Dear malariacontrol.net user,
As you have read in our latest status update, we are close to the end of the current phase of the project. We will soon stop sending out workunits for a few weeks, until the project scientists have prepared the input data for the next phase of research.
Update: We have stopped creating new workunits on Mon March 1st
This is a good time to explain what we have achieved with your help over the last few months, and how these results will be used in the future.
Work to Date
The main objective of this phase of the project was to optimize our mathematical models, to get them to predict the malaria situation in the field as accurately as possible. We use an “evolutionary” approach: Starting with an initial version of the model, we create a range of variations. Then we test all the variations against the real-world data to see which perform best. We then take the best-performing variations and repeat the process, over and over again... While extremely effective, this approach takes a great deal of computation; we couldn’t do it without your help.
By way of illustration, here's a chart of results from one of our recent runs: You can see how our measure of the discrepancy between model and real world (on the y-axis) decreases as the model is optimized through successive iterations (on the x-axis).
Another way of looking at the same data is by plotting the predicted outcomes of the model against real-world data. The following examples look at one outcome, the so-called age-incidence of malaria episodes: This is the rate at which humans of a certain age come down with malaria. Here it is shown as the number of episodes per person-year (py). The first of the two plots show an early stage of the “evolution” process, where field data and predictions are quite different. In the second image, plotted from a later stage, you can see the data and predictions converge: thus indicating that the model has become much more accurate.
If you're wondering what Ndiop and Dielmo means, these are the names of two Senegalese villages, that's where the field studies for this particular dataset were carried out.
We’ve gone through this type of evolutionary optimization process with quite a number of different model formulations. In technical terms, we have “parameterized a number of different models”. Some of these did not do a good job at predicting patterns like the age-incidence curves above, and we've abandoned these, the rest were kept in our growing model family. As a result, we now have a set of models which work in very different ways but which all appear effective. This gives us a strong basis from which to undertake the next phase of our project. The goal is to now compare the predicted impact of various control interventions (such as bed nets, indoor spraying, vaccines, case management,etc.) across our current families such model formulations. This is a way of learning about model uncertainties, and is done routinely these days with climate models.
The following -very preliminary- plots demonstrate where this is going: We have simulated an effective house spraying program that went on for twelve years and then stopped. We've then plotted the predictions from such a model ensemble (Prevalence, shown on the Y-axis, indicates the proportion of the human population with parasites in their blood):
One of our top priorities in the near future will be the design and analysis of more simulation experiments like this one, predicting the effect of a number of different malaria control interventions. Once optimized-- though your contributions-- these models can be used to develop intelligent and effective deployment strategies, hopefully saving a great many lives.
Thank you for your ongoing interest and support!
On behalf of the modeling team,
Swiss Tropical and Public Health Institute