Content uploaded by Emily Trudell Correa
Author content
All content in this area was uploaded by Emily Trudell Correa on Mar 22, 2022
Content may be subject to copyright.
Development and Validation of the Asthma
Exacerbation Risk Score Using Claims Data
Jonathan Hatoun, MD, MPH, MS; Emily Trudell Correa, MPH, MS;
Andrew J. MacGinnitie, MD, PhD; Jonathan M. Gaffin, MD, MMSc, MPH;
Louis Vernacchio, MD, MSc
From the Pediatric Physicians’ Organization at Children’s (J Hatoun, ET Correa, and LVernacchio), Wellesley, Mass; Division of General
Pediatrics, Boston Children’s Hospital (J Hatoun and L Vernacchio), Boston, Mass; Department of Pediatrics, Harvard Medical School (J
Hatoun, AJ MacGinnitie, JM Gaffin, and L Vernacchio), Boston, Mass; Division of Immunology, Boston Children’s Hospital (AJ MacGinnitie),
Boston, Mass; and Division of Pulmonary Medicine, Boston Children’s Hospital (JM Gaffin), Boston, Mass
None of the authors have a conflict of interest.
Address correspondence to Jonathan Hatoun, MD, MPH, MS, Pediatric Physicians’ Organization at Children’s, 112 Worcester St, Suite
300, Wellesley, MA 02482 (e-mail: jonathan.hatoun@childrens.harvard.edu).
Received for publication February 17, 2021; accepted July 3, 2021.
TAGGEDPABSTRACT
OBJECTIVE:Pediatric asthma is a costly and complex disease
with proven interventions to prevent exacerbations. Finding the
patients at highest risk of exacerbations is paramount given lim-
ited resources. Insurance claims identify all outpatient, inpatient,
emergency, pharmacy, and diagnostic services. The objective
was to develop a risk score indicating the likelihood of asthma
exacerbation within the next year based on prior utilization.
METHODS:A retrospective analysis of insurance claims for
patients 2 to 18 years in a network in Massachusetts with 3 years
of continuous enrollment in a commercial plan. Thirty-six poten-
tial predictors of exacerbation in the third year were assessed
with a stepwise regression. Retained predictors were weighted
relative to their contribution to asthma exacerbation risk and
summed to create the Asthma Exacerbation Risk (AER) score.
RESULTS:In a cohort of 28,196 patients, there were 10 predic-
tors associated with the outcome of having an asthma
exacerbation in the next year that depend on age, meeting the
Healthcare Effectiveness Data and Information Set persistent
asthma criteria, fill patterns of asthma medications and oral
steroids, counts of nonexacerbation outpatient visits, an
exacerbation in the last 6 months, and whether spirometry was
performed. The AER score is calculated monthly from a
claims database to identify potential patients for an asthma
home-visiting program.
CONCLUSIONS:The AER score assigns a risk of exacerbation
within the next 12 months using claims data to identify
patients in need of preventive services.
TAGGEDPKEYWORDS:asthma; asthma exacerbation; insurance claims;
risk score
ACADEMIC PEDIATRICS 2022;22:47−54
TAGGEDPWHAT’S NEW
Using the Asthma Exacerbation Risk score, an organi-
zation can efficiently risk stratify a population with
respect to their risk for an asthma exacerbation in the
next 12 months and assign resources accordingly.
TAGGEDPASTHMA IS ONE of the most costly pediatric conditions in
part because of the frequency with which patients suffer
acute exacerbations requiring urgent outpatient or emer-
gency department (ED) visits and inpatient admissions.
1−3
While asthma is diagnosed in nearly 8% of all children
under age 18, more than half of whom will have an asthma
exacerbation
1
, the precise timing of an exacerbation is
impossible to predict for any given patient. Still, certain
patients are more prone to experiencing exacerbations for
several reasons, including biologic predisposition, environ-
mental exposures, or difficulties adhering to a prescribed
treatment regimen.
4
Adding to the difficulty of predicting
exacerbations is the fact that a patient’s first asthma exacer-
bation often occurs without a prior formal diagnosis of
asthma, though it may be preceded by related diagnoses or
treatments (eg, diagnosis of wheezing or bronchiolitis, pre-
scription of bronchodilators).
Interventions such as interval checkups, step up in med-
ication therapy, specialty consultation, or in-home inter-
ventions may improve asthma management and minimize
risk of exacerbation.
4
Home visits for intensive education
and environmental remediation have particularly been
proven to reduce morbidity associated with asthma, but
are costly, difficult to scale, and can be hard to sustain.
4,5
Although home visiting programs have shown a positive
return on investment, they typically recruit patients who
have recently had a costly event, such as one or more ED
visits or a hospital admission, making it difficult to disen-
tangle the effect of the program from regression to the
mean.
6
Another, frequently used recruitment technique is
to use the Healthcare Effectiveness Data and Information
Set (HEDIS) definition of a patient with persistent asthma
to identify patients who would benefit from intervention.
7
However, the HEDIS definition itself relies heavily on
recent costly events and is overly specific for routine
ACADEMIC PEDIATRICS
Copyright ©2021 by Academic Pediatric Association 47
Volume 22, Number 1
January−February 2022
population health monitoring, as it fails to identify most
patients who will experience an asthma exacerbation in
the near future, as many as 80% in a recent analysis.
8
A
broader population health approach would seek to identify
those patients who are most at risk for an exacerbation,
even absent a recent costly event or prior diagnosis of
asthma, and then allocate appropriate resources toward
exacerbation prevention.
To this end, insurance claims data provide a potential
means to identify patterns of medical care utilization that
precede an asthma exacerbation, potentially even exacer-
bations prior to formal diagnosis. Insurance claims data
have the benefit of describing all the medical utilization
of an individual over time, including their pharmacy, pri-
mary care, specialist, emergency, and hospital encounters,
regardless of the institution at which care is provided.
Thus, we sought to produce a patient-level score indicat-
ing the risk of future asthma exacerbation for all patients,
irrespective of a history of asthma, that could be calcu-
lated from an evaluation of the asthma-related predictors
present in paid insurance claims databases. Given that
roughly one third of patients in our network suffer an
exacerbation without previously having been given a
diagnosis of asthma, we intentionally included children
with and without prior ICD codes for asthma in our analy-
sis. Our hope was to ensure a risk score that would be
applicable to an entire general pediatric population,
regardless of the accuracy of prior ICD coding of an
asthma diagnosis. It is important to cast a wide net when
considering pediatric patients, since exacerbations can
occur without preceding symptoms of persistent asthma,
more often in younger individuals than in older individu-
als.
4
We hypothesized that the relative contribution of a
specific utilization pattern would effectively predict a
patient’s overall risk of a subsequent asthma
exacerbation.
9,10
TAGGEDH1METHODSTAGGEDEND
The Pediatric Physicians’ Organization at Children’s is
an independent practice association of 85 pediatric pri-
mary care practices affiliated with Boston Children’s Hos-
pital. Using paid insurance claims data from 4
commercial insurance companies that share data with our
organization for quality improvement purposes, we identi-
fied all patients 2 to 18 years of age with continuous
enrollment over a 3-year period from January 2012 to
2014, irrespective of a prior diagnosis of asthma or an
asthma exacerbation. The first 2 years of the sample were
used to examine predictors for exacerbations occurring in
the third year of the sample.
We analyzed 36 claims-based predictors of asthma
exacerbation based on known clinical correlations and
other likely associated utilization patterns. Potential pre-
dictors were built by evaluating the claims database for
available demographic characteristics, pharmacy fill pat-
terns, diagnosis assignment (both for asthma and comor-
bidities) using ICD-9, procedure coding, visit frequency,
and timing and frequency of asthma exacerbations. While
the characteristics of most predictors are self-evident
from their titles, patients satisfied the “HEDIS Persistent
Criteria” predictor if they had any one of the following in
each of the prior 2 years: an ED visit with a principal diag-
nosis of asthma; a hospitalization with a principal diagno-
sis of asthma; 4 or more outpatient asthma visits with 2 or
more asthma medication-dispensing events in the last
year; or 4 or more asthma medication dispensing events.
We categorized visits to urgent care centers as
“outpatient” visits. Predictors were determined with
claims from 2012 to 2013, while the outcome (asthma
exacerbation) was evaluated in claims from 2014. The
outcome of an asthma exacerbation was defined as: 1) an
admission with a primary diagnosis of asthma; 2) an ED
visit with a primary diagnosis of asthma
11
; or 3) an outpa-
tient visit with a diagnosis of asthma and an oral steroid
prescribed.
12
Bivariate analyses of all potential predictors with
asthma exacerbations were performed with claims infor-
mation from 2012 to 2014, and all predictors were
included in a subsequent step-wise logistic regression
model as either dichotomized or categorical variables,
with categories determined by natural cutpoints in the
data and the authors’ clinical experience. Category sizes
of <1% of the population were combined to avoid over-
fitting the model. The stepwise model parameters were
alpha for entry = 0.3 and alpha for stay = 0.1. The retained
predictors of the step-wise regression were assigned
points weighted to their relative contribution to the out-
come, as determined by their beta coefficients in the
model. The variable with the smallest beta coefficient was
assigned a score of 1 and other variables were assigned
points based on the ratio of their beta coefficients to the
variable with the lowest beta coefficient, rounded to the
nearest whole number.
13
Leave-one-out cross validation,
which calculates the likelihood of the outcome to each
individual using a re-fit model derived from all prior data
except those contributed by the individual, was used to
determine the predictive accuracy of the model. Given the
imbalanced distribution of events, we also analyzed the
area under the precision-recall curve to assess the predic-
tive capacity of the model.
All analyses were performed in SAS v9.4 (SAS Insti-
tute Inc, Cary, NC). The project was approved by the Bos-
ton Children’s Hospital Committee on Clinical
Investigation. A waiver of individual informed consent
was obtained.
TAGGEDH1RESULTSTAGGEDEND
A cohort of 28,196 patients was used for the 3-year der-
ivation sample and the potentially predictive asthma utili-
zation patterns in the first 2 years of the derivation cohort
and their associated odds of exacerbation in the third year
are shown in Table 1.
Ten variables were retained in the step-wise model.
These variables and their specified regression parameters
(beta coefficients) are shown in Table 2 along with their
relative contributions to the final point system—the
TAGGEDEND48 HATOUN ET AL ACADEMIC PEDIATRICS
Table 1. Bivariate and Categorical Associations of the Potential Risk Factors for Asthma Exacerbations in Derivation Cohort
N(% Total) % With
Exacerbation
OR (95% CI)
All patients 28,196 (100) 1.4 - -
Demographics
Sex, male 14,317 (50.8) 1.6 1.4 (1.1, 1.7)
ZIP code, <3x federal poverty level 3260 (11.6) 1.7 1.2 (0.9, 1.7)
Age in years
2−6 7017 (24.9) 2.0 2.1 (1.6, 2.6)
7−11 10,249 (36.3) 1.4 1.5 (1.2, 1.9)
12−16 10,930 (38.8) 1.0 ref
Existing asthma identification criteria
HEDIS persistent criteria met (2-year lookback) 432 (1.5) 17.8 19.1 (14.6, 25.0)
Pharmacy
Asthma medications of any type
No fills 25,367 (90.0) 0.6 ref
≥1 asthma medication of any type filled 2829 (10.0) 8.3 14.7 (11.9, 18.0)
Rescue medications
No fills 25,743 (91.3) 0.7 ref
≥1 rescue medication fills 2453 (8.7) 8.4 12.9 (10.6, 15.9)
Controller medications
No fills 26,533 (94.1) 0.8 ref
≥1 controller medication fills 1663 (5.9) 10.7 15.0 (12.2, 18.4)
Inhaled ICS-LABA combination medications
No fills 28,111 (99.7) 1.3 ref
≥1 inhaled ICS-LABA combination medication fills 85 (0.3) 28.2 21.7 (15.1, 31.0)
LTRA medications fills
LTRA only (no other asthma medications) 109 (0.4) 9.2 15.0 (8.1, 27.2)
Oral steroid fills
No fills 26,904 (95.4) 1.0 ref
≥1 oral steroid filled 1292 (4.6) 9.7 9.9 (8.0, 12.1)
Diagnoses
Wheeze as any diagnosis 504 (1.8) 8.3 7.2 (5.1, 10.0)
Wheeze as primary diagnosis 252 (0.9) 7.5 6.1 (3.8, 9.8)
Cough as any diagnosis 2364 (8.4) 3.6 3.2 (2.5, 4.1)
Cough as primary diagnosis 1358 (4.8) 3.8 3.1 (2.3, 4.1)
Recurrent pneumonia as any diagnosis 18 (0.1) 11.1 9.0 (2.1, 39.2)
Atopic dermatitis as any diagnosis 1357 (4.8) 3.8 3.1 (2.3, 4.1)
Food allergy as any diagnosis 512 (1.8) 4.3 3.3 (2.2, 5.2)
Allergic rhinitis as any diagnosis 2864 (10.2) 3.5 3.2 (2.5, 4.0)
Allergic conjunctivitis as any diagnosis 263 (0.9) 4.6 3.5 (1.9, 6.3)
Other/Unspecified environmental allergies as any diagnosis 469 (1.7) 4.1 3.1 (2.0, 5.0)
Any environmental allergies (rhinitis, conjunctivitis, or other/unspecified)
as any diagnosis
3238 (11.5) 3.4 3.1 (2.5, 3.8)
Any allergies (environmental allergies, atopic dermatitis, or food aller-
gies) as any diagnosis
4084 (14.5) 3.1 2.9 (2.4, 3.6)
Obesity as any diagnosis 1385 (4.9) 2.1 1.6 (1.1, 2.3)
Procedures
MDI teaching performed 383 (1.4) 7.6 6.2 (4.2, 9.3)
Spirometry performed (any location) 735 (2.6) 6.1 5.1 (3.7, 7.0)
Spirometry perfor med in PCP office 207 (0.7) 5.8 4.5 (2.5, 8.1)
Flu shot administered 17,693 (62.8) 1.6 1.6 (1.3, 2.0)
Exacerbations
Outpatient
0 27,908 (99.0) 1.1 ref
≥1 outpatient 288 (1.0) 24.7 28.4 (21.2, 37.9)
Emergency department
0 28,115 (99.7) 1.3 ref
≥1 emergency department 81 (0.3) 18.5 16.9 (9.5, 29.8)
Inpatient
0 28,160 (99.9) 1.4 ref
≥1 Inpatient 36 (0.1) 16.7 14.5 (6.0, 35.1)
Any exacerbation setting
0 27,810 (98.6) 1.1 ref
≥1 any exacerbation setting 386 (1.4) 21.8 25.1 (19.2, 32.8)
(Continued)
TAGGEDENDACADEMIC PEDIATRICS DEVELOPMENT OF THE ASTHMA EXACERBATION RISK SCORE 49
Asthma Exacerbation Risk (AER) score. Predictors that
contributed points to the risk of an exacerbation in the fol-
lowing year were age; meeting the criteria for “persistent
asthma” as defined by HEDIS; pharmacy dispensation in
the last year for any asthma-related medication, inhaled
corticosteroid−long-acting beta-agonist combination
medication, and/or an oral steroid; having a billed office
visit for asthma with either the PCP or a specialist without
a concurrent exacerbation; and having had a recent
exacerbation. Having had spirometry performed was pro-
tective.
The correlation between weighted points and asthma
exacerbations experienced in the subsequent year is
shown in Figure 1. Individual patient scores range from 0
to 17. Patients with AER scores of 3 or lower, represent-
ing 84.7% of the cohort, had a negligible amount of
asthma exacerbations requiring medical intervention in
the subsequent year. Those with scores of 10 or more had
a high rate of exacerbation, with approximately 26% of
those patients experiencing an exacerbation. Table 3
details the proportion of patients by AER score value who
went on to have an exacerbation in the outcome year.
ROC curves for the AER score model with all data
(Derivation Model, areas under the curve = 0.8471, 95%
confidence interval: 0.8239, 0.8704) and using leave-one-
out cross validation (Validation Model, areas under the
curve = 0.8120, confidence interval: 0.7833, 0.8408) were
not significantly different, indicating that the model is not
over-fit to our data (Fig. 2). Both models performed sig-
nificantly better than a random classifier (P<.001). The
area under the precision-recall curve was 0.1600, an
increase over the true event rate, which was 0.0138.
TAGGEDH1DISCUSSIONTAGGEDEND
The AER score demonstrates that it is possible to iden-
tify which pediatric patients are most at risk for an asthma
exacerbation in the next 12 months using each patients’
prior utilization of medical services and medication fills at
pharmacies from an insurance claims database. In our der-
ivation cohort of over 28,000 patients, the AER score pro-
vides meaningful risk stratification and assists with
identification of patients with the highest risk of asthma
exacerbations. Furthermore, the AER score can be
calculated on a rolling basis, allowing for identification of
changes in risk profile as often as new claims information
become available, typically monthly.
The AER score has advantages over other means of
patient identification for asthma interventions. Many
asthma improvement projects rely on patient identification
through clinician recall, which is known to be inaccu-
rate
14
; through standard definitions such as the HEDIS
definition of persistent asthma, which misses many
patients who will have an asthma exacerbation in the near
future
8
; through the asthma medication ratio, which, while
predictive of ED visits and hospitalizations for
asthma,
15,16
is only calculated for patients meeting the
HEDIS definition of persistent asthma, and is thus may be
applied to an overly narrow subset of the population for
our purposes; or through review of reports of patients
with a recent asthma-related inpatient admission or ED
visit. The latter strategy, used by most programs with
which the authors are familiar, is not able to identify
patients before a costly event occurs, which would be pre-
ferred.
6
While the AER score does increase with recorded
exacerbations, it has other inputs that help to overcome
the limitations above. The derivation process allows for a
better understanding of the prospective risk of a large pop-
ulation of pediatric patients, because selected variables
were chosen from the utilization patterns of all patients,
regardless of whether or not they had previously been
assigned a diagnosis of asthma.
The AER score was derived from patterns of utilization
that were associated with future exacerbations and con-
structed via a step-wise logistic regression model that
weights the relative contribution of various predictors to
determine a summary risk score for a given patient, with a
higher AER score associated with a greater risk of the
patient experiencing an asthma exacerbation in the subse-
quent 12 months. It was important that the derivation of
selected predictors include all patients in our database,
regardless of assigned asthma diagnosis, in order to poten-
tially detect “rising risk” in a patient not yet identified as
having asthma. However, most of the selected predictors
contributing to the AER score have criteria that are only
likely to be satisfied by patients with known asthma.
Indeed, only 23 patients without a diagnosis of asthma
achieved a score >6, and only one suffered an
Nonexacerbation office visits
PCP and specialist visits
0 nonexacerbation asthma visits 25,487 (90.4) 0.6 ref
1 nonexacerbation asthma visit 1662 (5.9) 6.0 9.4 (7.3, 11.9)
≥2 nonexacerbation asthma visits 1047 (3.7) 11.9 18.6 (14.8, 23.2)
Specialist visits
0 nonexacerbation asthma visits 27,743 (98.4) 1.2 ref
≥1 nonexacerbation asthma visits 453 (1.6) 10.8 9.8 (7.1, 13.4)
Time since last exacerbation
>12 months 27,810 (98.6) 1.1 ref.
6−12 months 176 (0.6) 26.1 23.8 (18.2, 31.3)
0−6 months 210 (0.7) 18.1 16.5 (12.1, 22.5)
HEDIS indicates Healthcare Effectiveness Data and Information Set; LTRA, leukotriene receptor antagonist; ICS, inhaled corticosteroid;
LABA, long-acting beta-agonist; MDI, metered dose inhaler; PCP, primary care provider; and CI, confidence interval.
All lookback periods are within the last year, unless specified.
T
AGGEDEND50 HATOUN ET AL ACADEMIC PEDIATRICS
exacerbation in 2014. Nonetheless, the additional compu-
tational work of calculating the AER score for all patients
is minimal.
We calculate the AER score monthly for all patients
using SAS and our electronic claims database, which
aggregates claims data from multiple insurance plans, in
order to stratify patients’ asthma risk within our network
and use it to offer intensive preventive services to those
with the highest risk for exacerbation. Lists of patients
with high AER scores are reviewed with the primary care
medical homes to decide if additional interventions should
be offered. We do not use a specific threshold to deter-
mine which patients are offered interventions, though we
only produce lists for patients scoring 6 or more points,
which represent 6.7% of the population in this study while
accounting for 63% of exacerbations. This cutoff could be
adjusted based on the resources available. While patient
AER score lists are particularly useful for patient identifi-
cation in a network or in a large practice, there are some
providers in our network who routinely review the AER
scores of their primary patients.
A high AER score is not a guarantee of a future exacer-
bation, and not all patients with an elevated AER score
will be endorsed as potential candidates for additional
interventions by the medical home (nor will such patients
necessarily agree to receive additional supports). When
patients do consent, our program offers: an environmental
assessment of the home, distribution of free “green”
cleaning supplies, an assessment of social determinants of
health that may be contributing to difficult control, moni-
toring of asthma control and asthma-related quality of
life, additional asthma education, and coordination of care
between patients, specialists, and the medical home.
Because recent exacerbations are not the only factor con-
tributing to the AER score, patients are often identified
before they experience a significant exacerbation, unlike
current practice in other settings where an ED visit or hos-
pitalization is typically the trigger for home visits or other
intensive interventions.
6,17
Additionally, the AER score
could be used to offer case management support, as a con-
tributor to the assessment of overall medical risk in a net-
work, or as a quality improvement measure.
Factors included in the AER score, while they correlate
with increased risk of asthma exacerbation, are not
thought to be causative of asthma exacerbations. Indeed,
because the predictors are derived from utilization of
medical care, many represent elements of recommended
care, such as the prescription of appropriate asthma medi-
cations. However, escalating therapy with numerous inter-
ventions can reasonably act as proxy for an increased
intrinsic severity of a patient’s disease. The AER score
helps to quantify that severity in relation to the risk of
future exacerbation. It is interesting to note that while fre-
quency of nonexacerbation visits to a specialist was not
retained in the model, the frequency of nonexacerbation
visits to a PCP or specialist was retained. While visits to
the PCP have previously been shown to be protective of
exacerbations among patients with known asthma,
18
our
work analyzed both patients with a known diagnosis of
Table 2. Final Model With Beta-Coefficients and Assigned Score Weights
Predictor Beta Coefficient From Stepwise Selection Weighted Points in Asthma Exacerbation Score
Nonexacerbation office visits: ≥2 nonexacerbation PCP and/or specialist asthma visits 1.4982 4
Nonexacerbation office visits: 1 nonexacerbation PCP or specialist asthma visit 1.2740 3
Pharmacy: ≥1 asthma medication 1.4531 3
Pharmacy: ≥1 inhaled ICS-LABA combination medication fills 1.3041 3
Pharmacy: ≥1 oral steroid filled 0.9761 2
Time since last exacerbation: 0−6 months 0.6611 2
Demographics: age 2−6 in years 0.8068 2
Demographics: age 7−11 in years 0.4226 1
Existing criteria: HEDIS persistent asthma (2-year lookback) 0.5435 1
Procedures: spirometry (any location) -0.5571 -1
HEDIS indicates Healthcare Effectiveness Data and Information Set; ICS, inhaled corticosteroid; LABA, long-acting beta-agonist; and PCP, primary care provider.
All lookback periods are within the last year, unless specified.
The variable with the smallest beta coefficient was assigned a score of 1 and other variables were assigned points based on the ratio of their beta coefficients to the variable with the lowest beta coeffi-
cient, rounded to the nearest whole number.
TAGGEDENDACADEMIC PEDIATRICS DEVELOPMENT OF THE ASTHMA EXACERBATION RISK SCORE 51
asthma and those without. That the predictor with PCP
visits included was retained in the model suggests it is
functioning as a proxy for patients with a prior diagnosis
of asthma, while also emphasizing increased risk for
patients that have more visits. Also of note, having had
spirometry performed was determined to be protective (-1
point), but was found on subsequent analyses to correlate
with an office visit to either an allergist or a pulmonolo-
gist, potentially representing the benefit of comprehensive
asthma management.
While other scoring systems have been shown to
assess risk of exacerbation, we are not aware of any
prior work that uses an insurance claims database to
evaluate risk based on the entirety of medical care pro-
vided to a patient. Others have assigned risk scores
with methods like those applied here, but were limited
in their scope of candidate predictors without access to
a claims database.
19
Attempts to assign risk using
patient or family surveys necessitate patient participa-
tion and may be subject to biases.
20
The use of detailed
clinical information and biomarkers from study partic-
ipants with severe persistent asthma has contributed to
a seasonal risk stratification rule for that specific
population;
21,22
however, most asthma exacerbations
occur in patients with more mild disease. In contrast to
each of these methods, the AER score can identify
patients seeking all types of medical care within or
outside of the primary care network across a large pop-
ulation and promote outreach to at-risk individuals
before their asthma worsens.
0%
5%
10%
15%
20%
25%
30%
0123456789>=10
raeYgniwoll
o
fe
htninoitbr
ecaxehti
wstnei
tapfotnecreP
AER Score
Figure 1. Performance of the AER score: percent of patients with an exacerbation in the following year by AER score. AER indicates
Asthma Exacerbation Risk.
Table 3. Count and Proportion of Patients by AER Score With Count and Proportion in Each Category That Went on to Have an Exacerba-
tion in the Outcome Year
AER Score Patients Number (% of Total) Patients With Asthma Exacerbation Number (%)
0 9046 (32.1) 24 (0.3)
1 8588 (30.5) 34 (0.4)
2 6234 (22.1) 37 (0.6)
3 969 (3.4) 19 (2.0)
4 901 (3.2) 25 (2.8)
5 518 (1.8) 30 (5.8)
6 591 (2.1) 37 (6.3)
7 440 (1.6) 39 (8.9)
8 328 (1.2) 37 (11.3)
9 211 (0.7) 22 (10.4)
≥10 322 (1.1) 84 (26.1)
AER indicates Asthma Exacerbation Risk.
T
AGGEDEND52 HATOUN ET AL ACADEMIC PEDIATRICS
Our derivation of the AER score and the AER score
itself has certain limitations. Insurance claims data typi-
cally are available for analysis with a 3- to 4-month lag
from the time that care was delivered; thus scoring sys-
tems based on claims data may miss patients with rela-
tively recent events that correlate with increased risk.
Also, we derived the AER score using claims from com-
mercial insurance companies only. It is possible that a
score derived for publicly insured patients could differ,
and this would be a fruitful area for further study, as chil-
dren affected by asthma are disproportionately from low-
income families or minorities.
1
Whether commercially
insured or not, access to care may be unequal because of
structural racism, resulting in a subset of patients appear-
ing “healthier” or at less risk, due to their relatively lower
health care utilization. In reality, many patients with low
health care utilization may still require significant addi-
tional supports. Since race data were not available in the
insurance claims data set we used, it was not assessed as
an independent predictor.
The diagnostic code set used during the time period of
analysis was ICD-9, which did not account for the level of
severity of asthma as assigned by the clinician (ie, inter-
mittent, mild persistent, moderate persistent, or severe
persistent), but only “extrinsic asthma,” “intrinsic
asthma,” “chronic obstructive asthma,” “other forms of
asthma,” and “asthma unspecified.” Additionally, because
insurance claims databases have only the diagnoses
assigned by clinicians, important risk factors for patients
may be missing. For example, while second hand smoke
exposure is a known trigger for asthma and can be
assigned a diagnosis code on submitted claims, only 1
patient in our entire cohort had this code applied on a
claim, and therefore we could not use this as a potential
predictor.
4
Last, it is possible that our definition of an
exacerbation counts as exacerbations ED visits with a
concern for asthma but without exacerbation severe
enough to warrant a systemic corticosteroid prescription.
Derivation of the AER score while requiring an oral ste-
roid fill from a pharmacy after an ED visit does not
Figure 2. Validation of the model used to derive the AER score. ROC curves for both the model and a re-fit model using leave-one-out
cross-validation demonstrating similar areas under the curve (AUC). AER indicates Asthma Exacerbation Risk.
TAGGEDENDACADEMIC PEDIATRICS DEVELOPMENT OF THE ASTHMA EXACERBATION RISK SCORE 53
change the predictors identified, though does change their
relative contribution slightly (4 predictors change by only
1 point). We believe it is important to count ED visits
without a subsequent oral steroid fill because of the
increased use of ED-administered dexamethasone, which
would not be found in pharmacy claims, as well as the
known low fill rates of oral steroid prescriptions after ED
visits.
23−25
Attempting to confirm an exacerbation with a
pharmacy fill for a corticosteroid may overly exclude true
exacerbations, and is not in keeping with an expert panel
review of the definition of a pediatric ED visit for an
asthma exacerbation.
11
TAGGEDH1CONCLUSIONTAGGEDEND
The AER score provides an assessment of any child’s
risk of an asthma exacerbation in the coming year based
on prior health care utilization. While no risk stratification
tool can perfectly identify all patients at risk of exacerba-
tion, the AER score identifies high-risk patients and
allows for a relative ranking of risk.
TAGGEDH1ACKNOWLEDGMENTSTAGGEDEND
Financial statement: This study was funded by internal funds of the
Pediatric Physicians’ Organization at Children’s. This research did not
receive any specific grant from funding agencies in the public, commer-
cial, or not-for-profit sectors.
Authorship statement: J.H. conceptualized and designed the study,
drafted the initial manuscript, and approved the final manuscript as sub-
mitted. E.T.C. carried out the analyses, reviewed and revised the manu-
script, and approved the final manuscript as submitted. A.J.M. assisted
with the design of the study, critically reviewed the manuscript, and
approved the manuscript as submitted. J.M.G. assisted with the design of
the study, critically reviewed the manuscript, and approved the manu-
script as submitted. L.V. conceptualized and designed the study, criti-
cally reviewed the manuscript, and approved the final manuscript as
submitted. All authors approved the final manuscript as submitted and
agree to be accountable for all aspects of the work.
TAGGEDH1REFERENCESTAGGEDEND
1. CDC. Most recent asthma data. 2018. Available at: https://www.cdc.
gov/asthma/most_recent_data.htm. Accessed July 11, 2018.
2. Soni, A, Statistical Brief #434: the five most costly children’s condi-
tions, 2011: estimates for U.S. civilian noninstitutionalized children,
ages 0-17. Available at: https://meps.ahrq.gov/data_files/publica
tions/st434/stat434.shtml. Accessed February 17, 2017.
3. Nurmagambetov T, Khavjou O, Murphy L, et al. State-level medical
and absenteeism cost of asthma in the United States. J Asthma.
2017;54:357–370. https://doi.org/10.1080/02770903.2016.1218013.
4. National Asthma Education and Prevention Program. Expert
panel report 3 (EPR-3): guidelines for the diagnosis and man-
agement of asthma-summary report 2007. J Allergy Clin Immu-
nol. 2007;120(5 suppl):S94–138. https://doi.org/10.1016/j.jaci.
2007.09.043.
5. Hsu J, Wilhelm N, Lewis L, et al. Economic evidence for US asthma
self-management education and home-based interventions. J Allergy
Clin Immunol Pract. 2016;4:1123–1134.e27. https://doi.org/
10.1016/j.jaip.2016.05.012.
6. Bhaumik U, Sommer SJ, Giller-Leinwohl J, et al. Boston Children’s
Hospital community asthma initiative: five-year cost analyses of a
home visiting program. J Asthma. 2017;54:134–142. https://doi.org/
10.1080/02770903.2016.1201837.
7. NCQA. HEDIS 2018. Available at: http://www.ncqa.org/hedis-qual
ity-measurement/hedis-measures/hedis-2018. Accessed July 11,
2018.
8. Hatoun J, Trudell EK, Vernacchio L. Identifying children at risk of
asthma exacerbations: beyond HEDIS. Am J Manag Care. 2018;24:
e170–e174.
9. Royston P, Moons KGM, Altman DG, et al. Prognosis and prognos-
tic research: developing a prognostic model. BMJ. 2009;338. https://
doi.org/10.1136/bmj.b604. b604-b604.
10. Kharbanda AB. Validation and refinement of a prediction rule to
identify children at low risk for acute appendicitis. Arch Pediatr
Adolesc Med. 2012;166:738. https://doi.org/10.1001/archpediat-
rics.2012.490.
11. Fuhlbrigge A, Peden D, Apter AJ, et al. Asthma outcomes: exacer-
bations. J Allergy Clin Immunol. 2012;129(3 suppl). https://doi.org/
10.1016/j.jaci.2011.12.983.
12. Vernacchio L, Trudell EK, Muto JM. Correlation of care process
measures with childhood asthma exacerbations. Pediatrics.
2013;131:e136–e143. https://doi.org/10.1542/peds.2012-1144.
13. Kharbanda AB, Vazquez-Benitez G, Ballard DW, et al. Develop-
ment and validation of a novel pediatric appendicitis risk calculator
(pARC). Pediatrics. 2018;141: e20172699. https://doi.org/10.1542/
peds.2017-2699.
14. Hoeksema LJ, Bazzy-Asaad A, Lomotan E, et al. Accuracy of a
computerized clinical decision-support system for asthma assess-
ment and management. J Am Med Inform Assoc. 2011;18:243–250.
https://doi.org/10.1136/amiajnl-2010-000063.
15. Andrews AL, Brinton D, Simpson KN, et al. A longitudinal exami-
nation of the asthma medication ratio in children. Am J Manag
Care. 2018;24:294–300.
16. Andrews AL, Brinton DL, Simpson KN, et al. A longitudinal
examination of the asthma medication ratio in children with Med-
icaid. JAsthma. 2020;57:1083–1091. https://doi.org/10.1080/
02770903.2019.1640727.
17. Morgan WJ, Crain EF, Gruchalla RS, et al. Results of a home-based
environmental intervention among urban children with asthma. N
Engl J Med. 2004;351:1068–1080. https://doi.org/10.1056/NEJ-
Moa032097.
18. Smith SR, Wakefield DB, Cloutier MM. Relationship between pedi-
atric primary provider visits and acute asthma ED visits. Pediatr
Pulmonol. 2007;42:1041–1047. https://doi.org/10.1002/ppul.20694.
19. Bateman ED, Buhl R, O’Byrne PM, et al. Development and valida-
tion of a novel risk score for asthma exacerbations: the risk score for
exacerbations. J Allergy Clin Immunol. 2015;135:1457–1464.e4.
https://doi.org/10.1016/j.jaci.2014.08.015.
20. Forno E, Fuhlbrigge A, Soto-Quir
os ME, et al. Risk factors and pre-
dictive clinical scores for asthma exacerbations in childhood. Chest.
2010;138:1156–1165. https://doi.org/10.1378/chest.09-2426.
21. Teach SJ, Gergen PJ, Szefler SJ, et al. Seasonal risk factors for
asthma exacerbations among inner-city children. J Allergy Clin
Immunol. 2015;135:1465–1473.e5. https://doi.org/10.1016/j.
jaci.2014.12.1942.
22. Hoch HE, Calatroni A, West JB, et al. Can we predict fall asthma
exacerbations? Validation of the seasonal asthma exacerbation
index. J Allergy Clin Immunol. 2017;140:1130–1137.e5. https://doi.
org/10.1016/j.jaci.2017.01.026.
23. Kenyon CC, Rubin DM, Zorc JJ, et al. Childhood asthma hospital
discharge medication fills and risk of subsequent readmission. J
Pediatr. 2015;166:1121–1127. https://doi.org/10.1016/j.
jpeds.2014.12.019.
24. Qureshi F, Zaritsky A, Poirier MP. Comparative efficacy of oral dexa-
methasone versus oral prednisone in acute pediatric asthma. J Pediatr.
2001;139:20–26. https://doi.org/10.1067/mpd.2001.115021.
25. Cooper WO, Hickson GB. Corticosteroid prescription filling for chil-
dren covered by Medicaid following an emergency department visit
or a hospitalization for asthma. Arch Pediatr Adolesc Med.
2001;155:1111–1115. https://doi.org/10.1001/archpedi.155.10.1111.
T
AGGEDEND54 HATOUN ET AL ACADEMIC PEDIATRICS