Consistency versus completeness in medical decision-making: Exemplar of 155 patients autopsied after coronary artery bypass graft surgery

Diagnoses made at autopsy are usually yes-no (binary) decisions inferred from clinicopathologic data. A major conceptual problem in determining cause of death is that variables used in classifying some patients may be missing in other patients. A model with too few logical implications will be mathematically incomplete for small data sets; but a model with too many implications may be inconsistent with large data sets. We examined the 155 patients autopsied after coronary artery bypass surgery from The Johns Hopkins Hospital autopsy data base of 43,200 cases. Diagnoses entered on a word processor and transmitted to a minicomputer were solved by the Quine-McCluskey algorithm. Our analysis disclosed that 41% of patients suffered a fatal complication of cardiac surgery; 43% had established surgical complications or unrelated causes of death; and in 17% of cases the cause of death was unexplained. Computerized symbolic logic analysis of medical information is useful in testing the completeness of a proposed set of causes of death.