Biology Reference
In-Depth Information
Rodman, Frost, and Jakubowski 1998), hospital admissions (Davies and
Finch 2003; Hogan et al. 2003; Lazarus et al. 2002), and clinical visits—in
particular, those of certain syndromic categories and deaths (Lazarus et al.
2001; Tsui et al. 2002).
For biosurveillance, the usefulness of a signal is highly dependent on how
early the warning is received. The performance of a biosurveillance system
can be measured by three indices for the alarms generated: sensitivity, speci-
ficity, and timeliness (Buckeridge et al. 2005; Mandl et al. 2004).
In the context of biosurveillance, timeliness is the period between the
occurrence of an event, such as an outbreak, and its detection (Wagner,
Moore, and Aryel 2006). Timeliness does not have a well-established defini-
tion represented by a mathematical equation (Stoto et al. 2005). However,
researchers generally define timeliness as the difference between the time of
an event and that of the reference standard for that event. In the literature,
timeliness has been used interchangeably with the terms “earliness” and
“time lead” (Hogan et al. 2003; Magruder 2003).
Timeliness has been characterized through various methods of different
levels of complexity. These methods are peak comparison, aberration detec-
tion comparison, and correlation (see Figure 1.1).
Timeliness Measurements of Data Source A Compared to B
B
A
Time
1. Calculate the correlation at
different time lags
2. Graph time lag versus
correlation
3. Timeliness is when
correlation is at a maximum
1. Construct a time series for
each data source
2. Calculate the time differrence
between the peaks in each
data source
1. Apply a detection algorithm
to each time series
2. Calculate the time differrence
between the alerts in each
source
Peak Comparison
Aberration Detection Comparison
Correlation
B
B
A
A
Time
Time
Time
n
date
1
date
2
date
1
date
2
Ti Timeliness (A) = date
2
- date
1
Ti Timeliness (A) = date
2
- date
1
Ti Timeliness (A) = n
Figure 1.1
Three methods for determining the timeliness of data source A compared to data source B.
(From Dailey, Watkins, and Plant 2007. Timeliness of data sources used for influenza surveil-
lance.
J Am Med Inform Assoc
14: 626-31.)
