Access Type

Open Access Dissertation

Date of Award

January 2012

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

First Advisor

Hao Ying

Abstract

Several thousands of drugs are currently available on the U.S. market. A complete understanding of the safe use of drugs is not possible at the time when drug is developed or marketed. At that time, the safety information is only obtained from a few thousand people in a typical pre-marketing clinical trial. Clinical trials are not capable of detecting rare adverse drug reactions (ADRs) because of limitations in sample size and trial duration. Early detection of unknown ADRs could save lives and prevent unnecessary hospitalizations. Current methods largely rely on spontaneous reports which suffer from serious underreporting, latency, and inconsistent reporting. Thus they are not ideal for rapidly identifying rare ADRs. In this dissertation, I developed a team-based multi-agent intelligent system approach for proactively detecting potential ADR signal pairs (i.e., potential links between drugs and apparent adverse reactions). The basic idea is that intelligent agents are capable of collaborating with one another by sharing information and knowledge which will accelerate the process of detecting ADR signal pairs. Each agent is equipped with a fuzzy inference engine, which enables it to find the causal relationship between a drug and a potential ADR (i.e., a signal pair). The fuzzy inference uses detection rules developed by me in this dissertation. The detection rules are based on different factors. I have also developed a methodology to find similar patients in the multi-agents system. The developed methodology uses similarity fuzzy rules in order to find similar patients in each agent's patient database.

In this dissertation, I developed a cooperative learning mechanism that was used by the agents in identifying ADR signal pairs and finding similar patients. The basic idea is that the agents are capable of collaborating with one another by sharing their knowledge. The agents start collaboration by providing their knowledge (i.e. rules) to the other agents. Using confidence level, the most important and insightful detection rules will be found and used for the benefit of the entire agent system. The new updated rules will lead to improve the agents' decision performance. To evaluate our approach, I designed a four-agent system and implemented it using JADE and FuzzyJess software packages. I choose four because it is representative enough while computing time is still reasonable. To assess the performance of the developed system, I conducted two simulation experiments that involved over 20,000 patients treated at the Veterans Affairs Medical Center in Detroit between 2005 and 2008. From the software standpoint, the four agents collaboratively worked one another as designed. Two physicians on the team independently reviewed the multi-agent system results. The results indicate that the agents can successfully collaborate in finding ADR signal pairs and finding similar patients.

Download

Share

COinS