Professional Documents
Culture Documents
James M. Chamberlain, MD*‡§; Kantilal M. Patel, PhD*§; and Murray M. Pollack, MD, MBA*㛳
ABSTRACT. Objective. To develop and validate a ceiver operator characteristic curve was 0.82 ⴞ 0.01 (SE) in
second-generation severity-of-illness score that is appli- the development sample and 0.77 ⴞ 0.02 in the validation
cable to pediatric emergency patients. The Pediatric Risk sample. In ordered predicted risk intervals, the propor-
of Admission (PRISA) score was developed in a single tion of patients with admissions, mandatory admissions,
hospital and was recalibrated and validated in 2, previ- and ICU admissions increased in a linear manner.
ous, small studies from academic pediatric hospitals. Conclusions. The second-generation PRISA II score
This study was performed to develop and validate a score for pediatric ED patients has been developed and vali-
in a larger sample of diverse hospitals. dated in a large sample of diverse hospitals. Performance
Methods. Emergency departments (EDs) were block characteristics indicate that PRISA II will be useful for
randomly selected as part of a study on ED quality on the institutional comparisons, benchmarking, and controlling
basis of 3 care characteristics: annual patient volume for severity of illness when enrolling patients in clinical
(high or low compared with national median), presence trials. Pediatrics 2005;115:388–395; emergency medicine,
or absence of a pediatric emergency medicine subspecial- health services research, severity-of-illness index.
ist, and presence or absence of residents. Patients were
selected randomly on the basis of daily arrival logs. Med-
ical records were photocopied, and abstracted data in- ABBREVIATIONS. ED, emergency department; PRISA, Pediatric
cluded demographic, historical, physiologic, and thera- Risk of Admission; PEM, pediatric emergency medicine; ROC,
peutic information. The total sample was randomly receiver operator characteristic.
divided into a 75% development sample and a 25% val-
idation sample. Univariate and multivariate analyses
S
pecific components of the Emergency Medical
were used to model the risk of mandatory admission,
Services system and the Emergency Medical
admissions for which preidentified, inpatient medical
resources were used. The resulting multiple logistic re- Services–Children system have benefited from
gression model coefficients were converted to integer the development and use of severity assessment
scores. Calibration (Hosmer-Lemeshow goodness of fit) methods. For example, PICU severity methods have
and discrimination (area under the ROC curve) were enabled evaluation of quality of PICU care using
used to measure performance. As a measure of construct severity-adjusted mortality rates,1 identification of
validity, proportions of patients in ordered risk intervals PICU care characteristics associated with quality of
were correlated with the outcomes of admission, manda- care,2,3 measurement of efficiency of bed use,4 and
tory admission, and ICU admission. measurement of severity-adjusted length of stay.5
Results. Sixteen EDs enrolled 11 664 patients. Mean
Such methods have not been applied to pediatric or
patient age (ⴞSD) was 6.8 ⴞ 5.8 years, and 53% were
male. Nine percent arrived by emergency medical ser- adult emergency departments (EDs). Although se-
vices, and 6.9% were admitted. The most common diag- verity scales exist for specific diseases, there is no
noses were minor injuries, otitis media, and fever. The widely accepted method that is applicable to all pe-
multivariate analysis yielded a score with 7 historical diatric ED patients.
variables, 8 physiologic variables, 1 therapy (oxygen) Previous severity-of-illness methods applied to
term, and 1 interaction term. Calibration was excellent. In hospital and ED environments (eg, trauma patients)
the development sample, 442 mandatory admissions have used mortality as the outcome for quality-of-
were predicted and 442 were observed (total 2 ⴝ 2.275), care studies.6 However, mortality is not an appropri-
and in the validation sample, 136.6 were predicted and
ate outcome for the predominant ED activities and is
145 were observed (2 ⴝ 8.575). The area under the re-
so rare in the general pediatric ED population that it
cannot be used as a meaningful outcome measure.
From the *Department of Pediatrics, George Washington University School
One of the most important functions of ED practice is
of Medicine, Washington, DC; and ‡Department of Emergency Medicine, the correct disposition of patients to either inpatient
§Center for Health Services and Community Research, Children’s Research or outpatient care settings. Overadmission results in
Institute, and 㛳Center for Hospital-Based Specialties and Department of wasted resources and the risk of iatrogenic injury,
Critical Care Medicine, Children’s National Medical Center, Washington, whereas failure to admit when necessary results in
DC.
Accepted for publication Jul 6, 2004.
delayed care and avoidable complications. We pro-
doi:10.1542/peds.2004-0586 pose that the appropriate outcomes that are indica-
No conflict of interest declared. tive of pediatric ED quality are correct rates of hos-
Address correspondence to James M. Chamberlain, MD, Department of pital admission and ED discharge.
Pediatrics, George Washington University School of Medicine, 111 Michi-
gan Ave NW, Washington, DC 20010. E-mail: jchamber@cnmc.org
We developed a severity-of-illness model using
PEDIATRICS (ISSN 0031 4005). Copyright © 2005 by the American Acad- hospital admission as the outcome. Initially devel-
emy of Pediatrics. oped and validated at a single institution, the Pedi-
ARTICLES 389
Downloaded from pediatrics.aappublications.org at Univ Of Birmingham on June 2, 2015
ment of the PRISM III score.1 The patient sample was randomly the presence of that variable in an individual patient contributes to
divided into a 75% development sample and a 25% validation a lower risk of hospital admission.
sample. The independent variables for possible inclusion in the Standard performance measures were used to assess the PRISA
predictive model were selected using a multistep process. First, II score. The area under the curve for the receiver operator char-
variables were chosen for clinical relevance on the basis of clinical acteristic (ROC) curve was used to measure discrimination. Cali-
experience and previous work in developing severity adjustment bration compared observed outcomes with predicted outcomes,
models. Second, univariate analyses were used to determine as- where the predicted outcomes were the number predicted by the
sociations with the dependent variable. For categorical variables, PRISA II score. The Hosmer-Lemeshow goodness-of-fit tests (2)
this was performed using unadjusted odds ratios. Some categor- were used to measure calibration by comparing observed with
ical variables were uncommon in ED populations and therefore predicted admissions in 8 ordered risk intervals of predicted
were combined with other, clinically similar variables. For exam- admission probability. Risk intervals were chosen to be clinically
ple, the chronic disease variables “HIV” and “oncologic disease” relevant, to ensure at least 10 predicted admissions in each inter-
were combined with other conditions into a single category “im- val in the development sample, and to attempt to achieve a bal-
munodeficiency.” For continuous variables, patients were strati- anced distribution of admitted patients across the intervals. Over-
fied into the age groups of neonates (0 –1 month), infants (⬎1–12 all performance was also assessed with the standardized
months), children (⬎12–144 months), and adolescents (⬎144 admission ratio, the ratio of observed mandatory admissions to
months). Univariate logistic regression analysis was used to de- predicted mandatory admissions. Comparison of mean PRISA II
termine which variables and subranges significantly contributed score among different severity subgroups used the Kruskal-Wallis
to the probability of hospital admission. The risk of hospital test. Construct validity was assessed by comparing proportions of
admission relative to the midrange (40th– 60th percentiles) of dis- patients with admission, mandatory admission, and ICU admis-
charged patients was computed for the physiologic variables in sion in the ordered PRISA II risk intervals using the Cochran-
each age stratum. We assumed that deviations from the midrange Armitage test for linear trend in proportion.11
contributed to risk, with larger deviations reflecting higher risk.
We initially used deciles of physiologic values, then readjusted
these cutpoints on the basis of clinical relevance and performance RESULTS
in the multivariate model. Neighboring deciles were combined Sample and Patient Data
when the regression coefficients were similar.1 Variables with a
liberal significance level (P ⬍ .30) in univariate analysis were Institutional and patient characteristics are de-
included in the multivariate model. Third, forward and backward picted in Table 2. The 16 sites ranged in annual
stepwise inclusion of variables was used in the development of the pediatric ED volumes from 1740 to 31 000 patients
multiple logistic regression model. Variables were retained as long (median: 13 905). Four sites had pediatric volumes
as they contributed to a lower Akaike’s Information Criteria and
were statistically significant at P ⬍ .10. The regression coefficients
⬍8000, 4 sites had volumes between 8000 and 13 900,
in the model were expressed as relative percentages by dividing 4 sites had volumes between 13 900 and 17 500, and
each of the coefficients by the sum of the regression coefficients 4 sites had volumes ⬎17 500. Consistent with the
and multiplying by 100. The rounded percentages for each vari- study design for site enrollment, 8 sites had residents
able and variable range are the components of the PRISA II score, caring for patients in the ED, and 8 sites had at least
and the total for all variables is the PRISA II score for each patient.
Although most variables contribute to a higher risk of hospital 1 PEM specialist in the practice. The most common
admission, a variable may have a negative regression coefficient chief complaints were fever (17.5%), nonspecific re-
(and thus a corresponding negative integer score), indicating that spiratory complaints (14.5%), and injuries to the ex-
TABLE 3. Multiple Logistic Regression Modeling Mandatory Hospital Admission, Developed Using 75% of the Sample (N ⫽ 8782 ED
visits, 442 Mandatory Hospital Admissions)
Parameter Estimate Standard Error Adjusted Odds Ratio P
(95% Confidence Interval)
Intercept ⫺3.9724 0.0935 .0001
Historical factors
Age ⬍90 d 0.9646 0.1913 2.62 (1.80–3.82) .0001
Minor injury* ⫺0.4161 0.1689 0.66 (0.47–0.92) .0138
Abdominal pain in an adolescent† 2.6649 0.6226 14.37 (4.24–48.68) .0001
Immunodeficiency‡ 1.6308 0.2786 5.11 (2.96–8.82) .0001
Indwelling medical device§ 1.0540 0.2813 2.87 (1.65–4.98) .0001
Controller asthma medications储 1.3490 0.2512 3.85 (2.36–6.31) .0001
Referral status¶ 1.2533 0.1152 3.50 (2.79–4.39) .0001
Physiologic factors
Low temperature 1.0191 0.7460 2.77 (0.64–11.96) .1719
Neonates: ⬍35.5
Infants: ⬍35.5
Children: ⬍35.0
Adolescents: ⬍35.0
Decreased mental status# 3.3968 1.0940 29.87 (3.50–254.94) .0019
Low systolic BP 1.2262 0.2853 3.41 (1.95–5.96) .0001
Neonates: ⬍70
Infants: ⬍70
Children: ⬍83
Adolescents: ⬍100
High diastolic BP 0.7989 0.1389 2.22 (1.69–2.92) .0001
Neonates: ⬎59
Infants: ⬎59
Children: ⬎70
Adolescents: ⬎90
Low serum bicarbonate (⬍20) 2.2385 0.2278 9.38 (6.00–14.66) .0001
High potassium (⬎4.9) 2.3186 0.3090 10.16 (5.55–18.62) .0001
High BUN (⬎18) 1.5412 0.2986 4.67 (2.60–8.39) .0001
High WBC count (⬎20 000) 2.3985 0.2651 11.01 (6.55–18.51) .0001
Therapies
Oxygen** 2.0448 0.2100 7.49 (5.21–11.66) .0001
Interactions
Low bicarbonate and high potassium ⫺1.7213 0.6472 0.179 (0.05–0.64) .0077
BP indicates blood pressure, BUN, blood urea nitrogen; WBC, white blood cell.
* Strain, sprain, abrasion, laceration, and fracture without associated cranial, thoracic, or abdominal injury.
† Chief complaint of abdominal pain in a patient from that age group.
‡ HIV, oncologic disease, sickle cell hemoglobinopathy, and transplanted organ.
§ Tracheostomy, long-term venous catheter, dialysis catheter or shunt, ventricular drainage catheter, feeding tube, home apnea monitor,
and home oxygen.
储 Any asthma medication other than bronchodilators.
¶ Referral from a physician’s office or another ED.
# Listless, irritable, lethargic, obtunded, or comatose.
** Oxygen administered other than during inhaled bronchodilator treatments.
ARTICLES 391
Downloaded from pediatrics.aappublications.org at Univ Of Birmingham on June 2, 2015
TABLE 4. The PRISA II Score mental status to ⫺7 as the interaction term to adjust
Parameter Score for low bicarbonate and high potassium (total of 19).
The maximum number of PRISA II points is 96. The
Age ⬍90 d 4
Minor injury* ⫺2 probability (P) of mandatory admission is calculated
Abdominal pain in an adolescent† 11 using the formula P ⫽ 1/(1 ⫹ e⫺R), where R ⫽
Immunodeficiency‡ 7 ⫺4.0250 ⫹ 0.2888*(PRISA II score) ⫺ 0.00279*(PRISA
Indwelling medical device§ 4
Controller asthma medications储 6
II score)2.
Referral status¶ 5 Overall, the PRISA II score performed very well.
Low temperature 4 Figure 1 illustrates the increasing PRISA II score with
Neonates: ⬍35.5 cohorts of increasing illness severity: ED discharges,
Infants: ⬍35.5
Children: ⬍35.0 nonmandatory hospital admissions, mandatory non-
Adolescents: ⬍35.0 ICU hospital admissions, and ICU admissions. There
Decreased mental status# 14 was a significant trend in increasing PRISA II scores,
Low systolic BP 5 and each group’s score was significantly different
Neonates: ⬍70
Infants: ⬍70 from the others with the exception of ICU admis-
Children: ⬍83 sions compared with mandatory non-ICU admis-
Adolescents: ⬍100 sions (P ⫽ .085).
High diastolic BP 3
Neonates: ⬎59
The Hosmer-Lemeshow goodness-of-fit test indi-
Infants: ⬎59 cated excellent calibration in both the development
Children: ⬎70 sample (2 ⫽ 2.275, degrees of freedom, df ⫽ 6, P ⫽
Adolescents: ⬎90 .8928; Table 5, Fig 2) and validation sample (2 ⫽
Low serum bicarbonate (⬍20) 9
High potassium (⬎4.9) 10 8.575, df ⫽ 7, P ⫽ .2846; Table 6, Fig 2). The stan-
High BUN (⬎18) 6 dardized mandatory admission ratios for the total
High WBC (⬎ 20 000) 10 samples were 1.00 ⫾ 0.04 and 1.06 ⫾ 0.07 in the
Oxygen** 9
Low bicarbonate and high potassium ⫺7
development and validation samples. The area un-
der the ROC curve was 0.822 ⫾ 0.012 in the devel-
The probability of mandatory admission, P ⫽ 1/(1 ⫹ e⫺R), where opment sample and 0.774 ⫾ 0.023 in the validation
R ⫽ ⫺4.0250 ⫹ 0.2888*(PRISA II score) ⫺ 0.00279*(PRISA II
score)2. sample (Fig 3).
* Strain, sprain, abrasion, laceration, and fracture without associ- Construct validity is depicted in Fig. 4. The figure
ated cranial, thoracic, or abdominal injury. illustrates the increasing proportion of patients with
† Chief complaint of abdominal pain in a patient from that age
group. admissions, mandatory admissions, and ICU admis-
‡ HIV, oncologic disease, sickle cell hemoglobinopathy, and trans- sions as PRISA II risk intervals increase (P ⬍ .0001).
planted organ.
§ Tracheostomy, long-term venous catheter, dialysis catheter or
DISCUSSION
shunt, ventricular drainage catheter, feeding tube, home apnea
monitor, and home oxygen. Severity-of-illness methods are important for mea-
储 Any asthma medication other than bronchodilators. suring and controlling for case-mix variability in
¶ Referral from a physician’s office or another ED.
# Irritable, lethargic, obtunded, or comatose.
many situations, including scientific studies and in-
** Oxygen administered other than during inhaled bronchodilator stitutional comparisons.12–16 Such tools have been
treatments. used extensively in pediatric and neonatal critical
Fig 1. PRISA II scores for outcome groups of increasing severity of illness. There is a significant linear trend (*P ⬍ .0001, Kruskal-Wallis
test). Comparison among groups shows that each group is significantly (P ⬍ .0001) different from the others except non-ICU, mandatory
admissions group versus ICU admissions group comparison (P ⫽ .0856).
care and have allowed comparisons among individ- tients with specific diseases,17–19 there are no widely
ual hospitals and among cohorts of hospitals. Al- accepted general severity measures that are applica-
though there are clinical severity scores for ED pa- ble to all pediatric emergency patients. The PRISA
ARTICLES 393
Downloaded from pediatrics.aappublications.org at Univ Of Birmingham on June 2, 2015
Fig 3. ROC curve of the PRISA II score.
Fig 4. Increasing severity of illness of admitted patients with increasing PRISA risk intervals (P ⬍ .001 for all outcomes, Cochran-Armitage
test for linear trend in proportions).
score was initially developed in a single institution7 ologic factors, including patient age, chronic history,
and validated in 2 small samples of large academic acute history, and variations in individual practice
centers8,9; thus, it was necessary to revise the first- and health system factors. Therefore, we calibrated
generation PRISA score based on a contemporary set the score to mandatory admissions, a group of pa-
of more diverse PEM patients. tients for whom most clinicians would agree under
The development of a severity measure that is most circumstances would be best cared for as inpa-
applicable to pediatric ED patients has been difficult. tients. We recognize, however, that there is still some
A major barrier to developing an effective severity practice pattern variability in the use of the inpatient
method that is applicable to all ED patients has been therapies that compose the mandatory admission
lack of a relevant outcome measure that occurs with list. Most of the criteria for mandatory admission can
sufficient frequency. Because one of the most impor- be assessed on disposition from the ED, making this
tant functions of ED practice is the correct disposi- a practical assessment for future use.
tion of patients to either inpatient or outpatient care Despite these difficulties, PRISA II performs well.
settings, we used this outcome in developing a se- First, mean PRISA II scores increased in patients with
verity-of-illness method. In contrast to mortality and increasing clinical severity, and the PRISA II scores
its tight linkage with physiologic derangement, hos- were ordered correctly for the patient groups by
pital admission is influenced by several nonphysi- increasing severity. Second, the goodness-of-fit data
ARTICLES 395
Downloaded from pediatrics.aappublications.org at Univ Of Birmingham on June 2, 2015
The Pediatric Risk of Hospital Admission Score: A Second-Generation
Severity-of-Illness Score for Pediatric Emergency Patients
James M. Chamberlain, Kantilal M. Patel and Murray M. Pollack
Pediatrics 2005;115;388
DOI: 10.1542/peds.2004-0586
Updated Information & including high resolution figures, can be found at:
Services http://pediatrics.aappublications.org/content/115/2/388.full.ht
ml
Supplementary Material Supplementary material can be found at:
http://pediatrics.aappublications.org/content/suppl/2005/01/31
/115.2.388.DC1.html
Citations This article has been cited by 13 HighWire-hosted articles:
http://pediatrics.aappublications.org/content/115/2/388.full.ht
ml#related-urls
Subspecialty Collections This article, along with others on similar topics, appears in
the following collection(s):
Research Methods & Statistics
http://pediatrics.aappublications.org/cgi/collection/research_
methods_-_statistics_sub
Emergency Medicine
http://pediatrics.aappublications.org/cgi/collection/emergency
_medicine_sub
Permissions & Licensing Information about reproducing this article in parts (figures,
tables) or in its entirety can be found online at:
http://pediatrics.aappublications.org/site/misc/Permissions.xht
ml
Reprints Information about ordering reprints can be found online:
http://pediatrics.aappublications.org/site/misc/reprints.xhtml
The online version of this article, along with updated information and services, is
located on the World Wide Web at:
http://pediatrics.aappublications.org/content/115/2/388.full.html