Figure 2 is the computer display of one of the hypotheses for an actual case as generated by the program. The figure shows the causal explanation that the program is proposing as the most likely explanation of the patient's findings. The findings entered by the user are shown in lower case and the nodes of the hypothesis are in upper case. The four nodes and one finding that are considered primary or diagnostic are printed in bold. The numbers in parentheses following the names indicate the probability that the item is true without being caused by anything else - the probability of being primary. The arrows indicate the most important causal relationships. (Ones with low probability are left out of the display if there is a higher probability link in the hypothesis. For instance, digitalis can also cause nausea/vomiting, but renal insufficiency is a more likely cause.) The numbers on the arrows are the probability that the cause will produce the effect. The probabilities from nodes to findings have been left out to keep the display readable. The links with a W+ are worsening factors. For example, cardiac dilitation increases the probability of low left ventricular emptying, but can not cause it alone. Links with P- decrease the probability of the effect. For example, the use of furosemide (a diuretic) decreases the probability of high blood volume, but does not eliminate it. In this case there is direct and indirect evidence that the blood volume is still high.
There are several points to notice about the hypothesis. First, there are a number of findings that can only be explained by one node. For example, the finding that the patient is on digitalis or furosemide (therapies), causes the corresponding nodes to be true. Also, some test results always indicate the condition (at least for our purposes). In this case, left bundle branch block (LBBB) on electrocardiogram always indicates that LBBB exists. These findings cause the corresponding nodes to be asserted true. Second, the program will generate hypotheses with multiple causes. In this case renal insufficiency and congestive cardiomyopathy are used to account for the findings, even though it is actually possible to account for the findings with just congestive cardiomyopathy. Third, the program will find multiple paths to a node. There are paths to dyspnea at rest (shortness of breath) from both pulmonary congestion and high left atrial (LA) pressure. Pulmonary congestion has independent evidence and high LA pressure is needed to explain the pulmonary congestion.
One of the advantages of this method of generating the hypothesis is the detailed information it provides about the internal nodes of the hypothesis - the mechanism by which the causes are producing the findings. In the medical domain, this information is useful both as a way for the user to check that the hypothesis is reasonable and as a way of identifying therapies that may be beneficial.