Biosurveillance is a decidedly modern response to the age-old problem of epidemics. In today's world, the large-scale spread of infectious disease has been minimized thanks in large part to the ability to systematically track and predict outbreaks.
Biosurveillance techniques take into consideration countless variables, analyze the statistical data with respect to previously observed patterns and model the likely trends of burgeoning epidemics in hopes of minimizing the spread of contagion.
A recent study that included researchers from the Hopkins Applied Physics Laboratory examined a variety of statistical methods to determine whether they would be applicable for the next generation of biosurveillance models.
They argued that biosurveillance could be further improved by using data monitoring techniques that explicitly take into account trends of an epidemic over time. The current system, on the other hand, uses models that do not include time effects such as days of the week, which are known to affect the behavior of epidemics - in effect, they take only a single snapshot of an epidemic.
By using time forecasting methods, the researchers hoped to create a more holistic and accurate picture of incoming data.
The researchers settled on a mathematical approach known as the Holt-Winters method as an alternative way to model data. The Holt-Winters method uses a series of algorithms to predict patterns. What makes this method unique is that it uses a time series model and can be used to predict future patterns based off of the past. The model is currently used as part of weather prediction algorithms.
When applied to infectious disease epidemics, it could help public health officials predict where an outbreak will strike several days or weeks into the future.
The researchers found that the Holt-Winters method was the most accurate of three tested in predicting artificial scenarios. It actually turned out to be easier to manipulate in that it requires less data input than other methods.
Unfortunately the Holt-Winters method is unable to use information about external factors such as temperature, but this could easily by fixed by melding the method with other predictive models.
The ultimate goal of this research is to develop a reliable, quantitative model that can predict the behavior of epidemic diseases. This paper represents a step forward because it gives more reliable predictions with a smaller input of data, and it takes into account one of the most important variables when racing the clock against a deadly disease: time.
