ApplicationtoFieldSites
OpenMalaria can be used for simulating P. falciparum epidemiology in specific sites, using measured characteristics of the vectors in the site and information on the local prevalence of incidence to parameterise the models. This may be used both for: (i) validating the models by comparing their outputs to additional data from the site. (ii) projecting the likely effects of interventions in that specific location.
The model is written in C++. In order to run the model you need to download the simulator that contains all the necessary code in an executable version. This is updated from time to time. For instance, as of April 2011 the current version of the OpenMalaria schema was version 26. Check back periodically to ensure you have the most up to date version. If you are in doubt contact the OpenMalaria administrator. To run the OpenMalaria simulator you need to create an XML file that contains all the data on malaria in your study area (described in later steps).
A description of how to get started working with OpenMalaria can be found here.
Before doing anything else, collaborate with the research team and/or control program managers in the study area (if possible) to ensure field data is available for what you want to simulate. For example, if the field study is a cohort where people are tested for malaria using rapid diagnostic tests (RDTs), you will be simulating patent infections. If the field study is an entomological study where mosquitoes are trapped, dissected and analyzed for sporozoites, you will be simulating number of infectious mosquitoes.
The full list of outputs OpenMalaria is able to simulate can be found here. Try to keep the number of outputs limited to 3 or 4 maximum for ease of analysis. The models are driven by the simulated Entomological Inoculation Rate (EIR), which is the main measure of transmission so this should ideally be one of the outputs if you are interested in simulating the effects of interventions on transmission.
It is important for this step to happen early on because you don’t want to get to the end of running simulations and find out your analysis is not relevant in the field. It will help immensely if everyone is on the same page at the beginning with what you are simulating and why.
What you want to simulate (Step #2) will determine which model(s) to use from the following options:
-Transmission (vector or non-vector)
-Within-host (Drug action PKPD and/or asexual infection)
-Clinical/sickness (morbidity/pathogenesis or neonatal mortality)
-Interventions (human stage, mosquito stage, and/or “special” interventions)
Full descriptions of each model and how they work can be found by following links from here.
Depending on what model(s) you choose, the next step is to make a list of the inputs necessary for each of those models.
In general, the types of information include:
-Demography
-Monitoring system
-Interventions
-Health system
-Entomological data
-Drug description
-Model options
-Human availability
A full list of these parameters as well as their definitions can be found by following links from here as can descriptions of the XML syntax required for the scenario files.
The next step is to gather the necessary data to include in the scenario file to describe the situation in your study area as closely as possible to reality. There are more or less 90 separate pieces of data that need to go into the model. As many inputs as possible should come from studies in your particular study area, but realistically the overall percentage of inputs that come from site-specific data will be rather modest (about 30%). A sizeable number of inputs will come from previously published literature at the district, national or regional level through a literature review. There will be a number of inputs that will remain the same as the initial model parameterization1-2, 4-7 depending on which model(s) you choose, particularly data on the vector life cycle and drug action. It is helpful to document the source of the data you choose for each parameter and if there are multiple sources why you chose the value that will go into the model.
The easiest way to do this is to use an example file and replace the values with those you compiled in Step #5. Keep in mind that while you can replace the individual values, the overall structure of the file (i.e. the order the parameters are listed) cannot change or the simulator will not work. Examples of the XML syntax that can be adapted and edited to fit your particular scenario file can be found here
These files are updated on a continual basis so be sure you download a file with a recent date. For troubleshooting contact the administrator.
The run time of the model is dependent on the population size, so to make sure your scenario file works before sending it to BOINC (Step #10) test run your scenario with one seed value and a population of 100. If there is a problem with the syntax an error message will alert you to the problem. Once the file successfully completes, you can increase your population size and number of seed values.
The simulation results of the outputs you have chosen from Step #2 are produced in the form of two separate .txt files (ctsout.txt and output.txt) that can be imported into your program of choice for analysis (i.e. Excel, R, STATA, etc). A guide to interpreting the output files can be found here.
Creating graphs that overlay observed field data onto the simulations will give you an idea of how much stochasticity exists in the model. Because there are many runs, consider showing the minimum, maximum and median simulation values as well as the inter-quartile range and compare this to the observed field data with its 95% CIs. There is no consensus in the modeling community on how to quantify how well the model “fits” the data so in practice this will evolve over time. For example, there is currently a debate over whether you can average the simulation runs for analysis. Picking a method and making sure to acknowledge its limitations is the best we can do at this point. Visually comparing model outputs to observed field data is a first step, then determining what proportion of model runs fall within the 95% CIs of the field data, and finally the sum of least squares if relevant.
9. Conduct experiments – evaluate uncertainty with a sensitivity analysis and examining a mix of intervention options
By running simulations with multiple seed values you are addressing stochasticity of the model. However, once you have your fitted model, conducting a sensitivity analysis of parameters can address parameter uncertainty. Choose parameters that you believe are important to the model but have input values you are not completely certain about. For example, measures of EIR can vary in the same location when different methods for measuring sporozoites and biting rate are applied3, so this parameter is a good candidate for a sensitivity analysis. Choose a range of values for the parameter that makes sense given your study area context – usually a minimum, maximum and median value. Using the EIR example, if your estimated EIR for the study area is 4, for your range you may want to choose 0.4 as a lower value, 10 as a slightly higher value, and 40 and an extremely high value. Using the scenario file of the “best fit” model from your simulations above, only change the value you are conducting the sensitivity analysis with and create as many new scenario files as you have values (i.e. one file for the minimum EIR, one for the maximum EIR, etc).
To explore the effects of different interventions, take your “best fit” model and update the scenario file to simulate different coverage levels, timing, and target age groups. Set the scenario to run for an extended period time over the whole population of the study area to see the impact.
Once you are satisfied that the syntax in your scenario files is working and you are ready to run the simulations with all seed values and parameterizations in the chosen population, you may conclude that your simulations will take too long on your own computer.
Collaborators of the OpenMalaria development team can send the scenario files in .xml format to the project staff who will ensure the scenario file is in order and then submit them to an appropriate computing resource. If you have gone through the setup and analysis exercises on smaller simulated populations this part should run smoothly.
-
Chitnis, N., A. Schapira, et al. (2010). "Comparing the effectiveness of malaria vector-control interventions through a mathematical model." The American journal of tropical medicine and hygiene 83(2): 230-240.
-
Chitnis, N., T. Smith, et al. (2008). "A mathematical model for the dynamics of malaria in mosquitoes feeding on a heterogeneous host population." Journal of Biological Dynamics 2(3): 259 - 285.
-
Kelly-Hope, L. A. and F. E. McKenzie (2009). "The multiplicity of malaria transmission: a review of entomological inoculation rate measurements and methods across sub-Saharan Africa." Malaria journal 8: 19.
-
Maire, N., T. Smith, et al. (2006). "A model for natural immunity to asexual blood stages of Plasmodium falciparum malaria in endemic areas." Am J Trop Med Hyg 75(2 Suppl): 19-31.
-
Penny, M. A., N. Maire, et al. (2008). "What should vaccine developers ask? Simulation of the effectiveness of malaria vaccines." PLoS One 3(9): e3193.
-
Ross, A., M. Penny, et al. (2008). "Modelling the epidemiological impact of intermittent preventive treatment against malaria in infants." PLoS One 3(7): e2661.
-
Smith, T., N. Maire, et al. (2006). "Relationship between the entomologic inoculation rate and the force of infection for Plasmodium falciparum malaria." Am J Trop Med Hyg 75(2 Suppl): 11-18.
| Download openmalaria | Installation instructions | XML Schema Documentation |
| XML Schema Version | Program version | master |
develop |
|---|---|---|---|
| 43 | schema-43.0 |
 |
 |