Linking fine scale mobility and dynamic contacts to understand the spatial dimension of pathogen transmission Gonzalo Vazquez-Prokopec Department of Environmental Studies and Global Health Institute, Emory University. Fogarty International Center, National Institutes of Health. gmvazqu@emory.edu www.prokopeclab.org Ohio State University September 18, 2012
Outline Dimensions of human movement - Scales of movement - Movement and infectious diseases - Methods for quantifying movement - Importance of resource-poor settings Case studies - Human influenza virus - Dengue fever - Cryptosporidium spp. Conclusions
Human mobility and social contacts Significant factor determining disease dynamics and pathogen mixing, propagation, and evolution. Eames & Keeling, PNAS 2002
Dimensions of movement and transmission Human movement determines individual exposure and ID transmission dynamics Stoddard et al. 2009
Local movement Key scale for understanding exposure and local spread e.g. Within-city dynamics. Poorly studied due to data scarcity. Recently: Cell-phone and GPS. Eubank et al. 2004. Science Models of ID epidemic spread and containment What if…? Gonzalez et al. 2008. Science
Quantitative approaches Statistical vs mathematical. Contiguity analysis Distance from point i Statistical Network spatial (probability of characterization by analysis movement i-j) groups Lattice models, cellular Diffusion, dispersal Meta-population Network dynamic automata kernels models, gravity models models Analytical and computational complexity Adapted from Riley 2007. Science
Spatial GIS Analysis Modeling RS GPS
Case studies: -Dengue fever - Human influenza virus -Cryptosporidium spp.
Dengue epidemics: explosive and widespread Neff et al. 1967. American Journal of Epidemiology
Space-time analysis & modeling Spatial heterogeneity driven by housing type (queenslander) PP MN WC MB 250/383 cases (65%) belonged to 18 space-time clusters Slow local propagation Speed of virus due to propagation: mosquito flight Kernel of transmission Fast circulation due to human movement
Need for spatially explicit consideration of exposure and transmission patterns
Human movement and dengue transmission in Iquitos, Peru Immediate Goal of the Study Determine the locations most visited by participants and assess the risk of acquiring dengue in such locations Collaborators Thomas W. Scott, Amy Morrison, Steven Stoddard – UC Davis John Elder – San Diego State Valerie Paz Soldan – Tulane Gonzalo Vazquez-Prokopec, Uriel Kitron – Emory Tad Kochel – NMRCD (US Navy- NAMRU) Funded by NIH/NIAID
Human spatial behavior • Human geography, behavioral sciences, neurosciences, physics, mathematics • Activity space: the local areas within which people move or travel in the course of their daily activities • Can be used to summarize routines • Represented by Nodes (locations) connected by Paths (routes)
Study design • Two neighborhoods (Maynas and Tupac Amaru) M T • Pilot study: 120 participants • Final study: 2,500 participants • Retrospective Activity Space (Cluster Studies): Characterizes key locations visited when infected • Prospective Activity Space : Key locations visited in an individual’s monthly routine
Challenges in the estimation of activity spaces • Traditional methods: direct observations, diaries and interviews • Issues of recall, reliability, reproducibility, compliance, behavioral change, and privacy • Alternative: use of GPS technology – Used in the past – Costly and technology challenging
Using GPS to track human movements Key features: memory and battery life; durable and tamper-proof; light weight; design widely accepted by participants; little to no maintenance required of participants; low cost ($50). Accuracy: 4-10 m
Strong input from social scientists Pamphlet developed to provide information about GPS
What data do we obtain from GPS? • Latitude, longitude, date, time, elevation Participant A Participant B 61 participants
Representing movement data Contact networks – nodes represent individuals (or locations), links represent relationships allowing pathogen transmission Bipartite & spatial topology Vazquez-Prokopec & Bansal, in prep.
Mobility Networks Distribution of number of potential contacts a person has due to mobility Vazquez-Prokopec et al. Submitted.
Human mobililty and dengue transmission Sampled contact networks: - DENV-4 positive - DENV-4 negative Two seasons. IC Balanced design (positive and negative clusters) Stoddard et al. submitted
Human mobility influences dengue transmission DENV + DENV - Attack rates (# infected / total tested) and infestation (proportion of houses with ≥1 infection) significantly higher in networks of infected individuals Stoddard et al. submitted.
Tracking ~600 individuals with GPS GPS: latitude, longitude, elevation, time. 2,500,000 data points. Sample balanced between ages and sexes. Goal: estimate mobility parameters and dengue exposure. Vazquez-Prokopec et al. in prep.
Characterizing mobility parameters Mobility kernels (Pb of moving m meters from home Δd) Bank notes (US) Brokman et al. Nature. 2006 ? (m) Most ~75-82% movements within 1km from home. Cell phone Males: higher (Europe) probability of moving further than (m) females Gonzalez et al. Nature. 2008
Characterizing mobility parameters (2) Number and type of locations People routinely Transportation (US) visited, on average, 4-6 locations. Ages 36-45 had the highest dispersion in their ? visitation patterns. Eubank et al. Science 2004 Most places visited Transportation models were residential, predict people visit 2-4 followed by locations and a commercial maximum of 14 (stores, markets). locations.
Dynamic movements and contacts Song et al. Science 2010 Cell phone data shows people in developed cities Individuals present highly dynamic and follow highly unstructured routines. structured routines
Dynamic movements and contacts
Unstructured routines and influenza dynamics Agent-based stochastic model including “mobility rules” -Distance -Number of locations -Type of location Structure of routines-> μ -Distribution of hours/visit μ = 1 -> highly unstructured μ = 4 -> highly structured Iquitos μ=3 10,000 people & 4,000 locations
Simulating movements Model: Tractable contact structure. 50 simulations running hourly contacts and lasting for 200 days. Run in net-logo (first) and python (now). Perkins et al. in prep Bisanzio et al. in prep.
Modeling influenza dynamics The consideration of temporally unstructured routines increased both an epidemic’s final size and effective reproduction number by 20% in comparison to models assuming temporally structured contacts.
Cryptosporidium spp. spillover Introduction of human Crypto into Gombe Chimpanzee populations (Tanzania). Sources: -Humans -Domestic animals (dogs, goats, sheep) -Environmental overlap in crop-raiding areas -Drainage networks. Are areas of intense activity potential spillover hot-spots?
Are goats the “vectors”? Is the crop-raiding area the spillover hot-spot? How does the landscape impact persistence and transmission? Parsons et al. in prep
Conclusions • Empiric consideration of fine-scale mobility can help unveil inherent complexities of pathogen transmission (esp. in resource-poor settings). • GISscience can complement other quantitative methods, but will require adaptation to new challenges (big data, dynamic visualization and simulation). • Integration of good data and proper models allow: - Identifying sources of infection. - test impact of interventions (location/people) - Gain deeper eco-epidemiological understanding Multi-disciplinary approach
Acknowledgements Cairns: Funding: Scott Ritchie, Peter • Emory University Horne, Jeffrey Hanna, Brian Montgomery • NIH/NIAID Iquitos: Movement team; phlebotomists; entomologists; GIS/data entry; Iquitos residents
Recommend
More recommend
