IBM Israel Research Seminars

The process of reporting research results within the Bayesian paradigm is called Bayesian Communication. Its purpose is to provide a formal, rather than ad hoc, framework for readers to assess new research results in light of their prior knowledge, and to assist in the process of technology adoption. Despite the increasing popularity of Bayesian methods among statisticians, the Bayesian paradigm is still unavailable to clinical decision makers, because, at least, of the lack of appropriate computer-based platforms.

The current project, Article Assistant, attempts to deliver the Bayesian paradigm as it applies to the evaluation of research results expressed as proportions or percentages of patients with a desired or undesired outcome. The interface provides a tutorial of core concepts, allows the user to select an article of interest from a library of articles, and to express prior belief in key parameters (minimally clinically important difference, expected outcome in the control group, expected difference between experimental and control, expected outcome in experimental group, or relative percentage difference between experimental and control). The interface then displays the data, and shows the effect of combining the prior knowledge with that data, along with the conclusions that are implied by the combination (e.g., consider staying with control; consider switching to experimental; cannot make a recommendation). The interface also displays the sensitivity of the recommendation to the prior knowledge specified by the user.

The details of the front end were developed in an iterative cycle with opinion leader clinicians at Johns Hopkins. The back end comprises a Cold Fusion-based set of templates that interact with Flash- and html-front end components, article and session databases, and BUGS, a generic engine for processing Bayesian statistical models.

We will shortly commence a Web-based controlled trial, comparing Article Assistant, with traditional statistical measures.

Speaker Bio
Harold Lehmann received his MD from Columbia, his pediatric training at Columbia, general pediatric fellowship training at Johns Hopkins, and PhD in Medical Information Sciences at Stanford. He is currently the Director of Research and Training for the Division of Health Sciences Informatics at the Johns Hopkins School of Medicine. DHSI runs one of 18 federally funded training programs in informatics, with a special emphasis on broad application of informatics principles and public health. His research has focused on applying Bayesian methods to clinical practice, including decision analyses and systematic reviews.