What is computational epidemiology

I consider myself a researcher in computational epidemiology (or, at least, a starting one!). At this moment, I am part of the compepi group at the University of Iowa. But it seems this term is not widely popular. Epidemiology, as part of public health, seems much more akin to statistics (biostatistics, rather), and then to dynamical systems (differential equations, i.e. mathematics and physics). Of course, statistics is taken much more seriously and seems more productive, in terms of research and impact, than dynamical systems. But where is computer science?

Examples of research in computational epidemiology

Despite the popularity of statistical epidemiology, there are some lines of research in computational epidemiology that are very famous. For example:

• epidemic spread processes in complex networks,
• surveillance of diseases in social media (e.g. Google Flu),
• opinion dynamics in social media (they behave like viral epidemics).

But computational epidemiology goes further. Computational epidemiology is about using computational science to address problems of epidemiological concern. By computational science, I mean the use of computational methods to collect information and draw inferences for scientific use. It is a collective term for methods such as:

• discrete event simulation (for continuous or discrete time, it does not matter),
• numerical computation (the good-old scientific computing),
• Web crawling and surveillance (which belongs to the field of information retrieval),
• machine learning and data mining (statistics gone hyper),
• algorithms and optimization applied to scientific concerns,
• etc.

Current “prophets” of computational epidemiology

I have encountered myself a number of times with works that explicitly mention the term computational epidemiology. As of the time I write this (Oct 2013), I can mention the following works:

• Computational Epidemiology, by Marathe and Kumar, in Communications of the ACM 56(7), 2013.
• Digital epidemiology, by Salathe et al, in PLoS Computational Biology, July 2012.
• The Role of Computational Epidemiology and Risk Analysis in the Fight Against HIV/AIDS, by Tameru et al, Journal of AIDS & Clinical Research 3(6), July 2012.
• (A workflow for software development within computational epidemiology, by Cakici and Boman, in Journal of Computational Science 2(3), 2011.)
• Computational Epidemiology: Modeling the Human Equation, by Gorder, in IEEE/AIP Computing in Science and Engineering 12(1), 2010.
• Computational Epidemiology in a Connected World, by Apolloni, Kumar, Salathe, Swarup, in IEEE Computer 42(12), 2009.

Earlier prophets of computational epidemiology

Going back in time a bit, we still find some other related work:

• Advancing Epidemiological Science Through Computational Modeling: A Review with Novel Examples, by Duke-Sylvester et al, in Annales Zoologici Fennici, 45(5), 2008.
• Agent-Based Simulation Tools in Computational Epidemiology, by Patlolla et al, in Innovative Internet Community Systems, LNCS 3473(2006), 2006.
• Computational and Mathematical Epidemiology, by Roberts, in Science Careers, 2004.

But there are some older publications…

In the book Veterinary Epidemiology by Michael Thrusfield defines computational epidemiology as the application of computer science to epidemiological studies, and then cites the work of Habteriam et al 1988, which is quite old! I am going to transcribe some text from it:

To understand the behavior of complex biological systems, it is useful to devise computer based models by approximating the interactions, via biomathematical expressions. Without doubt, these models could be over simplifications of complex interactions but they would be useful in comparison to classical laboratory experimental approaches which may not be practical or feasible.

Nice! And not stopping here, the article then goes to define a little workflow for computational epidemiology:

• Stage 1: conceptual model + biomathematical methods. This is about understanding the dynamics of the system-problem at hand, and use mathematical models.
• Stage 2: knowledge base development. This is about collecting information of interest; in the work, it comes from literature, and consists of other models and statistical data.
• Stage 3: model development. The programming part, in all of its senses: conceptual, mathematical, computational.
• Stage 4: experimentation with the model. Simulations, testing, validation, sensitivity analysis, etc.

Stages 1 to 3 are interrelated, so this workflow was not a cascade model.

Of course, today’s activity in computational epidemiology is much broader. Data can be collected in many ways, and it may capture different aspects of the phenomena under study, so models also have to deal with the problem of partial or biased validation.

This was cute, was it not?