.” You could further define that “postoperative complication of anesthesia” has a “timestamp” and has a “causative factor” and that “hypotension during surgery” is a “causative factor.” What is created is an extraction of our knowledge into a format that computers can manipulate with some sense of the underlying meaning of the terms themselves. It means that we can now query our database for cases with events that are descendents of “postoperative complications of anesthesia” and be reasonably sure that the result will be a meaningful and complete set, provided they have been documented correctly. The beauty of creating the semantic network of a terminology is that it deals mainly with the meaning of terms called “concepts.” Different words used to describe the same concept are “synonyms.” Putting terms into the right slot in a terminology is a process called “modeling.” As the terminology grows, it becomes easier to see how it all fits together and to model new terms. If the modeling is done well, it does not need to be done again. Another advantage is that the concepts or meanings should be equivalent in all languages, so internationalization becomes the process of translating a new synonym for each language. The structure stays the same.

That we were not alone in our effort became important in two different ways. First we learned of considerable expertise and experience in creating anesthesia terminologies in the United Kingdom (U.K.). Second we became aware of the possibility of wider adoption by the U.S. government of a major medical terminology called SNOMED (Systematized NOmenclature of MEDicine), previously considerably limited in its usefulness by a steep licensing fee. In addition our colleagues who were creating anesthesia terms in the U.K. for the National Health Service were submitting their terms to SNOMED via the U.K.’s Clinical Terms Version 3 initiative (CTV3). In early 2002, the National Health Service entered into a national licensing agreement with SNOMED that incorporated CTV3, leading to the release of SNOMED CT. Terminologies were to be central to all new government health initiatives. Given that terms only need to be modeled once, the synergies became obvious, and in mid-2002, our effort in the United States became focused on leveraging and enhancing the work that was already started in the U.K.

It is to the great credit of the National Health Service Information Authority and their experts, Roger Tackley, M.D., Andrew Norton, M.D., and others, that their expertise has been so freely shared with that in the United States. We did, however, have something to bring to the table; early in our effort, we created a highly visual term-modeling tool that was recognized as an advance in making the intricate relationships of a terminology understandable to mere mortals. Our modeling tool was adopted as the means of extracting the knowledge of clinical and domain experts, and a working pattern developed in which the semantics of anesthesia concepts were gradually broken down and reconstructed. Our combined task has became one of searching for anesthesia terms in SNOMED CT and, where they do not exist, modeling them in our tool for submission to SNOMED CT. We are effectively creating the “anesthesia subset” of terms for SNOMED CT.

What has evolved is a collaborative international effort to develop an English-speaking terminology for anesthesia. The effort is supported by representatives from the U.K. and the Canadian Anaesthesiologists Society, with interest from other national representative bodies. Our work has been underpinned by the July 1, 2003, adoption of a national license for SNOMED CT in the United States by the National Library for Medicine, making it free for all U.S. medical entities. Regarding SNOMED CT, Secretary of Health and Human Services Tommy Thompson said, “It is free because we want you to use it.”

So how can we avoid wasting the data we collect? In short we can do so by adopting a standard way of documenting anesthesia events using a terminology linked to the totality of medical content. Other industries have been driven by the adoption of standards, and ours is no exception. As well as International Classification of Diseases (ICD) and Current Procedural Terminology™ (CPT), we will have a standard way of describing the essence of anesthesia linked to codes that can be meaningfully manipulated by computer. Of course all of this is only as good as the policies and training an institution adopts to ensure the quality of its documentation. If we can achieve and adopt this international effort at this early stage of the computerization of our specialty, however, the effect will be profound.

Iain C. Sanderson, M.D., is an Associate in the Department of Anesthesiology, Duke University Hospital, Durham, North Carolina.

Terri G. Monk, M.D., is Professor of Anesthesiology, University of Florida College of Medicine, Gainesville, Florida.