The main audience of patient reports consists of other physicians. For this reason, understanding what is said in a dictated medical report is difficult for a human reader without domain knowledge. Researchers compared laypeople against physicians at reading chest radiograph reports and judging whether the report described radiological evidence of acute bacterial pneumonia (Fiszman et al., 1999). Not surprisingly, laypeople performed much worse than physicians. As long as the report stated explicitly that the findings were consistent with pneumonia, the laypeople agreed with the physicians in their judgment, but pneumonia was mentioned in only one-third of the positive reports. In the remaining two-thirds of the reports, the evidence for pneumonia was inferred by the physicians and missed by the laypeople.

Implication in medical reports can occur at the sentence level and at the report level. A simple example is the sentence, "The patient had her influenza vaccine.'' If our SARS expert system had a variable for influenza, even a layperson reading the previous sentence could determine that the value for the variable would probably be no, because the patient was vaccinated. This inference requires domain knowledge that a vaccine generally prevents the target disease. In the radiology study reported above, evidence for pneumonia in positive reports was not always explicitly stated by the radiologist. Instead, the radiologist described "hazy opacities'' or "ill-defined densities'' in the lobes of the lung, which can be inferred to mean localized infiltrates. Once the inference at the sentence level has been correctly made, a physician reading the radiology report can integrate the findings described throughout the entire report and can infer that because the chest x-ray shows localized infiltrates not explained by other causes, there is evidence for acute bacterial pneumonia. Domain knowledge about words, combinations of words, and combinations of findings make it possible for a physician to make inferences from reports that a lay person—or an expert system—may not be able to make without training in knowledge of the domain.

Was this article helpful?

0 0

Post a comment