PEDIATRICS/ORIGINAL CONTRIBUTION
A Clinical Decision Rule for Identifying
From the Division of Emergency
Medicine, Department of Internal
Medicine,* the Department of
Radiology,‡ and the Department of
Pediatrics,§ University of California–
Davis School of Medicine,
Sacramento, CA.
Author contributions are provided
at the end of this article.
Received for publication June 6, 2001.
Revision received December 28, 2001.
Accepted for publication
January 10, 2002.
Presented at the Society for Academic
Emergency Medicine annual meeting,
Atlanta, GA, May 2001.
Address for reprints: James F.
Holmes, MD, University of California–
Davis Medical Center, Division of
Emergency Medicine, 2315 Stockton
Boulevard, PSSB 2100, Sacramento,
CA 95817-2282; 916-734-1539,
fax 916-734-7950; E-mail
jfholmes@ucdavis.edu.
Copyright © 2002 by the American
College of Emergency Physicians.
0196-0644/2002/$35.00 + 0
47/1/122901
doi:10.1067/mem.2002.122901
4 9 2
Children With Thoracic Injuries After Blunt
Torso Trauma
James F. Holmes, MD* See related article, p. 500, and editorial, p. 537.
Peter E. Sokolove, MD*
William E. Brant, MD‡
Study objective: We sought to determine the prevalence of
Nathan Kuppermann, MD, MPH*§
thoracic injuries in children sustaining blunt torso trauma and
to develop a clinical prediction rule to identify children with
these injuries.
Methods: We prospectively enrolled pediatric patients (<16
years) who presented to the emergency department of a Level I
trauma center with blunt torso trauma and underwent chest
radiography. Clinical findings were recorded in a standardized
fashion by the ED faculty physician. Thoracic injuries included
the following: pulmonary contusion, hemothorax, pneumotho-
rax, pneumomediastinum, tracheal-bronchial disruption, aortic
injury, hemopericardium, pneumopericardium, cardiac contu-
sion, rib fracture, sternal fracture, or any injury to the diaphragm.
Multiple logistic regression and recursive partitioning analyses
were performed to generate a clinical prediction rule for identi-
fying children with these injuries.
Results: Nine hundred eighty-six patients with a mean age of
8.3±4.8 years were enrolled. Eighty (8.1%; 95% confidence
interval [CI] 6.5% to 10.0%) patients sustained thoracic injuries.
Multiple logistic regression and recursive partitioning analyses
identified the following predictors of thoracic injuries: low sys-
tolic blood pressure (14% with injury versus 2% without injury;
adjusted odds ratio [OR] 4.6), elevated age-adjusted respiratory
rate (51% versus 16%; adjusted OR 2.9), abnormal results on
examination of the thorax (68% versus 36%; adjusted OR 3.6),
abnormal chest auscultation findings (14% versus 1%; adjusted
OR 8.6), femur fracture (13% versus 5%; adjusted OR 2.2), and
a Glasgow Coma Scale (GCS) score of less than 15 (61% versus
26%; adjusted OR 3.3). Seventy-eight (98%; 95% CI 91% to
100%) of the 80 patients with thoracic injuries had at least 1 of
these predictive factors. Three hundred thirty-six (37%) children
had none of these predictive factors, including 2 (0.6%; 95% CI
ANNALS OF EMERGENCY MEDICINE 39:5 MAY 2002
CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al
0.1% to 2.1%) with thoracic injuries. These 2 injuries, however,
did not require any intervention.
Conclusion: Predictors of thoracic injury in children sustaining
blunt torso trauma include low systolic blood pressure, elevated
respiratory rate, abnormal results on thoracic examination,
abnormal chest auscultation findings, femur fracture, and a GCS
score of less than 15. These predictors can be used to create a
sensible clinical decision rule for the identification of children
with thoracic injuries.
[Holmes JF, Sokolove PE, Brant WE, Kuppermann N. A clinical
decision rule for identifying children with thoracic injuries after
blunt torso trauma. Ann Emerg Med. May 2002;39:492-499.]
INTRODUCTION
Thoracic injuries account for 14% of pediatric trauma-
related deaths and are second only to head injuries as a
cause of death in pediatric patients experiencing blunt
trauma.1 After physical examination, chest radiography
is the initial diagnostic test for the identification of tho-
racic injuries.2 Chest radiographs, however, may not be
necessary for all pediatric patients with blunt trauma.
Although chest radiography is noninvasive and rela-
tively inexpensive, it becomes costly if universally applied
to patients at low risk for thoracic injuries. In addition,
chest radiography exposes the patient to radiation and
requires either transportation of the patient to the radio-
graphic suite or a radiology technician to be present at the
patient’s bedside. Identification of predictors of thoracic
injury, therefore, may provide clinicians with useful in-
formation to enhance clinical and cost efficiency.
The objectives of this study were to determine the
prevalence of thoracic injuries in a population of pedi-
atric patients presenting to the emergency department
after blunt torso trauma and to create a clinical decision
rule to identify children with these injuries.
M AT E R I A L S A N D M E T H O D S
The study was conducted at a Level I trauma center be-
tween April 1996 and September 1998. We prospectively
enrolled pediatric patients younger than 16 years old sus-
taining blunt trauma if they had any of the following and
underwent chest radiography during ED evaluation: blunt
torso trauma from a significant mechanism of injury
(motor vehicle crash, automobile versus pedestrian, falls
of >10 feet); decreased level of consciousness (Glasgow
Coma Scale [GCS] <15) in association with blunt torso
MAY 2002 39:5 ANNALS OF EMERGENCY MEDICINE
trauma; blunt traumatic event with extremity paralysis;
multiple bone fractures; a CRAMS score (measuring circu-
lation, respiration, abdomen, motor responses, and
speech) of 8 or less3; a physical examination suggestive of
torso injury after blunt trauma; or a Pediatric Trauma
Score of 8 or less after blunt trauma. Patients not under-
going chest radiography in the ED or transferred to our
facility after initial evaluation at an outside ED were
excluded.
Patients underwent a complete physical examination
by the faculty emergency physician. Vital signs and physi-
cal examination findings were documented on a struc-
tured data collection sheet after physical examination.
Data sheets were completed as soon as possible after the
physical examination and before chest radiography results
were known in the majority of cases. Low systolic blood
pressure was defined by the following age-adjusted cutoff
points for the initial ED systolic blood pressure: age 0 to 1
month, systolic blood pressure of less than 70 mm Hg; 1
month to 5 years, systolic blood pressure of less than 80
mm Hg; and 5 to 15 years, systolic blood pressure of less
than 90 mm Hg.4 Initial respiratory rates were considered
elevated if they were more than 2 SDs above the age-
adjusted mean on the basis of a previously defined scale.5
The results of the thoracic examination were consid-
ered abnormal if erythema, abrasions, contusions, lacera-
tions, crepitus, or tenderness were identified on the ante-
rior or posterior thorax. Findings on auscultation of the
chest were considered abnormal if breath sounds were
decreased or absent or if crackles or rhonchi were present.
The patients’ abdominal examination results were con-
sidered to be abnormal if tenderness to palpation was
present. The patients’ levels of consciousness were mea-
sured with the GCS for children at least 2 years old and
with the pediatric GCS for children younger than 2 years.6
Patients with a GCS score of less than 15 were considered
to have a decreased level of consciousness.
Thoracic injury was considered present if the patient
had an identified injury to a structure within the thorax,
including pulmonary contusion, hemothorax, pneumo-
thorax, pneumomediastinum, tracheal-bronchial disrup-
tion, aortic injury, hemopericardium, pneumoperi-
cardium, cardiac contusion, radiographically identified
rib fracture, sternal fracture, or any injury to the dia-
phragm. Isolated rib fractures were categorized as tho-
racic injuries because of the greater morbidity and mor-
tality associated with rib fractures in children compared
with in adults and because of the greater impact forces
required to fracture the less calcified ribs of children.7
Injuries were considered present if identified on any of
4 9 3
CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al
the following: plain chest radiography, abdominal or
chest computed tomography (CT), bronchoscopy, tho-
rascopy, tube thoracostomy, thoracotomy, laparotomy,
angiography, ECG, or echocardiography. All diagnostic
studies beyond chest radiography were obtained at the
discretion of the treating physicians. Faculty radiologists
interpreted all radiographic studies.
Patients were admitted to the hospital at the discretion
of the faculty emergency physicians in conjunction with
the trauma surgery team. All admitted patients were fol-
lowed through their hospital courses for identification of
thoracic injuries and therapy for those injuries. Patients
were considered to require specific therapy for a thoracic
injury if they underwent tube thoracostomy, thoracotomy,
therapeutic bronchoscopy, or endotracheal intubation.
Patients discharged to home after ED evaluations were
contacted by telephone 1 week after discharge to deter-
mine clinical status. Patients without symptoms of tho-
racic injury at the telephone follow-up were considered
not to have a thoracic injury. On completion of the study,
we reviewed the trauma registry and continuing quality-
improvement records of the participating institution to
determine whether any enrolled patients discharged from
the ED or inpatient ward without the diagnosis of thoracic
injury were later given a diagnosis of such injuries.
Continuous data were reported as the mean±1 SD.
Variables hypothesized a priori to be associated with chest
injury were entered into both multiple logistic regression
and binary recursive partitioning analyses. These variables
included the following: low systolic blood pressure, elevated
respiratory rate, abnormal thoracic examination results,
abnormal thoracic auscultation findings, abdominal tender-
ness, femur fracture, and a GCS score of less than 15.
We performed binary recursive partitioning in addi-
tion to multiple logistic regression to develop a model
that maximized sensitivity for identifying children with
thoracic injuries, while also maximizing specificity. We
planned to use those variables identified in either the
recursive partitioning or logistic regression analyses for
inclusion in the decision rule.
Binary recursive partitioning refers to an analytic tech-
nique by which patients are divided into subpopulations
(nodes) on the basis of risk profiles for the outcome of
interest.8 Each subpopulation is subsequently subdivided
to minimize misclassification of patients. The process is
continued until either a subpopulation is homogeneous
with regard to the outcome variable or too small to be sub-
divided further. The results of a recursive partitioning
analysis are typically displayed in a tree-like format that is
easy for the clinician to follow. The analysis was validated
4 9 4
internally by using 10-fold cross-validation. Recursive
partitioning may be preferable to multiple logistic regres-
sion when the objective is to develop a decision rule with
the highest sensitivity.9
We considered the same 7 variables for entry into a
multivariate logistic regression analysis, and variables
demonstrating a significant association (P<.05) with tho-
racic injury were considered independent predictors of
chest injury. To validate the logistic regression model, we
performed 2 separate bootstrap resampling procedures
with 1,000 iterations each. Bootstrap resampling ran-
domly selects patients from the original database with
replacement to create new databases of the same size as
the original database. These new databases are subse-
quently used to test the model attributes.10-12 This proce-
dure generates conservative estimates of confidence
intervals (CIs) and assesses the stability of the model. We
obtained 95% bias-corrected CIs of the investigated vari-
ables in the bootstrap analysis. We then repeated the mul-
tivariate analysis on 1,000 bootstrap databases to identify
the percentage of iterations in which each predictive vari-
able was identified as an independent predictor (P<.05)
of thoracic injury. We considered the selection of a vari-
able to be validated if it was selected as significant in
greater than 50% of the bootstrap iterations.11
We calculated the sensitivity, specificity, positive pre-
dictive value, negative predictive value, and positive and
negative likelihood ratios for those variables considered
important in both of these models. We then calculated the
risks of injury in the presence of any of the risk factors
versus the absence of all the risk factors.
Sample size was based on the requirement of 10 out-
comes of interest (thoracic injury) per predictor variable
analyzed in a logistic regression analysis.13,14 Data analy-
sis was performed with STATA 6.0 for Windows (STATA
Corporation, College Station, TX) statistical software.
Answer Tree statistical software (version 2.0, SPSS Inc.,
Chicago, IL) was used to conduct the recursive partition-
ing. The study was approved by the Human Subjects
Research Committee at our institution.
R E S U LT S
A total of 986 patients were enrolled during a 28-month
period. The mean age of the population was 8.3±4.8
years, and the median PediatricTrauma Score was 10
(interquartile range 9 to 11). Mechanisms of injury
included motor vehicle crash in 348 (35%), automobile
versus pedestrian in 244 (27%), fall in 160 (16%), auto-
mobile versus bicycle in 87 (9%), bicycle fall in 53 (5%),
ANNALS OF EMERGENCY MEDICINE 39:5 MAY 2002
CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al

crush injuries in 36 (4%), assault in 28 (3%), child abuse
in 14 (1%), and other in 16 (1%).
Eighty (8.1%; 95% CI 6.5% to 10.0%) patients were
determined to have thoracic injuries. The mean age of
these 80 patients was 8.9±5.2 years, and the median
Pediatric Trauma Score was 8 (interquartile range 5 to
10). Fifty-four (68%) patients with thoracic injuries had
abnormalities on their thoracic examinations, including
the following: erythema (n=1), abrasions (n=41), contu-
sions (n=9), lacerations (n=1), crepitus (n=8), and ten-
derness (n=24). Patients with thoracic injuries and a GCS
score of 15 were more likely to have thoracic tenderness
(18/31 [58%; 95% CI 39% to 75%]) than those patients
with thoracic injuries and a GCS score of 13 or 14 (2/12
[17%; 95% CI 2% to 48%]; odds ratio [OR] 6.9; 95% CI
1.3 to 37). Eleven patients with chest injuries had low
systolic blood pressures.
Six hundred twenty-one (63%) patients were admitted
to the hospital. Telephone follow-up was obtained for
329 (90%) of the 365 patients discharged to home, and all
were asymptomatic at the time of follow-up. There were
no apparent cases of missed thoracic injuries among
patients in this study on review of the trauma registry and
continuing quality improvement records.
The frequencies of specific thoracic injuries identified
are listed in Table 1. Thirty-eight (48%) patients had mul-
tiple thoracic injuries. Fifty (63%; 95% CI 51% to 73%) of
the 80 patients with thoracic injuries had an injury visual-
ized on initial ED chest radiography. The remaining thirty
(37%) patients had their thoracic injuries identified only
on further studies (CT, angiography, or echocardiography).
These 30 patients were ultimately given a diagnosis of the
following injuries: pneumothorax (n=11), hemothorax
(n=3), pulmonary contusion (n=20), rib fracture (n=9),
cardiac contusion (n=1), pericardial effusion (n=1), and
sternal fracture (n=1).
Five hundred sixty-one (57%) patients had imaging
studies in addition to chest radiography. Five hundred
fifty-five (56%) patients underwent abdominal CT, of
whom 56 (10%) demonstrated thoracic injuries. Eight
patients underwent chest CT scanning, of whom 3 (38%)
had thoracic injuries. Eleven patients underwent angiog-
raphy of the thoracic aorta, and 2 (18%) had thoracic aor-
tic injuries.
Fifty (63%; 95% CI 51% to 73%) of 80 patients with
chest injuries underwent specific therapies. Twenty-six
(33%; 95% CI 22% to 44%) patients underwent tube tho-
racostomy. Three (4%; 95% CI 1% to 11%) patients
underwent thoracotomy. Thirty-eight (45%; 95% CI 34%
to 57%) patients underwent endotracheal intubation.
Seven (9%; 95% CI 4% to 17%) patients underwent bron-
choscopy.
Results of the multiple logistic regression analysis are
shown in Table 2. Independent predictors of thoracic
injury included low systolic blood pressure, elevated res-
piratory rate, abnormal thoracic examination results,
abnormal chest auscultation findings, and a GCS score of
less than 15. Femur fracture showed a trend toward statis-
tical significance. The area under the model receiver
operating characteristic curve was 0.82. The model

Table 1.
Frequency of thoracic injuries in the study population. Table 2.
Results of multiple logistic regression.

Frequency Among Frequency

Patients With Among Total Frequency With
Thoracic Injuries Population (n=986), Which Variable
Type of Injury No. (n=80), % (95% CI) % (95% CI) Bias: Was Identified
Adjusted Corrected P as an Independent
Pulmonary contusion 57 71 (60–81) 5.8 (4.4–7.4) Variable OR 95% CI* Value Predictor,* %
Rib fracture 28 35 (25–46) 2.8 (1.9–4.1)
Isolated rib fracture 9 11 (5–20) 0.9 (0.4–1.7) Hypotension 4.6 1.0–13.8 .003 78
Pneumothorax 20 25 (16–36) 2.0 (1.2–3.1) GCS score <15 3.3 1.9–5.6 <.001 99
Hemothorax 9 11 (5–20) 0.9 (0.4–1.7) Abnormal thoracic 3.6 2.1–6.4 <.001 100
Hemopneumothorax 5 6 (2–14) 0.5 (0.2–1.2) examination results
Pneumomediastinum* 6 8 (3–16) 0.6 (0.2–1.3) Abnormal chest 8.6 1.9–31.3 <.001 91
Cardiac 5 6 (2–14) 0.5 (0.2–1.3) auscultation findings
Aortic 2 3 (0–9) 0.2 (0.0–0.7) Elevated respiratory rate 2.9 1.5–5.1 <.001 94
Diaphragmatic injury 1 1 (0–7) 0.1 (0.0–0.6) Femur fracture 2.2 0.9–5.1 .08 39
Sternal fracture 1 1 (0–7) 0.1 (0.0–0.6) Abdominal tenderness 1.1 0.6–1.9 .76 7
*Includes 2 patients with tracheal lacerations. *
In bootstrap validation, as described in the “Materials and Methods” section.

MAY 2002 39:5 ANNALS OF EMERGENCY MEDICINE 4 9 5

CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al

demonstrated satisfactory goodness-of-fit, as measured
with the Hosmer-Lemeshow test (P=.70). The bootstrap
analysis identified all 5 variables as independent predic-
tors of thoracic injury in more than 50% of the 1,000
bootstrap iterations (Table 2).
The results of the recursive partitioning analysis are
demonstrated in the Figure. In the resulting tree, the most
important variable, chest auscultation, is the top node of
the decision tree. As the tree branches, the risk of thoracic
injury can be estimated for particular combinations of
variables. This multivariate analysis identified the same 5
predictor variables as the multiple logistic regression
analysis but added femur fracture as an important vari-
able in the model.
The sensitivity, specificity, positive and negative pre-
dictive values, and positive and negative likelihood ratios
for each of the high-risk variables identified in either
analysis are presented in Table 3. At least 1 of the predic-
tors of thoracic injury identified in either analysis (low
systolic blood pressure, an elevated respiratory rate, an
abnormal thoracic examination, an abnormal chest aus-
cultation finding, femur fracture, or a GCS score of <15)
was present in 650 (66%; 95% CI 63% to 69%) patients
undergoing chest radiography. Included in this group
were 78 (sensitivity 98%; 95% CI 91% to 100%) of the 80
patients with thoracic injuries. Three hundred thirty-six
(34%) patients had none of these predictive factors dur-
ing the ED evaluation, including 2 (0.6%; 95% CI 0.1% to
2.1%) patients with thoracic injuries. These 2 patients are
described in Table 4.
One of the 2 injured patients without any of the high-
risk predictors experienced shortness of breath during
ED observation but after initial evaluation. This patient
had both a pulmonary contusion and a pneumothorax
identified on abdominal CT scan, but only the pulmonary
contusion was visualized on ED chest radiography. The
child was observed in the hospital for 3 days but without
specific therapy for these injuries. The second patient had
a pulmonary contusion visualized only on abdominal CT
scan. This patient was observed in the hospital for 1 day

Figure.
Results of the recursive
partitioning analysis.
Each box represents the
number of patients with
and without thoracic
injury given the particular
combination of variables.
No injury 906 (92%)
Thoracic injury 80 (8%)
Abnormal
Yes No
auscultation
No injury 9 (45%) No injury 897 (93%)
Thoracic injury 11 (55%) Thoracic injury 69 (7%)
Yes Low systolic No
blood pressure
No injury 12 (55%) No injury 885 (94%)
Thoracic injury 10 (45%) Thoracic injury 59 (6%)

Yes Tachypnea No

No injury 135 (84%) No injury 750 (96%)

Thoracic injury 26 (16%) Thoracic injury 33 (4%)

Yes GCS score <15 No

No injury 173 (91%) No injury 577 (97%)

Thoracic injury 17 (9%) Thoracic injury 16 (3%)

Yes Abnormal thoracic No

examination results
No injury 218 (94%) No injury 359 (99%)
Thoracic injury 13 (6%) Thoracic injury 3 (1%)

Yes Femure No
fracture
No injury 25 (96%) No injury 334 (99%)
Thoracic injury 1 (4%) Thoracic injury 2 (0.6%)

4 9 6 ANNALS OF EMERGENCY MEDICINE 39:5 MAY 2002

CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al

and was discharged home without further incident or
therapy.
DISCUSSION
In this series of pediatric patients with blunt trauma evalu-
ated in the ED, 8% of those undergoing chest radiography
had thoracic injuries. Clinical findings associated with
thoracic injuries included low systolic blood pressure, ele-
vated respiratory rate, abnormal results on thoracic exam-
ination, abnormal chest auscultation findings, femur frac-
ture, and a GCS score of less than 15. Twelve percent of
patients with any one of these factors had thoracic injury
versus 0.6% of patients without any of these risk factors.
Pulmonary contusion was the most prevalent thoracic
injury identified in this population, which is similar to
what has been seen in prior studies.1,15-18 After pul-
monary contusion, pneumothoraces and rib fractures
were the most prevalent injuries in our study population.
Rib fractures, however, were identified in only 3% of the
patient population. Rib fractures are considered to occur
less frequently in children than in adults yet are associ-
ated with other significant injuries in the pediatric popu-
lation.7 Similar to previous studies, aortic injuries,
diaphragmatic injuries, and sternal fractures were rare in
this population,1,15,19 and no patient enrolled in the
study had a bronchial injury identified.
Abnormal chest auscultation findings had the highest
predictive value for thoracic injury, with a positive likeli-
hood ratio indicating a large effect on posttest likelihood
of disease. In our study, 35% of patients with pneumotho-
races and 18% of patients with pulmonary contusions
had abnormal findings on auscultation of the chest. One
prior study demonstrated that 46% of children with pul-
monary contusions have abnormal findings on chest aus-
cultation.20
Tachypnea was present in more than half of the patients
with thoracic injuries when age-adjusted standards were
applied. One prior study reported that more than 90% of
children with pulmonary contusions have respiratory

Table 3.
Accuracy of clinical findings for predicting chest injury.

Sensitivity, % Specificity, % PPV, % NPV, %

Clinical Findings (95% CI) (95% CI) (95% CI) (95% CI) LR (+) LR (–)

Abnormal chest auscultation findings 14 (7–23) 99 (98–100) 55 (32–77) 93 (91–94) 14.0 0.9
Low systolic blood pressure 14 (7–23) 98 (97–99) 44 (24–65) 93 (91–94) 7.0 0.9
GCS score <15 61 (50–72) 74 (71–77) 17 (13–22) 96 (94–97) 2.3 0.5
Abnormal thoracic examination results 68 (56–78) 64 (60–67) 14 (11–18) 96 (94–97) 1.9 0.5
Elevated respiratory rate 51 (40–63) 84 (81–86) 22 (16–29) 95 (93–97) 3.2 0.6
Femur fracture 13 (6–22) 95 (93–96) 18 (9–30) 92 (91–94) 2.6 0.9
Any of the high-risk criteria* 98 (91–100) 37 (34–40) 12 (10–15) 99 (98–100) 1.6 0.05
PPV, Positive predictive value; NPV, negative predictive value; LR, likelihood ratio.
*
Includes any 1 of the 6 listed variables.

Table 4.
Findings in the 2 patients with thoracic injuries but without any of the high-risk criteria.*

Age, y Injury Method of Diagnosis Additional Clinical Findings Additional Injuries Therapy†

5 Pulmonary contusion and pneumothorax Chest radiograph, abdominal CT scan Developed shortness of Liver laceration Observation
breath in the ED
12 Pulmonary contusion Abdominal CT scan Microscopic hematuria None Observation
*
High-risk criteria included any 1 of the following: low systolic blood pressure, elevated respiratory rate, abnormal thoracic examination results, abnormal chest auscultation findings, femur fracture,
or GCS score of <15.
†
Therapy indicates treatment for the thoracic injury.

MAY 2002 39:5 ANNALS OF EMERGENCY MEDICINE 4 9 7

CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al
rates of greater than 20 breaths/min, and 46% have respi-
ratory rates of greater than 30 breaths/min, although this
study did not adjust respiratory rates for age.20 Because
normal respiratory rates in children vary by age, the
assessment and analysis of the respiratory rate should be
age adjusted.
Abnormalities of the thoracic examination had the
highest sensitivity for identifying thoracic injuries. These
abnormalities were present in more than two thirds of
patients with thoracic injuries. Some believe that thoracic
injuries may occur in children without tenderness on
examination of the thorax.2 It is hypothesized that, be-
cause children’s ribs are more pliable, they may not
demonstrate external evidence of injury, despite the
application of significant force. Our study also demon-
strates that alert children may frequently have thoracic
injuries, despite an apparent lack of chest-wall tenderness.
Of the 6 high-risk criteria, a decreased level of con-
sciousness had the second highest sensitivity but the low-
est specificity. It is not surprising that patients with a
mechanism of injury resulting in a traumatic brain injury
frequently also have experienced sufficient force to the
torso to put them at risk for injury to the chest. In addi-
tion, both children and adults with decreased levels of
consciousness often have unreliable physical examina-
tions.2,21,22 In the current study, patients with thoracic
injuries and a GCS score of 13 or 14 were significantly less
likely to have thoracic tenderness than injured patients
with a GCS score of 15.
Low systolic blood pressure was also independently
associated with thoracic injury and had a positive likeli-
hood ratio indicative of a test with a moderate-to-large
effect on posttest probability of disease. Hypotension in
the pediatric trauma patient, however, is an infrequent
occurrence because children are generally able to main-
tain normal blood pressures despite significant blood
loss.2 Hemothoraces, pneumothoraces, and cardiac and
aortic injuries are thoracic injures that may cause hypo-
tension. A chest radiograph is therefore imperative in the
hypotensive pediatric trauma patient to rapidly identify
such injuries.
Femur fracture was an important predictor in the re-
cursive partitioning analysis and approached significance
in the logistic regression analysis. The presence of a femur
fracture implies significant blunt force and causes sub-
stantial pain, which might distract the patient from not-
ing chest pain or tenderness.
Two patients with thoracic injuries were not identified
by these criteria; however, 1 patient developed shortness
of breath during ED observation and would therefore
4 9 8
have had chest radiography obtained. Thus, only one
patient with thoracic injury would have potentially been
missed. This patient’s injury (pulmonary contusion) was
not visualized on ED chest radiography and did not re-
quire therapy.
Only one prior study has attempted to define indica-
tions for chest radiography in pediatric patients sustain-
ing blunt trauma.23 That study found closed-head injury
and out-of-hospital intubation to be the only predictors
of abnormalities on chest radiography.23 The authors rec-
ommended obtaining chest radiographs in pediatric
patients who have undergone closed-head injury requir-
ing intubation, have undergone smoke inhalation, or
have sustained blunt thoracic or abdominal trauma.23
The conclusions of the study, however, may have been
limited by its small size and retrospective data collection.
There are several limitations to this study that are wor-
thy of comment. Although chest radiographs were ob-
tained on all patients, this imaging modality is not per-
fectly sensitive for identification of all patients with
thoracic injuries. Thirty percent to 40% of thoracic injuries
among children are not identified on the initial ED chest
radiograph.24-27 Pulmonary contusions, rib fractures,
and pneumothoraces are the most frequent chest injuries
not identified by means of plain radiography.11,24-27 In
our study, many patients did not have further evaluation
beyond chest radiography, and thus some chest injuries
may have gone undetected. Patients discharged from the
ED, however, had telephone follow-up. Therefore, if
injuries were missed, they were likely to be minor. In
addition, only patients undergoing chest radiography
were included in this study. It is also possible that some
patients presenting during the study period who did not
undergo chest radiography had thoracic injuries that
were never identified. Because these injuries would be
anticipated to be minor and clinically silent, the necessity
of identifying these additional patients is unclear.
We did not assess the presence of hypoxia (and abnor-
mal pulse oximetry) as a predictor of thoracic injury. Most
children sustaining blunt trauma receive supplemental
oxygen during their out-of-hospital evaluation and trans-
portation. This affects the presence and detection of
hypoxia once evaluated in the ED. Most would agree,
however, that patients who manifest hypoxia should have
a chest radiograph obtained.
The study was conducted at a Level I trauma center,
and many patients had high-force mechanisms of injury.
Therefore, the results may not be generalizable to all ED
settings. Physicians were requested to complete the data
sheets as soon as possible after the physical examination
ANNALS OF EMERGENCY MEDICINE 39:5 MAY 2002
CLINICAL DECISION RULE FOR PEDIATRIC TRAUMA
Holmes et al
and before chest radiography. Because of ethical consider-
ations, however, the treating physicians were likely aware
of the initial chest radiograph results in a small number of
patients. These patients who underwent bedside radiog-
raphy were likely the most severely injured and would
likely need chest radiography regardless of the physical
examination findings. In addition, because of the con-
straints of faculty availability, interrater reliability was not
assessed for physical findings. Finally, although the vari-
ables identified in these analyses were validated inter-
nally, external validation is necessary to further assess the
generalizability of our findings before specific recom-
mendations can be made.
This study provides information useful to the clinician
evaluating pediatric patients with blunt trauma. The pre-
dictive variables identified in this study are clinically
intuitive and easy to recall. Although these criteria should
not replace clinical judgment, the identification of these
factors should assist the clinician in managing pediatric
patients who have sustained blunt thoracic trauma.
Patients without any of these findings are at very low risk
for having clinically important thoracic injuries, and
chest radiography is unlikely to be useful.
In summary, predictors of thoracic injury in pediatric
patients with blunt trauma include low systolic blood
pressure, elevated respiratory rate, abnormal thoracic
examination results, abnormal chest auscultation find-
ings, femur fracture, and a GCS score of less than 15.
These predictors can be used to derive a sensible and sen-
sitive clinical decision rule. Children with any of these
predictors should undergo chest radiography. Children
without any of these predictors have a very low likelihood
of thoracic injury.
Author contributions: JFH conceived the study. JFH, PES, and NK designed the study. JFH,
WEB, PES, and NK participated in data collection. WEB participated in radiographic inter-
pretation. JFH and NK participated in data analysis. JFH and NK participated in manuscript
preparation and all authors in manuscript revision. All authors take responsibility for the
paper as a whole.
We thank Nicole Glaser, MD, for her critical review and helpful comments on this article.
REFERENCES
1. Cooper A, Barlow B, DiScala C, et al. Mortality and truncal injury: the pediatric perspective.
J Pediatr Surg. 1994;29:33-38.
2. American College of Surgeons. Advanced Trauma Life Support of the American College of
Surgeons: Advanced Trauma Life Support for Doctors. 6th ed. Chicago, IL: American College of
Surgeons; 1997.
3. Gormican SP. CRAMS scale: field triage of trauma victims. Ann Emerg Med. 1982;11:132-135.
4. Task Force on Blood Pressure Control in Children. Report of the Second Task Force on Blood
Pressure Control in Children-1987: National Heart, Lung, and Blood Institute, Bethesda, Maryland.
Pediatrics. 1987;79:1-25.
MAY 2002 39:5 ANNALS OF EMERGENCY MEDICINE
5. Illif A, Lee VA. Pulse rate, respiratory rate, and body temperature of children between two
months and eighteen years of age. Child Dev. 1952;23:237-245.
6. Reilly PL, Simpson DA, Sprod R, et al. Assessing the conscious level in infants and young
children: a paediatric version of the Glasgow Coma Scale. Childs Nerv Syst. 1988;4:30-33.
7. Garcia VF, Gotschall CS, Eichelberger MR, et al. Rib fractures in children: a marker of
severe trauma. J Trauma. 1990;30:695-700.
8. Brieman L, Friedman JH, Olshen RA, et al. Classification and Regression Trees. Washington,
DC: Champan & Hall; 1994.
9. Stiell IG, Wells GA. Methodologic standards for the development of clinical decision rules
in emergency medicine. Ann Emerg Med. 1999;33:437-447.
10. Altman DG, Andersen PK. Bootstrap investigation of the stability of a cox regression model.
Stat Med. 1989;8:771-783.
11. Chen CH, George SL. The bootstrap and identification of prognostic factors via Cox’s pro-
portional hazards regression model. Stat Med. 1985;4:39-46.
12. Efron B, Tibshirani R. Statistical data analysis in the computer age. Science. 1991;253:390-
395.
13. Harrell FE, Lee KL, Mark DB. Multivariate prognostic models: issues in developing models, eval-
uating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361-387.
14. Concato J, Feinstein AR, Holford TR. The risk of determining risk with multivariate models.
Ann Intern Med. 1993;118:201-210.
15. Nakayama DK, Ramenofsky ML, Rowe MI. Chest injuries in childhood. Ann Surg. 1989;210:770-
775.
16. Allen MS, Cox CS. Pulmonary contusion in children: diagnosis and management. South
Med J. 1998;91:1099-1106.
17. Roux P, Fisher RM. Chest injuries in children: an analysis of 100 cases of blunt chest trauma
from motor vehicle accidents. J Pediatr Surg. 1992;27:551-555.
18. Smyth BT. Chest trauma in children. J Pediatr Surg. 1979;14:41-47.
19. Reilly JP, Brandt ML, Mattox KL, et al. Thoracic trauma in children. J Trauma. 1993;34:329-
331.
20. Bonadio WA, Hellmich T. Post-traumatic pulmonary contusion in children. Ann Emerg Med.
1989;18:1050-1052.
21. Beaver BL, Colombani PM, Fal A, et al. The efficacy of computed tomography in evaluating
abdominal injuries in children with major head trauma. J Pediatr Surg. 1987;22:1117-1122.
22. Hoffman JR, Mower WR, Wolfson AB, et al. Validity of a set of clinical criteria to rule out
injury to the cervical spine in patients with blunt trauma. N Engl J Med. 2000;343:94-99.
23. Parish RA, Watson M, Rivara FP. Why obtain arterial blood gases, chest x-rays, and clotting
studies in injured children? Experience in a regional trauma center. Pediatr Emerg Care.
1986;2:218-221.
24. Karaaslan T, Meuli R, Androux R, et al. Traumatic chest lesions in patients with severe
head trauma: a comparative study with computed tomography and conventional chest
roentgenograms. J Trauma. 1995;39:1081-1086.
25. Sivit CJ, Taylor GA, Eichelberger MR. Visceral injury in battered children: a changing per-
spective. Radiology. 1994;173:559-561.
26. McClellan BA, Ali J, Towers MJ, et al. Role of the trauma-room chest x-ray film in assess-
ing the patient with severe blunt traumatic injury. Can J Surg. 1996;39:36-41.
27. Manson D, Babyn PS, Palder S, et al. CT of blunt chest trauma in children. Pediatr Radiol.
1993;23:1-5.
4 9 9