Software item


Epidemiology team

Monday, April 17, 2017

Binding epidemiology to visual elements: spatio-temporal presentation of EMOD simulation outputs

This year, IDM is excited to debut our Vis-Tools demo during the 5th Annual Disease Modeling Symposium.
If you are trying to understand the mechanisms driving the epidemiology of disease transmission, such as  over population centers and settlements of varying sizes,  through environments with varying infection risk, or those connected by the complex patterns of migrating people, pathogens and vectors, you are probably pouring over many static and dynamic plots for hours. When that is coupled with the variety of spatio-temporal intervention distributions, arriving at a coherent picture of effect-sizes, failure modes, and optimal intervention mixes over household, village, health facility catchment, district, and country scales becomes a formidable task: weeks of data wrangling, assembly of ad-hoc charts and maps, reconciling plotting libraries and styles, etc., etc. is expected. Even when dealing with the well-structured output of EMOD simulations, the process may be intimidating given the realism and respective complexity of the various disease models.
To alleviate this process and allow epidemiologists and modelers to focus on epidemiology and modeling, we are introducing Vis-Tools. Vis-Tools is a software framework which encompases an integrated set of modules that binds epidemiological components to visual elements,  and is rendered in a web-browser upon EMOD simulation completion. For instance, modelers and epidemiologists can now easily map migration of infected individuals and the mobility of pathogens, disease vectors, or other

mobile entities that are part of an EMOD disease-cycle to visual events – e.g. comet-like transitions of infected individuals between population nodes on a simulation map – as shown in the screenshot below. Scalability is important: this tool can handle visualization of dozens of thousands of events (note that only two migration events are shown here for clarity).
We take advantage of EMOD’s detailed reporting capabilities across diseases to automatically extract these events and create their spatio-temporal layout. All that’s left to users is selecting from a set of appearance options for the desired event type (e.g. map migration events of HIV infected male individuals to blue comets depending on CD4 counts; map female infected individuals to red comets; in the context of malaria, map individuals carrying p. falciparum Kelch 13 mutations to orange comets, etc.). If the available options do not sufficiently represent the user’s vision, they can create their own appearance options via Python scripts, based on the well-documented existing examples.

To enable users’ creation of a  complete-as-possible view of their disease spatio-temporal epidemiology patterns, we are working on a rich set of visual elements which  cover various epi components relevant to the diseases EMOD supports. For instance, many environments are characterized by a diffused infection risk that is dependent on environment, vectorial capacity, or human behavioral risk factors. Correspondingly, we provide diffused risk visual “clouds”. In the context of malaria, the clouds may represent vector habitats, respectively vectorial capacity, adult vector number, or infectious vector numbers, as shown in the screenshot below

In the case of malaria, vectors’ behavioral patterns, reflected in the movement of infectious mosquitoes, is often one of the primary transmission drivers at household and village scales. These patterns, their seasonal changes, or their response to interventions (e.g. insecticide residual spraying deployment) are visually reflected in the dynamics of vectorial spatial “clouds”. The human eye is a powerful analytic tool, and in our experience, seeing the dynamics and spatial changes in these patterns leads to better understanding of epidemiology and appropriate intervention adjustments. Drilling down in individual households/villages stats, by simply clicking on a node (as shown below), provides further insight in high-resolution dynamics in the context of the broader-area epidemiology, and may, for example, serve as visual assistance in debugging faulty behavior due to model misconfiguration.

We provide mapping of interventions to visual elements as well. In the previous and next screenshot, we show a day-snapshot of reactive case detection in the context of malaria resurgence in a Zambian village, within the malaria hot-spot Lake Kariba region.  Clinical cases are mapped to black triangles, and drug treatments administered as a response to reported clinical cases are marked with red crosses. Some malaria clinical episodes are left untreated for long times, which leads to increasing numbers of infectious mosquitoes (green “clouds”). That is readily seen in the Vis-Tools animated simulation visualization, which highlights the susceptibility of the area to re-importation of malaria and the importance of strong reactive case detection early in the outbreak.