Lincoln laboratory journal



Yüklə 2,73 Mb.
Pdf görüntüsü
səhifə1/3
tarix20.06.2017
ölçüsü2,73 Mb.
  1   2   3

VOLUME 17, NUMBER 1, 2007

n

LINCOLN LABORATORY JOURNAL



101

Health Surveillance and 

Diagnosis for Mitigating 

a Bioterror Attack

Adam Szpiro, Bernadette Johnson, and David Buckeridge

Lincoln Laboratory has been designing 

and developing bio-agent detection systems 

for both the Department of Defense and the 

Department of Homeland Security. Most of 

our work has focused on rapid environmen-

tal detection of a bio-aerosol attack. The attack informa-

tion is used in some cases to initiate protective responses 

such as evacuation or masking, and in others to direct 

appropriate treatment to the targeted population. For 

the most part, these systems are limited to use in specific 

locations, against specific threat scenarios and targets. 

Thus, until environmental detectors are widely deployed 

and integrated with the public health community, the first 

indication of a bio-agent attack is likely to be via exposed 

and infected individuals presenting at points of care.

Events following the release of Bacillus anthracis 

(the causative agent for anthrax) in the United States 

mail system in 2001 illustrate the importance of clini-

cal case findings in sensing and characterizing a bio-

terrorism incident. Ultimately there were 11 confirmed 

or suspected instances of inhalational anthrax, and all 

were identified through clinical diagnoses [1]. Informa-

tion from the initial diagnoses led to an unprecedented 

antibiotic prophylaxis campaign involving about 10,000  

individuals—an  effort  that  probably  prevented  many 

additional cases of disease [2].

In general, how quickly a person receives a correct 

diagnosis depends upon what symptoms are evident, how 

astute the treating physician is, and what access the treat-

ing facility has to advanced diagnostics. Anthrax’s initial 

presentation can closely resemble that of common respi-

ratory viruses, making the disease difficult to distinguish 

from flu-like illness [3]. Unless there were an unusual 

The most significant factor in minimizing the

adverse impact on public health of a biological-

warfare-agent aerosol attack is the time required

to receive appropriate treatment. We look at two

human-health surveillance techniques designed

to reduce the time window between attack and

the start of treatment. In the first technique,

syndromic surveillance, data are assembled from

a variety of sources, including primary reported

symptoms in emergency departments, calls to

911, pharmacy records of specific treatments,

and school absenteeism. Assimilation of these

data sources can provide an early indicator of a

regional or local outbreak of infectious disease.

The second approach, the Biological-Agent

Correlation Tracker (BACTrack), offers a means

of locating, in space and time, the probable

origin of an attack through the use of a volunteer

population who report their health status to a

central source and who also carry some kind of

location tracking device.

»


102

 LINCOLN LABORATORY JOURNAL

n

VOLUME 17, NUMBER 1, 2007



heAlth SurveillAnce AnD DiAgnoSiS for mitigAting A Bioterror AttAck

clustering of cases, or some other information leading to 

suspicion of an attack, days could elapse before the first 

cases were correctly diagnosed. Once even a few cases 

(perhaps even one) were definitively shown to result from 

bio-agent exposure, however, actions would be initiated 

to determine the probable origin and to establish whether 

the exposure was of natural or man-made cause.

Given the possibility of a delay in diagnosing the ini-

tial cases and determining the source of exposure, it is 

likely that many infected individuals would receive treat-

ment too late and that large numbers of people who were 

neither exposed nor infected would receive treatment. 

The public-health impact of an attack can be measured 

in the lives lost by lack of timely treatment, the adverse 

health effects associated with treating well individuals, 

the cost of acquisition and distribution of treatment to 

the population, and the economic and social disruption 

that such a campaign would engender.

timeline for clinical Diagnoses

We  have  performed  analyses  to  quantify  the  effect  of 

detection and identification time on the outcome of a 

bio-agent attack. We have selected Bacillus anthracis as 

an example agent for four reasons: it is regarded as a sig-

nificant bioterrorism threat, it is amenable to treatment, 

data exist on human exposure, and we have historical data 

on treatment efficacy and distribution logistics.

 Rapid diagnosis of the initial anthrax cases is critical 

in order to enable a timely medical response that can save 

the lives of individuals who are infected but not yet sick. 

Antibiotics are highly effective at preventing death if treat-

ment is initiated prior to the onset of symptoms. Because 

anthrax has an incubation period that ranges from several 

days to more than a month, there is a window of oppor-

tunity after an attack during which mass distribution of 

antibiotics can dramatically reduce the number of fatali-

ties. In a companion article in this issue (page 115), 

Diane 

Jamrog and colleagues



 present the results of analyses 

that predict how effective such a prophylactic campaign 

could be as a function of how soon it commences. Their 

results indicate that to decrease the number of fatalities 

by 95%, prophylaxis must begin less than three days after 

an attack, assuming an aggressive timeline to complete 

distribution to the entire population in the course of three 

days. The time available to initiate prophylaxis is even less 

if it takes longer to reach the entire population.

To find out whether traditional medical diagnosis 

provides sufficiently early warning for a mass prophylaxis 

campaign to be effective, we employed a mathematical 

model that focuses on microbiological blood culture as 

the primary means of diagnosis and that simulates the 

time-evolving state of each infected individual in terms 

of  disease  progression,  health-care-seeking  behavior, 

and diagnostic-test status. Because inhalational anthrax 

is rare in the United States, diagnosis of a single case 

without a known environmental risk factor would be suf-

ficient to conclude with high confidence that an attack 

has occurred. Still, a single case would not be sufficient 

basis on which to justify a mass prophylaxis response. 

We  regard  10  as  a  representative  number  of  cases  to 

declare that an attack is large scale. Depending on the 

size of the attack, our model indicates that the first diag-

nosis  would  occur  two  to  three  days  after  the  attack, 

with the tenth diagnosis following approximately one 

day later. As Jamrog shows, this timeline would severely 

limit the response effectiveness. Thus current gold-stan-

dard diagnostic assays are too slow to be of much help in  

characterizing a large-scale attack.

Experimental diagnostic techniques that could pro-

vide reliable identification data within minutes to hours 

do exist, but they are not available to the general medi-

cal community. In a study commissioned by the Defense 

Advanced  Research  Projects  Agency  and  the  Defense 

Threat Reduction Agency, Lincoln Laboratory examined 

the feasibility of building a Department of Defense health 

surveillance and biodefense system (HSBS) for protection 

against and mitigation of biological assaults (see sidebar, 

facing page). One objective of the study was to assess the 

feasibility of developing rapid, broad-spectrum diagnos-

tic tools for detecting and identifying pathogenic micro-

organisms. The 2002 HSBS summary report concluded 

that there were no obvious technological, logistical, or 

legal barriers to the development of such a system. 

Indeed,  because  emerging  diagnostic  technolo-

gies such as gene chips (DNA microarrays) and protein 

microarrays offer a way to provide thousands of concur-

rent assays in a single test format, it seemed reasonable 

to assume that investments in cost reduction or readout 

simplification could lead to practical diagnostic devices 

with the desired characteristics. The diagnostic technol-

ogy would also offer direct patient benefit, hastening its 

acceptance by the medical community. The creation of a 


VOLUME 17, NUMBER 1, 2007

n

LINCOLN LABORATORY JOURNAL



103

ADAm Szpiro, BernADette JohnSon, AnD DAviD BuckeriDge

it has been more than six years 

since a Defense Science Board

task force released the results of a

study that examined deficiencies

in the nation’s readiness against a

biological attack [a]. The study’s

three top recommendations were

to develop a database of signatures

of bio-agents that cause human dis-

ease; create a diagnostic device

(Z-chip), to be used during routine

clinical sampling, which would pro-

vide immediate diagnoses of dis-

eases in the database; and set up a

warning and communication system

that would alert military and civil-

ian health-care organizations in the

event of a confirmed bio-attack.

A follow-on study requested

by the Office of the Secretary

of Defense on the feasibility of a

health surveillance and biodefense

system (HSBS), led by Darryl

Greenwood at Lincoln Labora-

tory, conceptualized such a system

and proposed a feasibility demon-

stration. The health-care provider

interacts with the patient with tech-

nologies that improve speed, accu-

racy, and specificity. Data at the

local point-of-care level are entered

into terminals with patient identity

protected. Information assurance

implements measures to protect

privacy and to protect the system

from intrusion. The network can be

thought of as a secure web, access-

ing data from a variety of sources.

Data are mined in various ways and

presented to a command and con-

trol center, which has the responsi-

bility of informing authorities such

as the Centers for Disease Con-

trol or the White House. Data are

fed back through the system down

to the point-of-care level to aid in

diagnosis. The proposed system

would provide both direct patient

benefit as well as generating real-

time, assured disease-surveillance

information.

The HSBS is theoretically

feasible. The fundamental techni-

cal barrier to its development was

(and remains) the lack of rapid,

broad-spectrum diagnostic assays.

Today’s diagnostic capabilities are

largely symptom driven and rely on

culturing; answers are unavailable

for hours to days to even weeks.

Treatment is often initiated, and

even completed, without certain

knowledge of the infectious organ-

ism. The few diagnostic tests now

in use at the point of care (e.g.,

rapid immunoassays for strep)

offer an excellent starting point for

HSBS, but there are no technolo-

gies sufficiently mature to diagnose

more than a few diseases in a multi-

plex fashion. In addition, the output

of the few point-of-care tests that

are available now is typically cap-

tured only within the health-care

provider’s own information network,

limiting the ability to identify emerg-

ing infections across a population.

reference

a. “Protecting the Homeland: Defense

against Biological Warfare: Report of

the Defense Science Board 2000

Summer Study,” Defense Science

Board, Washington, D.C.



The Health Surveillance and 

Biodefense System

A recommended approach to defending against bio-attack has yet to be implemented.

Point-of-care

terminals

Data mining

tools


Direct patient benefit

Surveillance

Information assurance

••



Diagnostics

HCP


Patient

Network


Actionable

authorities

Other data

sources


Dynamic

data base

C2

figure A.

 In a system such as that proposed by Lincoln Laboratory in 

2002, data mined in various ways are presented to a command and control 

(C2) center that then informs authorities such as the Centers for Disease 

Control. Data are fed back to the point-of-care level to aid in diagnosis by 

health care providers (HCP). 



104

 LINCOLN LABORATORY JOURNAL

n

VOLUME 17, NUMBER 1, 2007



heAlth SurveillAnce AnD DiAgnoSiS for mitigAting A Bioterror AttAck

network that not only collected diagnostic and other dis-

ease data for attack detection but that would significantly 

raise disease-surveillance reporting rates and facilitate 

rapid  response  was  considered  an  important  feature 

in a bio-attack mitigation strategy. In short, the study 

endorsed HSBS as a valid, workable concept and strongly 

recommended that the Department of Defense undertake 

a program to demonstrate the technology.

Unfortunately, with the exception of an Air Force–

funded pilot demonstration program called Epidemic 

Outbreak Surveillance [4], little has been done to advance 

the HSBS concept. A few infectious-disease reporting sys-

tems have been developed, but these systems rely on exist-

ing diagnostic and confirmatory assays for common (or 

newly emerging) diseases. They do not provide for timely 

two-way reporting (up to public-health authorities and 

back down to the point of care), and their latency is highly 

variable. The private sector has little incentive to develop 

diagnostics for biowarfare agents (as they are considered 

rare diseases) and the civilian sector has neither the char-

ter nor coordination to implement a wide-scale surveil-

lance-based information-technology network.

Two Lincoln Laboratory activities are designed to 

reduce the time between exposure and initiation of cor-

rect treatment. The first is an analysis of the role of syn-

dromic surveillance, in which early reporting of common 

symptoms and patterns of therapeutic purchases can be 

used to characterize disease outbreaks. Data are assem-

bled from a variety of sources, including primary reported 

symptoms in emergency departments, calls to 911, phar-

macy records of specific treatments, and school absentee-

ism. The assimilation of these data sources can be used 

to provide timely information about a regional or local 

outbreak of infectious disease. The second approach, the 

Biological-Agent Correlation Tracker (BACTrack), offers 

a means of locating in space and time the probable origin 

of an attack through the use of a volunteer population 

who report their health status to a central source and who 

also carry some form of geolocation tracking. 



Syndromic Surveillance

The idea behind syndromic surveillance is to monitor 

population patterns of nonspecific symptoms to detect 

evidence of a bio-agent-induced epidemic. In part because 

initial symptoms of many bio-agents are nonspecific and 

can appear flu-like (e.g., elevated temperature and respi-

ratory distress), there is a significant delay between the 

onset of symptoms and the time at which cases are diag-

nosed. A single person exhibiting respiratory symptoms 

may not be immediately identified as an anthrax victim, 

for example, but a sudden and unexpected increase in 

individuals with such symptoms throughout the popula-

tion could be taken as an indication of a bioterrorism 

attack. A syndromic-surveillance alert is based on pat-

terns of nonspecific symptoms throughout the population 

and does not depend on even one individual being posi-

tively diagnosed with anthrax. The goal of syndromic sur-

veillance is to use combinations of multiple data sources 

to look for such patterns. Lincoln Laboratory’s primary 

interest is evaluating syndromic surveillance as part of 

an integrated system to defend against a bioterrorism 

attack, specifically focusing on detect-to-treat defense 

against an aerosol anthrax attack.

Many techniques have been devised to collect and 

process data that could be useful in bioterrorism defense 

[5]. Still, many in the public-health community remain 

skeptical about syndromic surveillance [6], for under-

standable reasons. One problem is that the information 

such a system provides tends to be nonspecific. There-

fore, if a syndromic-surveillance system detects an anom-

alous pattern in population health, it would be difficult 

to propose initiating a mass antibiotic campaign without 

independent confirmation that the outbreak is caused 

by anthrax. However, syndromic-surveillance detection 

could at least cue physicians to be more alert to anthrax 

and other bioterrorism-related diseases. This approach 

holds promise for speeding detection by clinical diagno-

sis, but it still depends on a syndromic-surveillance alert 

occurring before the first clinical case finding. 

Moreover, syndromic surveillance has a potentially 

important role in characterizing an attack after it has 

been detected. Diagnosis of a single case of anthrax would 

most likely be sufficient to conclude that there has been 

an attack, but it would not provide adequate information 

to determine that mass prophylaxis is the appropriate 

response. The default option in that circumstance is to 

delay the response until there are additional diagnosed 

cases. However, our analyses indicate that waiting for 

even the tenth case to be diagnosed could cause an addi-

tional delay of one day. Syndromic surveillance could 

augment medical diagnosis by providing information 

about the size of an attack.



VOLUME 17, NUMBER 1, 2007

n

LINCOLN LABORATORY JOURNAL



105

ADAm Szpiro, BernADette JohnSon, AnD DAviD BuckeriDge

We  have  conducted  a  modeling  simulation  study 

to assess the value of syndromic surveillance for early 

detection [7] and for attack characterization [8]. To con-

duct this evaluation, we require background syndromic 

data and simulated data corresponding to an anthrax 

attack. We start with authentic background data from 

the  Department  of  Defenses’  Tricare  health-care  sys-

tem in the Norfolk, Virginia, area, and we tabulate the 

clinic visits with upper-respiratory symptoms under the 

ICD-9 system of classification (ICD-9 stands for Interna-

tional Classification of Disease, 9th Revision). We con-

struct attack data by adding syndromic records resulting 

from patients in a simulated attack being assigned ICD-

9 codes that correspond to upper-respiratory distress 

such as runny nose, cough, and sore throat. We consider  

a syndromic-surveillance system that searches for anoma-

lous patterns of ICD-9 codes for such symptoms as an 

indicator of a biowarfare attack.

A key step in detecting or characterizing an attack 

is to compare the observed syndromic records to what 

would be expected without an attack. Our algorithm gen-

erates the expected number of syndromic visits by fore-

casting the historical trend by using a seasonally adjusted, 

autoregressive  integrated  moving  average  (SARIMA)  

statistical model. 



early Detection

We  begin  by  evaluating  the  capability  of  a  temporal 

algorithm to detect an anthrax attack and comparing 

the detection time to the first clinical case finding. We 

construct a Monte Carlo simulation of the time series of 

outbreak-related upper-respiratory syndromic records 

and add it to the authentic background data to obtain a 

time series corresponding to an attack. Figure 1 shows 

an example time series resulting from an attack affect-

ing 50,000 people. The total number of records includ-

ing anthrax cases evidently exceeds the forecast values, 

so we expect that a temporal-anomaly detection algo-

rithm would generate an alert. We employ a cumulative-

sum detection algorithm tuned to give about one false 

alarm  per  month,  and  test  it  on  simulated  outbreaks  

of different sizes. 

For attacks affecting 10,000 people, the algorithm 

detects the outbreak about 60% of the time; that value 

increases to 100% for attacks affecting 30,000 people or 

more. That is a good result, but keep in mind that our 

primary interest is in the timeliness of syndromic-sur-

veillance detection compared to the first clinical diag-

nosis.  By  that  measure,  syndromic  surveillance  still 

falls short. Even with a large attack, during which this 

approach has its best performance, temporal syndromic 

surveillance typically does not detect the attack prior to 

the first case finding. Note, however, that these results 

represent the median performance, and in some of the 

Monte Carlo runs, syndromic surveillance does provide an  

early-detection advantage.

One approach to improving syndromic surveillance 

is  to  include  spatial  information  with  the  syndromic 

records and to search for anomalous patterns of nonspe-

cific disease in spatial-temporal clusters. The idea is that 

an aerosol anthrax attack may involve releasing the agent 

either inside a building or from a specific area outdoors, 

and in either of these scenarios the exposed population 

would be concentrated in a small area. By looking for 

geographically focused disease clusters, we could rule out 

many of the anomalies that cause false alarms for pure  

temporal surveillance.

To evaluate the performance gained by adding spatial 

information, we extended the simulation study to include 

a spatial-anomaly detection algorithm using home zip 

codes and Poisson regression within each zip code. As 

Figure 2 makes clear, spatial-temporal surveillance offers 

an improvement over temporal surveillance, especially 

when the release occurs from a single point in space. In 

the median case, the syndromic-surveillance detection 

precedes the first clinical diagnosis for all size attacks con-

sidered, with a maximum advantage of a half day. Because 

we rely on home zip codes, these results are valid if an 

attack occurs at a time when most people are at home 

(e.g., early morning). The situation is significantly more 

complicated if we allow for the possibility that people are 

not at home when the attack occurs. 




Yüklə 2,73 Mb.

Dostları ilə paylaş:
  1   2   3




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©azkurs.org 2020
rəhbərliyinə müraciət

    Ana səhifə