Disease Prediction Models Aim To Guide Medical Decision Making

When performing a history and physical examination on a patient with a disorder such as chest pain, physicians try to gauge the likelihood that each differential diagnosis explains the chief complaint. Determining the likelihood of disease as an estimate of probability has important advantages, and this will become increasingly evident as medical information systems, such as those developed for hand-held personal digital assistants (PDAs), help physicians to estimate probability at the moment of decision making.

Clinical prediction rules provide a structured approach to calculating a probability estimate by incorporating the key variables that predict whether a disease is present or may develop in the future.

Clinical prediction rules provide a structured approach to calculating a probability estimate by incorporating the key variables that predict whether a disease is present or may develop in the future.

Many prediction rules assign a point score to each key clinical finding so that the physician can characterize the patients' clinical picture with a summary score that represents their probability of having or developing a given disease. Clinical prediction rules can help estimate a pretest probability of a disease, the probability of an adverse surgical outcome, or the probability that a patient will develop a given disease. Such rules can also help identify high-risk patients who need intervention.

The work-up of a patient with chronic, recurrent chest pain is a scenario in which probability estimates can be used to choose a diagnostic test and interpret the results. Current guidelines by the American Heart Association and other professional organizations for evaluation of suspected chronic ischemic heart disease emphasize the role of pretest probability. The foundation for this reasoning is summarized by Bayes theorem: Posttest probability of a disease depends on pretest probability. In other words, one must know the pretest probability to decide whether a test …