Database Analysis and Software Design for Medical Decision Making

Period of Performance: 04/24/1997 - 05/27/1999

$375K

Phase 2 SBIR

Recipient Firm

Chirp Corp.
8248 Sugarman Drive
La Jolla, CA 92037
Principal Investigator

Abstract

The objective is to design software tools for recommending efficacious medical protocols, derived from the information contained in a patient database. The approach combines software algorithms for diagnosis, protocol selection, and assessment of patient progress. The diagnostic software uses Bayesian updating of diagnosis probabilities with cost/time constrained selection of tests that are predicted to yield the most relevant information about the patient's condition. Individualized protocol selection uses generalized fuzzy nearest neighbor analysis to identify database patients who are similar to the current patient and to analyze database patient outcomes using the prefenced of the current patient's condition using dynamic model that describes the evolution of a patient's condition over time. The dynamic model is derived from analysis of recorded observations of data bases patients at a sequence of sampling times. In addition of recommending efficacious, individualized practice guidelines, the approach provides insight into identification of database knowledge gaps and straightforward utilization of experts to fill such gaps. The software can also be applied to cost-benefit analysis of proposed guidelines and to cost-utilization analysis for determining the best price for a treatment or test.