Diagnostic reasoning of high- and low-domain-knowledge clinicians: a reanalysis.

Thinking-aloud protocols provided by Joseph and Patel were reanalyzed to determine the extent to which their conclusions could be replicated by independently developed coding schemes. The data set consisted of protocols from four cardiologists (low domain knowledge = LDK) and four endocrinologists (high domain knowledge = HDK), individually working on a diagnostic problem in endocrinology. The two analyses agree that the HDK physicians related data to potential diagnoses more than did the LDK group and were more focused on the correct diagnostic components. However, the reanalysis found no meaningful difference between the groups in diagnostic accuracy, speed of diagnosis, or the breadth of the search space used to seek a solution. In the reanalysis, the HDK physicians employed more single-cue inference and less multiple-cue inference. The generalizability of results of protocol-analysis studies can be assessed by using several complementary coding schemes.

Computers as clinicians: an update.

Computers as clinicians entered medical settings relatively recently, but with limited success because they lack general intelligence--that is, though they can be experts in domain specific specialties, they cannot yet deal with clinical problems never before encountered. SOAR, a novel AI computer programming architecture, can learn from past encounters with prior problems and can generalize its learning to new ones. It may therefore take computers to a higher level of clinical performance.

Why we still use our heads instead of formulas: toward an integrative approach.

This review begins with a discussion of Meehl's (1957) query regarding when to use one's head (i.e., intuition) instead of the formula (i.e., statistical or mechanical procedure) for clinical prediction. It then describes the controversy that ensued and analyzes the complexity and contemporary relevance of the question itself. Going beyond clinical inference, it identifies select cognitive biases and constraints that cause decision errors, and proposes remedial correctives. Given that the evidence shows cognition to be flawed, the article discusses the linear regression, Bayesian, signal detection, and computer approaches as possible decision aids. Their cost-benefit trade-offs, when used either alone or as complements to one another, are examined and evaluated. The critique concludes with a note of cautious optimism regarding the formula's future role as a decision aid and offers several interim solutions.

Computer modeling of clinical judgment.

Four main developments leading to computer modeling of clinical judgment are described in this paper. These include information processing psychology, clinical vs. statistical prediction studies, behavioral decision theory, and Bayesian decision analysis approaches. One clear catalyst in these developments has been the computer, which has been used as an information management tool rather than a data-processing device. Future directions of these efforts are delineated, and problems as well as prospects of computerizing clinical judgment are described.

Statistical modeling of clinical intake decisions.

Modeled the intake decisions of four clinicians and a group of clinicians at a community mental health center by principal components-discriminant function analysis. The models of two clinicians and the group-based models withstood replication by the jackknife technique of discriminant analysis. Results showed that clinicians' written notes can be used to predict their decisions. We learned, for example, that the cues that are most important in arriving at intake decisions were therapy history, interview site, level of functioning, diagnosis, and behavioral disturbance.

Diagnostic problem solving by computer: a historical review and the current state of the science.

The current scientific interest in medical diagnostic problem solving originates in human judgment research among psychologists, statisticians, computer scientists, and physicians. The present state of this research is described and traced to its beginnings in the 1950s. Although the pivotal importance of the computer is emphasized, its impact on medical information management is recognized as not having been as great as in some other spheres of intelligent reasoning. But its future important role in medical diagnostic problem solving is optimistically anticipated.

Computational and noncomputational clinical information processing by computer.

Computers have been used in their computational as well as noncomputational modes for the scientific study of the clinical decision making. As computational tools, computers serve as storage devices for hospital file data so that these data can be analyzed statistically for a large variety of epidemiological and diagnostic purposes. As noncomputational devices, computers are functioning as knowledgable medical teachers that interface with students who can query the system. In this mode, computers are sometimes used as consultants to clinicians who need information on technical topics. Noncomputational computers are also used to simulate the inferences of clinicians. Byproducts of these simulations are computer programs that become formal statements or theories of the decision processes under investigation.