Next Generation Surveillance for the Next Pandemic - Abstracts
From Santa Fe Institute Events Wiki
Alessandro Vespignani - From nowcasting to forecasting the evolution of the next pandemic by using digital surveillance.
Mathematical and computational models have gained importance in the public-health domain, especially in infectious disease epidemiology, by providing quantitative analysis in support of the policy-making processes. These models are however data-hungry and needs novel surveillance schemes able to provide them with the required parametrization and initial conditions. I will discuss the integration of data-driven models with digital disease surveillance data and the successes and challenges that we should expect from forecasting the next pandemic.
Christian Stefansen - Six Years of Google Flu Trends – Learnings and the Future
Google Flu Trends, a service launched by Google.org in 2008, aims to provide real-time estimates (nowcasts) of flu activity data in more than 25 different countries and 40 languages. In the United States, for example, Google Flu Trends provides a nowcast of the influenza-like illness (ILI) data provided by the Centers for Disease Control. Google Flu Trends computes its estimates as the fraction of the Google searches associated with a set of query terms historically correlated with ILI data. The primary advantages of Google Flu Trends are finer granularity (trained on regional data, it can provide city-level estimates) and timeliness (estimates are available with less than 24 hours’ delay). In this talk we look back on six years of Google Flu Trends, paying particular attention to the ’12/’13 season, during which Google Flu Trends overestimated the peak of the flu season significantly in the United States. We discuss the factors that affect a search-based model’s ability to produce sound estimates and different ways to mitigate them.
David Heymann and David R Harper - Global surveillance: past, present and future
The continuous need for strengthened global surveillance of infectious diseases has long been recognised by the public health community and decision-makers around the world. Increased movements of people, expansion of trade, conflict and natural disasters all bring this into sharp focus. In this presentation, we will provide an overview of some of the most important aspects of surveillance, including the international legal framework. We will describe why global surveillance is considered a global good, and illustrate key features of how it enhances global public health security. We will examine the broader aspects of global influenza surveillance, and consider how it impacts on preparedness, response and recovery. In this context, the challenges of pandemic influenza risk management will be discussed, with particular attention paid to the lessons learned from the 2009 pandemic. Current and future requirements will be explored against the backdrop of human infections with avian influenza A (H7N9) and MERS-CoV.
David L. Swerdlow, MD - The OSTP, Pandemic Prediction and Forecasting Science and Technology Working Group: Update and discussion
The Pandemic Prediction and Forecasting Science and Technology (PPFST) Working Group (WG) was established by the Biological Defense Research and Development (BDRD) Subcommittee, which reports to the National Science and Technology Council (NSTC) Committee on Homeland and National Security (CHNS). The goal is to provide an effective mechanism for strengthening federal infectious disease prediction and forecasting capabilities and providing a mechanism to facilitate federal and non-federal cooperation in this important area. The working group is co-chaired by OSTP, CDC, and DoD. Activities will include a series of workshops and a pilot project that will use data from multiple sources to parameterize a model to predict the course of an infectious disease outbreak. After a brief update about the activities of the working group, the session’s discussion will allow participants to provide input on which pathogens should be modeled and what methodologies, data sources, and partnerships should be utilized for the pilot project and the activities of the working group.
Don Olson - Searching for Better Influenza Surveillance? Early experiences with Google Flu Trends at the State and Local Level
In November 2008, Google Flu Trends (GFT) was launched as an open tool for influenza surveillance. Engineered as a system for early detection and daily monitoring, GFT was developed as a disruptive technology using proprietary search data and a closed algorithm to provide a surrogate measure of influenza-like illness (ILI) in the population. The GFT system was presented as a more timely, no-cost supplement to traditional ILI surveillance. The original GFT model is the most highly cited paper in the field of syndromic surveillance or digital disease detection; numerous studies have “validated” the findings, ushering in a “big data revolution” in the field. From the beginning, however, GFT raised serious concerns among quantitative epidemiologists, disease modelers and public health authorities. These concerns were realized in the first season of prospective use when GFT missed the emergence of the 2009 pandemic, and again in the third full season of the updated model when GFT greatly overshot the 2012/2013 epidemic. Research papers and policy forums are now emerging from peer-review, dissecting the true operating characteristics of GFT. This talk will cover some the early experiences, with a focus on the use of electronic syndromic surveillance systems at the state and local level.
Edward Goldstein - Assessing the dynamics of influenza epidemics: syndromic surveillance and the role of open schools
During the first part of the talk, we examine the impact of school openings on the effective reproductive number of an influenza pandemic using the Summer/Fall 2009 surveillance data from several US states. In the second part of the talk we show how weekly web-based participatory syndromic surveillance data allows for the estimation of the incidence of influenza infections in population cohorts.
Elaine Nsoesie - Assessing Online Foodservice Reviews for Foodborne Illness Surveillance in the United States
Reports of foodborne illness submitted through online business review sites and social media have the potential to aid in foodborne illness surveillance. Based on reports submitted on Yelp.com (a business review site) from 2005-2013, we (1) assessed trends in foodborne illness reports for various cities and states, and (2) compared foods implicated in foodborne illness reports to those in outbreak reports from the U.S. Centers for Disease Control and Prevention (CDC). We noted a steady increase in the number of foodborne illness reports over time. However, this could be due to the increased usage of social media and related technologies. We also observed that foods implicated in foodborne illness reports on Yelp were similar to foods implicated in outbreak reports from the CDC. Broadly, the distribution of implicated foods across five categories was as follows: aquatic (16% Yelp, 12% CDC), dairy-eggs (23% Yelp, 23% CDC), fruits-nuts (7% Yelp, 7% CDC), meat-poultry (32% Yelp, 33% CDC), and vegetables (22% Yelp, 25% CDC). Although there are limitations to the data, we posit that if properly mined and filtered, online illness reports could complement traditional approaches to foodborne illness surveillance.
Gabriel Milinovich - Assessing performance of internet-based surveillance systems across infectious diseases
The potential for internet-based surveillance systems has increased with the growth in global internet usage and shifts in health-related information seeking behaviour. Whilst monitoring infectious diseases using internet-based data has been applied to single or small groups of infectious diseases, no study has systematically assessed the suitability of this approach for a wide range of infectious diseases of high public health importance and ranked them according to their suitability for monitoring using this approach. This study aimed to assess correlations between a wide spectrum of infectious diseases and Internet metrics for related search terms and to identify diseases for which internet-based data could be used to support early warning systems. The findings suggest that internet-based surveillance systems have broader applicability to monitoring infectious diseases than has previously been recognised. Furthermore, these internet-based surveillance systems have a potential role in forecasting emerging infectious disease events, especially for vaccine-preventable and vector-borne diseases.
Lauren Meyers - Goal-oriented design of surveillance systems
With public health facing increasing budget constraints and the explosion of internet-source data, disease surveillance is at a critical juncture. To design effective surveillance under resource constraints, we propose a four-step process that produces a surveillance system through systematic evaluation and integration of candidate data streams: 1) Formalize surveillance objectives, 2) Specify candidate data sources, 3) Simulate historical data where missing, and 4) Select the most informative data sources. This methodology quantifies the performance of individual data streams in terms of the specified surveillance objectives, and prioritizes them for incorporation into surveillance systems. In this talk, I will present the method and discuss the future of scientific-public health partnerships to advance infectious disease surveillance.
Lyn Finelli - How Might Technology-Enabled Data Sources Enhance Influenza Situational Awareness?
To accomplish pandemic and seasonal surveillance, CDC employs several systems: syndromic outpatient influenza-like illness provider network, influenza-associated hospitalizations, and mortality, and virologic surveillance. The current surveillance network was sufficient to meet the needs of decision makers during the 2009 H1N1 pandemic in the areas of defining the epidemiology of pandemic influenza, tracking national influenza activity, targeting risk groups for medical countermeasures, estimating disease burden, and characterizing changes in the virus. However, all of the current influenza surveillance systems are based on medically-attended visits, with attendant biases, and some of the systems are less timely than is optimal. It would be useful to make improvements in these areas with technology-enabled data were it robust enough to be an adjunct to traditional systems. This presentation will review some of the available technology-enabled data sources and discuss ongoing efforts to integrate these new data sources into the national influenza surveillance situational awareness.
Marisa Bargsten and Deborah Thompson - State of New Mexico: Current and Pandemic Influenza Surveillance Activities
New Mexico has a centralized department of health which directs surveillance activities across the state. Traditional surveillance is conducted for influenza using influenza-like illness (ILI) sentinel site reporting, monitoring of laboratory data, mortality surveillance, active hospitalization influenza surveillance, and US/Mexico border ILI surveillance. During the 2009 H1N1 influenza pandemic, New Mexico mobilized resources to conduct statewide active surveillance for influenza hospitalizations and deaths, while also conducting enhanced laboratory surveillance. Data collected during the pandemic was used for real-time situational awareness and to inform prevention and treatment messages as well as policy decisions. New Mexico surveillance data has been utilized for state- and national-level analyses to better elucidate and understand risk factors for severe influenza outcomes and inform policy.
Philip M. Polgreen, MD - Approaches for Aggregating Influenza Information from Healthcare Professionals
Most of the current approaches to influenza surveillance rely on the collection and aggregation of data derived from patients. Examples include data derived from microbiologic samples, death-certificate data, or healthcare-encounter data (e.g., diagnostic data related to admissions, clinic visits, or emergency department visits). Other approaches have focused on syndromic data (e.g., school absentee data, pharmacy related data). More recent syndromic approaches consider electronically available data from Internet search queries, Twitter, or from mobile devices. However, all of these approaches have strengths and limitations, and all of them still require post-processing and human interpretation of data. Inspired by work in traditional and experimental economics, we will consider surveillance approaches that can incorporate information from healthcare professionals in addition to the data streams mentioned above. We propose gathering and aggregating both subjective and objective information from a variety of healthcare professionals to supplement influenza-surveillance efforts. Finally, we will propose validation approaches.
Reid Priedhorsky - Measuring Disease with Wikipedia
Infectious disease is a leading threat to public health, economic stability, and other key social structures. Efforts to mitigate these impacts depend on accurate and timely monitoring to measure the risk and progress of disease. Traditional, biologically-focused monitoring techniques are accurate but costly and slow; in response, new techniques based on social Internet data such as social media and search queries are emerging. These efforts are promising, but important challenges in the areas of scientific peer review, breadth of diseases and countries, and forecasting hamper their operational usefulness. We propose a freely available, open data source previously unexplored for this use: access logs from the online encyclopedia Wikipedia. Using linear models, language as a proxy for location, and a simple article selection procedure, we demonstrate that these data feasibly support an approach which overcomes these challenges. Specifically, our proof-of-concept yields models with $r^2$ up to 0.92, forecasting value up to the 28 days tested, and several pairs of models similar enough to suggest that transferring models from one location to another without re-training is feasible. Based on these preliminary results, we close with a research agenda designed to overcome these challenges and produce a disease monitoring and forecasting system that is significantly more effective, robust, and globally comprehensive than the current state of the art.
Sara Del Valle - Measuring influenza with the Internet: A dim present and bright future
Infectious disease surveillance systems typically rely on clinical records to monitor disease activity. However, non-traditional streams, such as social media, have remained absent from such systems due to the scientific challenges associated with them such as volume, velocity, and specificity. In this talk, I will describe some of the advantages and limitations of Internet systems and how they could be used towards influenza surveillance activities. Specifically, I will describe how Twitter and Wikipedia can provide actionable health information that may be used to monitor and forecast influenza spread.
Scott Epperson - Enhancing legacy influenza surveillance systems
In light of current efforts to expand the timeliness and geographic coverage of influenza surveillance data using new technologies, we will review the enhancements to legacy influenza surveillance systems since the 2009 H1N1 pandemic which incorporated lessons learned and the availability and emphasis on alternate methods of data collection and transmission.
Stephen Eubank - Data fusion and novel data sources: lessons learned
Data, sophisticated machine learning algorithms, and raw computing power are so plentiful that many think we've entered a brave new world of surveillance. But the main stumbling blocks of data analysis -- dealing with noisy, nonstationary data, the curse of dimensionality, the ease of overfitting, and the difficulty of collecting longitudinal data and representative samples -- have not magically disappeared. Awash in an ocean of data, we still must learn to use it efficiently. One approach supplements data about the outcomes of interactions with procedural knowledge of how things interact. I will draw lessons from attempts to fuse diverse data sources sampled at different spatiotemporal resolutions into a self-consistent, harmonized form that supports both hypothesis-based situation assessment and counterfactual ("what-if") course-of-action analysis.
Todd Bodnar - The next generation of individual-based online surveillance
Social media has been considered as a data source for tracking disease. However, most analyses are based on models that prioritize strong correlation with population-level dis- ease rates over determining whether or not specific individual users are actually sick. Taking a different approach, we develop a novel system for social-media based disease detection at the individual level using a sample of professionally diagnosed individuals.