Milwaukee Biosurveillance Project: Real-Time = Syndromic Surveillance Using Secure Regional Internet

From: Subject: Milwaukee Biosurveillance Project: Real-Time Syndromic Surveillance Using Secure Regional Internet Date: Tue, 21 Oct 2003 10:26:34 +0300 MIME-Version: 1.0 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Location: http://216.239.59.104/search?q=cache:lB4Y0-C1xMgJ:www.nyam.org/events/syndromicconference/2002/Supplementpdf/Abstracts_SectionIII.pdf+craig+feied&hl=en&ie=UTF-8 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1165 =EF=BB=BF Milwaukee Biosurveillance Project: Real-Time = Syndromic Surveillance Using Secure Regional Internet

This = is the html=20 version of the file http://www.nyam.org/events/syndromicconference/2002/Suppleme= ntpdf/Abstracts_SectionIII.pdf.
G o = o g = l e=20 automatically generates html versions of documents as we = crawl the=20 web.
To link to or bookmark this page, use the following = url:=20 =

http://www.google.com/search?q=3Dcache:lB4Y0-C1xMgJ:www.nyam.org/ev=
ents/syndromicconference/2002/Supplementpdf/Abstracts_SectionIII.pdf+crai=
g+feied&hl=3Den&ie=3DUTF-8

Google is not affiliated with the = authors=20 of this page nor responsible for its=20 content.

These=20 search terms have been highlighted:

craig

feied

Page 1

i126

ABSTRACTS

earlier = versus=20 later, respectively. No significant deviations were observed using=20 CUSUM

methods. We found=20 higher ED-based respiratory-related chief complaints rates during a=20 time

of = intense forest=20 fire activity near Denver. UC visit rates for asthma were decreased=20 compared

to the = prior=20 period. Public health information alerts may have played a role in=20 decreasing

exposure and=20 avoiding increased HCU.

Milwaukee=20 Biosurveillance Project: Real-Time Syndromic = Surveillance

Using Secure=20 Regional Internet

Seth=20 Foldy,

1,2

Paul=20 Biedrzycki,

Edward=20 Barthell,

2,3

Nancy=20 Haney-Healey,

Bevan=20 Baker,

Donna=20 Howe,

Douglas = Gieryn,

and the = Milwaukee=20 Biosurveillance Workgroup

City = of=20 Milwaukee Health Department,

Medical=20 College of Wisconsin,

3 Infinity Healthcare, 4

Waukesha=20 County Public Health

Milwaukee,=20 Wisconsin, was visited by 1.2 million people for events, including the = All=20 Star

baseball game, in=20 July 2002. Eight emergency departments (EDs), four primary/urgent care=20 prac-

tices, = and one=20 medical examiner reported, using existing personnel, syndromes = associated=20 with

bioterrorism=20 agents to the Milwaukee Health Department daily for 4 weeks. Clinicians = were=20 to

complete a brief=20 symptom checklist during each patient encounter. In practice, some=20 EDs

screened only=20 selected patients, and many supplemented clinician reports with log = reviews.=20 Daily

ED = syndrome and=20 total visit volume reports were collected and displayed using the=20 EMSystem

secure = Web site.=20 Patient-identifying information was not sent to the Milwaukee Health=20 Depart-

ment, = but was=20 retained at the ED in case needed. Participating EDs were visited by=20 26,888

patients, and 314=20 patients were reported to meet syndrome criteria over the 4 weeks. The=20 rate

of = syndrome cases=20 to total visits ranged from 0.04% to 2.8% across the various EDs; EDs=20 that

relied=20 exclusively on physician checklists had lower syndrome-to-visit rates. = Mean ED=20 adminis-

trator = ratings of=20 implementation and reporting ease ranged from neutral to modestly=20 positive.

They = negatively=20 rated the ease of clinician involvement. Mean clinician ratings of their = experi-

ence = were neutral=20 to modestly negative. Estimated added patient time in ED averaged less=20 than

4 = minutes.=20 Estimated total additional staff time per patient approximated 10 = minutes.=20 Primary

care = practices=20 reported a higher syndrome rate (8.8% of 2,442 visits), which included a = camp-

associated=20 cluster of streptococcal pharyngitis. A Web-mounted = =E2=80=9Cdashboard=E2=80=9D facilitated=20 com-

parison = of=20 syndrome rates and other surveillance trends. The Web site facilitated=20 collection,

analysis, and=20 display of surveillance information.

SECTION III:=20 DATA TRANSFER AND TRANSFORMATION

The = Frontlines of Medicine Project: a Proposal for the = Standardized

Communication=20 of Emergency Department Data for Public Health

Uses=20 Including Syndromic Surveillance for Biological

and = Chemical=20 Terrorism

Edward = N.=20 Barthell,

1,2

William H.=20 Cordell,

John = C.=20 Moorhead,

Jonathan=20 Handler,

Craig Feied,

Mark = S.=20 Smith,

Dennis = G.=20 Cochrane,

Christopher W.=20 Felton,

Michael A.=20 Collins,

Kim R. = Pemble,

1,8

and = Brian F.=20 Keaton

Infinity=20 HealthCare, Inc., Wisconsin,

Medical=20 College of Wisconsin,

Indiana=20 University School of Medicine,

Oregon Health=20 and Sciences University,

Northwestern=20 University,

Washington=20 Hospital Center, Washington, DC,

Morristown=20 Memorial Hospital, New Jersey,

University of=20 Wisconsin=E2=80=93Milwaukee,

Summa Health=20 System, Akron, Ohio

Page = 2

ABSTRACTS i127

The = nation=E2=80=99s=20 emergency departments (EDs) are a potential source of surveillance=20 informa-

tion. = The=20 Frontlines of Medicine Project is a collaborative effort of emergency=20 medicine,

public = health,=20 emergency government, law enforcement, and informatics to develop=20 nonpro-

prietary,=20 standardized methods for reporting emergency department patient data. An = initial

proposal,=20 published in April 2002, proposed a standardized message structure based = on

XML = (Extensible=20 Markup Language) for reporting triage information from emergency=20 de-

partments to=20 regional surveillance centers and called for reader comments. = Subsequently,=20 a

consensus=20 conference, with attendees chosen through a modified nominal consensus=20 process,

was = held to=20 discuss the initial Frontlines proposal and provide recommendations for=20 next

steps. = Since the=20 consensus conference, an Internet-based Delphi survey technique has=20 been

used = to refine=20 further the Frontlines recommendations. The technique was utilized for=20 two

rounds = to yield=20 a consensus exceeding 75% acceptance of the proposed data elements=20 and

preferred=20 International Classification of Diseases, 9th Revision=20 (ICD-9)=E2=80=93coded chief com-

plaint = values.=20 The data elements for the triage surveillance report include provider=20 facility

ID, = patient ID,=20 encounter ID, patient age, age unit, gender, date/time first documented=20 in

ED, = date/time=20 symptom onset, chief complaint, first ED responsiveness assessment, = first=20 ED

systolic blood=20 pressure, first ED diastolic blood pressure, first ED heart rate, first = ED=20 temper-

ature, = ED=20 temperature unit, and ZIP codes for home, work, and incident site. The=20 preferred

chief = complaint=20 categories include 159 complaints arranged in 16 hierarchical categories = that

are = expected to=20 describe the reason for visit in greater than 99% of ED encounters.=20 Further

details are=20 available at www.frontlinesmed.org.

Foodborne=20 Outbreak Early Detection System = (FOEDS)

Paul = C.=20 Bartlett,

Holly=20 Wethington,

Bryan=20 DeZeeuw,

Sally=20 Bidol,

John=20 Tilden,

Theresa=20 Bernardo,

Lixin=20 Zhang,

Dean=20 Sienko,

and = Mary Grace=20 Stobierski

National=20 Center for Food Safety and Toxicology, Michigan State=20 University,

2 Ingham

County Health=20 Department,

Department of=20 Epidemiology, University of = Michigan,

Food and=20 Dairy Division. Michigan Department of=20 Agriculture,

5 Bureau

of=20 Epidemiology, Michigan Department of Community=20 Health,

6 Information

Technology,=20 Michigan State University College of Veterinary=20 Medicine

The = FOEDS=20 (Foodborne Outbreak Early Detection System) Forum = (www.RUsick2.msu

.edu) = is a=20 structured, Web-based forum that collects and shares data regarding a = 4-day=20 food

history, food=20 sources, animal contact, and other risk factors that are helpful in=20 establishing

the = existence of=20 a time-space cluster of possible foodborne origin. It is a syndromic=20 surveil-

lance = system=20 that allows users to search the database to evaluate the possibility = that a=20 group

of = people became=20 sick with the same symptoms at about the same time after eating the=20 same

food = from the=20 same source. The FOEDS Forum is designed to identify suspicious=20 time-space

disease clusters=20 that may, at the local health department=E2=80=99s discretion, be worthy = of=20 further

investigation.=20 As such, it can be viewed as a =E2=80=9Cfront end=E2=80=9D to our = current national system=20 for

identifying and=20 investigating foodborne outbreaks. Data collection was scheduled to=20 begin

in = October 2002=20 in the three-county area of Greater Lansing, Michigan. Clinic-based=20 and

population-based=20 advertisements were to encourage people with suspected foodborne=20 disease

to = visit the Web=20 site to determine if they ate the same food that others ate before = becoming=20 ill

with = similar=20 symptoms. Input screens and output reports will be presented, as will=20 program

implementation=20 in the three-county pilot area. The FOEDS Forum was developed by=20 epide-

miologists from=20 state and local governmental agencies and academic departments=20 working

under = the=20 umbrella of the National Food Safety and Toxicology Center at Michigan=20 State

University.

Page = 3

i128 ABSTRACTS

Conceptual=20 Models: Definitions, Construction, and Applications

in = Public=20 Health Surveillance

Rachel = Richesson=20 and James P. Turley

University of=20 Texas School of Health Information = Sciences

Conceptual=20 models are the core of robust classification schemes (e.g.,=20 SNOMED-RT,

GALEN) = and=20 emerging standards for the exchange of health care information (e.g.,=20 HL-7).

Conceptual=20 models are defined here as a conceptualization, simplification, or = abstraction=20 of

reality.=20 Community surveillance activities involve representing information for = sharing,=20 and

therefore they=20 rely heavily on conceptual models. Conceptual models, either implicit or = ex-

plicit, act to=20 guide processes of information exchange and as the context for=20 assimilating

heterogeneous=20 data. An explicit conceptual model is what guides aggregation of = =E2=80=9Cunits=20 of

information=E2=80=9D and=20 will be an essential component of successful surveillance systems.=20 The

Centers for=20 Disease Control and Prevention has labeled conceptual data modeling as=20 =E2=80=9Cone

of the = most=20 powerful and effective analytical techniques ever developed for=20 understanding

and = organizing=20 information required to support any enterprise.=E2=80=9D There is little = formal=20 discus-

sion = about what=20 constitutes strong conceptual models or a formal methodology for=20 how

they = are=20 constructed. This project examined qualitative data from two sources to = develop=20 a

generalized=20 methodology for constructing conceptual models. Narratives from the=20 literature

in = various=20 domains are explored, and these data were triangulated with content=20 analysis

from a = case=20 study. In this case study, domain experts reviewed and commented on a=20 concep-

tual = model=20 designed to represent pediatric asthma knowledge. This triangulation of=20 literature

and = expert=20 reviews identified a generalized methodology for the construction and=20 evaluation

of = conceptual=20 models. In addition to the discussion of conceptual model development,=20 this

poster = presents=20 potential applications of conceptual models for data aggregation and=20 manip-

ulation critical=20 to syndromic surveillance.

Tools to=20 Facilitate the Interchange and Analysis of = Nontraditional

Health=20 Surveillance Data

Zachary Pincus,=20 David L. Buckeridge, Michael K. Choy, Justin V. Graham,

Martin = J.=20 O=E2=80=99Connor, and Mark Musen

Stanford=20 Medical Informatics, Stanford University School of=20 Medicine

The = use of=20 =E2=80=9Cnontraditional=E2=80=9D health data, such as school absenteeism = and pharmacy=20 sales,

is = thought to=20 increase the sensitivity and timeliness of warnings from syndromic=20 surveillance

systems. A=20 common description scheme for nontraditional data and data sources will=20 greatly

increase the=20 utility of such data. Standards such as those proposed in the National=20 Electronic

Disease=20 Surveillance System and the Health Level 7 Reference Information Model=20 enable

interchange of=20 traditional epidemiological and clinical data. Unfortunately, such=20 standards

were = not=20 designed to accommodate the needs of nontraditional data and their = sources.=20 To

meet = these=20 needs, we have developed a flexible, extensible ontology (metadata = model).=20 Most

data = standards=20 attempt to preenumerate fields and formats for every needed piece of=20 infor-

mation. However,=20 it is not yet established which types of nontraditional data and=20 metadata

are = necessary or=20 useful. In addition, heterogeneity in the format and structure of=20 nontradi-

tional = data=20 makes it difficult to construct a single, comprehensive = =E2=80=9Cstandard.=E2=80=9D Instead,=20 our

ontology allows=20 users to construct detailed, customized, machine-readable descriptions=20 of

data = sources,=20 data formats, and relationships between data sources by building them up = from

simple = terms and=20 attributes provided by the ontology. New classes of data or data=20 sources,

and = individual=20 instances of these classes, can be entered through an open source=20 knowledge

acquisition=20 application. In an initial evaluation, we used the ontology to produce a = model=20 of

output = data from=20 a simulated outbreak. We are continuing to validate the ontology as=20 a

Page = 4

ABSTRACTS i129

robust = and=20 extensible structure for rapidly characterizing, describing, and=20 communicating

nontraditional=20 data.

A=20 Knowledge-Based Approach to Defining=20 Syndromes

Justin = V.=20 Graham, David L. Buckeridge, Zach Pincus, Michael K. Choy,

Martin = J,=20 O=E2=80=99Connor, and Mark A. Musen

Stanford=20 Medical Informatics, Stanford University School of=20 Medicine

Syndromic=20 surveillance can only produce meaningful results if there is a common=20 under-

standing of what=20 observations constitute a syndrome and consequently how a = syndrome

relates to=20 diseases that may cause those observations. However, the constituent = elements=20 of

syndromes, such=20 as =E2=80=9Cflulike illness,=E2=80=9D are poorly characterized and = rarely explicitly=20 defined

by = surveillance=20 system developers. We describe here a preliminary ontology for the=20 creation

of = bioterrorism=20 syndrome knowledge bases that will facilitate sharing and comparison=20 of

knowledge=20 independent of a particular system or research group. In addition, we = have=20 cre-

ated = an=20 inference heuristic problem-solving method that can relate indirect = measurements=20 of

disease to=20 diseases of interest. Our ontology enables precise enunciation of forms = of=20 evidence

required to=20 diagnose a syndrome. The ontology contains six major categories:=20 syndrome,

syndrome=20 modifier, system affected, sign/symptom, direct supporting evidence, and = indirect

supporting=20 evidence. We have instantiated the ontology for the syndrome = =E2=80=9Cbioweapon=20 respi-

ratory = illness.=E2=80=9D=20 The inference heuristic can use the elements of this ontology to=20 combine

direct = measurements into meaningful abstractions. Each sign and symptom has a=20 defined,

explicit=20 relationship to supporting direct and indirect evidence, like measured=20 temperature

or a = patient=E2=80=99s=20 chief complaint. Similarly, the presence of a syndrome can only be = inferred=20 if

illness within=20 requisite body systems can be substantiated by the presence of symptoms. = We

propose that all=20 developers of syndromic surveillance systems explicitly define their=20 syn-

drome = concepts=20 using a standard ontology. Syndrome definitions can be stored as=20 instanti-

ated = knowledge=20 bases in a common central repository, permitting knowledge sharing=20 and

reuse.

A=20 Knowledge-Based Method for = Surveillance

David = L.=20 Buckeridge, Martin O=E2=80=99Connor, Justin Graham, Michael K. Choy, = Zachary=20 Pincus,

and = Mark=20 Musen

Stanford=20 Medical Informatics, Stanford University School of=20 Medicine

Surveillance of=20 prediagnostic =E2=80=9Cnontraditional=E2=80=9D data sources (e.g., = school=20 absenteeism,

pharmaceutical=20 sales, emergency medical services calls) is expected to enhance the=20 timeliness

of = epidemic=20 detection. However, prediagnostic data are not as specific as diagnostic = data,=20 so

multiple sources=20 must be followed to reduce false-positive detections. Combined analysis=20 of

multiple=20 nontraditional data sources requires knowledge about the relationships=20 between

data = sources,=20 but knowledge of these relationships is often qualitative and uncertain. = Statisti-

cal = methods=20 perform well for focused analyses of quantitative data according to=20 well-defined

models. However,=20 statistical models do not readily incorporate qualitative data and can=20 be-

come = unwieldy as=20 the number of parameters grows. A knowledge-based approach = requires

explicit=20 representation of surveillance knowledge and tasks and enables knowledge = to=20 be

applied to=20 problem solving in a structured manner. Our research approach is to = model=20 the

tasks = involved=20 in public health surveillance and the knowledge required to accomplish=20 these

tasks. = Based on=20 these models, we identify or develop problem-solving methods (PSMs)=20 that

accomplish=20 surveillance tasks. This modular development approach enables controlled = evalu-

ation = of=20 different PSMs and knowledge representations in terms of epidemic = detection=20 and

impact = on=20 decision making around interventions. Prototype methods have been=20 implemented

Page = 5

i130 ABSTRACTS

for = detection of=20 syndromes in individuals using a heuristic approach, normalization=20 using

Kalman = Filtering, and epidemic detection using a Gaussian Bayesian belief = network.=20 Based

on our = experience with these prototypes, we are beginning to formally model=20 surveillance

knowledge=20 requirements, implement additional PSMs, and develop an evaluation=20 framework.

SECTION IV:=20 STATISTICAL METHODOLOGIES

A = Cumulative=20 Sum Approach to Syndromic Surveillance

in = Geographic=20 Regions

Peter = A.=20 Rogerson

Department of=20 Geography, University at Buffalo, Buffalo, New=20 York

A = primary=20 objective of syndromic surveillance is to find as quickly as possible = any=20 signifi-

cant = increase in=20 the rates of syndromes of interest. One way to achieve this is to use=20 cumula-

tive = sum=20 methods; such methods are commonly used in industry to detect unwanted=20 changes

in = industrial=20 processes, and they fall within the more general category of statistical = process

control methods.=20 Cumulative sum (CUSUM) methods operate by accumulating = deviations

between=20 observations and expectations. When these cumulated deviations exceed = some=20 pre-

defined=20 threshold, an alarm is sounded to indicate an increase in the mean of = the=20 underlying

variable of=20 interest. Daily data were available on the number of visits made by=20 patients

residing within=20 given census tracts to clinicians in the Boston, Massachusetts, area for = lower

respiratory=20 infections. Expected counts were modeled for the period = 1996=E2=80=931998 using=20 logis-

tic = regression=20 and using month, weekday/weekend, and a time trend as covariates. In=20 general,

weekends and=20 summer months result in lower odds of office visits in comparison with=20 week-

days = and winter=20 months. The observed counts were then compared with the expected=20 counts

using = the=20 cumulative sum method. The small daily counts suggest that a Poisson = CUSUM=20 be

employed, for=20 which the expectations vary over time. The method outlined can be=20 extended

to = account for=20 the fact that observed counts typically exhibit more variability than a=20 Poisson

model = would=20 suggest. The current approach detects increases in specific regions; a = next=20 step

is to = allow for=20 the possibility that small clusters of geographic regions witness = increases=20 in

rates.

An = Elliptic=20 Spatial Scan Statistic and Its Application to Breast = Cancer

Mortality=20 Data in Northeastern United States

Martin = Kulldorff, Lan Huang, and Linda Pickle

University of=20 Connecticut, National Cancer = Institute

The = spatial scan=20 statistic is commonly used for geographical cluster detection,=20 cluster

evaluation, and=20 disease surveillance. Recent use includes daily analyses of syndromic=20 emer-

gency = room data=20 for the early detection of disease outbreaks in New York City. Whether=20 an

outbreak is due=20 to a natural cause or a bioterrorism attack, this system enables city=20 health

officials to=20 investigate the outbreak as early as possible and, if necessary, to = rapidly=20 imple-

ment = disease=20 prevention and control measures. When applying the spatial scan = statistic, it=20 is

natural to use a=20 circular scanning window to define the potential cluster areas since the = circle

is the = most=20 compact shape that can be obtained. Other shapes are also possible, such = as