Abstract:
The specter of emerging infections and potential bioterrorist threats has lead to an increased interest in population-based syndromic surveillance systems. GIS-enabled syndromic surveillance enhances significantly the utility of such surveillance methods.
A simultaneous three-phased surveillance strategy, which includes spatio-temporal information collected through the Emergency Department Information System (EDIS) in a major metropolitan area, demonstrates no statistically significant historical spatio-temporal association between known vector spaces and the target syndrome population.
A baseline syndrome rate of .5% of 235,000 patient visits for approximately 4 years is being compared on daily, weekly, and monthly values for the year 2003 and prospectively 2004. The rapid availability of spatio-temporal mapping adds a critical informational component to both the clinician’s treatment plan and the public health decision support system.
Introduction:
West Nile Virus (WNV) has
yet to be detected in the indigenous population of Oregon. Oregon public health
authorities anticipate human cases will occur later this year. Surveillance
systems for detection in birds, equines, and mosquito vectors are already
operational. Public health officials are educating health workers on the signs
and symptoms of the WNV disease. In spite of these efforts, the initial
detection of a WNV case in our community is a major challenge. Early detection can be linked to early
intervention, which may limit the extent and impact of the disease on the
population. Currently, no automated surveillance program exists for WNV
infection in humans. Our investigation explores the feasibility of using an
emergency department information system (EDIS) as the foundation of a
surveillance system designed to detect WNV infection in our community.
An EDIS requires the
collection of patient specific symptomatology and diagnoses, as well as point
locational information related with each patient seen and treated. Automated
query and message generation, as well as automated and semi-automated
spatio-temporal analysis may provide a unique opportunity for early detection
of West Nile Virus,
The Oregon Health &
Science University Emergency Department (ED) is located near the downtown
district of the largest metropolitan area in Oregon and Southwest Washington.
This medical center is the only academic center in the state of Oregon. Each
year approximately 47,000 patients are seen in the ED. As part of the
registration process home addresses, as well as employer addresses, are
collected in the Hospital Information System (HIS), which is connected via
interface with the EDIS. This information, together with disease presentation
and diagnoses, are maintained in a single enterprise relational database
(Oracle). Queries are routinely run to detect diagnostic and treatment
patterns, as well as many other patterns and sentinel events, enabled by the
high degree of content and accessibility of the database. Due to its primary
function as a departmental management application, data must be weighted for
confidence, something best accomplished by investigators with a thorough understanding
of ED operations practices.
Methods:
This study uses five full
years (1998-2002) of data from the EDIS to create baseline values for disease
diagnoses and other patient characteristics, such as age, gender, complaint
categories, home and work locations, laboratory orders, and laboratory results.
Patient referral areas or “capture” areas were demonstrated as well through
home address matching and geocoding. Both Streets, Microsoft Corporation
(Bellevue, Washington), geocoding service and the Regional Land Information
System (RLIS) published by Metro, a Portland Area Regional government body,
were used in matching addresses.
Primary data was collected
through a leading industry EDIS (EMSTAT) from A4Healthsystems, Cary, North
Carolina. Oracle 7.3, Microsoft Access 2000, and SQL Server 2000 database
management systems were used to store and manage the data. Patient records were
analyzed using SPSS v11.0. Spatial data management and analysis were
accomplished with ArcGIS v. 8.3, Esri, Redlands, CA.
Department procedures
require triage and primary nurses to elicit patient information, including
complaints and medical history. Registrars collect and code demographic and
billing information either in the EDIS or Hospital Information System (HIS).
These systems are linked through a Health Level 7 (Hl7) interface. Prior to
patient discharge from the ED, OHSU emergency physicians are required to assign
at least one diagnosis, but may assign two and make notes for review at
discharge as well. For the period of the study the department had an average
time in department of 4 hours and 40 minutes, and a standard deviation of 8
hours and 59 minutes. This is the period of time from patient presentation
until the patient is admitted or discharged home. It also includes an
observation unit with stays of over 24 hours being not uncommon.
Case records were analyzed
with particular focus on diagnosis (ICD9 codes) and laboratory results (spinal
tap), home address location, and disposition.
Chief complaint and chief complaint codes are also available, but were
not found to be as consistent for the study period. One study has shown good
agreement between complaint and diagnosis overall (Begier, et al, 2003). An
expanded diagnosis group included both primary and secondary diagnoses deemed
to be consistent with possible WNV
Infection. Two groups were formed. The expanded group included headache,
altered mental status, fever of unknown origin, febrile illness, viral
meningitis, and viral encephalitis. The focal group included only viral
meningitis and viral encephalitis.
There are 18 emergency
departments in the Portland Metro area. While OHSU is not the only such ED in
the core area, an examination of the home addresses of patients presenting to
this ED over the last several years reveals a broad and highly dispersed array
of points. (Figure 3) This array indicates that, over time, the address of
residence of OHSU patients is representative of the entire Portland community.
The potential usefulness of EDIS data in disease surveillance is supported by
the continuous hours of operation, the lack of financial restrictions on access
to care, and the perception of the public of this particular hospital as a
place of high quality care (National Research Corporation, Lincoln, Nebraska
http://www.nationalresearch.com/ ccahosp02.html). An increased number of
indigent and uninsured seek care in this ED at a rate of 40% of admissions (P.
Southard, personal communication, 2003).
Figure 1: National Array of ED patient home addresses
Propinquity of diagnosed
patients home residence to water-born disease vectors is being explored as a
useful input for epidemiological studies (Kleinschmidt, I, et al). Home
residences may provide a proxy for exposure period to adjacent mosquito
breeding areas. While it is expected to be highly variable in the population,
some groups may be identifiable as relatively homebound, thus increasing the
relative risk of exposure in place. Both the old and the very young probably
tend to spend more time at home (Levinson, et al, 1997). By determining home
locations for patients presenting with the target syndrome, we are able to
provide map visualizations demonstrating patients who reside and/or work within
identified known potential mosquito vector ranges.
Figure 1: Regional array of patient locations
Using the target
diagnoses, a broad group totaling 4,570 patient visits over more than 5 years
was selected from the EDIS. These included diagnoses such as headache, febrile
illness, and fever of unknown etiology as well as more specific diagnoses such
as viral meningitis, and or viral encephalitis. Within this expanded group, a
subgroup was identified where either the primary or secondary diagnosis was
viral encephalitis or viral meningitis. Viral meningitis totaled 66%, and viral
encephalitis 13% of the primary diagnoses in the 71 cases in this subgroup.
Temporal or spatial clustering was not evident in direct observation of the
array. No single week with more than two cases of viral meningitis was present
in the data. Encephalitis diagnoses were rare. Headache diagnoses were
typically primary diagnoses, with viral encephalitis as the secondary diagnosis
in all but one case. Table 1 shows the diagnoses associated with these two
viral diagnoses and their ICD9 codes as coded into the EDIS. In 48 of these
cases, no secondary diagnosis was assigned.
Diagnosis |
ICD9 |
Count |
Altered mental status |
780.09 |
1 |
Encephalitis, viral |
049.9 |
9 |
Febrile Illness |
780.6 |
1 |
Headache |
784.0 |
10 |
Meningitis, viral |
047.9 |
47 |
Pneumonia |
486 |
1 |
Sepsis |
038.9 |
1 |
Viral syndrome |
079.99 |
1 |
Grand Total
|
|
71 |
Table
1: Diagnoses associated with focal group of viral encephalitis or meningitis
Outpatient (ED) diagnoses
correlated very highly with inpatient diagnoses on a representative sample of
admissions, with ICD9 codes matching in greater than 90% of the cases sampled,
and approaching 99% for viral meningitis.
Of the 71 patients in the
focused group, only 18 resided in the Metro area. Of these, 15 had valid
addresses in the EDIS. Two of the others had no home address and may have been
homeless. One did not match, though the address appeared valid. Patient’s home
locations were observed for spatial correlations, i.e. proximity to other
locations in the group or to known bodies of water. Dispersion over the Metro
area was high. No consistent observable spatial association with known water
features was found. The water feature dataset is limited to major and minor
waterways, with wetlands.
Two
one-week periods of cases with the target diagnoses were identified over the
entire five years of data to have more than 1 case diagnosed. Only one week had
3 cases. No confidence interval was computable. All cases in the focused group
were within a three-mile buffer of known water sources.
Figure 2: Temporal aggregation
of array of patient locations
Discussion:
The small numbers of
patients in the narrow target diagnosis group (viral meningitis and viral
encephalitis), consistently low at every level of temporal aggregation over the
study period, leads us to believe that any increase in the number of patients
presenting with these diagnoses will be detected by the system. The diagnostic
accuracy of the ED physicians with regards to the two prominent diagnoses (with
more than 50 physicians practicing during the period) is encouraging. We believe viral meningitis will be the
most likely ED discharge diagnosis of patients presenting with West Nile Viral
Infections. We base this conclusion on both our knowledge of the practice of
emergency medicine and rationale for diagnosis coding used by clinicians. Our
data suggests that when the diagnosis of “viral meningitis” is made, this
diagnosis is a good predictor of the definitive diagnosis in those patients
admitted to the hospital.
Because of severe time and
resource limitations, ED diagnoses can frequently represent the physician’s
best understanding of the patient’s disease or injury. We believe ED diagnosis
correlated with spinal taps were accurate because clinicians were aware of the
results of the spinal tap at the time the diagnoses were made.
An inherent limitation of
the use of ED diagnosis is that in many cases, the clinician will use
symptom-based diagnoses if a more specific diagnosis is not available. An
example is abdominal pain. As a diagnosis this accounts for as much as 40% of
patients discharged from the ED over any given period of time. However, even
with the non-specific nature of this diagnosis, when combined with spatial and
temporal analysis, these diagnoses may prove valuable in narrowing the focus of
an investigation.
The frequently generalized
nature of such diagnoses, descriptive analysis of EDIS discharge diagnoses,
combined with related medical, spatial, and temporal elements in the same data
set, will likely further delineate trends than can a more restricted system of
inputs. Such diagnoses may represent syndromes at their early stages, until
confirmed or further specified by inpatient discharge diagnoses or special
testing.
Prospectives:
The target diagnosis group
will be tracked on a daily basis with map visualizations produced where cases
indicate this would be useful to clinicians. Maps will be posted on an intranet
web site accessible from the EDIS. Because the data set includes inter-hospital
transfers from the entire state and region, the data can provide some potential
as a proxy for the region, most of which is rural.
The criteria used to
determine a spatial relationship between home locations and water will include
the known range of the representative mosquito species of the area, the
distance of each location from each other, as well as the distance of each case
from represented water sources. In addition, patient spatial relationships will
be compared with spatial analysis of “dead bird siting” a technique proven to
be of value in previous investigations.
An expanded water feature set will be used by state and local
authorities to determine if such relationships might exist, with de-identified
ED records provided to them routinely through a secure data Virtual Private
Network (VPN). This system of transmission is currently in place and
functioning as a method of transmitting reportable disease cases every 4 hours.
Locational identity will not be protected when cases are deemed reportable. As
a matter of course the cases will be sent only on request of responsible Public
Health authorities. As such they represent a public health concern and can be
escalated to a reportable status.
References:
National
Research Corporation, Lincoln, Nebraska http://www.nationalresearch.com/
ccahosp02.html
1997,
Levinson, David M. “Life-Cycle, Money, Space, and the Allocation of Time.”
Begier,
E.M.; Sockwell, D.; Branch, L.M.; Davies-Cole, J.O.; Jones, L.H.; Edwards, L.;
Casani, J.A.; and Blyth, D.: “The National Capitol Region’s Emergency
Department Syndromic Surveillance System: Do Chief Complaint and Discharge
Diagnosis Yield Different Results?” CDC, Vol. 9, No 3, March, 2003.
Kleinschmidt,
Immo, Sharp, Brian, Mueller, Ivo, Vounatsou, Penelope: Rise in Malaria
Incidence Rates in South Africa: A Small-Area Spatial Analysis of Variation in
Time Trends. Am. J. Epidemiol. 2002 155:257-264.
Author Information:
Christopher
Bangs, MS
Administrative
Manager/Instructor
Department
of Emergency Medicine
Oregon
Health & Science University
3181
SW Sam Jackson Park Way
Portland,
OR 97239
GH
239
Phone:
503-494-7473
Fax:
503-494-1470
bangsc@ohsu.edu