The kind of situation that calls for temporal reasoning is illustrated by the following example, which will be used as a running example for the rest of the paper. In an actual (and quite typical) case diagrammed in figure 1, a patient was admitted to the Emergency Department with chest pain of an hour duration beginning two hours prior and was given nitroglycerin. Four hours later, when the data was entered into the computer, the patient had a Swan-line in place providing cardiac pressure information. At that time, the patient had rales in the chest examination (fluid in the lungs) and a pulmonary capillary wedge pressure (PCWP) of 12, indicating a normal left atrial pressure.
In this patient, there is adequate evidence from the chest pain (and other findings which we will ignore for simplicity) indicating a myocardial infarction (MI). It can also be concluded from the rales that the patient has pulmonary congestion. An MI can cause pulmonary congestion (figure 2) by poor left ventricular function (LVF), which elevates the left atrial pressure (LAP), causing fluid to accumulate in the lungs. Prior to temporal reasoning, the HDP had difficulty accounting for the rales because the PCWP of 12 indicated that the LAP was normal, breaking the causal pathway from MI to pulmonary congestion. As a result, the program proposed pneumonia as an explanation for the rales.
Because pneumonia can also cause pulmonary congestion, this is a reasonable hypothesis, but it is not the best hypothesis. The nitroglycerin decreases the LAP as well as improving blood flow in the myocardium. Since the rales can take many hours to go away after the LAP has returned to normal, a better explanation is that the rales were caused by the MI, but the nitroglycerin has now decreased the LAP and the rales have not had time to clear. A third explanation is that the MI only transiently decreased the LVF. While the rales have not had time to clear, the causal mechanism is no longer present. Thus, there are three reasonable mechanisms to account for the findings, each with different implications for treatment, but unless the reasoning considers the time relationships only one explanation can be generated.