The first issue in designing an expert system is to determine an appropriate view of the domain and of the reasoning required in problem solving. It may be easier to take a solution and look for a suitable problem, but the potential for addressing the needs of the user are greater if we investigate the problem first and find or develop methods to address the problems that are there.
The medical domain for our program is the diagnosis of patients with heart failure. Heart failure is the state of the patient when some disease process makes cardiac output inadequate for the demands of the body. There are many possible causes of heart failure including diseases of or damage to the myocardium (heart muscle), dysfunction of a heart valve, restriction of the heart by the pericardium (enclosing membrane), electrical dysfunction, and so forth. The cardiovascular system has strong mechanisms that attempt to compensate for the inadequate output. One mechanism alters the capacitance and volume of fluid compartments to increase the heart input pressure (preload) in an attempt to increase cardiac output. A longer term mechanism also starts retaining fluid to increase the blood volume to also increase preload. If the patient has lost blood, these are appropriate responses. When the problem is not low blood volume, the mechanisms can produce high pressures in the lungs, stiffening them, and causing fluid to leak into the air space, producing the common symptoms of pulmonary congestion. When the pressures increase on the right side of the heart, the pressure balance changes in the systemic venous circulation and the patient gets edema in the ankles and up the legs. Low cardiac output also causes the systemic blood vessels to contract to maintain the blood pressure to the important organs. While this is appropriate when blood volume is low, it often puts excessive stress on an already weakened heart. Since many different causes produce some combination of the same overt symptoms, they are grouped together as the syndrome of heart failure.
The problem for the physician is not simply to determine what disease the patient has and order an appropriate therapy. Since many of the most important diseases are incurable short of replacing the heart, the physician must determine how the disease is causing the findings in order to identify points where therapy can reduce the undesirable effects. If the disease is correctable, it is important to recognize that - indeed to rule out all correctable diseases. Fortunately, with most of the incurable diseases the patient is usually able to live an active and relatively normal life if the treatable problems and compensatory mechanisms are appropriately controlled. Since these tend to be chronic problems, many of the cases admitted to the hospital have multiple diseases - often a new problem added to a chronic problem. In addition, the patients are usually on therapies which may change the symptoms of the diseases or cause symptoms of their own. The physician needs to determine what diseases the patient has and the causal mechanisms by which they are producing the observed symptoms in order to determine whether there are treatable causes and what therapies might limit or correct the undesirable effects of the other causes. Thus, the goal of diagnosis is to determine the pathophysiological state of the patient in enough detail to guide therapy.
These diagnostic requirements imply that we need a disease model with enough intermediate states to distinguish among situations where different therapies are appropriate. For example, a myocardial infarction (heart attack) can produce both diastolic (ventricular filling) and systolic (ventricular emptying) dysfunction, each of which has different implications for therapy. To make such distinctions, a simple associational scheme such as those used by DXplain[Barnett87], or QMR[First85] would not be sufficient.