In several cases the primary problem is the nature of the independence of diseases. Currently, each disease that is considered primary is assigned a prior probability, usually dependent on age or other attributes in the input. In a hypothesis with two or more primary diseases, these prior probabilities are multiplied together. This simple notion of independence overlooks the differences between diseases that are chronic and therefore have higher probability of coexisting with other diseases and those that are over by the end of the hospital stay. For example, in case 50, COPD with infection was considered less likely than pneumonia even though the COPD can be present in a patient for thirty or more years.
The independence assumption also overlooks the fact that a patient with chronic diseases is more likely to need hospitalization for a mild disease that would not require hospitalization in an otherwise healthy patient. As a result, the program can have probabilities that are too low on hypotheses involving multiple chronic diseases and probabilities that are too high on ones with multiple acute diseases. (This problem is addressed in a recent paper[5].)
Because of the low probability of primary diseases, the program has a tendency to invoke relatively unusual causal mechanisms rather than add a new primary disease. For example, in case 107 it explained an unusual prosthetic valve murmur as evidence for a ventricular septal defect caused by a known MI. In case 171, known aortic stenosis was used to account for unstable angina, rather than adding coronary artery disease. While aortic stenosis can cause exertional angina, it is rare that the angina would be unstable because of the fixed nature of the aortic stenosis, unless there were coexisting coronary artery disease. This problem is partially a question of independence and partially a more complex relationship among pathophysiologic states. In case 116, with known mitral stenosis HFP invoked the causal mechanism to pulmonic regurgitation to account for a murmur. Even though mitral stenosis can cause pulmonic regurgitation, this is sufficiently unusual that an additional primary disease, aortic regurgitation, is a better explanation. HFP missed that explanation because it underestimated the probability of the two chronic primary diseases coexisting. Because mitral regurgitation is often secondary to cardiac dilitation from low LV systolic function, the program sometimes will invoke this explanation to account for an underspecified murmur, as it did in case 183. Some of the problems with the degree of dependence among valvular lesions could also be addressed by including more of the disease processes in the KB that lead to multiple valvular lesions. These include rheumatic heart disease and endocarditis, especially that caused by intravenous drug abuse.