Infectious diseases such as tuberculosis, malaria and cholera are the leading cause of death globally. When there is an outbreak of an infectious disease – such as the Zika virus epidemic that began in 2015 – time and data is often too limited to build up an accurate model to map how a disease might progress, or to guide public health decisions. Experts and policymakers must work together to decide how best to control these outbreaks, with important decisions depending on surveillance and expert opinion. But it is a highly individual judgement.
“A lot of this work relies on their knowledge and experience as an expert in the field,” explains Ashlynn Daughton, graduate student at Los Alamos National Laboratory and PhD student at the University of Colorado, US. “This process can be a little subjective.
One way to add a more objective element is to simulate many possible outbreaks and the effects of lots of different control measures. This lets decision-makers experiment with many ‘what if’ scenarios, and then make decisions based on that cumulative knowledge,” she continues. “These models aren’t new. They’ve been in use for decades. However, they haven’t previously been very accessible to most individuals in public health, because they required some knowledge of computer programming to be able to run a model exactly the way one wants to.”
The decision-making process
According to Daughton, epidemiological models can plug the gaps in the decision-making process by using available information and data to yield a quantitative estimate of outbreak paths; determining where and how fast the infection is moving, so that medical supplies and staff can be deployed for maximum effect.
Effective deployment requires the collaboration of the modelling community and the public health policy community, but Daughton says such partnerships are rare, and as a result there is a lack of suitable models that actually meet the needs of the public health community.
Traditional epidemiological models categorise people based on their disease status: susceptible, infected or recovered (SIR). Other models are agent-based; an agent – often an individual – is identified and their potential interactions are mapped for a day. This takes into consideration people they have intermingled with and where, such as other children at school, or co-workers in an office. The model then extrapolates how each interaction might spread the disease. This latter model, however, requires significant expertise, computer power and large amounts of data, and so is often beyond the resources of the average health department.
Pure and simple
So Daughton focused her research on the simplest model. “What we’ve done is use the old models in a different way. The type of model we currently use is called a compartmental model because people are grouped into compartments based on their disease status. In order to run these models, one has to have some information about the outbreak, such as how long a person is infectious for, and how many new people one person can infect. This information is used to parametrise the model.
“However, determining these can be difficult,” she continues. “Our contribution to the model tries to solve this problem by running the model over a range of parameters, rather than one point estimate. We also make comparisons between a controlled outbreak and an identical uncontrolled outbreak, so someone can see the effects of a control measure given the range of possible model parameters. In epidemiology, this is called a counterfactual and describes an experimental set-up where the experiment and control are exactly the same, except for one thing – your variable of interest.”
The benefit of looking at ranges in the parameters is that it allows a public health individual or policymaker to explore possible control measures, while seeing how changes in parameters affect that outcome, Daughton explains.
The model in action
“If someone has some uncertainty about one of the parameters – perhaps they know the infectious period is five to eight days, but they aren’t sure what the exact number is – they can visualise how that impacts the number of people who become sick, or how effective the control measure might be at preventing illness,” she says. “It also allows someone to see how sensitive the model is to a particular parameter. For example, one might find that changing a parameter within the range of expected values doesn’t actually impact how large the model thinks the outbreak will be, or how well the control measure works. Knowing that a particular parameter ‘doesn’t matter’ would then allow a policy maker to look at other related and important aspects, such as financial concerns or the time required to assemble resources for control. This allows the policymakers or analysts to better understand possible trade-offs between, for example, time to implementation and coverage.”
Daughton and her team used their model to study the outbreak of measles, norovirus and influenza to show the feasibility of its use, and to detail a research agenda to further promote interactions between decision-makers and the modelling community. The research so far has been a proof of concept on a new way to use existing models, but they are not at a point where they can be operationally used for decision support, says Daughton.
During the research, the team were also able to develop a web interface. Currently, most compartmental models require the user be able to interact with the model on their computer, potentially in a complicated, programming-heavy way. This isn’t conducive for use by public health persons whose training typically doesn’t include computer-programming aspects, she explains. The team described this interface in their most recent paper, which has been published in the Scientific Reports journal, and plan to have it physically on the web within the next few months.
The importance of variety
The researchers’ initial work focused on one kind of compartmental model, but there are many types.
“Different types of compartmental models capture different aspects of diseases,” explains Daughton. “For example, the model we used describes diseases that are transmitted person-to-person and don’t have very long incubation periods. Other models include additional categories and parameters to describe other kinds of diseases, such as those that are spread using vectors, or have longer incubations. One of the things we’re interested in doing is expanding this analysis to include those models.”
The team is also interested in seeing if the models can be improved by adding other kinds of data, including access to healthcare. For example, perhaps if you have better access to doctors, you’re more likely to get yourself and your family vaccinated against diseases such as measles. If such parameters that describe these features could be included – albeit broadly – perhaps the models would better reflect reality.
“Lastly, we’re very interested in better describing when these models are appropriate to use,” explains Daughton. “There are some general guidelines in the literature, but it isn’t explicitly known for which parameter ranges these models might be appropriate approximations of reality. We’re currently looking into different validation techniques to try to define this more rigorously.”
Daughton believes there is an important balance in public health between providing a model that reflects reality, but is simple enough to run quickly, and uses data that is readily available and can be easily interpreted by public health personnel and decision-makers.
“There are other models that are being developed by this community that describe disease transmission at a much finer scale, and include many additional parameters,” she explains. “However, it’s difficult to adapt those models to a new situation because the data required to run them often isn’t available, and because they typically require a supercomputer. That’s where compartmental models can come in; a model like ours might be useful in situations where a possible answer is needed quickly in order to make effective decisions. Since the work so far has been a proof of concept, we’ve got a lot of ideas about ways to expand the tool and make it more robust. For example, we’re working on expanding the number of compartmental models used, validating those models and performing experiments to see if this way of using compartmental models reflects reality.
“Sometimes we’re asked why Los Alamos does this kind of research,” Daughton concludes. “Our mission is tied to national security, and it’s easy to see how tracking diseases and stemming their spread is vital to that security. Diseases don’t care about boundaries or respect borders and can clearly represent a serious threat to our national well-being. So the lab has been using mathematics and computer modelling since the early 2000s to track infectious diseases with the goal of improving disease prediction to stop their spread more quickly.”