skip to primary navigationskip to content

Digital Epidemiology: Modelling of Epidemic Spread in Developing Countries using Human Mobility Data

last modified May 20, 2015 05:43 PM
Eiko Yoneki, Computer Laboratory

Digital Epidemiology: Modelling of Epidemic Spread in Developing Countries using Human Mobility Data

Eiko Yoneki University of Cambridge Computer Laboratory

Respiratory and other close-contact infectious diseases, such as TB, measles and pneumonia, are major killers in much of the developing world. Mathematical models are essential for understanding how these diseases spread and for identifying how best to control them. Although central to the models, few quantitative data are available on relevant contact patterns, and no study to measure these factors has yet been attempted in rural Africa or South America. This situation is particularly problematic given the high burden of infectious diseases in the developing world and the resourcedriven requirements for optimally targeted interventions. The recent population shift in Africa and elsewhere from rural to urban areas adds additional challenges for the understanding of complex social networks. How people behave and interact during a large outbreak of an infectious disease directly impacts not only the spread of infection, but also the efficacy of control strategies, and it can also have wide-reaching economic implications. We will develop mathematical models based on these new data, which will help us gain valuable insight into the spread and control of diseases. Examples of diseases to be modelled are tuberculosis, pneumococcal disease, meningococcal disease, measles, and disease associated with Haemophilus influenza.

Improved knowledge of relevant contact patterns will hugely help the understanding of social and spatial patterns of infection spread. The recent emergence of wireless technologies (e.g., mobilephones and sensors) makes it possible to collect real-world data on human connectivity along with environmental and contextual information. Capturing human interactions with such devices will provide empirical, quantitative measurements of societal mixing patterns to underpin mathematical models of the spread of close-contact diseases. The use of sensors and mobile-phones for these purposes has distinct advantages over other methods of collecting contact data (such as diaries and interviews), because they can gather proximity data automatically, allowing detailed longitudinal with no possibilities of re-call bias, no barriers due to problems of literacy or understanding, and minimal disruption to the participants of the survey. Such approaches therefore offer an unparalleled and timely opportunity to collect information on social contact patterns that would allow a step-change in our understanding of the patterns of disease spread. The use of sensors and mobile-phones for this purpose has not been explored for this application before. We plan to develop a human mobility study framework for low cost RFID sensors and mobile-phones. In a rural environment, simple RFID sensors will be used together with low-cost computers as readers (e.g., Raspberry PI). We will integrate minimum satellite communication. Mobile-phones may be used more in urban environments, along with sensing movement, light, and humidity. Various sensors embedded in the phone will capture contextual changes in the environment to infer behavioural patterns of the phone carriers. The dynamic network of connections between participants will be used to investigate the topology of the social network, including: 1) duration-weighted pairs – time spent in close-proximity is a powerful determinant of infection risk and these can be considered as a weighted link between individuals with location and context associations; 2) number of encounters per person – are some individuals responsible for a disproportionate number of contacts? 3) social distances – betweenness and centrality measures describe how far apart individuals are in a network, and strongly impact disease dynamics; 4) community structure – identify individuals that form bridging links between otherwise distinct groups offers efficient targeted interventions.

Mathematical and computational models of social networks and epidemic spread, and methods to analyse the collected data are critical for public health and epidemiology. Advances in computing and large-scale data mining now create fresh opportunities to support a new generation of epidemiology, i.e. Digital Epidemiology.