Modelling symptom progression in individuals for disease surveillance

Peter Dawson, A Meehan, Cecile Aubron, A. Cheng, R. Gailis, W.M. Lau, Charlotte Pierce

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review


In order to create next generation disease surveillance systems for detecting biological attacks, which fully utilise patient records in an electronic health record, it is essential to be able to model how disease symptoms will appear in an attacked population with higher fidelity than has been done previously.
In this paper detailed models of symptom progression of people infected with inhalational anthrax, pneumonic plague and influenza pneumonia are constructed. They are used within an agent based epidemic simulation to predict the prevalence of symptoms within the infected community over time. Influenza
pneumonia was chosen as it is a common respiratory disease that presents with similar symptoms to inhalational anthrax and pneumonic plague.

The models were constructed through a combination of literature review and hospital record analysis, which advised on the nature of the symptoms of the diseases, the time lines of progression and the probabilities of progression, and using this information to construct mathematical models of how these diseases present in individuals (see Figure 1 for the influenza model). Due to imperfect knowledge of these diseases, due to lack of cases or constant mutation, most of the probability parameters associated with the models have an associated uncertainty range. In our models the middle values of these ranges have been chosen, however knowledge of the uncertainty ranges is retained for users of the models to sample from randomly, if required. The simplifying assumption that comorbidity with other diseases does not affect parameter values is made, when modelling individuals with multiple infections.

The disease progression is modelled as moving through a series of states, each with a lognormal probability distribution function describing duration, and with a range of possibilities for the next state. Each state has associated symptoms. As an approximation, all developed symptoms begin at the start of their associated state. These models were then used in an agent based outbreak
simulation of 1000 agents to model the prevalence of various symptoms, to illustrate their different appearance in a health record system. Ensemble results
highlight that time dependent ratios of the prevalence of key symptoms are different between each disease. An anthrax attack, being non-contagious, produces a lognormal symptom onset curve, with fulminant stage symptoms time delayed. An influenza outbreak produces a longer epidemic with few cases with serious symptoms. A pneumonic plague attack produces a double peaked epidemic between primary cases and those that follow.

These models are now being used to parameterise a dynamic Bayesian network based electronic disease surveillance system, which combines probabilistic analysis of medical records and outbreak timelines.
Original languageEnglish
Title of host publication20th International Congress on Modelling and Simulation (MODSIM)
Number of pages1
Publication statusPublished - 2013
EventInternational Congress on Modelling and Simulation 2013: Adapting to Change: the multiple roles of modelling - Adelaide Convention Centre, Adelaide, Australia
Duration: 1 Dec 20136 Dec 2013
Conference number: 20th


ConferenceInternational Congress on Modelling and Simulation 2013
Abbreviated titleMODSIM 2013
Internet address

Cite this