Dear malariacontrol.net user,
In our last posting, we described the approach that we have used, with your help, to optimize our mathematical models of malaria. Starting from a base model, we have developed a range, or ensemble, of alternative models components, each one representing a different set of assumptions about malaria transmission and epidemiology. We tested and compared the results from each one against field data until we identified the parameter values that gave the best possible fit of model projections to the field data. Once each model is fit, it can be used to predict outcomes of interest under different conditions, for example, following implementation of an antimalarial intervention. Ideally, these predictions should be validated against additional field data, if it is available.
The motivation for using ensembles of models, rather than just one, is to assess whether or not our understanding of a particular phenomenon is sufficient to obtain valid predictions across a range of assumptions. If so, we can be fairly confident that we have adequately captured the phenomenon. If, on the other hand, different models yield very different predictions, this would indicate a need for further study to determine which assumptions are the correct ones.
The two models components described below (we refer to them simply as “models” for brevity) represent important points of scientific uncertainty in several areas important for malaria transmission and burden. We will use the ensemble approach to explore both to an extent that would be impossible without the computing resources provided by you, the volunteers.
You may know that malaria is caused by parasites, which enter the body through the bite of a mosquito, and replicate in the red blood cells over the course of an infection. The number of parasites per microliter of blood, or “parasite density”, is a very important quantity which influences how infective a person is to mosquitoes and how likely he or she is to become ill with malaria. This is, therefore, the starting point for our project: a model that predicts parasite densities in an individual’s blood over time following inoculation by an infected mosquito. Humans are “hosts” for malaria parasites, and so this is called the “within-host model”.
The base within-host model uses as a starting point a statistical description of the parasite densities in artificially infected humans. We have good data on this from the times when malaria was used to induce fevers to treat neuro-syphilis in the pre-antibiotic era. On top of this, we added a model, also based on field data, to account for acquired immunity to the parasites. Applied together, these models are perfectly adequate for simulation studies that investigate the effect of malaria control interventions intended to prevent new infections: bed nets, spraying of houses, or infection-blocking vaccines would be examples.
However, these models are less appropriate for studying control measures targeted at infected humans – like an investment in improved case management as discussed in the next section. In order to do this, one would like a model which captures the mechanisms generating the observed patterns of parasite densities, rather than relying on a statistical description. We have therefore developed an improved within-host model, based on a more detailed representation of both the parasite life cycle and the human immune system. If you are a MalariaControl.net or Africa@home user, this is the “within-host model” you are now running.
Here’s one important question this will allow us to address: a key unknown in malaria control today is how long the current generation of antimalarial drugs will remain effective. Unfortunately, drug resistance develops over time to most widely-used drugs (and has, in the past, to previous malaria drugs); and this resistance can be accelerated if drugs are not used properly. In the absence of alternative drugs, this would be a disaster for malaria control in much of the world, as there would be no effective cure for people suffering from the disease. We have therefore developed a model for how drugs act to reduce parasite densities. This model, together with the new within-host model, will enable us to investigate drug resistance and the conditions under which it develops. Such results should provide guidance as to how to slow the development and spread of drug resistance and also inform development of new generations of antimalarial drugs.
The new within-host model predicts parasite densities each day, rather than every five days as in the initial model, which makes it more computationally intensive. However, the ability to predict parasite densities each day allows a finer-grained simulation of the action of drug treatment that people receive, since potent antimalarial drugs generally act to significantly reduce parasite densities within hours of being administered. It also allows a more realistic simulation of people’s response to illness and treatment (of which, more below).
Case Management Model
The model ensembles we're describing are actually more than just variations of the within-host model: another major component we have developed recently is concerned with human behaviour and how cases of malaria are managed – that is, what people do when they become ill, what kind of treatment they receive, and how they take the treatment. Case management is important to consider because, in addition to reducing death and illness, treating people with malaria parasites reduces the number of people that can infect others. This affects future transmission of the parasite causing disease.
An important objective of malaria control programmes is to achieve high antimalarial treatment coverage levels, which means to make sure that quality treatment is provided to as many people as need it, as soon as they have symptoms. However, in many countries, in particular those with weak health systems, barriers to effective timely treatment exist, and treatment coverage levels remain low.
The initial case management model investigated how different antimalarial treatment coverage levels would affect transmission, morbidity and mortality. However, antimalarial treatment coverage is determined by a number of sequential decisions, by both patients and health care providers, and a breakdown, or conversely, an improvement at any step in this sequence can greatly alter coverage. The alternative model simulates these decisions, enabling investigation of strategies and interventions to raise antimalarial treatment coverage and quality. The new model considers, for example, antimalarial treatment in the informal sector, which is an important alternative to the public health sector in many countries; health worker practices, which research suggests often deviate from national guidelines; and delays to seeking treatment, which can result in progression from uncomplicated to severe disease.
The new case management model simulates individuals decisions daily; this is more realistic than the previous model, which allowed a decision only every five days. Furthermore, in combination with the drug action and within-host models mentioned above, we can examine the effect of treatment on daily parasite densities. We can simulate sub-curative treatment (where people recover from illness but still harbor parasites), and the effect of misuse of drugs on the growth of drug resistance.
With this alternative case management model, we’ll be able to predict how antimalarial treatment coverage levels might be increased using various health system interventions, and suggest whether these are a good use of resources compared to, or in combination with, other interventions (such as distribution of bed nets). We include the use of new diagnostic tools, which make it possible to target treatment more effectively. We can model the feedback effects within health systems, whereby an intervention affects not only biological and epidemiological parameters, but the health system itself. Finally, we have compiled data from across Africa to parameterize the model, so we can tailor predictions to different settings.
Finally, in addition to the alternative models components described above, as part of our work on ensembles we are also exploring the effects of other variations to the base model. We’ll describe this in a future posting.
We hope to be able to offer soon the most comprehensive and policy-relevant set of malaria models to date and ensure they are available to and used by planners and policy makers.
Because these new simulations are so computationally intensive, they could not be undertaken without your participation; we really appreciate your ongoing support and contributions!