Mathematical models have become invaluable decision-making tools for public health officials. As demonstrated during the United Kingdom's foot-and-mouth epidemic of 2001, models can be useful in two ways: they can reveal the underlying characteristics of an infection and they can allow the comparison of alternative control measures. Often, however, such models make implicit assumptions that may systematically bias their predictions, say researchers at the University of Georgia.
In a paper published today in the online open-access journal PLoS (Public Library of Science) Medicine, Helen Wearing and Pejman Rohani of the Institute of Ecology at the University of Georgia and Matt Keeling of the University of Warwick, United Kingdom, showed that commonly used disease models may risk making overly optimistic predictions about the levels of public health interventions needed to bring a disease under control.
Wearing and her colleagues found that many off-the-shelf models used in infection management do not realistically account for the length of time that people harbor infections. The simplest models entirely ignore the latent period of a disease: the period of time when an individual is infected but not yet infectious. Other models often assume that the rate of progression from latent to infectious, and infectious to recovered, is constant, irrespective of the time already spent in that status. In such models, for example, many people have a very short infectious period while a few have a very long infectious period. In reality, most people are infectious for an average period of time. For the flu the average infectious period is around 4-5 days, with incredibly few people infectious for less than a couple of days or more than a week.
"Models which do not incorporate the latent period or assume unrealistic distributions of the latent and infectious period," said the researchers, "always resulted in underestimating the transmission potential of an infection when fitted to initial outbreak data."
The idea for the project came about when the Centers for Disease Control and Prevention approached Rohani to assist in the modeling of potential smallpox introduction. "I noticed that when we assumed different distributions of the infectious period in the models, we observed very different epidemic curves," said the researcher. Conversations with Wearing and Keeling followed, alternative models were compared and data from an influenza outbreak were used to illustrate the potential bias in model predictions.
The impact of such differences on predicting the spread of new, highly transmissible diseases could be important to public health workers, said Wearing. Underestimating the transmissibility of an infection could lead to predicting inadequate levels of disease management, such as contact tracing and quarantining of those exposed. However, Wearing was careful to stress that current health protection programs have shown their effectiveness in such situations as the recent SARS outbreak in Southeast Asia.
"We are highlighting what 'may happen' if we don't pay careful attention to the inherent assumptions in the models that we fit to data," said Wearing. "The key point is that uncertainty about the latent and infectious period distributions should be taken into account when making quantitative predictions for public health use."
Source: Eurekalert & othersLast reviewed: By John M. Grohol, Psy.D. on 21 Feb 2009
Published on PsychCentral.com. All rights reserved.
I have not failed 10,000 times. I found 10,000 ways that won't work.
~ Thomas Edison