Hybrid agent-based modeling of healthcare surveillance networks

Future cancer registry systems will be provided on elastic and scalable systems in order to cost effectively allow for data sharing agreements to be realized. In order to develop systems to link the various disciplines of the health care industry, it is proposed that modeling and simulation techniques be used to reduce risk and improve the rate of adoption. A novel hybrid agent-based modeling simulation engine, PyModel, is developed to simulate load data for basic workflow operations in existing systems to produce models of larger scale information systems without the requirement to build full-scale prototype systems. This may reduce the risks involved with implementing large-scale health informatics systems. A number of experiments are conducted in which a virtual machine encapsulated system is constructed to provide either web or Grid based servers, acted upon by web or Grid based clients. Workflow load data were recorded and used as a basic load building block to model larger and more complex system configurations. Additional tests were performed to indicate other performance features of a Grid based interface's suitability for providing scalable database services for public health surveillance networks. The experiments resulted in three core results. The first illustrates that the computer model closely matches calibration data for multiple web servers retrieving data from a single web server. The second instance of multiple client web services interfacing iv against multiple web service appliances also calibrates against real world observed data. The third significant result shows a close correlation between real world observed data against a model comprising of multiple Grid clients interfacing to multiple Grid services in both appliance and standalone configurations. The experiments show that the configurations of both web and Grid clients interfacing to respective appliances calibrates closely with real world measurements for the same configurations. The observed data supplies the model with building blocks for simulating parallelized models of both web and Grid based services. The use of black box performance and load data to simulate systems of a more complex configuration in a scalable manner is more effective than building a traditional agent based model.