Data Analytics

Booz Allen Hamilton is developing a novel bio-surveillance prototype tool, the Digital Disease Detection Dashboard (D4) to address the questions fundamental to daily biosurveillance analysis and decision making: is something unusual happening (e.g., is an outbreak or novel disease emerging)?, What is the probability that what I’m seeing is by chance?, How confident am I that this data is really detecting a signal?, Why is this happening and can I explain it?; and How many cases should I expect? (e.g., magnitude of event over time). These questions focus on detection, confidence, variance, and forecasting and D4 integrates a number of diverse analytical tools and methods that are crucial to a complete biosurveillance program.

Objective

To develop a web-enabled Digital Disease Detection Dashboard (D4) that allows users to statistically model and forecast multiple data streams for public health biosurveillance. D4 is a user-friendly, cloudenabled, and R Shiny-powered application that provides intuitive visualization enabling immediate situational awareness through interactive data displays and multi-factor analysis of traditional and non-traditional data feeds. The objective of D4 is to support public health decision making with high confidence across all four aspects of the biosurveillance continuum—detection, investigation, response, and prevention.

We implemented the CDC EARS algorithms in our DADAR (Data Analysis, Detection, and Response) situational awareness platform. We encountered some skepticism among some of our partners about the efficacy of these algorithms for more than the simplest tracking of seasonal flu.

We analyzed several flu outbreaks observed in our data, including the H1N1 outbreaks in 2009, and noted that, using the C1 algorithm, even with our adjustable alerting thresholds, there was an uncomfortable number of false alarms in the noisy steady-state data, when the number of reported cases of flu-like symptoms was less than five per day.

We developed an algorithm, RecentMax, that could offer better performance in analyzing our flu data.

Objective

To develop an algorithm for detecting outbreaks of typical transmissible diseases in time series data that offers better sensitivity and specificity than the CDC EARS C1/C2/C3 algorithms while offering much better noise handling performance.

This year’s conference theme is “Harnessing Data to Advance Health Equity” – and Washington State researchers and practitioners at the university, state, and local levels are leading the way in especially novel approaches to visualize health inequity and the effective translation of evidence into surveillance practice.

Objective

Washington is leading the way in especially novel approaches. Our goal is to share some of these innovative methods and discuss how these are used in State and Local monitoring of Health

NPDS is a near real-time surveillance system and national database operated by the American Association of Poison Control Centers. NPDS receives records of all calls made to the 55 regional US poison centers (PCs). The Centers for Disease Control and Prevention (CDC) use NPDS to 1) provide public health surveillance for chemical, radiological and biological exposures and illnesses, 2) identify early markers of chemical, radiological, and biological incidents, and 3) find potential cases and enhance situational awareness during a known incident. Anomalies are reviewed daily by a distributed team of PC medical and clinical toxicologists for potential incidents of public health significance (IPHS). Information on anomalies elevated to IPHS is promptly relayed to state epidemiologists or other designated officials for situational awareness and public health response.

Current NPDS surveillance algorithms utilize the Historical Limits Method, which identifies a data anomaly when call volumes exceed a statistical threshold derived from multiple years of historical data. Alternative analysis tools such as those employed by ESSENCE and other computerized data surveillance systems have been sought to enhance NPDS signal analysis capability. Technical improvements have been implemented in 2013 to expand NPDS surveillance capabilities but have not been thoroughly tested. Moreover, other data aberration detection algorithms, such as temporal scan statistics, have not yet been tested on real-time poison center data.

Objective

To compare the effectiveness of current surveillance algorithms used in the National Poison Data System (NPDS) to identify incidents of potential public health significance with 1) new algorithms using expanded NPDS surveillance capabilities and 2) methods beyond the NPDS’ generalized historical limits model.

Investigation of cases, clusters, and outbreaks of infectious disease is a complex process requiring substantial support from protocols, distributed and cooperative work, and information systems. We set out to identify public health information needs, the types of data required to meet these needs, and the potential alignment with visualizations of this data.

Objective

The goal of this work is to identify specific work practices in disease investigation that would be supported by data visualization, such as identifying exposure, contact, and spatiotemporal clustering.

Currently, there is an abundance of data coming from most of the surveillance environments and applications. Identification and filtering of responsive messages from this big data ocean and then processing these informative datasets to gain knowledge are the two real challenges in today’s applications.

Use of Analytics has revolutionized many areas. At LongRiver Infotech, we have used various Machine Learning techniques (Regression, Classification, Text Analytics, Decision Trees, Clustering etc.) in different types of applications. These methodologies are abstracted in a generic platform, which can be put to use in many public health and surveillance applications, which are enumerated here.

Objective

To summarize ways in which Analytics, Machine Learning (ML) and Natural Language Processing (NLP) can improve accuracy and efficiency in bio surveillance and public health practices. We also discuss the use of this framework in typical surveillance applications (Integration with Devices/Sensors, Web/Mobile, Clinical Records, Internet queries, Social/News media).

Syndromic surveillance systems often produce large numbers of detections due to excess activity (alarms) in their indicators. Few alarms are classified as alerts (public health events that may require a response). Decision-making in syndromic surveillance as to whether an alarm requires a response (alert) is often entirely based on expert knowledge. These approaches (known as heuristics) may work well and produce faster results than automated processes (known as normative), but usually rely on the expertise of a small group of experts who hold much of their knowledge implicitly. The effectiveness of syndromic surveillance systems could be compromised in the absence of experts, which may significantly impact their response during a public health emergency. Also, there may be patterns and relations in the data not recognised by the experts. Structural learning provides a mechanism to identify relations between syndromic indicators and the relations between these indicators and alerts. Their outputs could be used to help decision makers determine more effectively which alarms are most likely to lead to alerts. A normative approach may reduce the reliance of the decision making process on key individuals

Objective

To analyse the use of Bayesian network structural learning to identify relations between syndromic indicators which could inform decision-making processes