Server: Netscape-Commerce/1.12 Date: Tuesday, 26-Nov-96 00:06:57 GMT Last-modified: Thursday, 15-Jun-95 00:33:41 GMT Content-length: 3857 Content-type: text/html CLINICAL DECISION MAKING GROUP

Peter Szolovits,
Professor of Computer Science and Engineering
Jon Doyle,
Principal Research Scientist
William J. Long,
Principal Research Scientist

Computer researchers have long sought to understand the process of medical decision-making and to use artificial intelligence (AI) methods to reproduce it in computer programs. Such programs could help prevent cognitive errors in health care, bring medical expertise to underserved parts of the world, and improve medical education and training. Accordingly, the Laboratory's Clinical Decision Making Group seeks to learn how doctors think about the diagnosis and treatment of various medical conditions, and how their approaches may be captured by AI methods.

The greatest challenge in building computer programs designed to mimic medical reasoning is that they must deal with many complexities. Diseases may appear in unusual combinations, for example, and exhibit variable symptoms. Medical care also is rife with basic uncertainties because of our incomplete understanding of disease and because of normal variations among patients. Further complicating the issues, AI programs must reflect individual patient values and preferences regarding decisions about potentially risky diagnostics and treatments.

Working with physicians from nearby institutions (including New England Medical Center and Children's Hospital), we are exploring the ways in which medical knowledge is represented, and how disease may be explained as physiological "derangements" of the healthy individual. Theoretical efforts develop general techniques for representing physiological, anatomical, and biochemical knowledge; exploring the consequences of disease pathologies; evaluating time-dependent data; and eliciting and representing patient preferences. Empirical studies involve implementation of computer programs that solve specific medical challenges. Evaluation of the capabilities and usefulness of such programs helps lead to a better understanding of future research needs.

Specific projects within the Group are selected based on the clinical importance of various disorders and on their usefulness in designing new AI-related technologies. Currently we are working on applications to congestive heart failure, genetic counseling, the detection of abnormal growth in children, and monitoring treatment in the intensive care unit (ICU). Earlier studies include the diagnosis and treatment of coronary artery disease, fluid and electrolyte disorders, and Hodgkin's disease (a sometimes fatal disease of the lymphatic system). We also have studied the design of clinical trial protocols for lung cancer therapies; problems in administering digitalis; diagnosis of the causes of acute renal failure and other kidney disorders, and the management of ventricular arrhythmias (irregular heart function) in the cardiac ICU.