When the program has produced a hypothesis that the user has accepted as the diagnosis, there still may be unknown aspects of the patient state. Perhaps the user accepted only part of the hypothesis because of uncertainties about data or because additional complicating factors have not yet been ruled out. Resolving these issues may uncover easily treatable aspects of the disease or affect the selection of appropriate therapy.
The mechanism we use for examining these possibilities and looking for additional patient data that would be useful in refining the diagnosis is quite simple. For a single diagnosis it traverses the diagnostic causal chains of the hypothesis looking for unknown states with links into or out of the chain that are likely. For a differential it starts with the nodes that are different among the hypotheses. From these it searches for measurements (observations or tests) that might clarify the state of the node (positive or negative). If there are no measurements, it examines the possible causes of the node for measurements to clarify the situation at that level. This mechanism identifies the measures that could have a bearing on the situation and lists them for the user. The justification for these suggestions are the nodes that they will clarify. Thus, this operator uses the PSM with the diagnosis to produce a list of suggested measurements with justifications.