The notion of clinical significance permeates evaluations of clinical research results, plays an important role in changing physician behavior, and enters implicitly into the development of rules for knowledge-based computer systems. To date, however, this concept has received no formal examination and is left to the intuition of the reasoning agent:.
The aims of this project are (1) to explore the role that clinical significance plays in physicians' evaluation of clinical-research results; (2) to propose formal methods that satisfy those roles; (3) to create a computer- based system that would implement those methods; and (4) to validate the proposed methods. As the test domain, we shall focus on the choice between two therapies as evaluated in randomized clinical trials. (1) We shall explore, through a survey process, three central questions regarding clinical significance: physicians' perception of its relative importance in evaluating study results, physicians' need for assistance in its evaluation, and the variability among physicians in its assessment. (2) We shall determine measures within three formal frameworks for expressing the factors that the survey ascertained as important. The frameworks are frequentist statistics, Bayesian statistics, and decision theory. The central challenge in applying these formal frameworks and their measures into the clinical setting are the difficulty physicians have with numerical measures, regardless of framework. To meet this challenge, we propose to develop accessible methods for using the frameworks and to develop computer-based tools for using the methods. The strategy we propose for developing accessible methods is to establish canonical models within each framework. We plan to propose a set of canonical decision models implicit in conclusions of clinical studies and to validate the set by a review of a random set of clinical trials gleaned from the medical research literature. To use these models, physicians need novel nonnumerical methods-graphical and qualitative techniques- that translate the statistical results into clinically meaningful terms and concepts. (3) Because these novel methods will be computationally more difficult than current practice, we propose to construct a set of computer-based tools that will implement them. As basis for building these tools, we shall use a set of novel artificial intelligence techniques. (4) We propose to validate each of the following propositions: that clinicians can use these advanced methods, that they prefer one framework over the other, and that use of these methods reduces or explains the variability established by the questionnaire process.
| Lehmann, H P; Goodman, S N (2000) Bayesian communication: a clinically significant paradigm for electronic publication. J Am Med Inform Assoc 7:254-66 |
| Lehmann, H P; Fleisher, L A; Lam, J et al. (1999) Patient preferences for early discharge after laparoscopic cholecystectomy. Anesth Analg 88:1280-5 |
| Lehmann, H P; Nguyen, B (1997) Bayesian communication of research results over the World Wide Web. MD Comput 14:353-9 |
| Lehmann, H P; Wachter, M R (1996) Implementing the Bayesian paradigm: reporting research results over the World-Wide Web. Proc AMIA Annu Fall Symp :433-7 |
