The following is a knowledge base fragment sufficient to cover the example MI and nitroglycerin problem. Not all of the temporal parameters are needed in this example, but it will give the flavor of the temporal reasoning. The definitions have been simplified to keep the example understandable:
(defnode MI full-name myocardial-infarction caused-by (primary prob 0.001) persist (time 0 1hr 0.5 1day) measure (chest-pain prob 0.9))For purposes of this example, there is no cause for an MI. It just has a probability of 0.001 of being true in a patient. The persistence time is the time from the end of a cause being true to the end of the node being true. Since in this case the cause is random chance, the persistence is the same as the length of time the condition is true. The time clause indicates that the MI is never over in less than an hour (probability 0) and always over if more than a day has gone by. If the time is between, there is a 0.5 probability that the MI is over. This is clearly an approximation since the length of an MI is actually a smooth curve, but this is sufficient. For simplicity, this ignores everything but the most acute effects of an MI. The measure clause has the possible findings for the node. In this case an MI has a 0.9 probability of producing chest pain. There are a number of other possible findings, but this is sufficient for the example.
(defnode (low LVF) full-name (low left-ventricular-systolic-function) caused-by (MI prob 0.6) persist (time 0.3 6hr 0.8 inf))An MI causes low left ventricular systolic function with a probability of 0.6. 30%of the time the low LVF returns to normal immediately or within six hours after the end of the MI. After six hours 80%of the cases of low LVF have returned to normal and the rest are permanently impaired. There is no direct measure of LVF for the example.
(defnode nitroglycerin caused-by (therapy))Nitroglycerin is a therapy. Since therapies are under the control of the physician there is an implicit measure in the therapy history.
(defnode (high LAP) full-name (high left-atrial-pressure) caused-by ((low LVF) prob 0.8) corrected-by (nitroglycerin prob 0.8) measure (PCWP prob (range 0 15 .1 18 .9)))High LAP is caused by low LVF with a probability of 0.8. It can be corrected by nitroglycerin and the probability that a LAP that is high will be brought down to normal by nitroglycerin is 0.8. The measure for LAP is the PCWP. Since that is a continuous variable, the range statement breaks it into ranges in which the probability can be specified. A PCWP less than 15 never indicates a high LAP. 10%of high LAPs will produce a PCWP between 15 and 18 and the rest will produce a PCWP above 18.
(defnode pneumonia caused-by (primary prob 0.01) persist (time 0 2day 0.5 2week))Pneumonia is another primary node, with a probability of 0.01, that takes 2 days to 2 weeks to return to normal.
(defnode PC
full-name pulmonary-congestion
caused-by ((high LAP) prob (onset now 0.2 1hr 0.5 6hr 0.8))
(pneumonia prob (onset 2day 0.5))
persist (time 0 6hr 0.5 1day)
measure (rales prob 0.9))
Pulmonary congestion can be caused by either high LAP or pneumonia. The
high LAP may take up to 6 hours to produce the congestion if it is going
to produce it at all. In 20%of cases the congestion occurs in less
than an hour. Pneumonia takes 2 days before half the cases produce
pulmonary congestion. Persistence on the other hand is always between 6
hours and a day.
There must also be knowledge about the findings including how often they might be falsely positive, but we will assume for now that they are perfect.