BioSense Initiative to Improve Early Outbreak Detection

Johns Hopkins University, Baltimore MD

Investigators

Linked publications & trials

Abstract

The broad project objective is to establish a robust technical link in the process of automated outbreak[unreadable] detection to complement and backup the traditional sentinel surveillance system. Means to this objective[unreadable] include development and efficient combination of data-driven statistical alerting algorithms and[unreadable] implementation of higher level decision-support tools for fusing algorithmic results with external information[unreadable] and epidemiologist judgment. Specific aims are to establish and exercise a context-sensitive testbed for[unreadable] standardized algorithm development and evaluation, to develop an information-sharing methodology for[unreadable] jurisdictional situations that preclude data-sharing, and to create decision-support tools by combining[unreadable] heuristic methods used by experience health monitors with Bayesian Belief Net representation. Research[unreadable] within the testbed will advance the state of the art in detection algorithms, stressing the adaptations required[unreadable] to make them relevant and effective for monitoring on a daily or, depending on input data rates, a near-realtime[unreadable] basis. Algorithms will broadly include univariate, multivariate hypothesis tests and data mining[unreadable] techniques more general automated learning and we will seek to determine the appropriate niche for each[unreadable] approach found useful. For univariate data, hypothesis test research will investigate means of combining[unreadable] data modeling and process control for optimal detection performance depending on the data background.[unreadable] Multivariate algorithm research will include both fully multivariate methods and multiple univariate methods[unreadable] and will blend the two for optimal monitoring capability. Structured testbed design and development will[unreadable] establish standards for algorithm evaluation and comparison based on health monitoring effectiveness,[unreadable] measured by sensitivity, specificity, and timeliness of event detection. These standards will make this public[unreadable] health research area more accessible and relevant to local heath department users. Potentially, the[unreadable] decision-support tools may make complex data scenarios understandable to users of these systems.[unreadable] Acceptance of potential of automated surveillance systems will stimulate increased cooperation of health[unreadable] departments, data providers, and policymakers, further improving public health monitoring capability.