Info

www

27 31 35 39 4;

S 47 51 i 7 II 15 » 21 21 31 35 }<• 43 J7 51 3

figure 21.1 Weekly counts for several types of routinely collected data for various time periods around the December 1999 influenza outbreak in Pittsburgh. Each data type is plotted on a normalized scale. Lab, influenza cultures from the University of Pittsburgh Medical Center Health System; WebMD, counts of queries to a national web health site using words such as "cold" and "flu"; Cough and cold and cough syrup, grocery chain point of purchase counts; School, school nurse influenza reporting; Resp. and Viral, categories of emergency department ICD-9-coded chief complaints.

figure 21.1 Weekly counts for several types of routinely collected data for various time periods around the December 1999 influenza outbreak in Pittsburgh. Each data type is plotted on a normalized scale. Lab, influenza cultures from the University of Pittsburgh Medical Center Health System; WebMD, counts of queries to a national web health site using words such as "cold" and "flu"; Cough and cold and cough syrup, grocery chain point of purchase counts; School, school nurse influenza reporting; Resp. and Viral, categories of emergency department ICD-9-coded chief complaints.

can transform the daily or weekly counts into units of "standard deviations from the mean'' (y axis), which enables him to measure the signal-to-noise ratio. This transformation is a standard way of normalizing time-series data in the field of signal analysis (e.g., Lobanov [1971] used this method to compare two time series in the domain of speech analysis). The normalization also allows an evaluator to plot multiple types of surveillance data, which may vary in scaling considerably, on the same graph.

The signal-to-noise ratio is the strength of signal in surveillance data during an outbreak, divided by the level of the signal at baseline in the absence of an outbreak. For example, Hogan plotted the mean and standard deviation of weekly sales of pediatric electrolytes for a four year period (Hogan et al., 2003). The signal-to-noise ratio of sales of pediatric electroytes for winter outbreaks in children was 18.3 standard deviations.

The signal-to-noise ratio is easy to calculate and is informative about the informational value of surveillance data. If the evaluator finds no signal in the surveillance data, there is little point in further study.1 An evaluator can compare the signal-to-noise ratios of multiple types of data to determine which one will provide the most sensitive outbreak detection.

This method cannot provide quantitative answers to the questions of whether and how early an outbreak can be detected. It only answers the question of whether there is any outbreak effect in the data and suggests how strong and early the effect is.

0 0

Post a comment