Keywords
clinical documentation - quality of care - interoperability - electronic health records
Background and Significance
Background and Significance
Accurate documentation of adverse reactions to medications, foods, or other substances
is important to patient safety and quality of care. Most electronic health record
(EHR) systems have an allergy module where providers document patients' allergies
and other adverse reactions (e.g., intolerances and contraindications). The EHR allergy
module relies on underlying terminologies to represent and encode allergy and adverse
drug reaction (ADR) information.[1] However, health care institutions in the United States utilize EHR systems supplied
by hundreds of certified health information technology developers, each siloed in
nature and often equipped with different features and clinical terminologies. Allergy
reaction “picklists” (i.e., coded reaction options presented to the health care provider
when entering the allergy) are often provided by third-party content vendors and/or
are customized to institution-specific dictionaries; even health care institutions
that use the same EHR vendor system can have different reaction picklists, ranging
from a dozen to a hundred reactions.[2]
[3]
The standard terminologies and value sets (a list of codes and corresponding terms
used to describe specific clinical information) within each EHR system serve as “building
blocks” that enable data exchange and sharing.[4]
[5] While initiatives for interoperable EHR systems have made progress in some domains,
such as medications, this has not held true in others, such as allergic reactions.[6] The lack of standardization and interoperability has downstream consequences on
data exchange between different providers, organizations, and research studies that
rely on the accuracy and consistency of coded data.[7]
[8]
[9] Standardization of allergy reaction picklists is fundamental to the utility of these
data for patient care, clinical research, drug safety, and postmarketing pharmacovigilance.[10] For example, many EHR systems have functions for clinicians to reconcile allergy
and ADR information from outside sources; however, different picklists and coding
mechanisms for reactions pose challenges for exchanging this information. In addition,
if data collection mechanisms vary by sites, the use of allergy and ADR EHR findings
from one site may not be generalizable to other populations. However, it is currently
unknown whether the lack of standardized reaction picklists impacts health care provider
data entry.
Objective
In the present study, we investigated differences in ADRs documented in the EHR allergy
list between two large U.S. health care delivery networks over a 5-year period. We
hypothesized that the prevalence of reported drug reactions would be similar but that
differences between sites could be influenced by differences between predefined reaction
picklists.
Methods
Clinical Settings and Data Collection
We obtained ADR data from Brigham and Women's Hospital (BWH, Boston, MA, United States)
and University of Colorado Hospital (UCH, Aurora, CO, United States). BWH and UCH
are both large tertiary academic hospitals and members of the integrated health care
delivery networks Mass General Brigham (formerly Partners HealthCare) and University
of Colorado Health, respectively. BWH and UCH were considered ideal comparison sites
given that they utilize the same commercial EHR system (e.g., Epic Systems, Verona,
WI, United States) but have institution-specific reaction picklists. At both sites,
patient ADR information can be documented by any health care team member via the EHR
allergy module.
In this study, patients who visited the emergency department and/or outpatient clinics
at BWH or UCH between 2013 and 2018 were included. We extracted patients' demographics
(e.g., sex and racial/ethnic group) and ADR information, including allergen (e.g.,
culprit drug), allergy status (active, inactive, or deleted), date/time of entry/update,
coded reaction(s), and role of the documenting health care team member (e.g., physician,
nurse, and medical assistant) from the EHR data warehouses at each site.
Data Analysis
In EHRs, the documented drug allergen can be a specific drug or drug class. Using
an approach similar to our previous work,[11] we classified drugs into corresponding drug classes using the American Hospital
Formulary Service (AHFS) Pharmacologic-Therapeutic Classification; we further classified
some drug classes into broader classes (e.g., “cephalosporin antibiotics—first generation”
and “cephalosporin antibiotics—second generation” were combined into “cephalosporins”).[12] We included only the most common drug class allergens that comprised at least 0.5%
of all reported drug allergies. Note: Within the EHR, the allergy module is where clinicians document both true allergies
and other adverse reactions (i.e., ADRs). The reaction picklist for documenting allergies
and ADRs is identical. While the field on the electronic form is called an “allergy
field,” this article focuses on specific ADRs rather than true allergies. Throughout
this text, we focus on ADRs, and any use of the term “allergy” refers to the EHR module
and common reaction picklist. Furthermore, the term “drug class allergen” refers to
ADRs reported in reference to a particular drug agent.
We compared patient demographics of the overall population and those with allergies
between BWH and UCH. We also examined the number of ADR records entered by health
care provider role. Because BWH transitioned to a new EHR system during the study
period, some ADR records were updated via a conversion process. We, therefore, used
a subset of the records postconversion from August 1, 2018 to December 31, 2018 to
estimate the proportions of reactions entered by different types of providers. Reported
ADR prevalences were calculated as the number of patients with an active ADR to a
drug class considering the total study population. We compared the proportions of
the 40 most frequently reported reactions at both institutions. Reaction proportions
were calculated out of the total number of reported reactions at each institution.
We further examined and compared the top 10 reported coded reactions for each drug
class at each institution. In all comparable analyses, we considered 1% a notable
difference; thus 1% was used as the difference threshold. At each institution, because
gastrointestinal (GI) reactions were documented using a group of specific reaction
codes (e.g., BWH has “GI Upset,” “Nausea Only,” “Vomiting,” “Nausea And Vomiting,”
and “Nausea and/or Vomiting” and UCH has “GI Reaction,” “Nausea,” and “Vomiting”),
we merged GI reactions into a single group. We compared frequencies using a chi-square
test. All reported p-values with type I error (α) of <0.05 were considered to be statistically significant.
Statistical analyses were performed using Microsoft SQL Management Studio Version
18.4. This study was approved by the Mass General Brigham Human Research Committee
and Colorado Multiple Institutional Review Board.
Results
General Description of the Patient Population
A total of 2,160,116 patients were included in this study, with 1,530,641 (71%) from
BWH and 629,475 (29%) from UCH ([Table 1]). Approximately, one-third of the populations at BWH (30%, n = 454,011) and UCH (30%, n = 186,433) had at least one EHR ADRs documented. In total, there were 705,413 active
drug ADR records with 1,230,165 reactions at BWH and 223,560 ADR records with 586,750
reactions at UCH. Among patients with allergic reactions, the majority of patients
at both institutions were White (76%) and female (60%).
Table 1
Demographics of overall patient population and patients with allergies at each institution
and health care team members role who documented allergies in the EHR
|
|
Institution
|
|
|
Total
n (%)
|
BWH
n (%)
|
UCH
n (%)
|
|
|
All patient demographics[e]
|
(n = 2,160,116)
|
(n = 1,530,641)
|
(n = 629,475)
|
c[2]
|
|
Sex
|
|
|
|
994.8
|
|
Female
|
1,299,274 (60)
|
930,421 (61)
|
368,853 (59)
|
|
|
Male
|
857,389 (40)
|
596,878 (39)
|
260,511 (41)
|
|
|
Race/Ethnic group
|
|
|
|
20,556.6
|
|
White
|
1,635,350 (76)
|
1,171,107 (77)
|
464,243 (74)
|
|
|
Non-White
|
483,984 (24)
|
294,139 (23)
|
189,845 (26)
|
|
|
Hispanic
|
189,582 (9)
|
116,990 (8)
|
72,592 (12)
|
|
|
Black
|
138,460 (6)
|
99,460 (7)
|
39,000 (6)
|
|
|
Asian
|
76,133 (4)
|
61,213 (4)
|
14,920 (2)
|
|
|
Other[a]
[b]
|
79,809 (4)
|
16,476 (1)
|
63,333 (10)
|
|
|
Patients with reported allergies demographics[e]
|
(n = 640,433)
|
(n = 454,011)
|
(n = 186,433)
|
|
|
Sex
|
|
|
|
3,787.1
|
|
Female
|
448,259 (70)
|
327,739 (72)
|
120,520 (65)
|
|
|
Male
|
190,832 (30)
|
124,934 (28)
|
65,898 (35)
|
|
|
Race/Ethnic group
|
0
|
|
|
6,798.1
|
|
White
|
530,420 (83)
|
380,545 (84)
|
149,875 (80)
|
|
|
Non-White
|
99,799 (17)
|
58,549 (16)
|
41,250 (20)
|
|
|
Hispanic
|
37,218 (6)
|
21,931 (5)
|
15,287 (8)
|
|
|
Black
|
31,115 (5)
|
21,435 (5)
|
9,680 (5)
|
|
|
Asian
|
14,589 (2)
|
11,220 (2)
|
3,369 (2)
|
|
|
Other[a]
[b]
|
16,877 (3)
|
3,963 (1)
|
12,914
|
|
|
Number reactions entered by provider role[c]
[e]
|
(n = 70,915)
|
(n = 48,335)
|
(n = 22,580)
|
|
|
Registered nurse
|
21,487 (30.3)
|
14,328 (29.6)
|
7,159 (31.7)
|
4,933.4
|
|
Medical assistant
|
16,042 (22.6)
|
7,828 (16.2)
|
8,214 (36.4)
|
|
|
Physician
|
10,278 (14.5)
|
8,774 (18.2)
|
1,504 (6.7)
|
|
|
Nurse practitioner
|
4,014 (5.7)
|
3,526 (7.3)
|
488 (2.2)
|
|
|
Physician assistant
pharmacist
|
1,418 (2.0)
825 (1.2)
|
1,148 (2.4)
436 (0.9)
|
270 (1.2)
389 (1.7)
|
|
|
Others[d]
|
9,267 (13.1)
|
5,293 (11.0)
|
3,974 (17.6)
|
|
|
Unknown
|
7,584 (10.7)
|
7,002 (14.5)
|
582 (2.6)
|
|
Abbreviations: BWH, Brigham and Women's Hospital; EHR, electronic health record; UCH,
University of Colorado Hospital.
a Other (BWH) includes Hawaiian, Mixed, Native American.
b Other (UCH) includes American Indian and Alaska Native, more than one race, Native
Hawaiian and other Pacific Islander, and others.
c Because BWH transitioned to a new EHR system during the study period, some allergy
records were updated via a conversion process. We, therefore, used a subset of the
records postconversion from August 1, 2018 to December 31, 2018 to estimate the proportions
of reactions entered by different types of providers.
d Other refers to provider roles not listed (i.e., pharmacist, pharmacy technician,
resident, etc).
e
p-Values comparing BWH and UCH by chi-square analysis were significant for all analyses
(sex, race/ethnic group, number reactions entered by provider role, and p < 0.001).
ADR information was primarily documented by medical assistants and registered nurses
with a higher proportion at UCH than at BWH (36 vs. 16% and 32 vs. 30%, respectively,
p < 0.0001). In contrast, physicians and nurse practitioners documented allergies considerably
more at BWH than at UCH (18 vs. 7% and 7 vs. 2%, respectively, p < 0.0001).
Most Commonly Reported Drug Class Allergens
Across both institutions, the most commonly reported drug class allergens were penicillins
(14%), opioids (10%), sulfonamide antibiotics (9%), and nonsteroidal anti-inflammatory
drugs (NSAIDs) (5%) ([Table 2]). The frequency of the most commonly reported drug class allergens was similar at
both institutions, with the exception of two classes that exhibited greater than 1%
difference across sites: opioids (BWH: 10 vs. UCH: 12%) and angiotensin-converting-enzyme
(ACE) inhibitors (BWH: 3 vs. UCH: 2%).
Table 2
Frequently reported drug class allergens among the total patient population at each
institution
|
Allergen drug class[a]
[e]
|
Total patients
(n = 2,160,116)
n (%)
|
BWH
(n = 1,530,641)
|
UCH
(n = 629,475)
|
|
|
n (%)
|
Ranking
|
n (%)
|
Ranking
|
c[2]
|
|
Penicillins
|
297,354 (13.8)
|
210,215 (13.7)
|
1
|
87,139 (13.8)
|
1
|
50,941.6
|
|
Opioids
|
225,116 (10.4)
|
147,783 (9.7)
|
2
|
77,333 (12.3)
|
2
|
22,047.3
|
|
Sulfonamides
|
192,994 (8.9)
|
135,026 (8.8)
|
3
|
57,968 (9.2)
|
3
|
30,767.5
|
|
NSAIDs[b]
|
102,574 (4.7)
|
73,596 (4.8)
|
4
|
28,978 (4.6)
|
4
|
19,408.1
|
|
Macrolides
|
65,792 (3.0)
|
49,475 (3.2)
|
5
|
16,317 (2.6)
|
6
|
16,711.0
|
|
ACE inhibitors[c]
|
60,734 (2.8)
|
48,963 (3.2)
|
6
|
11,771 (1.9)
|
8
|
22,775.5
|
|
Cephalosporins
|
59,875 (2.8)
|
43,130 (2.8)
|
7
|
16,745 (2.7)
|
5
|
11,627.0
|
|
Fluoroquinolones
|
49,268 (2.3)
|
35,338 (2.3)
|
8
|
13,930 (2.2)
|
7
|
9,302.2
|
|
Statins
|
42,115 (1.9)
|
33,448 (2.2)
|
9
|
8,667 (1.4)
|
10
|
14,581.5
|
|
Tetracyclines
|
34,150 (1.6)
|
25,115 (1.6)
|
10
|
9,035 (1.4)
|
9
|
7,571.5
|
|
Phenothiazines
|
18,816 (0.9)
|
12,680 (0.8)
|
11
|
6,136 (1.0)
|
11
|
2,275.9
|
|
Thiazide diuretics
|
13,406 (0.6)
|
10,846 (0.7)
|
12
|
2,560 (0.4)
|
13
|
5,121.4
|
|
Lincosamides
|
15,124 (0.7)
|
10,839 (0.7)
|
13
|
4,285 (0.7)
|
12
|
2,840.23
|
|
Other[d]
|
182,985 (8.5)
|
103,654 (6.8)
|
n/a
|
79,331 (12.6)
|
n/a
|
3,233.1
|
Abbreviations: ACE, angiotensin-converting-enzyme; BWH, Brigham and Women's Hospital;
n/a, not available; NSAID, nonsteroidal anti-inflammatory drugs; UCH, University of
Colorado Hospital.
a Reported drug allergy frequencies were calculated as the number of patients with
an active allergy to the drug class considering the total study population.
b NSAIDs is an abbreviation for nonsteroidal anti-inflammatory drugs.
c ACE inhibitors is an abbreviation for angiotensin-converting enzyme inhibitors.
d “Other” aggregates patients with reported allergies to all other drug classes not
presented.
e
p-Values comparing BWH and UCH by chi-square analysis were significant for all allergen
drug classes (penicillins, p = 0.0341; lincosamides, p = 0.0281; all other allergen drug classes, p < 0.0001). Note that as our dataset is very large, p-values are getting very small, even though the differences are not really significant
clinically based on the percentages.
Diversity in Prevalences of Adverse Drug Reactions between Two Institutions
While BWH's reaction picklist had only 48 reactions, UCH's had 160. Of the 179 unique
reactions, 29 (16%) were common to both picklists, 19 (11%) only on BWH's, and 131
(74%) only on UCH's ([Supplementary Table 1], available in the online version). [Table 3] displays the top 40 coded reactions reported at both sites, including “other (see
comments),” which indicates there was a free-text reaction documented, “unknown,”
and “null” (e.g., no coded reaction). The most commonly reported reactions at both
institutions were similar, including “rash,” “GI upset,” “hives,” “itching” and “anaphylaxis.”
Table 3
Top 40 reported drug allergen reactions institution wide and considering picklist
reactions not shared by both institutions (bold)
|
Ranking
|
BWH (n = 1,230,165)
|
UCH (n = 586,750)
|
|
Reaction[a]
|
n (%)
|
Reaction[a]
|
n (%)[b]
|
|
1
|
Rash
|
169,288 (13.8)
|
Rash
|
77,689 (13.2)
|
|
2
|
Hives
|
108,233 (8.8)
|
Hives
|
50,434 (8.6)
|
|
3
|
GI upset
|
66,036 (5.4)
|
Nausea/vomiting
|
47,433 (8.1)
|
|
4
|
Nausea/vomiting
|
60,448 (4.9)
|
Itching
|
26,337 (4.5)
|
|
5
|
Itching
|
50,560 (4.1)
|
Swelling
|
20,345 (3.5)
|
|
6
|
Anaphylaxis
|
39,949 (3.2)
|
Anaphylaxis
|
18,435 (3.1)
|
|
7
|
Swelling
|
29,767 (2.4)
|
Shortness of breath
|
10,935 (1.9)
|
|
8
|
Cough
|
21,873 (1.8)
|
Diarrhea
|
7,540 (1.3)
|
|
9
|
Angioedema
|
18,816 (1.5)
|
Headache
|
6,929 (1.2)
|
|
10
|
Mental status change
|
17,492 (1.4)
|
GI reaction[c]
|
6,769 (1.2)
|
|
|
|
Throat swelling
|
6,079 (1.0)
|
|
11
|
Shortness of breath
|
14,767 (1.2)
|
Abdominal cramping
|
5,972 (1.0)
|
|
12
|
Diarrhea
|
13,396 (1.1)
|
Cough
|
5,903 (1.0)
|
|
13
|
Headache
|
10,184 (0.8)
|
Hallucination
|
5,416 (0.9)
|
|
14
|
Myalgia
|
9,714 (0.8)
|
Sneezing
|
5,329 (0.9)
|
|
15
|
Sneezing
|
7,137 (0.6)
|
Dizziness
|
4,980 (0.8)
|
|
16
|
Musculoskeletal pain
|
6,262 (0.5)
|
Blistering
|
4,113 (0.7)
|
|
17
|
Palpitations
|
4,877 (0.4)
|
Anxiety
|
3,781 (0.6)
|
|
18
|
Flushing
|
4,812 (0.4)
|
Itchy watery eyes
|
3,423 (0.6)
|
|
19
|
Bronchospasm
|
4,726 (0.4)
|
Myalgia
|
3,217 (0.5)
|
|
20
|
Fever
|
4,191 (0.3)
|
Swollen tongue
|
2,421 (0.4)
|
|
21
|
Hypotension
|
3,751 (0.3)
|
Palpitations
|
2,403 (0.4)
|
|
22
|
Dystonia
|
2,991 (0.2)
|
Fever
|
2,274 (0.4)
|
|
23
|
Anxiety
|
2,968 (0.2)
|
Confusion
|
2,198 (0.4)
|
|
24
|
Wheezing
|
2,960 (0.2)
|
Congestion nose
|
2,114 (0.4)
|
|
25
|
Dizziness
|
2,698 (0.2)
|
Arrhythmia
|
2,089 (0.4)
|
|
26
|
Renal toxicity
|
2,340 (0.2)
|
Agitation
|
2,088 (0.4)
|
|
27
|
Hepatoxicity
|
2,272 (0.2)
|
Edema
|
2,014 (0.3)
|
|
28
|
Dermatitis
|
2,210 (0.2)
|
Watering eyes
|
1,644 (0.3)
|
|
29
|
Seizures
|
1,570 (0.1)
|
Seizures
|
1,514 (0.4)
|
|
30
|
Arrhythmia
|
1,174 (0.1)
|
Respiratory distress
|
1,453 (0.2)
|
|
31
|
Maculopapular rash
|
1,137 (0.09)
|
Asthma
|
1,429 (0.2)
|
|
32
|
Thrombocytopenia
|
741 (0.06)
|
Flushing
|
1,315 (0.2)
|
|
33
|
Rigor
|
685 (0.06)
|
Itching of mouth
|
1,261 (0.2)
|
|
34
|
Erythema multiforme
|
430 (0.03)
|
Bleeding
|
1,247 (0.2)
|
|
35
|
Acute generalized exanthematous pustulosis
|
422 (0.03)
|
Fatigue
|
1,230 (0.2)
|
|
36
|
Lightheadedness
|
414 (0.03)
|
Airway obstruction
|
1,212 (0.2)
|
|
37
|
Photosensitivity
|
388 (0.03)
|
Chest pain
|
1,135 (0.2)
|
|
38
|
Tinnitus
|
338 (0.03)
|
Hypotension
|
1,069 (0.2)
|
|
39
|
Anemia
|
330 (0.03)
|
Eye swelling
|
1,037 (0.2)
|
|
40
|
Fixed drug eruption
|
244 (0.02)
|
|
|
|
Other (see comments)[d]
|
217,904 (17.7)
|
Other (see comments)[d]
|
19,444 (3.3)
|
|
Unknown
|
171,529 (13.9)
|
Unknown
|
9,693 (1.7)
|
|
Null
|
147,720 (12.0)
|
Null
|
171,871 (29.3)
|
Abbreviations: BWH, Brigham and Women's Hospital; GI, gastrointestinal; UCH, University
of Colorado Hospital.
a Bold formatting denotes structured reaction terms that do appear on the other institution's
reaction picklist.
b Reported allergic drug reaction prevalences were calculated as the number of patients
with a documented reaction considering the study population with at least one reported
reaction.
c GI is an abbreviation for gastrointestinal.
d “Other (see comments)” refers to other reactions where free text was entered.
The naming and prevalences of the remaining reactions on the picklists are more diverse.
For example, reactions related to “swelling” and “angioedema” were coded differently
at both sites. “Swelling” was reported with a noticeable difference, 2.4% at BWH versus
3.5% at UCH. “Angioedema” was on BWH's picklist but not on UCH's, while UCH included
“throat swelling,” “swollen tongue,” and “edema.” Another example of the differences
in the reactions was “myalgia,” which appeared on both picklists, but “musculoskeletal
pain,” appeared only on BWH's.
Of the top 25 reactions, “mental status change,” “musculoskeletal pain', “bronchospasm,”
“dystonia,” “wheezing,” and “renal toxicity” were only reported at BWH. In contrast,
“abdominal cramping,” “hallucination,” “blistering,” “itchy watery eyes,” “confusion,”
and “congestion nose” were only reported at UCH. In addition, BWH's picklist included
severe and rare hypersensitivity reactions, such as “acute generalized exanthematous
pustulosis” and “rash with skin desquamation.” Health care team members more frequently
entered “other (see comments)” at BWH than at UCH (18 vs. 3%, respectively). On the
contrary, UCH had considerably more “NULL” (e.g., left blank) reactions than BWH (29
vs. 12%, respectively).
Furthermore, if an institution does not have a structured entry option for a certain
reaction within its picklist, the reaction must be entered as free text in the comments
field. [Table 4] shows several examples of commonly found reactions that are available as a structured
entry in one institution's picklist but not in that of the other institution's, thus
requiring the use of free-text entry. Reactions that were already available as a structured
entry option in an institution's picklist were recorded at a higher frequency in the
EHR when compared with another institution whose picklist did not contain a structured
entry for the reaction, thus requiring free-text entry ([Table 4]).
Table 4
Examples of allergen drug reactions being recorded as structured entry versus free
text between the two institutions[a]
|
BWH
(n = 3,174,702)
|
UCH
(n = 846,465)
|
|
Structured entry
n (%)
|
Free-text entry
n (%)
|
|
Angioedema
|
18,816 (0.6)
|
1,126 (0.1)
|
|
Mental status change
|
17,492 (0.6)
|
1,957 (0.2)
|
|
Musculoskeletal pain
|
6,262 (0.2)
|
2 (0.0002)
|
|
Bronchospasm
|
4,726 (0.1)
|
863 (0.1)
|
|
Wheezing
|
2,960 (0.09)
|
455 (0.05)
|
|
Free-text entry
n
(%)
|
Structured entry
n
(%)
|
|
Throat swelling
|
11,107 (0.3)
|
6,079 (0.7)
|
|
Abdominal cramping/pain
|
7,846 (0.2)
|
5,972 (0.7)
|
|
Hallucination
|
15,756 (0.5)
|
5,416 (0.6)
|
|
Blistering
|
7,243 (0.2)
|
4,113 (0.4)
|
|
Itchy watery eyes
|
4,548 (0.1)
|
3,423 (0.4)
|
Abbreviations: BWH, Brigham and Women's Hospital; UCH, University of Colorado Hospital.
a “n” delineates the total number of instances an allergen drug reaction is mentioned,
within both structured entry and free-text entry, from that specific institution.
Diversity in Proportion of Adverse Drug Reactions by Drug Class between Two Institutions
Overall, the majority of reported reactions to antibiotic drug classes were potential
hypersensitivity reactions (e.g., rash and hives) which appear on both institutions'
picklists ([Fig. 1A]). Penicillins, sulfonamides, cephalosporins, and lincosamides displayed similar
reaction distribution across sites; however, rash was reported more at BWH than at
UCH across all antibiotic drug classes. Musculoskeletal pain to fluoroquinolones at
BWH was comparable in prevalence to myalgia to fluoroquinolones at UCH.
Fig. 1 (A) For the antibiotics listed below, we demonstrate the most common reactions at BWH
(B) and UCH (U). A majority of the reported reactions to antibiotic drug classes were
hypersensitivity reactions (e.g., rash and hives) which appear on both institutions'
picklists. Across all antibiotic drug classes, rash was reported more at BWH than
at UCH. While angioedema was reported across all drug classes at BWH, it was not reported
at UCH as it is not a coded reaction term on UCH's picklist. “Other reactions” includes
coded reactions that do not comprise the top 10 frequently reported reactions, excluding
“null,” “other: see comments,” and “unknown.” (B) For the drugs listed below, we demonstrate the most common reactions at BWH (B)
and UCH (U). Rash to opioids, NSAIDs, and thiazide diuretics was reported considerably
more at BWH than at UCH. While mental status change to opioids was reported at BWH,
this term is not included on UCH's picklist. Instead, hallucinations to opioids were
reported at UCH. The sum of “swelling” and “angioedema” reported reactions to ACE
inhibitors at BWH is comparable to “swelling” at UCH, as angioedema does not exist
on UCH's reaction picklist. In addition to myalgia, musculoskeletal pain was reported
in response to statins at BWH but does not exist on UCH's picklist. Dystonia to phenothiazines
was reported considerably more at BWH than at UCH. Other Reactions includes coded
reactions that do not comprise the top 10 frequently reported reactions, excluding
“null,” “other: see comments,” and “unknown.”
Reported reactions to nonantibiotics exhibited greater variability ([Fig. 1B]). Rash to opioids, NSAIDs, and thiazide diuretics was reported considerably more
at BWH than at UCH. While mental status change to opioids was reported at BWH, this
term is not included on UCH's picklist. Instead, hallucinations to opioids were reported
at UCH. The sum of “swelling” and “angioedema” reported reactions to ACE inhibitors
at BWH is comparable to “swelling” at UCH, as angioedema does not exist on UCH's reaction
picklist. For NSAIDs, bronchospasm and renal toxicity were among the top 10 reactions
at only BWH as they do not appear on UCH's picklist, while bleeding was only reported
at UCH as it is absent from BWH's picklist. For statins, both myalgia and musculoskeletal
pain comprised the top 10 reactions at BWH, which were comparable in prevalence to
myalgia at UCH, as their picklist does not include musculoskeletal pain. Dystonia
to phenothiazines was reported considerably more at BWH than at UCH. Cough encompassed
most of the reported reactions to ACE inhibitors at both BWH and UCH.
Discussion
This study presents an investigation of the diversity of drug allergies and reactions
documented in the EHR across two large sites. Few previous studies have focused on
reported drug allergies by patients and patient characteristics,[11] but no study to date has focused on reaction differences across institutions with
the same commercial EHR vendor but different reaction picklists. We found that the
top reported drug allergens were largely similar between BWH and UCH. Antibiotics,
opiates, and sulfonamides continue to represent a large proportion of drug allergies
across multiple institutions. On the contrary, we found greater variability in the
commonly reported ADRs across the two institutions. This variation is, perhaps, a
product of each institution having its own reaction picklist. Clinicians' reporting
of drug allergen reactions is indeed picklist-driven and influenced by the available
coded entries. The EHR design and what is available in the picklist may not only influence
reaction documentation but also downstream clinical decision-making and analyses of
EHR ADR data for drug surveillance and research. For studies involving secondary use
of EHR data, researchers should consider potential biases in clinical documentation
due to EHR design when interpreting results related to reactions.
With respect to documented causative drugs, our findings are consistent with prior
studies. Historically, antibiotics have accounted for a majority of documented drug
allergies.[13] A previous study conducted utilizing EHR data from 1990 to 2013 showed the most
frequently reported drug class allergen was penicillins, followed by sulfonamide antibiotics.[11] Our study found that this is still true today for penicillins; however, opioids
replaced sulfonamides as the second most commonly reported drug class allergen at
both sites. This shift may be due to higher overall opiate exposure in patients in
the United States.[14] Interestingly, we determined that the frequency of documented opioid allergies was
considerably higher at UCH than at BWH (11.9 vs. 9.3%, respectively), and the number
of opioid prescriptions per 100 persons in Colorado (52.9) is greater than that in
Massachusetts (40.1).[15] Overall, the rates of all reported drug class allergens were comparable between
the two institutions. This similarity could potentially be attributed to BWH and UCH
using the same commercial medication data dictionary for drug allergens (i.e., First
Databank Inc.), which may prompt similar documentation patterns, enabling more accurate
and feasible comparisons across sites.
While BWH and UCH share a common dictionary for allergen documentation, the two health
care centers, despite utilizing the same EHR system, have institution-specific reaction
picklists. At the time of this study, there were 48 reactions on BWH's picklist but
160 on UCH's. Perhaps due to the picklist differences, there were several notable
differences in the specific coded reactions reported at the two sites. For example,
swelling was more commonly reported across all the top drug classes at UCH than at
BWH. The current reaction picklist at BWH had both “angioedema” and “swelling” as
coded reactions, while UCH had “swelling,” “throat swelling,” and “swollen tongue.”
Thus, clinicians at BWH may differentiate between etiologies of swelling (angioedema
vs. edema) when entering reaction details, whereas clinicians at UCH differentiate
swelling by location (e.g., “throat swelling” and “swollen tongue”). Indeed, when
comparing only the “swelling” coded reactions between BWH and UCH for ACE inhibitors,
it initially appears that swelling is reported considerably more frequently at UCH
for that drug class. However, after aggregating the “swelling” and “angioedema” coded
reactions at BWH, the sum was comparable to “swelling” at UCH in response to ACE inhibitors.
Many reactions were entered in the allergy EHR module only as free-text comments.
At BWH, 17% of reactions were entered as free text compared with only 3% at UCH. A
possible reason for the higher number of free-text reactions at BWH may be due in
part to the diversity of reactions available in the picklist (40 vs. 160 at UCH).
There are certain reactions that have a structured entry option in one institution's
picklist but not in that of the other institution. Because of this, clinicians in
the other institution, whose picklist does not contain the structured entry option,
must utilize free-text comments to record the specific reaction. We conducted free-text
extraction and analysis to compare reactions' presence as structured entry versus
as free text between the two institutions. This was conducted using a natural language
processing tool and a reaction lexicon consisting of 469 uniquely identified reaction
concepts.[16] As shown, having a structured entry option for an ADR may promote increased documentation
for that specific reaction.
Further, free-text comments are encouraged for complete documentation by the Allergy
Clinical Consensus Group within BWH's health care system.[17] Along this line, a drug allergy practice parameter—developed by the American Academy
of Allergy, Asthma and Immunology, the American College of Allergy, Asthma and Immunology,
and the Joint Council of Allergy, Asthma and Immunology—advises that a relevant drug-allergy
history should include ample details such as the symptoms' timing, onset, duration,
relationship with medication use while also discussing the history of previous reactions,
and their management.[17] Documenting these details requires documentation beyond Epic's coded fields. Furthermore,
it is critical to consider the balance between documenting enough details to guide
proper management and doing so in a means that is easy enough for patients to fill
out and without contributing to clinician burnout.
Another possible explanation for the variation is different allergy documentation
policies at the two institutions. Registered nurses and physicians are the primary
clinicians who enter reactions at BWH, while this role is primarily held by medical
assistants or nurses at UCH. A previous study conducted at Mass General Brigham using
EHR data from 2005 to 2012 found that it was most often nonallergist physicians who
interfaced with the EHR allergy module and that most of the team members lack sufficient
drug allergy knowledge. The allergy/ADR classification was customized from the AHFS,
which was developed by pharmacists who serve as patient care providers in hospitals,
health systems, ambulatory clinics, and other health care settings. This highlights
a potential knowledge gap between pharmacists, who are developing the allergy/ADR
classification found within the EHR module and the nonpharmacist clinicians who are
most frequently interfacing with the EHRs allergy modules and documenting ADRs. Our
data indicate that pharmacists only accounted for 1.5% of free-text entries at BWH
and 3.1% at UCH, while the majority of free text was entered by registered nurses,
physicians, and medical assistants (see [Appendix]). Standardization of the nomenclature and clear understanding of the classification
between pharmacists and the clinicians in charge of documentation is critical. The
different roles might, therefore, contribute to the differences in documentation.
To improve the quality and utility of the EHR allergy module, more allergy education
and training of best practices for allergy documentation is needed.[18]
Standardization of reaction picklists would substantially enhance the interoperability
of EHR systems and allow providers and organizations to exchange electronic health
information easily and accurately.[19] Having a common reaction terminology and picklist is important on both the patient
and population level. Many patients visit more than one hospital or clinic throughout
their course of care,[8] and incomplete or inaccurate carryover of reaction information may result in duplicated
documentation or clinical testing.[20] On the population level, standardized reaction picklists enable accurate comparison
and pooling of EHR data from multiple institutions to study true epidemiological trends,
as opposed to differences observed due to picklist variation. Furthermore, reaction
list standardization as a gateway for more accurate documentation is particularly
relevant for drug safety surveillance, including the Food and Drug Administrations'
Adverse Event Reporting System.[10]
Development of a standardized reaction picklist that captures the most frequently
documented reactions would be valuable for improving ADR documentation and promoting
the use of coded fields over free-text entries which will help improve the specificity
of allergy alerts. Current drug-allergy alerting mechanisms in the EHR do not consider
the type and severity of the reaction (e.g., mild nausea vs. anaphylaxis). Future
improvement on alerting that considers these factors would allow more specificity
and potentially reduce the number of alerts and clinician alert fatigue.
However, a picklist that is granular and lengthy forces clinicians to scroll, which
is time-consuming. To mitigate the concerns with a lengthy picklist, we developed
a dynamic, data-driven reaction picklist that automatically generates commonly reported
reactions to a particular allergen—based on statistical association methods (e.g.,
support, lift, derived term frequency inverse document frequency).[16] Thus, optimizing the user interface of the ADR field, together with the picklist,
could facilitate the accuracy of ADR documentation. Multiple factors including the
granularity and length of the picklist, as well as the coded reactions available,
impact the accuracy of documentation. Future research could utilize natural language
processing to compare notes and laboratory data (i.e., renal function and liver function)
within the EHR with the coded ADR list to detect discrepancies and prompt clinicians
to reconcile such discrepancies and thus improve the accuracy of ADR documentation.
Although this study highlighted the diversity of reactions documented in EHRs that
utilize institution-specific picklists, there are several limitations worth noting.
First, many different types of ADRs are documented in the EHR, such as common side
effects (e.g., diarrhea), intolerances (e.g., GI upset), and immune-mediated hypersensitivities
(e.g., anaphylaxis). Although it is possible to specify the “reaction type” in the
EHR, this field is rarely used by clinicians. Even when a documented reaction is potentially
immune mediated, it is rarely confirmed with specialist assessment or diagnostic testing.
Thus, the EHR allergy module houses a myriad of reactions that may not be “true” allergies.
Furthermore, the quality of documentation is reliant on the knowledge of the person
entering the data.[17] Prior studies evaluating the accuracy and effectiveness of ADR documentation have
identified discrepancies in ADR documentation within hospital systems and the process
of transferring ADR information into the EHR.[21] Additional studies have demonstrated widespread under-reporting of ADRs including
serious or severe ADRs.[22]
Thus, it is important to consider the accuracy of the EHR allergy records and the
distinction between true allergies versus ADRs. This discussion raises the question
about whether ADRs should be documented within allergy fields and how to best utilize
the module to differentiate among the entries. One possible solution is to promote
and train clinicians documenting allergies, ADRs and other adverse reactions to utilize
the “type” field within the allergy module. Currently, four types of reactions (allergy,
intolerance, contraindication, unknown) are available and can help differentiate from
true allergies.
These data additionally include just two large academic health care centers that are
not representative of the broader patient population. Future studies of reaction picklists
across more academic and community hospitals are needed to confirm these initial observations.
Conclusion
Even when using the same commercial EHR system, hospitals and health care systems
often adopt institution-specific allergy reaction picklists. The availability and
granularity of the reactions in the picklist likely influence ADR documentation and
contribute to the variations of frequencies and proportions of reported reactions
across EHR systems. Different picklists make it difficult to share records or conduct
comparisons across multiple institutions and, hence, affect downstream analyses of
EHR data. Future work must focus on methods to standardize reaction picklists to improve
the accuracy and completeness of ADR documentation as it relates to clinical care
and patient safety. Furthermore, it is critical to develop methods to improve documentation
without contributing to clinician burnout.
Clinical Relevance Statement
Clinical Relevance Statement
Accurate documentation of adverse reactions to medications, foods, and other substances
is critical to ensure quality and safety of care. Differences in coverage and granularity
of reaction picklists impact how health care providers document ADRs in the EHR and
may contribute to variations in reported ADRs across health care systems. Standardized
reaction picklists may facilitate more efficient sharing and comparison of allergy
and ADR information across sites.
Multiple Choice Questions
Multiple Choice Questions
-
Which of the following describes a benefit of standardized reaction picklists compared
with proprietary picklists?
-
Standardized picklists increase the specificity of reaction documentation.
-
Standardized picklists increase interoperability between EHRs.
-
Standardized picklists are shorter.
-
Standardized picklists are better at capturing reaction severity and type.
Correct Answer: The correct answer is option b. Standardized coded reaction lists are not necessarily
more or less specific than existing coded reaction lists, which may be more or less
specific/granular by location. Standardized picklists do increase interoperability
and facilitate sharing, transfer, and comparison of allergy documentation across different
sites. Because proprietary picklists currently vary greatly in length, a standardized
picklist may be longer or shorter depending on the location. A standardized reaction
picklist would only include reaction names; reaction severity and type (e.g., allergy
or intolerance) would still need to be documented separately.
-
Which of the following describes a difference between coded reactions compared with
reactions documented in free text?
-
Coded reactions are more accurate.
-
Coded reactions are automatically visible in the EHR, while free-text reactions are
not.
-
Coded reactions are easier to use in downstream applications.
-
Coded reactions can be entered by clinicians of all roles, while only physicians can
add free-text comments.
Correct Answer: The correct answer is option c. Coded reactions are not necessarily more or less
accurate than reactions documented in free text. The visibility of coded reactions
and reactions documented in free-text comments depends on the EHR system in use and
is not inherent to the method of documentation. Coded reactions can more easily be
used in downstream applications such as allergy alerts and other automated clinical
decision support tools in the EHR. Whether or not clinicians of different roles (e.g.,
nurses, medical assistants, or physicians) can enter reaction information is determined
by institution-specific policies and is not inherent to the method of documentation.
