Original Contributions
Prediction of Traumatic Wound Infection With a
Neural Network-Derived Decision Model
RICHARD L. LAMMERS, MD,* DONNA L. HUDSON, P
H
D,†
AND MATTHEW E. SEAMAN, MD‡
The objective of this study was to develop and validate a decision model,
usingan artificial neural network, that predicts infection in uncompli-
cated, traumatic, sutured wounds. The study was a prospective, cohort
study of all patients presentingto the emergency department of a county
teachinghospital with uncomplicated wounds that required suturing. In
evaluatingand treatingwounds, emergency medicine (EM) faculty and
residents, resident physicians in primary-care specialties, and super-
vised medical students on EM clerkships followed a standardized
wound-management protocol. Clinicians estimated the likelihood of sub-
sequent infection usinga 5-point scale. Wound healingwas followed
until sutures were removed. Wound outcome data were collected by
medical personnel blinded to the initial prediction. Student’s t-tests and
Pearson’s chi-square statistic were used to identify independent predic-
tors that served as input variables. Wound infection was the single
output variable. Neural network analysis was used to assign weights to
input variables and derive a decision equation. A total of 1,142 wounds
were analyzed in the study. The overall infection rate was 7.2%. The most
predictive factors for wound infection were wound location, wound age,
depth, configuration, contamination, and patient age. To derive a deci-
sion equation for the model, the network was trained on data from half of
the subjects and tested on the remainder. When used as a diagnostic test
for wound infection, the decision model had a sensitivity of 70%, as
compared to 54% for physicians, and a specificity of 76%, as compared
to 78% for physicians. We conclude that through the use of combina-
tions of 7 clinical variables available at the time of initial wound man-
agement, a neural network-derived decision model may be used to iden-
tify uncomplicated, traumatic wounds at higher risk for infection. (Am J
EmergMed 2003;21:000-000. Copyright 2003, Elsevier Science (USA). All
rights reserved.)
Despite optimal management in the emergency depart-
ment (ED), some “uncomplicated” traumatic wounds will
become infected. The reported infection rate varies from
about 1to 31%, depending on wound characteristics, treat-
ment, definition of infection, follow-up rate, and method of
wound assessment.
1-14
Many individual risk factors for
wound infection have been identified, including wound
age,
10,11,15,16
anatomic site and tissue vascularity,
4,11,17-19
visible contamination,
8,10,11,17,19
types of soil contamina-
tion,
20-23
wound configuration,
24
amount of devitalized tis-
sue,
25
wound length or number of sutures,
8,9,18
use of epi-
nephrine solutions,
26,27
foreign bodies,
28
hematoma
formation,
29
and experience of the treating physician.
10,30
Host factors such as tissue ischemia from hypotension or
arteriosclerosis, lymphedema, uncontrolled diabetes melli-
tus, uremia, obesity, malnutrition, remote infection, immu-
nocompromising illnesses and therapies, irradiation, the ex-
tremes of patient age, and multiple trauma also predispose
wounds to infection.
31-39
The presence of only one of these risk factors is seldom
an indication for changing wound management. Some
authors recommend primary wound closure only if the
wound is less than 4 to 6 hours old; others apply a
19-hour rule.
6,40-44
However, other factors can alter the
wound-age rule. A contaminated, stellate laceration on
the foot of an elderly diabetic individual who presents for
care within 30 minutes is likely to be at higher risk for
infection than a 24-hour old, clean, tidy, facial laceration
in a young, healthy patient. Clinicians must consider
complex combinations of clinical factors in making
wound-treatment decisions.
Quantitative bacteriologic analysis has been used exper-
imentally with 95% accuracy to determine the safety of
wound closure, but this test is not available for routine use
in most institutions.
3,45-47
If it were available, it would be
impractical to biopsy every wound. The ability of emer-
gency physicians to accurately predict wound infection has
not been studied. If physicians could identify wounds that
appear uncomplicated but are actually at high risk for in-
fection when multiple, minor clinical risk factors are con-
sidered, they could selectively use antibiotics or delayed
primary closure for these wounds.
The objective of this study was to identify important
clinical risk factors for wound infection, and, through use of
an artificial neural network, to develop and validate a deci-
sion model that predicts infection in uncomplicated, trau-
matic, sutured wounds.
From the *Department ofEmergency Medicine, Michigan State
University/Kalamazoo Center for Medical Studies, Kalamazoo, MI;
†Fresno Medical Education Program, University ofCalifornia, San
Francisco, Fresno, CA; and ‡Department ofEmergency Medicine,
Yakima Medical Center, Yakima, WA.
Manuscript received August 20, 2000; accepted January 1, 2001.
Supported by the Valley Medical Center Foundation, Fresno Med-
ical Education Program, University ofCalifornia, San Francisco,
Fresno, CA.
Address reprint requests to Richard L. Lammers, MD, Michigan
State University/Kalamazoo Center for Medical Studies, Depart-
ment ofEmergency Medicine, 1000 Oakland Drive, Kalamazoo, MI
49008. E-mail: Lammers@KCMS.msu.edu
Key Words: Wound, wound management, wound infection, pat-
tern recognition, neural network.
Copyright 2003, Elsevier Science (USA). All rights reserved.
0735-6757/03/2101-0001$35.00/0
doi:10.1053/ajem.2003.50026
1