INTERNIST-1/CADUCEUS: Problems Facing Expert Consultant Programs

Randolph A. Miller

Methods of Information in Medicine · 1984 · 81 citations · 6 references

Concepts

TL;DR

INTERNIST‑1 is an experimental general internal medicine consultation program that, despite matching staff physicians on test cases, suffers from limited anatomical, temporal, severity, and causal reasoning, and illustrates broader challenges in interface design, knowledge base development, algorithmic complexity, and clinical validation. The study seeks to correct these deficiencies to enable the successor system CADUCEUS. INTERNIST‑1 performs comparably to university hospital staff physicians, yet CADUCEUS is projected to be unavailable to the medical community for five to ten years.

Abstract

Summary INTERNIST-1 is an experimental computer program for consultation in general internal medicine. On a series of test cases, its performance has been shown to be similar to that of staff physicians at a university hospital. Despite INTERNIST-1’s apparent success in dealing with complex cases involving multiple diagnoses in the same patient, many shortcomings in both its knowledge representation schemes and its diagnostic algorithms still remain. Among the known problems are lack of anatomical and temporal reasoning, inadequate representation of degrees of severity of findings and illnesses, and failure to reason properly about causality. These drawbacks must be corrected before INTERNIST-1’s successor program, CADUCEUS, can be used. It is estimated that CADUCEUS will not be ready for release to the general medical community for five to ten years. Broader problems faced by all medical diagnostic consultant systems are: design of an efficient human interface; development and completion of medical knowledge bases; expansion of diagnostic algorithms from simple heuristic rules to include a range of complex reasoning strategies, and development of a method for validating computer programs for clinical use.

References

6