Landscape Epidemiology Application

EXPLORING REMOTELY DETECTED ENVIRONMENTAL FACTORS ASSOCIATED WITH DISEASE OUTBREAK AND SPREAD, THEREBY IMPROVING OPPORTUNITIES FOR DISEASE CONTROL AND PREVENTION.
 

Traditional approaches to providing operational commanders assessments of disease threats to forces in garrison and in theater rely on historical data or syndromic surveillance. Historical data doesn’t reflect current conditions or may be too coarse to be of any practical use. Syndromic surveillance is close observation for emergence of a group of symptoms that collectively indicate or characterize a disease or abnormal condition within a population. This approach is hampered by inaccurate and late reporting of symptoms by medical and public health functions. In addition, syndromic surveillance forces a delayed response as one cannot react to a disease outbreak until symptoms emerge in the population.

A wealth of sensor data collected by orbiting satellites is available today from commercial and government sources. This information is accompanied by a vast body of knowledge on how to use this data to model numerous on-ground conditions. Part of this growing research explores remotely detected environmental factors associated with disease outbreak and spread, thereby improving opportunities for disease control and prevention.

A team consisting of Weston Solutions Inc., Adolos Strategic and INCELL Corporation performed for the Air Force Center for Environmental Excellence a proof-of-concept demonstration using remotely sensed data to associate mosquito habitat and mosquito-born disease transmission risks in an interactive geographical information system. The resulting prototype system, Landscape Epidemiology Application (LEAP), provides a mosquito-borne disease risk index for a scaleable geographical area within a given time period. This preemptive epidemiological intelligence will enable commanders to assess disease risks endemic to a deployed area and take appropriate mitigation actions or assess conditions while in the field and react if factors warrant.

LEAP was developed with two tiers of users in mind. The first are headquarters or theater-level epidemiologists, planners and analysts with access to significant computing resources. These users would employ the application for deliberate and contingency planning, casualty estimates for simulation or training purposes, or other medical intelligence needs. The members of the second user group are deployed epidemiologists or medical intelligence officers who are tasked to provide localized epidemiological intelligence to operational commanders anywhere, any time. These personnel would be operating in austere environments with limited computing resources. They will benefit from an intelligent decision aid providing immediate risk assessments based on local observed conditions combined with remotely sensed observations.

To develop the predictive capability for LEAP, research was derived from research meta-data and Bayesian inference models. Bayes’ theorem is a branch of mathematical probability theory that models uncertainty about the world and outcomes of interest by combining common-sense knowledge and observational evidence. Applying Bayes’ theorem, expressed a degree of belief in the possible values a parameter takes as a probability distribution or a deterministic variable based on the inferences made regarding conditional probabilities derived from observed data.

The LEAP Bayesian inference models defined the confidence or belief of the risk of being exposed to a mosquito-born disease in the context of observed environmental evidence derived from orbiting satellites. These models also have the benefit of being readily converted into an algorithmic form suitable for implementation in an operational system. Furthermore, Bayesian inference emulates closely the cognitive processes by which risk estimates are now made by military epidemiologists.

The analysis determined three remotely sensed environmental factors were the most reliable predictors of mosquito habitat and mosquitoborne disease risk: vegetation index, land surface temperature, and land use/land cover. One of the best predictors was the Normalized Digital Vegetation Index (NDVI). NDVI is a measure of photosynthesizing vegetation and incorporates several other environmental factors such as precipitation, land uses, and temperatures. In this application, instead of the NDVI, the Enhanced Vegetation Index (EVI) was used. The EVI lessens the impacts of cloud cover and other atmospheric interferences suffered by the NDVI.

In addition to the EVI, land surface temperature (LST) provided additional knowledge under certain conditions. The LST is a general index of the apparent environmental temperature (soil or vegetative) and radiometric surface temperature. It corresponds to temperatures observed from ground meteorological stations, but has the advantage of being available for many more geographical locations. Finally, it was discovered that knowledge of a land cover or land use further improved the risk estimate given prior knowledge of EVI and LST.

Commercially available data was obtained to derive the three landscape variables from the NASA Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard the Terra satellite. The Terra MODIS instrument views the entire surface of the earth every one to two days. Data was imported and accomplished the necessary transformations for Belize, Kenya, Democratic People’s Republic of Korea, Mexico and China for time composites during June 2005.

The Bayesian inference model requires a prior probability of disease risk. The national prevalence or the endemic rate of mosquito-borne diseases in the geographical location given limited or no other pertinent information was used. Disease prevalence data was obtained from sources such as the World Health Organization and the Mapping Malaria in Africa project. Despite many influencing factors, research makes it entirely reasonable to assert that the higher the disease prevalence in the surrounding human population, the more likely it is that a non-infected person among or near that population may be exposed to the disease if the conditions are favorable for the vector.

LEAP will enable operational commanders to proactively assess disease risks endemic to a deployed area and take appropriate mitigation actions or assess conditions while in the field and react if factors warrant. The relevant operational question then is what are the chances a significant number of persons in a deployed force will be exposed to a disease? Using accepted definitions for levels of endemic disease prevalence provided a basis to derive operationally significant categories for visual presentation in terms understandable by non-scientific users of the information. This allowed the conversion of the continuous risk index into more meaningful categories related to appropriate actions. These are illustrated in the table on this page.

Another practical aspect of the application is the time frame to which the risk estimates apply. LEAP must allow a user to make reasonable estimates within a future period of time in which operations may occur. Typically this can range from immediate to 60 days or more. The environmental factors used in LEAP lag one to two months from the time of the observed value to the time the applicable risk category would apply. Therefore, the risk categories are applicable to forces in that location for 30 to 60 days of the composite time period the remotely sensed data was derived. If operations occur beyond that time, a user would need to obtain new composite values to update the risk for that geographical location, or rely on the previously determined risk estimate.

One of the more attractive features of LEAP is the resolution of the risk estimates available to a user. During development it was necessary to select a reasonable resolution corresponding to operational considerations and the assumptions applied to interpretation of the results. The data allows risk estimates at a scale of less than 30 meters. However, since an operational decision support tool was being developed, a resolution linked to actions was needed. It was concluded it unlikely a commander would alter his/her decisions or actions in regards to force protection based on resolution of 30 m. To illustrate, only a platoon-sized unit would operate within this small an area, and even so, would be constantly moving from area to area at this resolution. A more appropriate minimal operational area for which force protection decisions would be made is a 200 meter to one kilometer resolution. Therefore, a nominal scale of one pixel equals 250 meters was selected as the spatial resolution.

To transform the above meta-data and Bayesian inference models into an interactive geographical information system, the Leica Geosystems ERDAS IMAGINE Professional, Versions 8.7 and 9.0., was used. IMAGINE includes a knowledge engineer that aided incorporation of the Bayesian inference models into the system. LEAP includes a distributed Web interface that utilizes an interactive mapping environment to display knowledge engineer data outputs in a user-friendly layout, and allows the user to modify parameters specific to the occurrence of the disease in a particular region.

The software suite consists of ESRI ArcIMS 9.1 (Internet map server), ESRI ArcSDE 9.1 (spatial database engine), and Oracle Database 10g. Oracle Database 10g is a back-end software product used to store and maintain the DEWS image data. ArcIMS 9.1 is a server-based product that provides the necessary framework for distributing LEAP risk data over the Web. The ArcIMS Website contains functionally that allows the client to navigate and query risk image data through Microsoft’s Internet Explorer (5.5 or greater) internet browser. ArcSDE is a server-based middleware, which connects the LEAP image data stored in Oracle’s 10g relational database management system to the ArcIMS website. The ArcSDE interface allows for the advanced and efficient management of LEAP image data.

LEAP applies advanced environmental data computing methods to improve decision making and epidemiological analysis in the face of rapidly evolving WMD and naturally occurring pathological threats in military and civilian sectors. The application also has obvious advantages to public health initiatives such as inoculation programs and insecticide application schedules.

LEAP provides an alternative to current disease threat assessment methods by providing a safe and systematic approach for identifying specific risks based on well-known and readily accessible remotely acquired data. Additional benefits are its abilities to quantify degree of risks, produce answers at a higher resolution than current methods, and produce answers more quickly than conventional methods. The tools are scalable so “what if” scenarios can be performed immediately by users at multiple levels. Furthermore, the models and knowledge engineer can be easily modified to accommodate changing input variables, data sources, and risk interpretations. ♦

Back_To_Top