Future Directions in Analytics for Infectious Disease Intelligence Toward an Integrated Warning System for Emerging Pathogens
Total Page:16
File Type:pdf, Size:1020Kb
Science & Society Future directions in analytics for infectious disease intelligence Toward an integrated warning system for emerging pathogens Barbara A Han1 & John M Drake2,3 merging infectious diseases are among our vigilance in order to better respond to about future risk and epidemic trajectories, the most destructive and costly natu- the next outbreak so that only a handful and characterizing possible losses under a E ral forces [1]: In terms of human and people fall ill instead of hundreds, thou- range of intervention scenarios. Infectious monetary losses, epidemics and pandemics sands, or tens of thousands. Our ability to disease intelligence therefore relies funda- rank with other major natural disasters, put out these proverbial fires has indeed mentally on data from multiple sources to such as earthquakes or tsunamis. And like become formidable over time, but it is still provide a stream of information that can be earthquakes and tsunamis, much of the reactive. The outbreak of Ebola in West inspected by modeling and real-time analyt- destructive potential of infectious diseases Africa did not become a worldwide ics to make decisions about prevention, stems from the fact that they often strike pandemic, but it nonetheless wreaked havoc surveillance, or emergency responses to unexpectedly, leaving little time for prepara- in West Africa. A total of 26,000 humans outbreaks. tion. The best countermeasure is therefore were infected, 11,300 died, and the outbreak ...................................................... an early warning to give affected regions or caused losses of about US$2B in the short communities more time to prepare for the term [2] with up to US$15B in estimated “...like earthquakes and impact. After the devastating earthquake losses to investment, trade, and tourism tsunamis, much of the and tsunami in the Indian Ocean that killed over the next couple of years. There are also destructive potential of 230,000 people in December 2004, the concerns about the long-term knock-on Indian Ocean Tsunami Warning System was effects on the political and economic stabil- infectious diseases stems from installed in 2005 and became operable in ity in Guinea, Liberia, and Sierra Leone [1]. the fact that they often strike 2006: It demonstrated its value after the unexpectedly, leaving little Banda Aceh earthquake in 2012 when it more efficient approach to emerging time for preparation.” alerted the affected islands within minutes infectious disease threats would be ...................................................... of the danger. There are some systems for A anticipatory: responding to disease tracking infectious diseases, such as CDC’s risk rather than occurrence by managing What data and analytics are most PulseNet that monitors disease outbreaks and reacting to the ebb and flow of risks in urgently needed to prepare for spillover across the USA or the global Influenza real time. Such a strategy would maintain from animal reservoirs and subsequent Surveillance and Response System, but these vigilance while simultaneously assessing spread of infectious diseases? For what are focused on particular geographic areas vulnerabilities: identifying where disease populations and regions should these be or on specific diseases. As new diseases risk is high, and providing decision support collected? The answers to these questions emerge and old diseases re-emerge, as analysis (see Sidebar A) to identify which vary according to where a particular infec- pathogens and their vectors are transported actions could prevent outbreaks or contain tious disease falls along a continuum of risks worldwide through trade and travel, it is epidemics at the outset. (Fig 1). To guide the collection of intelli- now time to improve global warning Anticipating and responding to disease gence, we envision the riskscape—the distri- systems for emerging infectious diseases in risk requires interpreting disease events— bution of risk in space—that consists of general. outbreaks and epidemics—as emergent three threat levels. At risk level I (yellow), To mitigate the threat of infectious properties of a complex system from which there are no detectable human cases, diseases, our main strategy so far has been a to gather infectious disease intelligence. although there may be sources of infection strong defense after emergence. Once an The production of intelligence involves iden- in proximity to human populations. At risk outbreak is under control, we improve tifying actionable and biologically meaning- level II (orange), human cases of an infec- infrastructure, develop vaccines, and refine ful data patterns, developing predictions tious disease have been verified. At risk 1 Cary Institute of Ecosystem Studies, Millbrook, NY, USA. E-mail: [email protected] 2 Odum School of Ecology, University of Georgia, Athens, GA, USA 3 Center for the Ecology of Infectious Diseases, University of Georgia, Athens, GA, USA DOI 10.15252/embr.201642534 ª 2016 The Authors EMBO reports 1 EMBO reports Infectious disease intelligence Barbara A Han & John M Drake I WATCH II WARNINGIII EMERGENCY PREVENTATIVE MANAGEMENT INTERVENTION AND CONTAINMENT CONTAINMENT AND DAMAGE CONTROL Human infection has not been detected, Spillover infection has occurred. Human Disease outbreak threatens to overrun but infection sources are present. existing efforts to control spread and have Management efforts aimed at preventing Management focused on human the potential to exact high levels of human spillover from latent sources. intervention. morbidity and fatality. Emergency responses are mounted for damage control. AUDIENCE AUDIENCE AUDIENCE • Funding agencies • Health agencies • Decision makers • (WHO, CDC, HHS, …) • Planners • Vaccine developers • • Logistics • Reinsurance • The public • Researchers ANALYTICS ANALYTICS ANALYTICS • Disease risk mapping • Transmission dynamics • Forecasting • Data mining • Phylodynamics • Scenario analysis • Statistical modeling RISK DATA DATA DATA HUMAN POPULATION • Outbreak investigations • Interventions • Census data • Seroprevalence surveys in • Case counts • Business data humans and animals • Genetic sequences CTACTGCCGTACC • Transportation data • Transmission pathways • Resources (e.g., • Vector monitoring (e.g., • Mobility equipment, lab facilities, etc) (e.g., call data records) … … ! • Resistance × ? × × ANIMALS × • Wild animal testing ! … ! • Livestock surveillance ……… ? ENVIRONMENT ? • Climate/weather … HUMAN DATA JFMAMJJASOND HUMAN DATA ORGANISMAL DATA ENVIRONMENTAL DATA SOCIETAL RESPONSES SOCIETAL CONDITIONS (e.g. poverty) Figure 1. A model for a global warning system for infectious diseases. 2 EMBO reports ª 2016 The Authors Barbara A Han & John M Drake Infectious disease intelligence EMBO reports level III (red), the number of human cases is uring the watch phase, we are chains (prevention), developing vaccines, or growing so large that it pushes the limits of primarily concerned with assessing improving facilities to better respond to spill- disease control. D the baseline risk of spillovers from over events in high-risk areas. Such capacity- wild animal sources (disease reservoirs) into building would also add value to ongoing he data and modeling required to humans. This would include knowing, for pathogen reconnaissance projects and many assess this riskscape are analogous to example, which reservoir and vector species investigator-initiated research programs T that needed for predicting extreme occur in an area, and what zoonotic infec- across a productive and globally distributed weather events or wildland fires. The risk of tions they are known to carry [4]. While scientific community. a wildfire is quantified at various scales, many of these pathogens may not pose an updated and tracked through time, and coor- immediate risk, particularly if there is mini- t the warning region of the riskscape dinated actions are executed in response to mal contact between humans and wildlife (risk level II), we are primarily analysis of data from multiple sources [3]. A in this area, quantifying the underlying A concerned with generating predic- fire watch is assigned based on the risk that, zoonotic potential is analogous to empirical tions that inform ministries of health and if a spark occurs, a fire will catch. If a fire horizon scanning to assess conditions that other responders such as the WHO or medi- has already started, a warning is issued with would favor the development of a wildfire. cal NGOs to react and respond to a disease continued vigilance and control actions Quantifying zoonotic potential would inform that has already emerged in a human popu- based on the fire’s speed and spread. If the management practices and developments lation. Interventions during the warning fire gets out of control, emergency measures that could disturb regions with high but phase focus on reducing transmission and are executed and a coordinated response unrealized risk of spillover infections. mitigating human mortality, economic costs concentrates on containment and damage To identify conditions that would favor of treatment, and lost productivity. Modeling control. To apply this analogy to infectious the emergence of an infectious disease, in the warning phase therefore needs to diseases, a watch may be assigned based on statistical (machine) learning algorithms address various objectives. For example, empirical quantification of the potential of (see Sidebar A) that are trained on a wide mathematically modeling transmission dyna- zoonotic events