Improving Breast Cancer Research Through Automated Matching of Patients to Clinical Trials
Abstract
An enhanced Web based prototype intelligent agent/expert system for matching breast cancer patients to clinical trials has been built. It allows for cost preferences to be entered. Therefore, the system user can choose to rule patients out of trials as quickly as possible without regard to the cost of tests necessary to do this. A user can choose to have questions appear so that the patient is ruled out of the trial with the minimal set of costs (tests) or can choose some combination of approaches. The system has been tested with 15 protocols and designed for maximal responsiveness and scalability as new protocols are added. The files of 178 former patients have been used to test the accuracy of the system. Additionally, the files of 213 current patients have been tested for eligibility. Patients for each of the protocols were correctly found eligible for one or more trials. We found 240 new matching clinical trials for the 213 current patients. A probabilistic prototype system has been developed to reorder questions based on the probability they will determine the patient is ineligible for trial and preliminary experiments have shown up to 13% less questions will be required on average. It can also indicate the probability of patients being eligible for protocols. We have also developed a prototype system to quickly add new clinical trials. This has been successfully used by novices to enter new trials.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2004
- Accession Number
- ADA427957
Entities
People
- Dmitry B. Goldgof
- Jeffrey P Krischer
- Lawrence O. Hall
Organizations
- University of South Florida