Hypothesis test for causal explanations in human pathology: evaluation of pulmonary edema in 181 autopsied patients with leukemia

Abstract

Symbolic logic, as used in the formal theory of scientific explanation proposed by Hempel and Oppenheim, has been suggested as the basis for automated medical diagnosis. In human autopsy pathology the determination of cause-and-effect relationships is a major area subject to logical analysis. We propose a modification of the Hempel-Oppenheim schema in which the logical relationships must only be satisfied “much” of the time, as determined by binomial significance tests. The analysis employs “certainty levels” logic with a more limited consistency requirement than classical logic. The analysis is applied to a series of 181 autopsied patients with leukemia in an attempt to determine a possible role of chemotherapeutic agents in the etiology of pulmonary edema. Among 51 patients who had received cytosine arabinoside (Ara-C) within 30 days of death, there was significantly more unexplained moderate or massive pulmonary edema than among patients with no or remote therapy ( p<0.001). The results suggest that a symbolic logical analysis combined with a binomial significance test can elucidate cause-and-effect relationships observed at autopsy, especially when there are multiple possible explanations for the same effect.