Revised Pediatric Emergency Assessment Tool
(RePEAT): A Severity Index for Pediatric
Marc H. Gorelick, MD, MSCE, Evaline A. Alessandrini, MD, MSCE, Kathleen Cronan, MD, Justine Shults, PhD
Objectives: To develop and validate a multivariable model, using information available at the time of
patient triage, to predict the level of care provided to pediatric emergency patients for use as a severity
of illness measure.
Methods: This was a retrospective cohort study of 5,521 children 18 years of age or younger treated at four
emergency departments (EDs) over a 12-month period. Data were obtained from abstraction of patient
records. Logistic regression was used to develop (75% of sample) and validate (25% of sample) models
to predict any nonroutine diagnostic or therapeutic intervention in the ED and admission to the hospital.
Data on ED length of stay and hospital costs were also obtained.
Results: Eight predictor variables were included in the final models: presenting complaint, age, triage acu-
ity category, arrival by emergency medical services, current use of prescription medications, and three tri-
age vital signs (heart rate, respiratory rate, and temperature). The resulting models had adequate goodness
of fit in both derivation and validation samples. The area under the receiver operating characteristic curve
was 0.73 for the ED intervention model and 0.85 for the admission model. The Revised Pediatric Emergency
Assessment Tool (RePEAT) score was then calculated as the sum of the predicted probability of receiving
intervention and twice the predicted probability of admission. The RePEAT score had a significant univar-
iate association with ED costs (r = 0.44) and with ED length of stay (r = 0.27) and contributed significantly to
the fit of multivariable models comparing these outcomes across sites.
Conclusions: The RePEAT score accurately predicts level of care provided for pediatric emergency
patients and may provide a useful means of risk adjustment when benchmarking outcomes.
ACADEMIC EMERGENCY MEDICINE 2007; 14:316–323 ª 2007 by the Society for Academic Emergency
Keywords: severity of illness index, risk adjustment, emergency medical service
often include comparisons between providers, settings,
valuating quality and outcomes of care is increas-
ingly recognized as an essential activity for pedi-
atric emergency care.1,2Such evaluations, which
or systems of care, must account for differences between
patient populations in underlying illness severity, which
may be an important confounder of outcomes.3,4A useful
severity of illness index for a given setting should have
several properties, including relative simplicity, data
availability, clinical credibility, and validity.4–7
A number of severity measures have been developed
that may be useful in risk adjustment for general hospital,
critical care, and trauma patients.5,8–11The presence of
comorbid illnesses is a standard method of risk adjust-
ment that is not readily applicable in pediatric patients,
who have relatively few comorbid conditions. Risk of
mortality is commonly used in other settings as a mea-
sure of severity but is poorly suited to pediatric emer-
gency care because death is rare in that setting.12,13
Alternative severity measures have been proposed for
the pediatric emergency setting in which acuity is re-
flected by intensity of resource use. One such measure
is the Pediatric Risk of Admission (PRISA) score, which has
been validated as a predictor of hospital admission.14,15
From the Department of Pediatrics, MedicalCollegeof Wisconsin,
and Children’s Research Institute(MHG),Milwaukee, WI;Division
Philadelphia, PA; Division of Emergency Medicine, A. I. duPont
tistics, Center for Clinical Epidemiology and Biostatistics, Univer-
sity of Pennsylvania (JS), Philadelphia, PA.
Received July 3, 2006; revision received September 13, 2006;
accepted November 10, 2006.
Presented at the Pediatric Academic Societies’ annual meeting,
Washington, DC, May 2005.
Supported by grant R03 HS11359 from the Agency for Health-
care Research and Quality.
Contact for correspondence and reprints: Marc H. Gorelick,
MD, MSCE; e-mail: email@example.com.
PII ISSN 1069-6563583
ª 2007 by the Society for Academic Emergency Medicine
The data requirements of this tool, however, are substan-
tial, with 16 variables including historical, physiologic,
and therapeutic variables. Moreover, admission is gener-
ally uncommon for children seen in emergency depart-
ments (EDs); risk of admission may therefore not be
discriminative in a large proportion of these patients.
We have previously demonstrated, in a single pediatric
ED, the feasibility of developing a severity of illness score
using a limited set of variables routinely collected at the
time of patient triage.16This model, the Pediatric Emer-
gency Assessment Tool (PEAT), accurately predicted
resource utilization, and the predicted probabilities in
turn were associated with other outcomes of care, sug-
gesting this approach could be useful in developing a
valid, practical severity of illness measure broadly appli-
cable to pediatric emergency patients. A severity score
derived at a single center, however, while supporting
the conceptual approach, cannot necessarily be extrapo-
lated to other settings. The purpose of the present study
was to extend this prior work in a multicenter fashion to
develop and validate a multivariable model to predict the
level of care provided to a diverse population of pediatric
emergency patients (the Revised Pediatric Assessment
Tool [RePEAT]) and to evaluate the usefulness of the pre-
dicted probability of care as a risk adjustment index in
comparing outcomes used in benchmarking emergency
care across sites. We hypothesized that there would be
differences in two outcomes, length of stay (LOS) and
ED costs, among the four EDs and that a substantial pro-
portion of the variation in these outcomes would be
explained by differences in patient severity, as measured
by the predicted resource use (RePEAT score).
This was a retrospective cohort study in which children
presenting to an ED for care were followed through
the course of the visit and data on predictor and outcome
variables were collected. Data on both predictors and
outcomes were abstracted from existing records. The
study was reviewed and approved by the institutional re-
view boards of the participating hospitals, with a waiver
of the requirement for informed consent.
Study Setting and Population
The study was conducted at the EDs of four hospitals.
The EDs were chosen by convenience but intended to
represent the spectrum of emergency care settings.
There were two EDs from each of two states, including
an urban children’s hospital ED, an urban general ED,
a rural general ED, and a suburban pediatric ED.
All children, aged 0–18 years, presenting to the EDs
during a 12-month period were eligible for inclusion in
the study. A systematic sample of eligible visits, where
every nth visit was chosen, was selected at each of the
participating hospitals. The sampling fraction was calcu-
lated differently at each hospital to provide the desired
sample size (see Data Analysis). There was no attempt
to identify unique patients; it is possible that some pa-
tients could be included more than once in the data set,
and the unit of analysis was the visit, not the patient.
Measurements. The predictor variables consisted of
data that are routinely obtained during the triage pro-
cess. Age was grouped by the investigators a priori
into three groups based on clinical sensibility: younger
than 3 months, 3–24 months, and older than 24 months.
Reason for visit, or presenting complaint, was catego-
rized using the Pediatric Emergency Reason for Visit
Clusters (PERC).17Up to three complaints could be re-
corded for each visit. Patients or parents were asked
about the presence of any significant past medical history
(a list of qualifying conditions was developed by the in-
vestigators in advance of the study) and current use of
any prescription medications. Triage acuity level was de-
termined according to the criteria in use at each hospital.
At the time of the study, all four EDs used a three-level
triage system. Mode of arrival was categorized as emer-
gency medical services (air or ground) versus self-trans-
port. Vital signs, pulse oximetry, and Glasgow Coma
Scale (GCS) score were measured and recorded accord-
ing to the routine practice at each ED. Heart rate, respi-
ratory rate, and blood pressure were converted to a
percentage of the published age-specific norm.18A list
of conditions for which measurement of oxygen satura-
tion or GCS score would be typically indicated was de-
veloped a priori.
Outcome Measures. For the severity model, the out-
come variable of interest was the level of care provided
during the ED visit, classified into three levels chosen
to reflect an increasing need for care (see Table 1). The
three levels were as follows: routine nursing and physi-
cian assessment (including noninvasive monitoring and
use of nonprescription medications), ED intervention
(having diagnostic or therapeutic procedures performed
in the ED, including prescription medications adminis-
tered, but leading to discharge to home), and admission
to the hospital or transfer to another facility. Death in
the ED is combined with hospitalization in the most
severe outcome category.
The two other benchmark outcomes for which we
wished to demonstrate the utility of the RePEAT score
as a risk adjuster were ED costs and LOS. LOS was de-
fined as the time from arrival at the ED to discharge
from the ED (either to home or to another care location).
ED charges for each visit were obtained from the hos-
pital billing office, including charges for nursing care,
supplies, and laboratory and other ancillary testing;
physician professional fees were not included because
two of the institutions would not provide these data.
These charges were divided by each hospital’s global
charge-to-cost ratio to yield an ED cost.19
Data Collection. Patient logs were reviewed at each site
daily. Every nth visit, according to the sampling fraction
for that site, was chosen and the chart abstracted using
a standardized form. Abstraction was performed by a
single nurse at three sites and by a team of two research
assistants at one site. All abstractors were blinded to
the study hypotheses. During the training phase, charts
were abstracted again by the principal investigator to en-
sure accuracy, and the investigators met quarterly with
abstractors to monitor abstraction. Because we have
ACAD EMERG MED?April 2007, Vol. 14, No. 4?www.aemj.org
previously demonstrated excellent interobserver reliabil-
ity in abstraction of these data elements, formal interob-
server reliability analysis was not repeated.16
Scannable data forms (Teleforms; Cardiff Software,
Vista, CA) were completed and mailed to the data man-
agement center for data entry. Range checks were incor-
porated into the data entry program to minimize entry
errors, and all scanned forms were manually reviewed
The total study sample was randomly divided into a der-
ivation set (75% of the records) and a validation set (25%
of the records). Prediction models were developed in the
derivation set only but evaluated in both sets. Multivari-
able logistic regression was used to develop predictive
models for the level of care.20,21Two separate models
were estimated: one for any ED intervention (whether
discharged to home or admitted) versus routine care
and the second for admission versus any discharge.
Although univariate screening is often used to reduce
the number of variables to be included in a model, this
practice is controversial. We chose to limit the number
of candidate predictors a priori, based on clinical sensi-
bility and anticipated data availability, and to include all
of the potential predictors in the initial model. However,
because we wished to develop a model incorporating
data that are routinely recorded, any given variable miss-
ing in more than 15% of records was excluded from the
modeling process. We then sequentially removed varia-
bles to attain the final model.
Two variables were included in the model as part of an
interaction term that includes an indicator of whether the
measurement of the variable was clinically indicated:
pulse oximetry (relevant only for subjects with respira-
tory illness) and GCS score (relevant only for those
with head trauma or a neurologic complaint).
Because the reason for the visit was being used to pre-
dict the probability of requiring different levels of emer-
gency care, the most useful categorization would group
complaints according to their likelihood of requiring
care. To do this, we used an independent data source:
the 1998 and 2000 National Hospital Ambulatory Medical
Care Survey (NHAMCS) ED data sets. For each of the
PERCs, the percentage of pediatric patients with that
complaint having each of the two outcomes, ED interven-
tion and admission, was calculated.17Each PERC was
then assigned an ordinal ranking from 1 (lowest risk) to
5, for each of the two outcomes, based on the quintile
of risk for that outcome, as shown in Table 2. For each
complaint listed for a given visit in the study data set,
the complaint was then assigned the NHAMCS-derived
quintile ranking for each of the two outcomes. For exam-
ple, a child with a complaint of fever would have a value
of 4 for that complaint in the model for ED intervention
and a value of 2 in the model for admission. For patients
with more than one complaint, each complaint was as-
signed the appropriate ranking, and the highest value
for all the complaints for a given patient was entered
into the model.
For all logistic regression analyses, parameters were
estimated using maximum likelihood techniques. Be-
cause of the complex sampling scheme, with different
sampling fractions at each site, appropriate survey
weighting methods were used. Analyses were performed
using the svy logit commands in Stata 8.2 (Stata Corp.,
College Station, TX). Probability weights were used,
with clustering by site, to yield robust estimates of the
standard errors. First, the full model was fitted. Predictor
variables with a nonsignificant (p > 0.10) association with
model, and the resulting model was compared with the
more complete one. The Bayesian information criterion
was calculated for each model, and a difference of <6
was indicative of insufficient evidence to reject the more
parsimonious model. However, if dropping a variable
would result in a change in value of the coefficients of
another variable by more than 20%, the clinical relevance
Definition of Levels of Care
Routine nursing and medical care
Discharged to home from ED; no diagnostic tests or therapeutic procedures performed
Discharged to home from ED; one or more of the following performed
? Imaging studies (x-rays, ultrasonography, computed tomography, magnetic
? Laboratory tests on body fluids (including blood tests, urine tests, lumbar puncture)—
exclude urine pregnancy test or throat culture/rapid strep antigen test
? Tests not on body fluids (e.g., electrocardiography, slit-lamp examination)
? Intravenous fluids
? Prescription medications administered in ED (oral, intravenous, or inhaled)
? Wound management (suture repair, Steri-Strip placement, burn dressing)
? Treatment of an orthopedic problem by splinting or casting, knee immobilizer, or
? Specialty consultation
? Invasive diagnostic procedures (e.g., arthrocentesis, thoracentesis)
? Resuscitation (cardiopulmonary resuscitation, bag-valve-mask ventilation)
Admitted to hospital, transferred to another facility, or died in the ED Hospitalized
Gorelick et al.?RePEAT SEVERITY INDEX FOR PEDIATRIC EMERGENCY CARE
of that variable was reevaluated by the investigators to
determine if the variable should be retained.
Calibration of the models was determined by the
Hosmer–Lemeshow goodness-of-fit test. For this test, a
p-value less than 0.05 was considered evidence of lack
of fit. Discriminative ability of the models was assessed
using the c statistic, or area under the receiver operating
RePEAT Score. The two logistic regression models were
used to derive a RePEAT score, defined as the predicted
probability of receiving ED intervention plus twice the
predicted probability of admission. This weighting was
selected in advance to reflect the higher severity inherent
in the need for hospitalization. Because patients who are
admitted by definition also receive intervention, the pos-
sible range of scores is 0 (0% predicted probability of re-
ceiving ED intervention) to 3 (100% predicted probability
of admission). To evaluate the potential usefulness of the
RePEAT score as a risk-adjustment tool, we evaluated the
association between RePEAT score and two other out-
comes: LOS and ED cost. For each of these outcomes,
analysis of variance was performed with site of care
as the independent variable and the model-adjusted R2
calculated. A second analysis of variance was then
performed with RePEAT score added as another inde-
Sample Size. The projected sample size was calculated
based on the ability to develop a model for the least com-
mon outcome: admission. Using the general rule of
thumb of needing at least ten outcomes for each degree
of freedom included in a logistic regression model,21
we would need at least 210 subjects admitted in the der-
ivation subsample to accommodate all of the potential
predictors. Assuming an overall admission rate of 7%,
and 80% of subjects with complete data on the predic-
tors, we would need 3,750 subjects in the derivation
sample, or a total of 5,000 subjects. To ensure adequate
representation of subjects from all sites, a target of
1,000 subjects was set for the two general hospitals and
1,500 from each of the pediatric hospitals (a somewhat
larger sample was targeted from the pediatric hospitals
because they were believed to be relatively more infor-
mative due to the greater spectrum of illness of pediatric
patients at those locations).
A total of 5,521 subjects were enrolled (Table 3) and split
randomly into a derivation set (n = 4,184) and a validation
set (n = 1,337). Overall, 66.8% of subjects received ED
intervention, and 8.5% were admitted; these numbers
were very similar in the two sets.
Model Derivation and Assessment
Two variables were excluded from further consideration
due to >15% missing data: blood pressure was measured
only 45% of the time overall, including 66% of the time
for children 3 years and older and 20% for those younger
than 3 years, and GCS score was documented in only 55%
of cases in which measurement of GCS would have been
indicated. Pulse oximetry was recorded in only 30% of
visits but in 89% of those with an indication for pulse ox-
imetry. The other variables were recorded in 94%–99%
of visits, and 85% of visits had complete data on all
Quintiles of Risk for PERCs
Quintile for ED
Altered mental status
Extremity pain or injury
Foreign body (ears, nose,
Foreign body (skin)
Motor vehicle collision
PERC = Pediatric Emergency Reason for Visit Clusters.
ACAD EMERG MED?April 2007, Vol. 14, No. 4?www.aemj.org
predictors. The remaining predictor variables were in-
cluded in the logistic regression models. Two variables
(past medical history and pulse oximetry) were subse-
The final results for the two models are shown in Table 4.
For both temperature and respiratory rate, the univariate
relationship with the outcomes appeared linear. For heart
rate, there was a slightly U-shaped relationship, with an
initially higher rate of ED intervention and admission at
the lowest heart rate values and then a linear increase.
However, the number of visits with the lowest heart rates
leading to a higher risk was very small (<0.5% of all visits).
imprecise estimates for the low heart rate coefficient, so a
single linear term was retained.
The goodness of fit of each model was determined to
be adequate in both the derivation and validation sam-
ples (Table 4). Model discrimination was also good. For
the ED intervention model, the c statistic was 0.73 (95%
confidence interval [CI] = 0.71 to 0.75) for the derivation
set and 0.75 (95% CI = 0.72 to 0.78) for the validation
set. The derivation and validation c statistics for the ad-
mission model were 0.85 (95% CI = 0.82 to 0.87) and
0.86 (95% CI = 0.82 to 0.90), respectively.
RePEAT Scores and Risk-adjusted Outcomes
The mean (?SD) RePEAT score was 0.842 (?0.387). The
median score was 0.809, with an interquartile range of
0.532 to 1.039. The total range was 0.233–2.735.
Each of the two care outcomes, ED costs and LOS, was
Study Site Characteristics
Race/ethnicity* (%)Insurancey (%)
*As reported to ED clerical personnel; other categories not reported due to small numbers.
yAll others were privately insured.
Logistic Regression Models for ED Intervention and Admission
ED Intervention Model*Admission Modely
ORVariable Logit CoefficientOR95% CI for ORLogit Coefficient95% CI for OR
Complaint quintile ranking
Younger than 3 mo
Older than 24 mo
Arrival via emergency medical
Currently taking prescription
Heart rate (as multiple of
Respiratory rate (as multiple of
0.325 1.381.12, 1.720.287 1.33 1.24, 1.43
0.648 1.910.85, 4.300.9342.54 1.59, 4.06
0.4931.64 1.22, 2.200.2831.320.70, 2.52
Hosmer–Lemeshow goodness-of-fit statistics: *ED treatment model, derivation sample: chi-square8 df= 12.1, p = 0.21; validation sample, chi-square10 df=
14.5, p = 0.21. yAdmission model: derivation sample, chi-square8 df= 14.6, p = 0.09; validation sample: chi-square10 df= 13.3, p = 0.28.
Gorelick et al.?RePEAT SEVERITY INDEX FOR PEDIATRIC EMERGENCY CARE
Spearman correlation coefficient for ED cost was 0.44
(95% CI = 0.40 to 0.47) and for LOS was 0.27 (95% CI =
0.24 to 0.30). In Table 5, we demonstrate the potential use-
fulness of risk adjusting the comparisons of outcomes
across sites using the RePEAT score. For each outcome,
the difference between sites was modeled using linear re-
explanatory power of the model was improved with the
addition of the severity score. In addition, the magnitude
of the differences is substantially different when adjusted
for severity. Site B, for example, has markedly lower costs
than the reference A (which is in the same geographic re-
gion). However, the difference is much smaller when ad-
justed for severity. The cost differences between sites C
for severity. Similarly, sites B and D appear to be much
more efficient in moving patients through the ED com-
pared with site A, but the risk-adjusted difference shows
the sites to be more comparable. If we examine site C ver-
sus A, a different pattern emerges. Both are similar set-
tings, but the number of beds per patient is smaller at
site C, which would therefore be expected to be more
crowded. The total throughput time is indeed higher at
site C, but the risk-adjusted difference is even greater
than the raw numbers suggest.
Severity is an inherent characteristic of patients and their
illnesses that reflects the natural history, that is, the prog-
nosis in the absence of intervention.22As such, it is closely
related to the concept of acuity or the need for care; pa-
tients with a higher level of severity require greater levels
of care and require care more urgently. We used a re-
source-based approach that relates severity to the level
of care required.23Severity indices that use this approach
are based at least implicitly on the assumption that only
necessary services will be rendered, and therefore ser-
vices provided are a reasonable proxy for services re-
quired.24Although practice variability may in fact lead
to some ‘‘unnecessary’’ care that is rendered, we believe
that when averaged over a large number of physicians,
the care provided to patients is likely to reflect an average,
reasonable judgment that such care was necessary. In
fact, severity scores based on risk of mortality have
been shown to correlate extremely well with resources
used, indicating reasonable construct validity of the
resource-based approach to measuring severity.4,25
predictsthelevel of resource utilizationin theED forpedi-
atric patients. The predicted probabilities from this model
can be used as a marker of severity; patients with higher
predicted resource utilization are presumably more se-
verely ill. Moreover, the predicted probabilities can be
used to adjust for differences in baseline risk when com-
paring outcome measures and benchmarks, such as costs
and LOS, across settings. Such risk adjustment is critical
in benchmarking and evaluating quality of care.3
It may be argued that this approach is tautological; ED
costs are determined by the resources used, so it should
be no surprise that a model made up of resource use
would be correlated with costs. However, the RePEAT
model does not consist of resources used. Rather, it is a
model consisting of patient characteristics that predicts
resource use and, as such, is a measure of illness severity.
The correlation of such a model with ED costs or LOS is
not a foregone conclusion. Moreover, it is not whether or
not the RePEAT model is correlated with these other out-
comes of care that is of primary interest, but the magni-
tude of the correlation: the amount of variation in cost or
LOS that can be explained by this severity index. The R2
coefficients observed for RePEAT compare favorably
with those of other severity of illness measures com-
monly used for risk adjustment in large data sets.3
Advantages of the RePEAT score include the small
number of variables and completeness of data availabil-
ity. We acknowledge that the score is computationally
complex but can be easily calculated by inputting the
variables into a simple spreadsheet. The minimal data
requirement makes the RePEAT potentially applicable
in a wide variety of settings and may be amenable to
application to large electronic data sets.
Several alternative resource utilization–based severity
indices exist for pediatric emergency visits. The Emer-
gency Severity Index version 3 is a prospectively applied
algorithm that categorizes patients into five levels based
on the predicted number and intensity of resources
used.26It also includes a limited number of variables
and has been demonstrated to predict not only ED re-
source utilization but LOS, admission, and six-month
mortality. However, in addition to objective findings
such as pain score and vital signs, the algorithm includes
subjective nurse judgment as to the number of resources
needed, which would preclude its being applied retro-
spectively to an existing data set. The PRISA15uses an
approach similar to ours in that it predicts probability
of admission. The PRISA score includes 16 variables, in-
cluding five laboratory variables and one treatment vari-
able. We believe that the RePEAT has some advantages
over the PRISA. First, the laboratory variables included
in the PRISA are obtained in only a very small minority
of all ED patients. The PRISA assumes an unmeasured
variable to be normal, which may lead to substantial
bias.27Moreover, performance of laboratory tests and
therapeutic interventions determine, in part, the PRISA
score; using such a score to adjust for differences in costs
and time variables (both of which are related to the per-
formance of such procedures) would not be appropriate.
Risk-adjusted Comparison of Outcomes
Hospital Costs ($)ED Length of Stay (hr)
Site A is reference category.
RePEAT = Revised Pediatric Emergency Assessment Tool.
ACAD EMERG MED?April 2007, Vol. 14, No. 4?www.aemj.org
Finally, admission is a relatively uncommon outcome. A
score based only on admission risk may not discriminate
well among the large proportion of patients at low risk.
However, there are likely situations where the more
detailed clinical information contained in the PRISA
may be preferable. Among a population of children at
reasonably high risk of admission, the basic triage infor-
mation included in the RePEAT may be insufficient to
discriminate various levels of risk. Thus, the scores may
provide complementary information in certain situations,
such as subsets of patients with high-acuity diagnoses or
in higher-acuity settings. The relative usefulness of these
two risk adjustment tools for different purposes would
be a fruitful area for future study.
Any retrospective study is limited by the availability and
quality of data recorded in the medical record. Important
predictors may not be recorded routinely; missing data
may lead to bias or loss of relevant predictive informa-
tion. Because the RePEAT would most likely be applied
retrospectively to existing data sources, however, the
study procedure is relevant to actual intended practice.
Although model discrimination is good, particularly for
admission, it is likely not sufficiently high to allow pre-
dictions for individual patients. This, however, is not
the intended use for this score. Rather, it is intended to
be applied to groups of patients, which is justified by the
goodness of fit of the models. More importantly, we
recognize that severity of illness is only one of many de-
terminants of resources used and level of care provided.
Individual variations in risk thresholds among physicians,
parent preferences, and financial incentives are some of
the other factors influencing management. By looking at
broad categories such as any ED intervention, we hoped
to minimize unimportant variation (conversely, such an
expansive definition of ED intervention may have de-
creased our ability to distinguish differences in severity
within this outcome category). Additionally, we believe
that when averaged over large numbers of providers, pa-
tients with greater severity will have greater resource uti-
lization. Thus, resource utilization as a proxy for severity
is more appropriate for comparing care at different insti-
tutions, considering physicians in aggregate rather than
comparing individual physicians.
Our study was conducted at only four institutions. Al-
though there is substantial diversity in patient popula-
tions and settings, the sample of EDs in this study is
not necessarily representative. If the practice at these
hospitals is not representative of generally accepted stan-
dards, our results may not be generalizable. This is some-
what mitigated by the use of an external, nationally
representative data set (NHAMCS) to derive quantiles
of risk for the chief complaints. However, further exter-
nal validation of the RePEAT score is necessary before
widespread adoption can be recommended.
We conclude that the RePEAT score is a valid predictor
of level of care. Although not applicable to decision mak-
ing at the level of the individual patient, it is a potentially
valuable tool for risk adjustment in evaluations of out-
comes and quality of pediatric emergency care.
1. Seidel JS, Henderson DP, eds. Emergency Medical
Services for Children: A Report to the Nation. Wash-
ington, DC: National Center for Education in Mater-
nal and Child Health, 1991.
2. Durch JS, Lohr KN, eds. Emergency Medical Services
for Children. Institute of Medicine Report. Washing-
ton, DC: National Academy Press, 1993.
3. Iezzoni LI, ed. Risk Adjustment for Measuring Health-
care Outcomes. 3rd ed. Chicago, IL: Health Adminis-
tration Press, 2003.
4. Blumberg MS. Risk adjusting health care outcomes: a
methodologic review. Med Care Rev. 1986; 43:351–93.
5. Cales RH. Injury severity determination: require-
ments, approaches, and applications. Ann Emerg
Med. 1986; 15:1427–33.
6. Gibson G. Indices of severity for emergency medical
evaluative studies: reliability, validity, and data re-
quirements. Int J Health Serv. 1981; 11:597–630.
7. Gorelick MH. Severity of illness measures for pediat-
ric emergency care: are we there yet? [editorial]. Ann
Emerg Med. 2003; 41:639–43.
alternatives. Ann Rev Public Health. 1988; 9:333–59.
9. Gross PA, Beyt BE, Decker MD, et al. Description of
case-mix adjusters by the severity of illness working
group of the Society of Hospital Epidemiologists of
America (SHEA). Infect Control Hosp Epidemiol.
10. Pollack MM, Patel KM, Ruttiman UE. PRISM III: an
updated pediatric risk of mortality score. Crit Care
Med. 1996; 24:743–52.
11. Slater A, Shann F, Pearson G, Paediatric Index of
Mortality (PIM) Study Group. PIM2: a revised version
of the Paediatric Index of Mortality. Intensive Care
Med. 2003; 29:278–85.
12. Orr RA, Venkataraman ST, Cinoman MI, Hogue BL,
Singelton CA, McCloskey KA. Pretransport Pediatric
Risk of Mortality (PRISM) score underestimates the
requirement for intensive care of major interventions
during interhospital transport. Crit Care Med. 1994;
13. Strauss RH, Rooney B. Critical care pediatrician-led
aeromedical transports: physician interventions and
predictiveness of outcome. Pediatr Emerg Care.
14. Chamberlain JM, Patel KM, Pollack MM, et al., for
the Collaborative Research Committee of the Emer-
gency Medicine Section of the American Academy of
Pediatrics. Recalibration of the Pediatric Risk of
Admission score using a multi-institutional sample.
Ann Emerg Med. 2004; 43:461–8.
15. Chamberlain JM, Patel KM, Pollack MM. The Pediat-
ric Risk of Hospital Admission score: a second-
generation severity-of-illness score for pediatric
emergency patients. Pediatrics. 2005; 115:388–95.
16. Gorelick MH, Lee C, Cronan K, Kost K, Palmer K.
Pediatric emergency assessment tool (PEAT): a
Gorelick et al.?RePEAT SEVERITY INDEX FOR PEDIATRIC EMERGENCY CARE
risk-adjustment measure for pediatric emergency
patients. Acad Emerg Med. 2001; 8:156–62.
17. Gorelick MH, Alpern ER, Alessandrini EA. A system
for grouping presenting complaints: the Pediatric
Emergency Reason for Visit Clusters (PERC). Acad
Emerg Med. 2005; 12:723–31.
18. Gunn VL, Christian N. The Harriet Lane Handbook.
16th ed. St. Louis, MO: CV Mosby, 2002.
19. MaWhinney S, Brown ER, Malcolm J, et al. Identifi-
cation of risk factors for increased cost, charges,
and length of stay for cardiac patients. Ann Thorac
Surg. 2000; 70:702–10.
20. Ruttiman UE. Statistical approaches to development
and validation of predictive instruments. Crit Care
Clin. 1994; 10:19–35.
21. Harrell FE, Lee KL, Mark DB. Multivariable prognos-
tic models: issues in developing models, evaluating
assumptions and adequacy, and measuring and
reducing errors. Stat Med. 1996; 15:361–87.
22. Pollack MM, Ruttiman UE, Getson PR. Pediatric risk
of mortality (PRISM) score. Crit Care Med. 1988; 16:
23. Maslanka AM. Scoring systems and triage from the
field. Emerg Med Clin North Am. 1993; 11:15–27.
24. Cullen DJ, Civetta JM, Briggs BA, Ferrara LC. Thera-
peutic intervention scoring system: a method for
quantitative comparison of patient care. Crit Care
Med. 1974; 2:57–60.
25. Yeh TS, Pollack MM, Holbrook PR, Fields AI. Assess-
ment of pediatric intensive care: application of the
Therapeutic Intervention Scoring System. Crit Care
Med. 1982; 10:497–500.
26. Tanabe P, Gimbel R, Yarnold PR, Kyriacou DN,
Adams JG. Reliability and validity of scores on the
Emergency Severity Index version 3. Acad Emerg
Med. 2004; 11:59–65.
27. Gorelick MH. Bias arising from missing data in pre-
dictive models. J Clin Epidemiol. 2006; 59:1115–23.
Please see accompanying video Data Supplement available at www.aemj.org.
Ultrasound Use in the Diagnosis of Abdominal
Aneurysmal disease of the abdominal aorta is common and may
be found in up to 10% of male smokers older than 65 years of
age. About 80% of patients who present with a ruptured abdom-
inal aortic aneurysm have no previous diagnosis. When rupture
occurs, mortality is very high. Rapid diagnosis and repair have
been shown to decrease this mortality. Bedside evaluation using
ultrasound is relatively straightforward (Figure 1). See the video
supplement (available as a Data Supplement at http://www.aemj.
org/cgi/content/full/j.aem.2007.01.001/DC1) for a 5-minute over-
view of ultrasound use in the diagnosis of abdominal aortic
Scott Davarn, MD
Rob Reardon, MD
Scott Joing, MD
Department of Emergency Medicine
Hennepin County Medical Center
Figure 1. Bedside ultrasound of abdominal aorta with transverse probe position and normal image. (Color version of this
figure available online at www.aemj.org.)
ACAD EMERG MED?April 2007, Vol. 14, No. 4?www.aemj.org