The source of temporal information about a case is the user input. From the viewpoint of temporal properties, findings can be grouped into four classes: observations, symptoms, history, and tests. The observations include the results of the physical exam. These only provide information about what is true at the time of the exam and say nothing about how long the findings might have been true. The symptoms are typically reported by the patient and include time information. For example, the patient may say that the chest pain started two hours ago and lasted for an hour. In practice descriptions can be much more complex, such as complaining of shortness of breath that only occurs at night. However, these more complex descriptions can either be handled as associations of findings or as specialized findings with names of their own. Important associations include descriptions such as having palpitations with shortness of breath. The shortness of breath at night is called paroxysmal nocturnal dyspnea or PND. History information has essentially the same temporal properties as symptoms. Patients could have coronary artery disease for five years or have had an episode of endocarditis a year ago. Tests occur at a specific time. They are essentially observations except that often they were done at a time in the past and, unlike physical exam observations, will not be repeated unless there is some specific need. For example, an echocardiogram done five years ago may provide useful information about the patient's present condition. Observations of the past are either summarized as patient history or are considered irrelevant to the current situation.
Thus, the user can provide the system with the appropriate temporal information with a small number of additional capabilities in the input interface. Observations are assumed to refer to the current time. Symptoms and history need to have durations, event times, and pertinent associations. Tests need to have event times. With these attributes it is possible in the heart disease domain for the user to provide the pertinent temporal details for diagnosis.
In the example, the user provides the following input:
chest pain: anginal at rest 6hr ago for 1hr, therapy: nitroglycerin for 4hrs, chest: rales, PCWP: 12 now
Given this input, the case provides the following facts: (In the
following discussion we will use the convention that times such as
will mean 6 hours in the past and now refers to the current time.)
The causal relations and the initial facts are shown graphically in figure 4. The nodes are shown with initial probabilities at the top and the persistence time at the bottom. The links have the causal probabilities.
