The software system presented above is operated through the GLEaMviz client, which provides the user interface: the part of the tool actually experienced on the user side. The GLEaMviz client integrates different modules that allow the management of the entire process flow from the definition of the model to the visualization of the results. In the following we will describe the various components and provide the reader with a user study example.
Model Builder
The Model Builder provides a visual modeling tool for designing arbitrary compartmental models, ranging from simple SIR models to complex compartmentalization in which multiple interventions can be considered along with disease-associated complications and other effects. (An example can be found in previous work [37].) A snapshot of the Model Builder window is shown in Figure 4.
The models are represented as flow diagrams with stylized box shapes that represent compartments and directed edges that represent transitions, which is consistent with standard representations of compartmental models in the literature. Through simple operations like 'click and drag' it is possible to create any structure with full flexibility in the design of the compartmentalization; the user is not restricted to a given set of pre-loaded compartments or transition dynamics. The interactive interface provided by the Model Builder enables the user to define the compartment label, the mobility constraints that apply (e.g. allowed/not allowed to travel by air or by ground), whether the compartment refers to clinical cases, as well as the color and position of their representation in the diagram (see Figure 5). This allows the user to model many kinds of human-to-human infectious diseases, in particular respiratory and influenza-like diseases.
Transitions from one compartment to another can be of two types: infection transitions or spontaneous transitions. Examples of spontaneous transitions include the transitions from latent to infectious individuals and from infectious to recovered individuals. An infection transition, on the other hand, generates new infected individuals. In GLEaM we consider a homogeneous mixing assumption in each subpopulation so that the expected number of new infections generated by I infectious individuals in a population with S susceptible individuals is equal to , where N is the total size of the subpopulation. The GLEaM simulation engine considers discrete individuals. All its transition processes are both stochastic and discrete, and are modeled through binomial and multinomial processes.
Transitions can be visually added by dragging a marker from the source to the target compartment. Spontaneous transitions are annotated with their rates, which can be modified interactively. Infection transitions are accompanied with a representation of the infection's source compartment and the applicable rate (i.e. β in the example above), which can also be modified in an interactive way. The rates can be expressed in terms of a constant value or in terms of a variable whose value needs to be specified in the variables table, as shown in Figure 4. The value can also be expressed by simple algebraic expressions.
The client automatically checks for and reports inconsistencies in the model in order to assist the user in the design process (see bottom right window in Figure 4).
Models can be exported to XML files and stored locally, allowing the user to load a model later, modify it, and share it with other users. The diagram representation can be exported as a PDF or SVG file for use in documentation or publications. A few examples of compartmental models are available for download from the Simulator website.
Simulation Wizard
The Simulation Wizard provides a sequence of panels that leads the user through the definition of several configuration parameters that characterize the simulation. Figure 6 shows some of these panels. The consecutive steps of the configuration are as follows:
The user is prompted with three options: create a new single-run simulation or a new multi-run simulation from scratch, or a new one based on a saved simulation previously stored in a file.
The user can design a new compartmental model, modify the current compartmental model (when deriving it from an existing simulation), or load a model compartmentalization from a file.
The user is asked to specify various settings and parameter values for the simulation, including, e.g., the number of runs to perform (only accessible in the case of a multi-run), the initial date of the simulation, the length of the simulation (in terms of days), whether or not seasonality effects should be considered, the airplane occupancy rate, the commuting time, the conditions for the definition of an outbreak, and others.
Here the user assigns the initial distribution of the population amongst compartments, defining the immunity profile of the global population on the starting date.
This panel allows the user to define the initial conditions of the epidemic by selecting the city (or cities) seeded with the infection.
Here the user selects the compartments that will constitute the output provided by the client at the end of the simulation. The corresponding data will be shown in the Visualization Window and made available for download.
When all the above configuration settings are defined, the user can submit the simulation to the GLEaMviz server for execution. This will automatically add the simulation to the user's Simulations History. It is furthermore possible to save the definition of the simulation setup to a local file, which can be imported again later or shared with other users.
Simulations History
The Simulations History is the main window of the client and provides an overview of the simulations that the user has designed and/or submitted, in addition to providing access to the Model Builder, the Simulation Wizard, and the Visualization Component. The overview panel shown in Figure 7 lists the simulation identifier, the submission date and time, the simulation type (i.e., single or multi-run), the execution status (i.e., initialized, start pending, started, aborted, complete, failed, or stop pending) and the results status (i.e., none, retrieve pending, retrieving, stop retrieve pending, complete, or stored locally). Additional File 1 provides a detailed explanation of all these values.
A number of context-dependent command buttons are available once a simulation from the list is selected. Those buttons allow the user to control the simulation execution, retrieve the results from the server and visualize them, clone and edit the simulation to perform a new execution, save the simulation definition or the output data to the local machine (in order to analyze the obtained data with other tools, for example), and remove the simulation. In addition to exporting the compartmental model (see the "Model Builder" Subsection) the user can export a complete configuration of a simulation that includes the compartmental model and the entire simulation setup to a local file, which can be imported again later or shared with other users.
Visualization interface
Once the execution of a simulation is finished and the results have been retrieved from the server, the client can display the results in the form of an interactive visualization of the geo-temporal evolution of the epidemic. This visualization consists of a temporal and geographic mapping of the results accompanied by a set of graphs (see Figure 8). The geographic mapping involves a zoomable multi-scale map on which the cells of the population layer are colored according to the number of new cases of the quantity that is being displayed. Several visualization features can be customized by clicking on the gear icon and opening the settings widget. It is possible to zoom in and out and pan by means of the interface at the top left of the map. Dragging the map with the mouse (on a location where there are no basin marks) can also pan the visualization. All the widgets and the graphs displayed over the map can be re-positioned according to the user's preferences by clicking and dragging the unused space in the title bar.
The color coding of the map represents the number of cases on a particular day. The time evolution of the epidemic can be shown as a movie, or in the form of daily states by moving forward or backward by one day at a time. For single-run simulations it is also possible to show the airline transportation of the 'seeding' individuals by drawing the traveling edge between the origin and destination cities. In the case where the output quantity is a subset of the infectious compartments, the edges show the actual seeding of the infection. Note that the evolution of the epidemic depends strongly on the model definition. For example, it is possible that some basins are infected by a latent individual that later develops the disease. In this case no seeding flight will be shown if only infectious compartments are selected as output.
Beside the geographical map, the Visualization Window displays two charts. One chart shows the number of new cases per 1,000 over time (incidence), and the other shows the cumulative number of new cases per 1,000 over time (size). For multi-run simulations, median values and corresponding 95% confidence intervals are shown. The menu above each chart combo lets the user choose the context for which the corresponding charts show incidence and size data. This context is either: global, one of three hemispheres, one continent, one region, one country, or one city. The currently selected day is marked by a vertical line in these plots, and the day number, counted from the initial date selected for the simulation, is shown by side of the time slider.
User study example
Here we present an example application of the GLEaMviz tool to study a realistic scenario for the mitigation of an emerging influenza pandemic. Disease-control programs foresee the use of antiviral drugs for treatment and short-term prophylaxis until a vaccine becomes available [38]. The implementation of these interventions rely both on logistical constraints [21, 39] - related, e.g., to the availability of drugs - and on the characteristics of the infection, including the severity of the disease and the virus's potential to develop resistance to the drugs [40].
Here we focus on the mitigation effects of systematic antiviral (AV) treatment in delaying the activity peak and reducing attack rate [41–43, 7, 8, 39, 40, 3], and assume that all countries have access to AV stockpiles. We consider a scenario based on the 2009 H1N1 influenza pandemic outbreak and feed the Simulator with the set of parameters and initial conditions that have been estimated for that outbreak through a Maximum Likelihood Estimate by using the GLEaM model [3]. The results provided by the present example are not meant to be compared with those contained in the full analysis carried out with GLEaM [3] due to the fact that in the present example we do not consider additional mitigation strategies that were put in place during the early phase of the outbreak, such as the sanitary control measures implemented in Mexico [3, 44], or the observed reduction in international travel to/from Mexico [45]. Indeed, the current version of GLEaMviz does not allow for interventions that are geographically and/or temporally dependent. However, these features are currently under development and will be available in the next software release. For this reason the simulation scenario that we study in this application of the Simulator does not aim to realistically reproduce the timing of the spreading pattern of the 2009 H1N1 pandemic. The results reported here ought to be considered as an assessment of the mitigating impact of AV treatment alone, based on the initial conditions estimated for the H1N1 outbreak, and assuming the implementation of the same AV protocol in all countries of the world.
We adopt a SEIR-like compartmentalization to model influenza-like illnesses [29] in which we include the systematic successful treatment of 30% of the symptomatic infectious individuals (see Figure 9). The efficacy of the AV treatment is accounted for in the model by a 62% reduction in the transmissibility of the disease by an infected person under AV treatment when AV drugs are administered in a timely fashion [41, 42]. We assume that the drugs are administered within 1 day of the onset of symptoms and that the AV treatment reduces the infectious period by 1 day [41, 42]. The scenario with AV treatment is compared to the baseline case in which no intervention is considered, i.e. the probability of treatment is set equal to 0 in all countries.
The GLEaMviz simulation results are shown in Figure 10 where the incidence profiles in two different regions of the world, North America and Western Europe, are shown for both the baseline case and the intervention scenario with AV treatment. The results refer to the median (solid line) and 95% reference range (shaded area) obtained from 100 stochastic realizations of each scenario starting from the same initial conditions. The resulting incidence profiles of the baseline case peak at around mid-November and the end of November 2009 in the US and Western Europe, respectively. These results show an anticipated peak of activity for the Northern Hemisphere with respect to the expected peak time of seasonal influenza. In order to make a more accurate comparison with the surveillance data in these regions, we should rely on the predictions provided by models that can take into account the full spectrum of strategies that were put in place during the 2009 H1N1 outbreak, viz. the predictions obtained by GLEaM [3].
In the case of a rapid and efficient implementation of the AV treatment protocol at the worldwide level, a delay of about 6 weeks would be obtained in the regions under study, a result that could be essential in gaining time to deploy vaccination campaigns targeting high-risk groups and essential services. In addition, the GLEaMviz tool provides simulated results for the number of AV drugs used during the evolution of the outbreak. If we assume treatment delivery and successful administration of the drugs to 30% of the symptomatic cases per day, the number of AV drugs required at the activity peak in Western Europe would be 4.5 courses per 1,000 persons, and the size of the stockpile needed after the first year since the start of the pandemic would be about 18% of the population. Again, we assume a homogeneous treatment protocol for all countries in the world; results may vary from country to country depending on the specific evolution of the pandemic at the national level.