The AFHSC-Division of GEIS Operations Predictive Surveillance Program: a multidisciplinary approach for the early detection and response to disease outbreaks

Abstract

The Armed Forces Health Surveillance Center, Division of Global Emerging Infections Surveillance and Response System Operations (AFHSC-GEIS) initiated a coordinated, multidisciplinary program to link data sets and information derived from eco-climatic remote sensing activities, ecologic niche modeling, arthropod vector, animal disease-host/reservoir, and human disease surveillance for febrile illnesses, into a predictive surveillance program that generates advisories and alerts on emerging infectious disease outbreaks. The program's ultimate goal is pro-active public health practice through pre-event preparedness, prevention and control, and response decision-making and prioritization. This multidisciplinary program is rooted in over 10 years experience in predictive surveillance for Rift Valley fever outbreaks in Eastern Africa. The AFHSC-GEIS Rift Valley fever project is based on the identification and use of disease-emergence critical detection points as reliable signals for increased outbreak risk. The AFHSC-GEIS predictive surveillance program has formalized the Rift Valley fever project into a structured template for extending predictive surveillance capability to other Department of Defense (DoD)-priority vector- and water-borne, and zoonotic diseases and geographic areas. These include leishmaniasis, malaria, and Crimea-Congo and other viral hemorrhagic fevers in Central Asia and Africa, dengue fever in Asia and the Americas, Japanese encephalitis (JE) and chikungunya fever in Asia, and rickettsial and other tick-borne infections in the U.S., Africa and Asia.

Figure 1

AFHSC-GEIS Predictive Surveillance Program Boxes represent model components: Green – eco-climate, yellow – vector, orange – animal, red – human. Horizontal arrows represent communications and feedback: blue – inter-component communication and coordination, orange – collaboration on generated data and information, red – progression of certainty between reports and feedback for program priorities. Vertical arrows represent component sequence of activities and production of reports.

Figure 2

Predictive Surveillance Model. Black arrow – timeline for a hypothetical vector borne, zoonotic disease outbreak. The outbreak progresses from left to right. Green bars – natural events in an outbreak’s emergence. Blue bars – the predictive surveillance program activities. Red bars –predictive surveillance program products. The blue and red bars together represent surveillance efforts that detect and follow the progression of the outbreak’s emergence over time. The sequence over time of the green, blue and red bars is set in the model’s configuration, but the spacing between bars varies, according to the outbreak’s pathogen etiology and location.

Figure 3

Land Surface Temperature (LST) Anomaly for Eastern Africa, December 2009. Blue – cooler than average land surface temperatures. Yellow/orange/red – warmer than average LST.

Figure 4

AVHRR-N17 NDVI Anomaly for Eastern Africa, December 2009. Green – greater vegetation growth as a surrogate for sufficient and prolonged rainfall. Brown – less vegetation growth signaling less than normal rainfall.

Figure 5

Global Sea Surface Temperature (SST) Anomaly for December 2009. Yellow and orange – warmer than normal SST in the equatorial Pacific and Indian Oceans. This is characteristic of a warm ENSO (El Niño) event.

Figure 6

Global Outgoing Longwave Radiation (OLR) Anomaly for December 2009. Blue areas - negative OLR anomalies in the eastern Pacific and Indian oceans and in east central Africa are indicative of higher than normal rainfall in these areas. Yellow to orange – lower than normal rainfall.

Figure 7

Cumulative Rainfall Anomaly for Eastern Africa, Sept. 1 to Dec. 31, 2009. Blue and green – Rainfall totals of between 100 and 300 mm above normal being detected in nearly all of the region except in north and central Kenya where November rains were light.

Figure 8

Cumulative Rainfall Totals for Marigat, Kenya. The two most recent El Niño seasons are represented in green – 1997-1998, and blue – 2006-2007. Red – the long-term rainfall mean. Purple – rainfall amount during the 2009-2010 season through Dec. 31, 2009. Not shown – heavy rainfall occurred during the last week of December 2009 at this site and others throughout the central Rift Valley.

Figure 9

The AFHSC-GEIS and NASA Predictive Surveillance Advisory, August 2009. Predictive surveillance advisories are sent to AFHSC-GEIS Predictive Surveillance program partners, DoD public health authorities, and other national and international organizations. The advisories also are publicly shared on an open-access website (ftp://rvf:geis@pengimms.gsfc.nasa.gov) when eco-climatic events and trends suggest that disease outbreaks may arise in the coming several months. The August 2009 advisory was not geographically limited because the expected El Niño conditions were expected to have global impact.

Figure 10

Predicted Distribution of Cx tritaeniorhychus in the Republic of Korea. Red and orange – higher probability of Cx tritaeniorhychus occurrence. Blue – lower probability of Cx tritaeniorhychus occurrence (corresponds to higher elevations). Purple and white dots – mosquito collection sites. White dots were used for model building (training points) while purple dots were withheld from the model and used for testing the accuracy of the model. This work demonstrates that cropland conditions, primarily rice fields and minimum temperature in the summer, are positive factors in predicting this mosquito’s distribution; whereas forested hills and mountains >1000 m elevation, are negative factors.

Figure 11

Overview of MosquitoMap. A client or prospective data provider accesses the web server hosting MosquitoMap. A downloadable spreadsheet is available with instructions regarding data requirements. Data can be submitted via email. A link is also available to the map viewer hosted on an application server running ArcGIS Server 9.3. The client can map and search the geodata-base, or use the Mal-area calculator (MAC) to quantify the overlap of models of vector, disease and human distribution. Output is available to the client in various formats. Figure reproduced from Foley et al. (2010)

Figure 12

Distribution of 2009 CHIKV Cases Showing a Clear Concentration of Cases in the Southern Peninsula. Data reflects suspected cases (per 100,000 province inhabitants) reported by the Bureau of Epidemiology, Ministry of Public Health, Thailand. Data available at http://203.157.15.4/chikun/chikun/situationy52/chikun_20091231520.pdf (in Thai).

Figure 13

Inspection of Water Cisterns in Southern Thailand as Part of a Chikungunya Virus Outbreak Investigation. AFRIMS personnel are collecting mosquitoes throughout Thailand as part of a niche modeling effort of CHIKV vector distributions. Vector density and distribution are the likely key drivers of the restricted distribution of CHIKV cases observed in Thailand in 2009.

Figure 14

Sand Fly Surveillance in Kenya. Traps are set on the base of a termite mound in Marigat, Baringo district, Kenya. Termite mounds are one of the natural habitats for Phlebotomus martini, a vector of visceral leishmania in East Africa.

Figure 15

Map of Kenya Showing Surveillance Sites: Arthropod Vectors and Animal Hosts. Lieshmania major, a protozoan parasite causing zoonotic cutaneous leishmaniasis, has been detected in Isiolo in the central part of Kenya and Lamu along the northern coast. The visceral leishmaniasis vectors, Phlebotomus orientalis, and P. martini were detected in Garissa in Kenya’s Northeast Province and in the Rift Valley village of Marigat respectively. Hantaviruses have been found in both Marigat and Garissa.