Biomedical Engineering Reference
In-Depth Information
using disparate data streams in multicenter data
is a substantial challenge. Marshall et al. [33]
considered the problem of assessing control charts
applied to multiple care providers or hospitals.
They adjusted for several sources of between-unit
variation and treated the multiple testing problem
using false discovery rate estimates. The search
for significant spatiotemporal clusters of cases
using multicenter data has generated several algo-
rithmic approaches [34], notably in scan statis-
tics widely used in cancer epidemiology [35].
Kulldorff et al. [36] investigated the power of
this approach with syndromic data records from
New York City, organized by patient zipcode.
Burkom [37] extended the approach to multiple
data types, controlling for differences in scale and
allowing for different spatial organization between
data types.
Aberration detection algorithms have been
developed to aid in hospital surveillance.
Researchers have adapted process control charts
and other tools to solve routine decision prob-
lems, and the resulting methods are being applied
to monitor for outbreaks of unknown description
that could result from a deliberate, covert act of
biological warfare. New case definition and data
adjustment procedures are producing time series
to be monitored for this purpose, and both data
modeling and hypothesis testing are being refined
to obtain the sensitivity, specificity, and timeliness
that bioterrorism surveillance will require. Even
so, the resulting algorithms can find only statis-
tical anomalies; follow-up investigation must effi-
ciently explain away irrelevant events that cannot
be modeled and examine those that remain for
public health significance.
3.11 Issues in Aberration Detection
The appropriate chart for any particular process
depends on both the in-control data behavior and
the effect expected from special cause variation.
For example, Shewhart charts are most effective
for rapidly detecting sudden changes, while Expo-
nentially Weighted Moving Averages (EWMA)
and Cumulative Summation (CUSUM) charts are
used to detect gradual changes. A thorough discus-
sion of these issues has been presented by Morton
et al. [24].
Other data complications, including varying
risks, small counts, and lack of data history
have been treated successfully in practical contexts
[25-28], and the respective adaptations are being
evaluated for
the detection of outbreaks
in
syndromic surveillance systems.
The application of these methods to bioterrorism
surveillance poses several challenges. Alibek [29]
reported biowarfare research aimed at altering
disease symptomatology by genetically altering the
antigenic presentation in the host. Other sections of
this chapter discuss the aggregation of diagnosis-
coded and chief-complaint data into syndrome
group categories to increase the sensitivity to an
unknown data signal. The time series behavior of
the counts or rates of these non-specific syndrome
groups must be understood in order to manage false
alarm rates in routine health monitoring. Mandel
observed [30] in 1969 that control chart perfor-
mance could be improved by replacing monitored
quantities by the residuals of regression using inde-
pendent covariates. Regression modeling has since
been applied to reduce the effects of day-of-week,
seasonality, late reporting, and other covariates
[23,31,32], and research is ongoing to produce
optimal combinations of data modeling and control
chart methods.
3.13 Summary
Syndromic surveillance is a new and rapidly
evolving approach to detecting important changes
in community health status as early as possible. It
depends upon the identification of opportunities to
use data in new ways as well as the identification
of novel data sources and statistical methods. It
is intended to enhance the traditional surveillance
systems already in place, such as traditional
notifiable disease surveillance and the anecdotal
3.12 Upcoming Challenges in
Hospital-Based Aberration Detection
Increasing concern over the threat of bioterrorism
is driving research for decision-making capability
well beyond interest in the change points of a
single time series. Multiple hypothesis testing
