Assessing risks associated with infectious diseases using remotely sensed data
This post is co-authored by Tiara Jones and Ann Facey
Outbreaks happen — but the spread and impact of the Zika virus surprised us
Pathogens (e.g., influenza, SARS, Ebola) can emerge and spread globally very quickly, often before the routes of disease transmission and methods of prevention are identified. The rapid global spread of the Zika virus [2] is a recent, frightening example because it can cause significant neurological disorders (e.g., microcephaly) in newborns, among other presentations. A protective vaccine is not currently available for the Zika virus and likely will not be for some time. Without these protective vaccines, limiting people’s exposure to the pathogen is critical.
To limit exposure, public health officials need to know (1) what is the major route of transmission, (2) who is at risk of becoming infected, symptomatic and infectious, (3) where they are, (4) what geographic areas are at risk, and (5) what can be done to mitigate the risk. For vector-borne diseases, understanding the vector — mosquitoes in the case of the Zika virus — will help delimit where and when transmission risks are the greatest.
Remotely sensed data can help assess infectious disease risks
Knowing where mosquitoes live tells us where transmission is likely to occur, and where response efforts need to be focused, such as risk communication. From an ecological perspective, every organism fills a niche in an ecosystem that is defined by biological (e.g., what it eats, what eats it, who it competes with for resources) and physiological constraints (e.g., temperature, moisture). Fortunately, remote sensing can and has been used reliably to collect environmental data globally. Therefore, data pertaining to the physiological constraints of a disease vector, in this case the Aedes aepyti mosquito and other competent vectors, can be collected via remote sensing and used to map the geographical areas and seasonal periods that are conducive physiologically to vector survival.
Correlating environmental data with known data on the presence and absence of mosquitoes allows one to build models that can be used to predict environments where one would expect to find the mosquitoes [3]. This process of correlating vector presence with environmental variables is called ecological niche mapping or disease mapping. Disease mapping was used effectively to define risk areas during the recent outbreaks of the Ebola virus disease [4] in West Africa and the Zika virus [5] in the United States.
This is where Weather Analytics comes into the picture. Weather Analytics has built a web-enabled global database with more than three decades of gap-free historic, current, and forecast atmospheric and soil information. Spatial resolutions range from 10 meters to 30 kilometers, depending on data type and location. Products leverage hundreds of proprietary algorithms to draw from more than 62 trillion data measurements worldwide. They feed the Weather Analytics decision-support tools — including Beacon forecasts, Dexter post-event forensics, and Gauge risk assessments. Weather Analytics provides its customers with answers to weather and non-weather related questions for a range of applications. For example, Weather Analytics scores probabilistic likelihood of natural hazard and structural risks. Insurance underwriters utilize these scores to select and price risks, ranging from hail and earthquakes to worn roofs and crop failures.
What we did — a pilot project on anticipating disease risks
Given this interesting capability to provide weather data and analytics at relatively fine spatial resolution, B.Next became interested in exploring if Weather Analytics could assess and map risks associated with vector-borne diseases such as those associated with the Zika virus. So, B.Next partnered with Weather Analytics to develop a proof-of-concept for mapping the risk of the spread of Zika virus in parts of South America.
For this project, Weather Analytics developed statistical models that correlated historical weather and non-weather variables with incidence of disease reported by various health organizations. These models synthesized the numerous possible predictive variables into key risk covariates for different geographies. The models provided statistical estimates for the risk of spread of the Zika virus in the near-term future (3 months from the study start date) within targeted geographies (Ecuador, Colombia, Guyana, Suriname, Rio de Janeiro, and Venezuela). The models incorporated historical disease occurrence data, demographic data, data representing socioeconomic conditions, mosquito population data, and climatological data, together with epidemiological and statistical principles. Final risk scores were derived from the statistical model output and categorized from low (1) to high (3), indicating how favorable conditions are for the spread of the Zika virus. Final risk forecasts were visualized in a web-enabled dashboard with access to the top predictors for each country (see figure below). As such, the statistical modeling and visualizations used by Weather Analytics demonstrated an approach to anticipate areas of high exposure risks.
How well did we do?
Weather Analytics assessed how well their models performed by comparing their forecasts against actual risk assessments based on case reports once they were available (Table 1).
Overall, the forecasts did a fair job; however, there were significant discrepancies in two locations, Suriname and Rio de Janeiro. Quantity and quality of data undoubtedly had an impact on the forecast results because there were few case reports from Suriname over the forecasted period. Improving the data is a perpetual challenge and consistent need across many public health jurisdictions. Incidentally, improving data and analytics to improve outbreak responses is a core focus for B.Next.
Placing these forecasts into a public health operational context will be key in any future iterations. Understanding what would be sufficient for a public health locality to issue a warning using these forecasts will be needed to ensure usability for future efforts. While this proof-of-concept resulted in some discrepancies, it offered a glimpse into a possible tool that can be used to develop disease risk models, map them onto a user friendly interface, and provide data-driven decision making during epidemics.
B.Next is designing a biodefense technology strategy, demonstrating the potential that innovative tools and techniques can provide, and supporting the investment strategies of these innovations.
Check out our work at www.bnext.org and follow us on Twitter: @_b_next
References:
[1] Hay SI, George DB, Moyes CL, Brownstein JS (2013) Big Data Opportunities for Global Infectious Disease Surveillance. PLoS Med 10(4): e1001413. doi:10.1371/journal.pmed.1001413
[2] http://www.who.int/emergencies/zika-virus/timeline/en/
[3] As an example see, Carney et al. Early Warning System for West Nile Virus Risk Areas, California, USA. Emerging Infectious Diseases. Vol. 17, №8, August 2011.
[4] http://www.healthdata.org/data-visualization/spatial-epidemiology-ebola-virus-disease
[5] Bogoch, I. I., Brady, O. J., Kraemer, M. U., German, M., Creatore, M. I., Kulkarni, M. A. et al. (2016). Anticipating the international spread of Zika virus from Brazil. The Lancet.
