G. Anthony Gorry

Decision analysis and clinical judgment

Abstract Sound clinical judgments derive both from the command of a sufficient body of facts and from the skill to combine such facts appropriately. Most undergraduate and graduate medical education concentrates on the first of these elements, the acquisition of knowledge; little formal effort is directed to the logic of dealing with clinical problems. In this discussion we suggest that the theory and technics of decision analysis provide new and useful strategies appropriate for dealing with complex clinical situations. In their qualitative aspects these formal strategies closely resemble those that the expert clinician employs informally, but which he is often unable to communicate explicitly. When applied quantitatively, the formalism affords greater precision than is otherwise readily attainable. To illustrate the application and utility of decision analysis we have considered the problems posed by severely hypertensive patients with possible functional renal artery stenosis, and have examined, both qualitatively and quantitatively, the alternative courses of action available to the clinician.

Experience with a model of sequential diagnosis.

Abstract A physician must have available two different classes of information when he attempts to deal with a diagnostic problem: the presenting signs and symptoms of the patient, and the pertinent medical knowledge about the disease state. The physician must then use this information in arriving at certain decisions: what additional information about the condition of the patient should be collected; what disease states are most probable; what are the potential consequences of misdiagnosis; what are the therapeutic and prognostic implications of the particular clinical situation. The possibility of significant errors in a diagnostic problem coupled with the decision-making activities involved have motivated the development of a wide range of tools to improve the performance of the diagnostician. This paper reports on an effort to exploit computer technology as such a tool for assistance in the diagnostic activity.

Mechanizing temporal knowledge

The importance that an understanding of time plays in many problem-solving situations requires that intelligent programs be equipped with extensive temporal knowledge. This paper discusses one route to that goal, namely the construction of a time specialist, a program knowledgable about time in general which can be used by a higher level program to deal with the temporal aspects of its problem-solving. Some examples are given of such a use of a time specialist. The principal issues addressed in this paper are how the time specialist organizes statements involving temporal references, checks them for consistency, and uses them in answering questions.

Toward a theory of clinical expertise

We believe that the outline of a promising approach to the study of expert clinical problem-solving is emerging. This process involves collecting as verbal transcripts the thoughts of experienced clinicians engaged in simulated clinical tasks and analyzing these transcripts to formulate theories of problem-solving tactics and strategies. In turn, these theories are embodied in a computer program that simulates the problem-solving process. The computer programs response to fresh clinical problems can be compared with physician behavior in order to revise, refute, or confirm the problem-solving theories. A scientific understanding of the expert clinicians reasoning strategies could prove exceptionally useful in improving the way we teach and practice medicine and in developing practical computer aids for medical decision-making.

Creating a knowledge-sharing community: if you build it, will they come?

Technology may support a knowledge-sharing environment, but getting users to participate in effective ways is key.

Sequential diagnosis by computer.

A computer program for sequential diagnosis has been developed for use in a variety of problem areas. The program employs sequential decision-making to balance the risk of making a diagnosis against the cost of further testing and the value of the evidence which can be obtained. Basically the program consists of an information structure which describes the problem area, an inference function which interprets signs and symptoms in terms of the medical knowledge in the information structure, and a test selection function which continually reevaluates the potential value of diagnostic tests in the light of increased information about the patient and the costs of tests and possible misdiagnoses.

Strategies for computer-aided diagnosis

Abstract A system consisting of a diagnostic program and a variety of support routines is described. The diagnostic program differs from ones previously reported in several important aspects. The program is based on a model diagnostic problem that subsumes the principal features of a number of real diagnostic problems. The user specializes the program to a specific area by providing information derived from past experience in the area. The program operates in an interactive mode. It performs sequential diagnosis, obtaining additional information from tests that it selects for the user to perform. In evaluating tests, the program considers both the cost of tests and the seriousness of possible misdiagnoses. The inference function employs Bayesian analysis of observed attributes, and can accomodate a variety of interattribute dependencies. The system facilitates the study of a variety of heuristics for various program functions. Some results from the application of the program to a problem in medical diagnosis are presented.

An Adaptive Group Theoretic Algorithm for Integer Programming Problems

Group theory is used to integrate a wide variety of integer programming methods into a common computational process. Included are group optimization algorithms, Lagrangian methods, the cutting plane method, and the method of surrogate constraints. These methods are controlled by a supervisor which performs four main functions: set-up, directed search, subproblem analysis, and prognosis. Some computational experience is given. One appendix contains an algorithm for dynamically solving unconstrained group problems. A second appendix gives an algorithm for solving zero-one group problems.

Computational experience with a group theoretic integer programming algorithm

This paper gives specific computational details and experience with a group theoretic integer programming algorithm. Included among the subroutines are a matrix reduction scheme for obtaining group representations, network algorithms for solving group optimization problems, and a branch and bound search for finding optimal integer programming solutions. The innovative subroutines are shown to be efficient to compute and effective in finding good integer programming solutions and providing strong lower bounds for the branch and bound search.

Sharing knowledge in the public sector: two case studies

In important areas of the public sector, client service depends on cooperation and collaboration among workers in different organizations or agencies. Examples include social services, education and health care. By sharing knowledge across organizations, workers can improve the quality of the service they provide. Technology to facilitate this knowledge sharing is widely available and a number of businesses have demonstrated how to put it to productive use. But the diffuse nature of many public sector settings imposes particular demands on knowledge management. Here I recount two experiments to promote knowledge sharing in the public sector: an early, unsuccessful effort to link social services workers; and a recent, substantially more successful undertaking to interconnect a large number of public school teachers. These experiences suggest the considerable potential of knowledge sharing in the public sector and underscore some of the challenges faced in that arena.