Key points are not available for this paper at this time.
A "conversational-mode" computer program can search a stored "library" of clinical experience for prognostic information. The library contains coded data for details of the clinical course of a "base population" of 678 patients with primary lung cancer. In estimating prognosis, the clinician at the computer terminal enters data to describe the new patient. The computer determines how many base-population members are identical in the cited characteristics. This first search seldom yields enough identical patients for prognosis. The clinician can reduce the specificity of resemblance by eliminating properties and by converting others into ranges and clusters. After each such modification, the computer indicates the number of library patients with the specified similarities. When the "resemblance group" is large enough, the computer lists the treatment of those patients and the results. All of the critical decisions are made by the clinician, who must choose the properties for a resemblance group and who decides when the group characteristics and size are adequate for prognostic extrapolations. The computer is merely a device for storage, retrieval, quantification, and display of the data needed for decisions.
Feinstein et al. (Thu,) studied this question.