A collaborative multi-agent system approach to postmarketing surveillance of unknown adverse drug reactions

Several thousands of drugs are currently available on the U.S. market. These drugs can occasionally cause adverse drug reactions (ADRs) in patients. To assure the safety and efficacy of all regulated marketed drugs, the Food and Drug Administration (FDA) has set up a spontaneous postmarketing surveillance system called MedWatch to monitor drugs, especially new ones, for unknown ADRs. However, as a passive reporting system, this method suffers from underreporting, inconsistent reporting and latency.; We propose a collaborative multi-agent software system, named ADRMonitor, for actively monitoring and detecting ADR signal pairs based on electronic patient data stored in different healthcare systems. To empower intelligent agents with effective decision-making capability, we have developed a fuzzy logic-based computational recognition-primed decision (RPD) model that utilizes fuzzy sets, fuzzy rules, and fuzzy reasoning to represent, interpret, and compute imprecise and subjective information. The fuzzy RPD model has been validated by using it to calculate the extent of causality between a drug (Cisapride, withdrawn by the FDA from the market in 2000 due to ADRs) and some of its adverse effects for 100 hypothetical patients. The simulated patients were created based on the profiles of over 1,000 real patients treated with the drug at our Veterans Affairs Medical Center in Detroit before its withdrawal. The model validity was demonstrated by comparing the decisions made by the proposed model and those by two independent experienced internists. The levels of agreement were established by the weighted Kappa statistics and the results suggested good to excellent agreement between the physicians and the model.; To establish the advantages of our multi-agent approach over the current spontaneous postmarketing surveillance approach, we have investigated several major questions at the system level. We performed a simulation study using 275,400 hypothetical patient cases that were created on the basis of the real veteran patients. In the simulation, healthcare professionals and their intelligent agents were simulated by fuzzy RPD models with different decision-making capabilities. The simulation shows that (1) the number of ADRs detected by the multi-agent system is 7.8 times of the spontaneous reporting strategy; (2) the detection rate of an intelligent agent with moderate decision-making skills is 5 times higher than that of spontaneous reporting; (3) if the number of connected medical centers (each of them is comparable to our local medical center in terms of the number of patient cases) increases from 30 to 70, ADRs could be detected 15 month earlier. These results suggest that this novel multi-agent system has the potential to identify unknown ADRs much earlier than the current passive reporting system.