The Histone Code

CMT Disease

Medical Informatics

Students in the Community
Research Briefs
Around & About

Clinical guidelines, recommendations for the workup or treatment of various conditions, have been shown to improve the quality of patient care. But approximately 2,000 guidelines exist, covering everything from the best practices for vaccinating babies to recommending which patients should have coronary angiography. Clinicians can no longer remember all the information contained in the many guidelines and then apply the recommendations to their patients.

Dr. Edward Shortliffe, chairman of medical informatics and professor of medicine and computer science, thinks the solution is to give clinicians access to the guidelines at more relevant point-of-care moments—for example, when a physician accesses patient data or places an order from a computer. But instead of just providing a link on the computer to the National Guideline Clearinghouse or Medline, which would require the clinician find the relevant guideline and read a 10- to 30-page document, Dr. Shortliffe's goal is to have the computer look at the patient data and automatically present the clinician with custom-tailored recommendations based on the applicable guideline.

Dr. Shortliffe's InterMed group, a collaboration among Columbia, Stanford, and Harvard informatics researchers (see, is now a step closer to integrating recommendations and patient data with the recent construction of an "execution engine" by Dongwen Wang, a Ph.D. candidate in medical informatics at P&S.

Mr. Wang's execution engine is just one of several steps on the way to a functioning clinical decision-support system. One of the first steps was allowing computers to handle the logical content derived from guidelines, so in the mid-1990s InterMed created a language called GLIF, for GuideLine Interchange Format, to encode a guideline's information.

From immunization guidelines, for example, GLIF creates a flowchart that represents when children should receive various vaccinations and boosters, which products are recommended, and what to do if the child's family has a history of seizures. Because GLIF is based on common structures from different types of guidelines, such as decision-making points and recommendations, it can encode many guidelines, like ones for cancer treatment or how a doctor should manage a cough.

From the GLIF-encoded guidelines, the next step is integrating the guidelines into a hospital's clinical information system, such as Columbia Presbyterian Medical Center's WebCIS. Once integrated, a clinician can retrieve such patient data as lab results and prescribed medications and also see what treatment step for that patient currently is recommended by the latest guideline.

Integration between the guidelines and patient data is accomplished with the help of an execution engine. Mr. Wang describes the engine as middleware between what the clinician sees on the computer screen and the underlying patient database (the electronic medical record).

The execution engine is able to retrieve specific patient data requested by a guideline and then, based on those data, present the recommended treatment. Unlike other systems in development, Mr. Wang says his engine is able to keep track of where a patient is in the management of a chronic disease. In other systems, a clinician often has to scroll through the entire flowchart of recommendations before getting to the patient's current stage.

The engine and its generalized version, which took Mr. Wang about two years to develop, can also understand a few other languages to take advantage of guidelines already encoded in other formats besides GLIF. The main challenge in making the engine multilingual, Mr. Wang says, was figuring out the features each language shared and developing a way the engine could access each language by using the common features.

Mr. Wang and Dr. Shortliffe, who also is deputy vice president for information technology in the Health Sciences Division and director of medical informatics services at New York-Presbyterian Hospital, recently tested the engine using a GLIF-encoded immunization guideline in a laboratory setting with hypothetical patients. The test compared their system to EzVac, another validated computer-based system that keeps track of immunization records, to assess how well the computer-generated recommendations complied with the written form of the guidelines.

"The GLIF-based system basically reached the same level of performance as EzVac and gave the same result more than 98 percent of the time," Dr. Shortliffe says. "Of the patients who need immunization, 99 percent are correctly identified by the GLIF system and of the patients who don't need immunization, 81 percent are correctly identified." The system was designed to find as many patients who needed immunizations as possible, even though that strategy leads to a modest decrease in accurately identifying those who don't need an immunization.

"But the big advantage of our system is it can be generalized to other guidelines and to other institutions," Dr. Shortliffe says. InterMed's method is general enough that other hospitals, with different patient database systems, can also use the execution engine. Ultimately, a standard system will save each hospital from having to spend the time and money to develop its own system, which may be incompatible with others. InterMed has submitted GLIF to a national standards organization, Health Level Seven, for consideration as a national standard for computer-based guideline representation.

The next step for Mr. Wang, Dr. Shortliffe, and InterMed is testing GLIF and the engine in a real clinical setting and, barring any problems, integrating the system with WebCIS. "A misperception is that we"re trying to replace clinicians," Mr. Wang says. "We're trying to help clinicians improve the quality of care without adding to their workload."

InterMed and Mr. Wang are supported by a joint grant from the National Library of Medicine, the Agency for Healthcare Research and Quality, and the U.S. Army. A recent paper on this research may be found in the Proceedings of the American Medical Informatics Association Fall Symposium by Wang, D. and Shortliffe E.H., and called GLEE: A model-driven execution system for computer-based implementation of clinical practice guidelines.