The Use of Electronic Health Records to Study Drug-Induced Hypersensitivity Reactions from 2000 to 2021: A Systematic Review

SYNOPSIS

Electronic health records (EHRs) have revolutionized the field of drug hypersensitivity reaction (DHR) research. In this systematic review, we assessed 140 articles from 2000-2021, classifying them under six themes: observational studies (n=61), clinical documentation (n=27), case management (n=22), clinical decision support (CDS) (n=18), case identification (n=9), and genetic studies (n=3). EHRs provide convenient access to millions of medical records, facilitating epidemiological studies of DHRs. Though the goal of CDS is to promote safe drug prescribing, allergy alerts must be designed and used in a way that supports this effort. Ultimately, accurate allergy documentation is essential for DHR prevention.

INTRODUCTION

The severe morbidity and mortality caused by drug hypersensitivity reactions (DHRs) make them a serious public health concern.¹ For example, penicillin allergies are estimated to be responsible for approximately 75% of fatal anaphylactic cases in the USA, which cause 500-1000 deaths per year.² While there has been increasing recognition of the importance of DHRs, their true incidence remains largely unknown.

With the rapid growth of information technology, the adoption and use of electronic health records (EHR) have greatly increased over the past several decades. The 2009 federal stimulus plan’s Meaningful Use initiative significantly accelerated the EHR adoption process.³ EHRs have since transformed the field of DHR research, significantly improving our ability to study the epidemiology, prevention, case identification, and management of DHRs. With accumulated longitudinal data, EHRs have made it possible to identify large patient cohorts for clinical epidemiological and pharmacogenetic studies and efficiently identify risk factors of DHRs.⁴ The ability to identify large study cohorts would not have been possible without the centralized EHR and advanced computing techniques. Furthermore, the use of EHRs has paved the way for clinical decision support systems (CDSS), which have significantly enhanced safe drug prescribing practices and antimicrobial stewardship. Nevertheless, challenges still exist in using EHRs for clinical practice and research. We continue to see challenges and opportunities for improvements in drug allergy documentation, drug allergy delabeling, CDSS (e.g., the counterproductive nature of overly frequent allergy alerts that contribute to clinician fatigue), and the epidemiology of DHRs.

Although prior efforts have been made to review the use of EHRs in specific areas of DHR research, there are no systematic reviews focused on the general utilization of EHR in this field. Our goal is to provide a comprehensive systematic review of the use of EHRs over the past two decades in DHR research across a broad spectrum of research topics.

MATERIALS AND METHODS

Data Sources and Searches

This review was conducted in compliance with the 2009 Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) statement.⁵ We conducted systematic database searches to retrieve relevant articles from several databases, including PubMed, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science, ScienceDirect, and Ovid. Bias could not be evaluated according to the PRISMA guidelines due to the heterogeneity of study types included in the review. Through an iterative process, we refined our search queries and included our final search queries and articles retrieved in eTable 1 in the Supplement.

Inclusion and Exclusion Criteria

Inclusion criteria for this review required that all articles are (1) published between January 1, 2000 to June 30, 2021; (2) written in the English language; (3) included title, author(s), and abstract; and (4) mentioned EHR use and DHR in the abstract. Reviews and conference abstracts were excluded.

Study Selection:

Search results from each query were exported, and duplicates were eliminated. The retrieved studies were screened by three independent reviewers (FB, SV, and YC) to assess whether they met the inclusion criteria. Via an iterative process, the reviewers identified major research topics across included studies. Then, two reviewers used the abstract to classify each study into one of the research topics. Any disagreements were adjudicated by a third reviewer (LW) via discussion until consensus was reached. Full-text reviews were conducted to extract relevant information from articles that were included in the review, including research topic(s), DHR discussed, and EHR database used if applicable.

RESULTS

We identified a total of 433 articles through our database search and included 140 articles in this systematic review (Figure 1). Six main research topics were identified: (1) Epidemiological analysis of drug allergies (Observational studies), (2) documentation of drug hypersensitivity (clinical documentation), (3) case management, (4) clinical decision support, (5) case identification, and (6) genetic studies. Table 1 shows the definition and distribution of the articles to the six research topics.

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2009 flow diagram. “Out of scope” were articles which were off topic, not written in English, or did not have an abstract.

Table 1.

Overview of research topics for included articles (n=140).

Research Topic	Description of Relevant Articles	n (%)
Observational Studies	Epidemiological analysis of drug hypersensitivity, including incidence, prevalence, and risk factors of hypersensitivity reactions and drug-induced reactions using EHRs Excluded: articles primarily focusing on clinical treatment comparison	61 (43.6%)
Clinical Documentation	Documentation of drug hypersensitivity, such as evaluation of how well drug allergies were recorded or labelled in EHR	27 (19.3%)
Case Management	Review of allergy challenge testing and therefore delabeling in EHRs	22 (15.7%)
Clinical Decision Support	Overview of CDSS, including allergy alerts and clinician prescription order entry systems Excluded: If study did not focus on clinical utility or drug hypersensitivity	18 (12.9%)
Case Identification	Studies used EHR data to identify diverse patient cohorts or clinical cases with drug hypersensitivity	9 (6.4%)
Genetic Studies	Analysis of the relationship between genetic variation and drug hypersensitivity reactions	3 (2.1%]

Research Trends Over Time

Most of the articles (n=129, 92.1%) were published in or after 2012 (Figure 2), with the number of articles peaking in 2019 (n=25, 17.9%). Articles discussing clinical documentation and observational studies were published throughout the review period. Dissimilarly, articles studying case management and genetic associations were published later in the study time window, starting from 2016 and 2019, respectively. The largest category of studies was observational studies (61 [43.6%]) followed by clinical documentation (27 [19.3%]), case management (22 [15.7%]) clinical decision support (CDS) (18 [12.9%]), case identification (9 [6.4%]) and genetic studies (3 [2.1%]).

Figure 2. The distribution of eligible articles by year (n=140).

*2021 only includes articles published between January 1, 2021 and June 30^st, 2021.

RESEARCH TOPICS

Observational studies

Observational studies, the largest category, included 61 (43.6%) epidemiological studies of DHRs (Table 2). They were primarily designed as retrospective cohort analyses, using EHR data to identify patient cohorts, report the prevalence of adverse drug reactions (ADRs), and elucidate clinical associations with medication use. The main subcategory was dedicated to the discussion of the antibiotic drug class (n=15, 24.6%), examining the prevalence of ADRs to B-lactams and secondary antibiotics.^6-13 Notably, the literature consistently reported that a significant proportion of ADRs (15.3%¹⁴-53.5%¹⁵) in hospitalized patients were attributed to antibiotics, specifically penicillin. The preponderance of research on penicillin allergies indicates the need for an accurate account of penicillin-related reactions, and improved care in the face of growing antimicrobial resistance.⁹ Seven studies broadly examined the general incidence of DHRs in hospital systems, reporting rates as high as 35.5%¹⁶; however, the incidence rates among studies are not easily comparable as the study population and inclusion criteria varied. Common risk factors are sex, age, and having at least one documented allergy.^6,10,17 Females and older-aged patients in nearly all study populations exhibited higher rates of reported drug allergies.^{12,14,15,18-20}

Table 2.

Summary of articles categorized as observational, organized by specific drug classes, general or specific adverse drug reactions, or both

Article	EHR Database/Data Source	Sample Size	Primary EHR Components*	Findings: Prevalence/Incidence, Risk Factors, Outcome, Other
DRUG: Antibiotics (n=15)
Apter et al. (2004) 21	UK General Practice Research Database	3.4 million	D	57 out of 3,014 (0.15%) patients who had an allergic-like event after the first prescription experienced another event to the second prescription.
Albin et al. (2014) 7	Internal Medicine Associates Clinic of Mount Sinai Hospital/ Epic EHR	1,348	A, D	Documented reactions to penicillin allergy are rash (37%), hives (18.9%), and swelling (11.8%) and was most prevalent in African Americans, then Caucasians and Asians. Risk factors for penicillin allergy: sex (female)
May et al. (2016) 22	Mayo Clinic	927	A, D	IV penicillin did not increase risk of allergy in children (OR: 0.84).
Crotty et al. (2017) 10	NSLIJ Huntington Hospital, Huntington, NY	175	A, D	89% of 175 patients who received at least one dose of cefepime, ceftriaxone, cefoxitin, cephalexin, or meropenem had self-reported allergy to penicillin. 20% reported incidence of rash while 63% said unknown reaction. 8 patients had an allergic reaction to penicillin while 2 patients had an adverse drug reaction (ADR) to amoxicillin and piperacillin/tazobactam. Risk factor: history of self-reported allergy can increase risk of cross-sensitivity reactions, type of drug (cephalosporin)
West et al. (2019) 12	ResearchOne, UK	2.3 million	A, D	The prevalence of penicillin allergy was 5.9%. Risk factors: age (older), sex (female), comorbidities
Liang et al. (2020) 13	Kaiser Permanente Southern California	6.1 million	A, B	More patients who received parenteral penicillin reported new allergic reactions (0.84%) than those who received oral penicillin (0.74%). 0.097% and 0.065% of parenteral and oral exposures respectively were confirmed anaphylaxis cases.
Lager et al. (2009) 6	University of Michigan Health System	211	A, D	Incidence of allergic-type reaction to carbapenem was 11% in patients with reported penicillin allergy, 5.2x greater than those who did not report a penicillin allergy. Risk factor: documented Penicillin allergy
Beltran et al. (2015) 8	Nationwide Children's Hospital Enterprise Data Warehouse (Epic EHR)	513	A, D	Cephalosporin resulted in one documented case of nonanaphylactic reaction when used as surgical prophylaxis. Clindamycin, most common cephalosporin, produced an adverse rate of 1.5% in patients with Penicillin allergy.
Macy et al. (2015) 23	Kaiser Permanente Southern California (Health Connect)	1.0 million	A, B, D	There were higher reports of allergy to cephalosporin among women (0.56%) than among men (0.43%). Anaphylaxis occurred in 5 oral exposures and 8 parenteral exposures. Clostridium difficile infection within 90 days (0.91%), nephropathy (0.15%), and all-cause death within 1 day (0.10%) were most common, serious ADR. No correlation with drug allergy history.
Blumenthal et al. (2016) 9	Partners Healthcare System-Mass General Brigham (Epic EHR)	96	A, D	ADR was observed in 21% of inpatient patients who received ceftaroline. No increased risk of ADR for patients with reported B-lactam allergy.
Alvarez-Arango et al. (2021) 24	Johns Hopkins Health System Corporation (JHHS) and Mass General Brigham (MGB) (Epic EHR)	4.5 million	A, D	0.3% of patients had documented vancomycin allergy with 42.1% reporting immediate phenotypes and 20.7% delayed reactions. 32% hypersensitivity reactions presented as rash and 16% as Red Man Syndrome.
Butler et al. (2018) 11	Seattle Children’s Hospital, University of Washington School of Medicine	17,496	B, D	1% of penicillin allergic patients who received cefazolin experienced perioperative adverse drug reactions, Vancomycin associated with greater rates of ADR as prophylaxis.
Fosnot et al. (2021) 25	UCHealth System-Health Data Compass Data Warehouse project	690	A, D	Probable DHRs occurred in 0.9% patients who received cefazolin, 1.4% patients who received clindamycin, and 1.1% patients who received vancomycin, not varying significantly.
Macy et al. (2009) 14	Kaiser Permanente	1,127	A	15.3% of drug allergy reports reported at least one allergy to one antibiotic class. Risk factors: female, older age, drug type (highest incidence w/ sulfa class)
Salden et al. (2015) 17	Academic Health Care Centre Terwijde at Leidsche Rijn Utrecht	8,288	A	2.0% of sample had recorded allergy to B-lactams in Dutch primary care system. Risk factors: age (very young), sex (female), comorbidities (asthma, allergies, skin disorders)
DRUG: Cancer therapy-related drugs (n=7)
Lal et al. (2009) 26	University of Texas M. D.- Anderson Cancer Center	3,746	D	Weekly paclitaxel infusions result in 1.5% rate of DHRs.
Kim et al. (2012) 27	Seoul National University Bundang Hospital	393	D	The prevalence rate of DHRs to oxaliplatin is 10.7%. Risk factors: higher dosage of oxaliplatin and lower dosage of dexamethasone
Jung et al. (2014) 28	Seoul National University Hospital	658	B, D	49.5% of patients who received rituximab experienced infusion related reactions. Risk factors: certain types of lymphoma (CLL, intravascular B cell lymphoma), high dosage and rate of injection in 1st 30 min
Levin et al. (2017) 29	Partners Healthcare System- Mass General Brigham	67	A, D	51% of patients who experienced grade 1 reaction to rituximab can be safely rechallenged.
Welborn et al. (2018) 30	University of Texas MD Anderson Cancer Center	17	D	70.6% of patients with cutaneous ADE to tremelimumab experienced pruritis. Other reactions include eczematous dermatitis, morbilliform rash, vitiligo, xerosis, acneiform rash, and psoriasiform dermatitis.
Shazib et al. (2020) 31	Dana-Farber/Brigham and Women's Cancer Center	13	D	Of 13 patients, 4 had oral-only immune-related ADEs to programmed cell death-1 inhibitors. 10 had lichenoid lesions, 2 with erythema multiforme, 1 with graft vs host disease reactivation, and 8 with or without ulcerations.
Keiser et al. (2021) 32	University of Texas MD Anderson Cancer Center	64	B, D	91% of patients with cutaneous adverse events to immune checkpoint inhibitors treated with topical steroids, oral antihistamines, or topical antihistamines and 70% recovered from rash over 4 months.
DRUG: Non-steroidal anti-inflammatory drugs (NSAIDs) (n=2)
Blumenthal et al. (2017) 33	Partners HealthCare System (PHS) - Partners Enterprise Allergy Repository (PEAR)	62,719	A, B, C, D	1.7% had an ADR to prescription NSAIDs, 18.3% of which were hypersensitivity reactions. Risk factors: drug hypersensitivity reaction history, sex (female), autoimmune disease, and those who were prescribed the maximum standing NSAIDs dose
Li et al. (2021) 34	Partners HealthCare System	47,114	A, B, D	7.7% of patients with chronic back pain had active aspirin or NSAID adverse reaction.
DRUG: Opioids (n=1)
Inglis et al. (2021) 35	Royal Adelaide Hospital, Australia (Sunrise)	231,623	A, C, D	15.9% of ADR reports were due to opioids, with 64.7% reported as allergy and 35.3% as intolerance.
DRUG: Radiocontrast media (RCM)-related compounds (n=4)
Dillman et al. (2007) 36	University of Michigan Health System	78,353	D	The reaction frequency to IV gadolinium is 007%. 74% of reactions were mild, 19% were moderate and 7% were severe. 50% had at least one risk factor to IV gadolinium reaction. Risk factors: allergic-like contrast reaction, a prior allergic reaction to a substance other than contrast media, or documented asthma
Power et al. (2016) 37	University Health Network, Toronto	19,074	D	The reaction rate to gadobutrol is 0.32% and the per patient reaction basis was 0.43%. Risk factors: previous allergic-like reaction to gadolinium, previous reaction to any substance, history of asthma
Young et al. (2019) 38	NHS Tayside, Health Informatics Centre, University of Dundee	22,897	D	0.01% of patients had DHRs to gadolinium-based contrast agents.
Lakshmanadoss et al. (2012) 39	Johns Hopkins Health System	234	D	71% of patients had previous recorded allergy to iodinated contrast agents, 24% to iodine, and 5% to both. Most patients (77%) had skin rashes or unspecified reaction while 8.5% had anaphylaxis.
DRUG: Statin (n=1)
Robison et al. (2014) 40	Intermountain Healthcare system (Murray, UT)	10,789	B, D	Patients with statin intolerance had a higher history of hypothyroidism (30.2%) compared to the control group (21.5%).
DRUG: Other (n=3)
Paisansinsup et al. (2013) 41	Park Nicollet Health Services EHR	1,268	D	3.79% of patients prescribed allopurinol experienced an ADR. Risk factors: sex (female), older age, diabetes mellitus, diuretic use, presence of tophi associated with possible ADR
Hall et al. (2018) 42	The Health Improvement Network, UK	70	C, D	Observed-to-expected ratio was 3.3 and 1.5 for convulsions and thrombocytopenia for those who received Optaflu, a trivalent seasonal influenza vaccine.
Laird et al. (2020) 43	UCHealth System	868	B, D	0.461% of patients who received Fosaprepitant as a prophylaxis for chemotherapy related nausea had a systemic hypersensitivity reaction.
Adverse Drug Reaction (ADR): Broad/general overview (n=6)
Kidon and See (2004) 44	KK Women’s and Children’s Hospital, Singapore	672	D	2.2% of pediatric patients had recorded ADR, 70% of which were due to antibiotics, specifically B-lactams. 18.5% were due to NSAIDs in Singapore. Risk factors: older age, male gender, presence of asthma or other chronic disease
Macy et al. (2014) 15	Kaiser Permanente Southern California, Health Connect	105,61 4	B, D	The most common allergies in hospitalized patients were due to penicillin (16.7%), other nonantibiotics (12.7%), narcotics (11.7%), sulfonamides antibiotics (10.2%) and NSAIDs (7.1%), which altogether accounted for 58.4% of reported allergies. Risk factors: sex (female), age (older)
Saager et al. (2015) 45	Cleveland Clinic Perioperative Health Documentation System	264	D	Overall incidence of intraoperative hypersensitivity reaction was 0.148%. 2 out of 10,000 operations resulted in severe hypersensitivity reactions.
Zhou et al. (2016) 16	Partners HealthCare System/ Partners’ Enterprise-wide Allergy Repository (PEAR)	1.8 million	A, D	35.5% of patients reported at least one drug allergy with 1.95 allergies/patient. Risk factors: female sex, Caucasian race, antibiotic, statin, and ACE inhibitor use
Mendes et al. (2019) 46	Portuguese Catalogue of Allergies and Other Adverse Reactions (CPARA)	380	D	0.4% inpatients (380 patients) had DHRs over 5-year period, 52.8% of which were associated with antibiotics, mainly beta lactam antibacterial use. 47.6% ADR were skin and subcutaneous tissue disorders and 41.3% were immune system disorders, specifically anaphylactic Reactions (37.4%).
Wong et al. (2019) 20	Partners HealthCare System/ Partners’ Enterprise-wide Allergy Repository (PEAR)	2.7 million	A, C, D	13.8% of patient population had documented DHRs. 53.1% were associated with immediate reaction phenotypes. Risk factors: female sex, race/ethnicity, drug type
ADR: Anaphylaxis (n=3)
Goh et al. (2018) 47	National University Hospital, KK Women’s and Children’s Hospital, and Tan Tock Seng Hospital, Singapore	426	B	45% of anaphylaxis cases were children in Singapore. Risk factors: food type, drug type
Dhopeshwarkar et al. (2019) 48	Partners HealthCare System/ Partners’ Enterprise-wide Allergy Repository (PEAR)	1.7 million	A, D	1.1% of patient population had at least one drug-induced anaphylaxis reaction. Risk factor: female sex, white race, drug type
Rangkakulnuwat et al. (2020) 49	Chiang Mai University (CMU) Hospital, Chiang Mai, Thailand	433	B	Overall incidence of anaphylaxis in Asia was 3.9 cases out of 100,000 visits, 84% of which were experienced in adults. Drug-induced anaphylaxis was more common in adults than children (19.8% vs. 8.1%). NSAIDs (7.4%) and antimicrobials were most common culprit drugs. 11.4% of cases had unknown cause. Risk factors: food, drug exposure, age, sex
ADR: severe cutaneous adverse reactions (SCAR) (n=6)
Ou-Yang et al. (2013) 50	Taiwan National Health Insurance Research Database	554	D	15.5% of outpatient and hospitalized patients in Taiwan were hospitalized twice due to SJS/TEN. Penicillin and cephalosporin (27%, 27%) were the main culprit agents for 1st hospitalization.
Micheletti et al. (2018) 51	18 US academic medical centers	377	D	89.7% of SJS/TEN cases were due to medication, mainly trimethoprim/sulfamethoxazole (26.3%) and b- lactams (12.4%).
Park et al. (2019) 52	Multicenter, Gangnam Severance Hospital, South Korea	745	D	Allopurinol was the causative drug in 14.2% of SCAR cases. Other culprit drugs are anticonvulsants (22.5%), b-lactams (21.1%) and NSAIDs (10.3%).
De Bustros et al. (2021) 53	Loyola University Medical Center	163	D	Anticonvulsants (30%), Trimethoprim-Sulfamethoxazole (19%), Beta-lactams (11%), NSAIDs (8.4%) and allopurinol (8.4%) were identified the most probable culprit in SJS/TEN cases.
Zhang et al. (2019) 54	Penn State Hershey Medical Center	35	B, D	35 patients (5.9% of larger patient cohort) had documented SCAR, including 54.3% of DRESS, 22.8% of SJS, 17.1% of AGEP, 2.9% of TEN and SJS/TEN overlap.
Ma et al. (2021) 55	Chang Gung Memorial Hospital Linkou Branch, Taiwan	119	D	Of patients with SJS/TEN and overlap syndrome, 46.2% had severe ocular complications.
ADR: Other (n=7)
Macy et al. (2012) 18	Kaiser Permanente Southern California (Health Connect)	2.4 million	A	2.1% of health plan members had 3 or more allergies reported and can be diagnosed with multiple drug intolerance syndrome. Risk factors: female sex, age, drug type, anxiety associated
Banerji et al. (2017) 56	Partners HealthCare System	135,000	A, C, D	Incidence of ACE inhibitor angioedema is 0.07% and 0.23% within the 1st month and 1st year of use, respectively.
Read et al. (2017) 57	Royal Brisbane and Women's Hospital, Gold Coast University Hospital, Australia	70	D	Only 9 cases of 70 reported erythema multiforme diagnoses in children met criteria for erythema multiforme, with most being misdiagnosed.
Blumenthal et al. (2018) 58	Partners HealthCare System/ Partners’ Enterprise-wide Allergy Repository (PEAR)	746,88 8	A, B, C, D	The overall prevalence rate was multiple drug intolerance syndrome was 6.4% and 1.2% for multiple drug allergy syndrome. Risk factors for MDIS: female sex, older age, greater weight, prior hospitalizations, and multiple medical comorbidities
Braswell et al. (2019) 59	University of Florida, the Medical College of Wisconsin, and Inform Diagnostics Research Institute	56	D	Of patients with lichenoid granulomatous dermatitis, most were diagnosed with drug eruption (39.3%, n = 22) and lichenoid keratosis (19.6%, n = 11).
Jimenez et al. (2019) 19	Cleveland Clinic	2.0 million	D	70.9% had no allergies; 27.4% had 1-4 allergies; 1.5% had 5-9 allergies; 0.22% had >9 allergies in the patient population. Rates of mental health and somatic syndrome disorder increased with more allergies. Risk factors: female sex
Leigh et al. (2019) 60	University of Pennsylvania Health Systems	1,218	A, B, D	Overall incidence of Eosinophilic esophagitis is 0.034% in patient population. There may be a correlation with smoking.
BOTH Drug/ADR (n=6)
Silverman et al. (2016) 61	University of Pennsylvania Health Systems	220	A	12.4% of patients without chronic urticaria had self-reported penicillin allergy. there was a 14.5% rate and 4.6% of chronic urticaria in patients with and without self-reported penicillin allergy, respectively.
Lin et al. (2017) 62	Penn State College of Medicine/ Medical Center	138	D	In 78 pediatric reactions to vancomycin, 92% were consistent with Red Man Syndrome. Of 60 children prescribed linezolid, 82% were unnecessarily avoiding vancomycin without prior reaction to vancomycin.
Coleman et al. (2019) 63	Yale-New Haven Hospital	98	D	Most rashes were associated with immune checkpoint inhibitors: pembrolizumab (35/103 rashes), nivolumab (33/103 rashes), and ipilimumab/nivolumab (17/103 rashes).
Jung et al. (2019) 64	Yonsei University Wonju College of Medicine, South Korea	1,253	D	The prevalence of DRESS cases among patients prescribed antituberculosis drugs is 1.2%. Ethambol (53.5%) and rifampin (26.7%) are the most common culprit drugs.
Fukasawa et al. (2021) 65	JMDC Claims Database, Japan	355	B, D	In Japan the odds ratio of SJS/TEN for anti-convulsant, including carbamazepine (OR 68.00) and lamotrigine (OR 36.00) were significantly increased.
Sim et al. (2021) 66	Chonnam National University Hospital, South Korea	27	D	48% of 27 patients with drug fever or maculopapular exanthem had Multiple Drug Hypersensitivity Syndrome. Most common culprit agent was ethambutol and rifampin, followed by pyrazinamide and isoniazid.

^*(A=Allergy List/Problem List, B=ICD Codes, C=free-text or structured data in the allergy list, D=Other/Unspecified)

Abbreviations: NSAIDs, Non-steroidal anti-inflammatory drugs; RCM, Radiocontrast media; ADR, adverse drug reaction; SCAR, severe cutaneous adverse reactions; DRESS, drug reaction with eosinophilia and systemic symptoms

Observational articles applied different methods to extract information from select EHR sections. Nearly all antibiotic-focused studies^{7-11, 13-14, 31, 33, 38-41} among other studies^{18, 26-28, 30, 34, 36, 49, 57, 59, 61-62} (n=25, 41.0%) explicitly referenced documented allergies and the problem list in EHRs. Several studies (n=15, 24.6%) referenced using ICD-9 or ICD-10 codes for the purposes of identifying study cohorts or measuring the prevalence of specific comorbidities.^{12, 14, 17, 26-27, 29, 32, 39, 44, 48, 55, 59, 61, 66} Six studies (9.8%) also used free-text comments and structured allergy entries with coded reactions to extract information.^{26, 28, 36, 43, 57, 59}

Clinical Documentation

Clinical documentation, a primary subject of this review, included a total of 27 (19.3%) articles (Table 3). Seventeen studies assessed the accuracy of documentation and found that drug hypersensitivity records were incomplete and inconsistent with incorrectly labeled or categorized DHRs.^67-83 The documentation inaccuracies were mostly identified by retrospective chart reviews; however, five articles assessed documentation accuracy through patient interviews^72-74 and qualitative interviews with clinical staff.⁷⁶,⁷⁷ There was significant heterogeneity in the delegation of drug hypersensitivity documentation among intake staff, pharmacists, physicians, registered nurses, nurse practitioners, physician assistants, and medical record technicians.

Table 3.

Summary of articles related to clinical documentation

Drugs of interest assessed	Study design	Articles	Summary of findings
Assessing documentation accuracy (n=17)
General	Retrospective review	Hsu et al. (2011)67 Goldblatt et al. (2017)68 Blumenthal et al. (2017)71 Foremen et al. (2020)69 Rukasin et al. (2020)70	Drug allergy histories on smart cards are incomplete in many cases and have inconsistent formats 67. Only 27% of SJS/TEN patients had all implicated drugs noted in the outpatient (primary care) record 68. Allergies tend to accumulate over time with comparatively few allergy deletions 71. Only 45.1% (n = 1671/3705) of reactions consistent with intolerance (e.g., “nausea,” “diarrhea”) were correctly categorized as such 69. EHR transitions pose a significant risk for EHR-related errors which can be compounded by human error. 70
	Cross-sectional	Reinhart et al. (2008)75 Lyons et al. (2015)73 Kiechle et al. (2018)72 Kabakov et al. (2019)74	Allergy information was successfully entered in 84.6 % of hospital admissions with a significantly lower rate (37.5%) among whose ethnicity groups, on average, have lower rates of English fluency 75. Three studies 72-74 reported that discrepancies between medication allergies recorded in EHRs and those elicited in interviews are common.
	Qualitative interview	Fernando et al. (2014)77 De Clercq et al. (2020)76	Most drug reactions are likely to go unreported to and/or unrecognized by healthcare professionals or are inaccurately recorded. 77 Family physicians and pharmacists perceive that few documented antibiotic allergies are in fact correct 76
Beta-lactam	Retrospective review	Moskow et al. (2016)78	Among all patients with a documented beta-lactam allergy, 36.2% had an empty or missing allergy reaction description in their EHR. 78
Penicillin	Retrospective review	Rimawi et al. (2013)79 Inglis et al. (2017)80	36% of the 55 patients with proven penicillin tolerance who revisited the hospital within a year had penicillin allergy redocumented. 79 Penicillin adverse drug reaction categorization was inconsistent. 10.1% of reports entered as allergy had reaction descriptions that were consistent with intolerance and 31.0% of the entered intolerances had descriptions consistent with allergy. 80
	Prospective/ interventional	Staicu et al. (2017)81	Severe or life-threatening penicillin allergies were underreported in nearly half of patients (43%). 81
Contrast agents		Deng et al. (2019)82 Ananthakrishnan et al. (2021)83	The majority of the 40,669 contrast allergen records were low quality (69.1%) rather than intermediate (19.4%) or high quality (11.5%). 82 Iodinated contrast media premedication prompts in the EHR are often erroneous because of inaccurate coding, incomplete data, and reaction misclassification. 83
Improving drug allergy documentation (n=10)
General	Prospective/interventional	Young et al. (2011)87 Burrell et al. (2013)84 Lesselroth et al. (2015)85 Masaharu et al. (2018)86 Goss et al. (2018)93 Soyer et al. (2019)89 Wang et al. (2020)90	Using an electronic version of a drug calendar considerably increased the ease and efficiency of completing dermatology consultations. 87 A pharmacy-driven initiative intended to improve the completeness of drug allergy/intolerance documentation was associated with modest success. 84 A patient-facing medication reconciliation and allergy review kiosk. 85 A novel standard format for recording allergy information resulted in increased allergy documentation and decreased adverse drug events. 86 A comprehensive value set to improve the consistency and accuracy of adverse reaction documentation in the allergy module was developed including 1106 concepts. 93 Structured intervention led to an increase in quality of coding and reduction in discrepancies coded by medical record technicians and pharmacovigilance teams. 89 A dynamic reaction picklist developed using EHR data and a statistical measure was superior to the static picklist and suggested proper reactions for allergy documentation. 90
	Retrospective review	Vethody et al. (2021)91	Patients with multiple drug allergy labels can be safely delabeled to multiple drugs in 1 visit. 91
Beta-lactam	Prospective/interventional	Wright et al. (2019)88	Allergy documentation of antibiotic test dose results increased with use of CDS. The addition of electronic alerting increased allergy documentation to 66.7% from 51.3% in the pre alert period. In addition to a greater likelihood of updating, updates were made significantly faster in the post alert period. 88
Penicillin	Retrospective review	Lachover-Roth et al. (2019)92	Penicillin allergy annulling via oral challenge test proved to be safe and effective. 92

Abbreviations: SJS, Stevens Johnson Syndrome; TEN, Toxic Epidermal Necrolysis

Five studies assessed approaches for improving drug allergy documentation, in particular, documentation completeness.^84-88 This was done by a pharmacist driven protocol,⁸⁴ patient-facing medication reconciliation,⁸⁵ a novel standard format for recording drug allergies,⁸⁶ the use of CDS alerts based on antibiotic test dose results⁸⁸, and an electronic version of a drug calendar.⁸⁷ Two articles discussed improving the quality of drug allergy documentation,^89,90 and three articles focused on allergy testing and delabeling inappropriate allergies.^88,91,92 Quality improvement of drug allergy documentation was done through designing a dynamic reaction pick list,⁹⁰ a comprehensive value set for encoding reactions,⁹³ and a structured intervention for medical record technicians and pharmacovigilance teams.⁸⁹ Two articles addressed the safety of delabeling,^91,92 and another study focused on the likelihood and speed of updating allergies using CDS alerts based on antibiotic test dose results.⁸⁸

Case Management

There were 22 (15.7%) articles discussing the management of DHRs using the EHR to identify and stratify patients who would benefit from the evaluation of drug allergies and to increase the efficiency and safety of desensitization. Predominantly, research focused on antimicrobial stewardship including fifteen studies on delabeling inappropriate antibiotic allergy labels and optimizing treatment,^94-110 one evaluating the safety and efficiency of antibiotic desensitization,¹¹¹ and one assessing antibiotic use in patients with antibiotic allergy documentation.¹¹² Delabeling and treatment optimization efforts were accomplished through retrospective review of antibiotic evaluation history,⁹⁹ a computerized guideline application with decision support,¹⁰⁶ patient interview,¹⁰⁷ and the stratification of patients in the EHR with antibiotic allergy labels as candidates for penicillin skin testing (PST),^94,95,108 oral challenges, ^98,102 or both.^{96,97,103,110} Other studies focused on the management of DHRs induced by cancer treatments including one study which retrospectively assessed chemotherapy drug rechallenges¹¹³ and one which stratified patients with taxane-related DHRs for skin testing.¹¹⁴ Finally, one study retrospectively reviewed patients undergoing desensitization and found that it was safe for patients with no alternatives for therapy but noted that urticaria and labored breathing were risk factors for having a reaction during desensitization.¹¹⁵

CDS

Eighteen studies (12.9%) reviewed the use of automated CDS in EHRs, including drug allergy alerting mechanisms and override rates (Table 5). Eleven of these studies measured the prevalence of allergy alerts, specifically, drug allergy and drug-drug interaction alerts, the most common types of CDS alerts.¹¹⁶ Medication-related alerts made up nearly three-quarters of all inpatient alert¹¹⁶ with most drug-allergy alerts triggered by narcotics.^{117, 118} These studies also evaluated how frequently clinicians overrode drug alerts, reporting override rates as high as 93%.¹¹⁹ Of these overrides, approximately 80% of these allergy alert overrides were deemed appropriate. Even so, Wong et al., 2018 determined that inappropriately overridden allergy alerts are six times more likely to trigger ADRs compared to those that are appropriately overridden.¹²⁰

Table 5.

Summary of articles related to clin ical decision support systems (n=18).

Focus/Intervention	Allergens	Alert Type	Articles	Clinical Setting	Summary of Findings
Descriptive Study Design (n=11)
Measure of allergy alerts and overrides	Food, drug, environment	Food/drug allergy and intolerance alerts	González-Gregori et al. 2012116	Inpatient	Alerts were mainly caused by drugs (74.4%), followed by foods (12.6%) and materials (4.8%).
	Drug	Drug allergy alerts, drug-drug interaction alerts	Lin et al. (2008)127	Inpatient	Clinicians indicated alerts were overly frequent, with low specificity and high sensitivity. 93% drug alerts were overridden. More drug-drug alerts were overridden (87- 95.1%) compared to drug allergy alerts (81-90.9%).
			Weingart et al. (2009)128	Inpatient; Outpatient
			Carspecken et al. (2013)129	Inpatient; Pediatric hospital
			Bryant et al. (2014)119	Inpatient
			Topaz et al. (2016)117	Inpatient	Alerts containing immune mediated (72.8%) and life-threatening reactions (74.1%) were overridden. Narcotics triggered most drug alerts (48%).
			Wong, A; Seger, DL; Slight, SP, et al. (2018)120	Inpatient, Outpatient	46.0% and 68.8% of definite anaphylaxis drug allergy interaction alerts were overridden in inpatient and outpatient settings respectively. 83.9% of inpatient overrides and 100% of outpatient overrides were appropriate.
		drug-allergy, drug-drug interaction, geriatric and renal alerts	Wong et al. (2017)130	Inpatient; Intensive Care Unit	Between commercial and internally developed EHR, physicians experienced more alerts and overrode more alerts with the commercial EHR. (commercial: n=5,535; legacy: n=1,030).
			Wong et al. (2018)131	Inpatient; Intensive Care Unit	81.6% of alert overrides were appropriate in the intensive care unit. However inappropriate overrides were 6 times more likely to result in an ADE compared to appropriate overridden alerts.
		Opioid allergy alerts	Ariosto, D. (2014)118	Inpatient	At least 89% of opioid allergy alerts that make up almost a third of visible alerts were overridden, Physicians are more likely to override opioid alerts than advanced practice nurses.
			Genco et al. (2016)132	Inpatient; Emergency department	34.6% of visible alerts are opioid alerts. Of these alerts, 96.3% were overridden.
Interventional Study Design (n=7)
Create Clinical Decision Support	Radiocontrast Media Agents	Premedication Alerts	Bae et al. (2013)122	Inpatient	There was a significant increase in premedication rates; however, only Bae et al. noticed a significant reduction in breakthrough reactions.
			Benson et al. (2017)123
	Drug	Drug-gene interaction alerts	Dolin et al. (2018)124	NA	With the use clinical decision support and pharmacogenetic sequencing data, genomics-EHR integration can lead to drug-gene interaction alerts.
	NA	NA	Garabedian et al. (2019)133	Outpatient	Redesigning CPOE structure to allow physicians to enter the indication, or reason for medication first, before prescribing will improve usability and user satisfaction while minimizing medication error.
Modified Alerting Rules	B-lactam antibiotics	B-lactam alerts	Macy et al. (2021)125	Inpatient, Outpatient	Elimination of cephalosporin alerts increased cephalosporin use, decreased 2nd line of antibiotic treatment without significantly increasing anaphylaxis
			Buffone et al. (2021)126	Inpatient	7.7% patients were alerted for a B-lactam prescription while 92.3% patients were no longer alerted under the adjusted rules when prescribed a B-lactam antibiotic with a different side chain. They did not report any incidence of anaphylaxis.
Evaluation Tool for Clinical Decision Support		Therapeutic Duplication, Drug-Dose (single and daily), Drug-Allergy, Drug-Route, Drug-Drug, Drug-Diagnosis, Drug-Age, Drug-Labs, Drug-Renal, Monitoring, Nuisance Orders	Cho et al. (2015)121	NA	Using the Leapfrog CPOE evaluation tool, errors were captured in Therapeutic Duplication and Drug-Drug Interaction alerts, mainly.

Abbreviations: CPOE, computerized provider order entry

Seven studies explored tools that can be used to evaluate¹²¹ and modify CDS by adding or eliminating certain alerts.^122-126 Two studies evaluated clinician behavior after reducing allergy alerts for beta-lactam antibiotics, a commonly documented patient allergy. ^125,126 Interestingly, both studies noted that the reduction¹²⁶ and even the elimination of allergy alerts¹²⁵ did not lead to a significant increase in DHRs. Three other studies created CDS alerts, including premedication alerts for when patients were prescribed radiocontrast media^122,123 and drug-gene interaction alerts to prevent future DHRs,¹²⁴ thereby promoting tailored alerts for specific drug interactions. Altogether, these studies intended to improve CDSS while reducing ADRs and medication errors in clinical settings.

Case Identification

Among 9 (6.4%) articles allocated to the case identification category, three focused on identifying DHRs using unstructured data,^134-136 two using structured data,^137,138 and four using a combination of both (Table 6).^139-142 Studies using unstructured data utilized text processing techniques like natural language processing (NLP) and free-text searches. Wolfson et al. used free-text searches of drug reaction with eosinophilia and systemic symptoms (DRESS) syndrome-related keywords to identify a large DRESS syndrome cohort (n= 69) from a database consisting of 3.1 million patients.¹³⁴ Similarly, Epstein et al. developed an algorithm that uses RxNorm¹⁴³ and NLP to identify drug allergies with an accuracy, precision, recall, and F-measure of above 97%.¹³⁶ Case identification research using structured data used ICD-9 codes and E codes.^137,138 Davis et al. and Saff et al. used ICD-9 codes to identify inpatient ADRs and Steven Johnson Syndrome (SJS)/Toxic Epidermal Necrolysis (TEN) cases, respectively from large patient populations.^137,138

Table 6.

Summary of articles related to Case Identification

Article	Search method	Reaction analyzed	Sample size	Summary of findings
Unstructured data (n=3)
Epstein et al. (2013) 136	RxNorm and natural language processing (NLP)	Adverse drug events	N/A	A high performing algorithm was used to identify medication allergies with a specificity of 90.3% and 85% in the training and testing data respectively. Accuracy, precision, recall, and F-measure for medication allergy matches were all above 98% in the training dataset and above 97% in the testing dataset for all allergy entries
Wolfson et al. (2019) 134	Free text keyword search of allergy module	DRESS syndrome*	69	Of 538 hypersensitivity reactions identified, 69 patients (2.18 in 100,000 patients) had DRESS syndrome.
DeLozier et al. (2021) 135	Text processing system	SJS/TEN* and torsades de pointes	138	The automated recruitment system resulted in the capture of 138 true cases of drug induced rare events, improving recall from 43% to 93%
Structured data (n=2)
Davis et al. (2015) 137	ICD-9 codes	SJS/TEN	475-875	Patients with the ICD-9 codes introduced after 2008 were more likely to be confirmed as cases (OR 3.32; 95%CI 0.82, 13.47) than those identified in earlier years. Likelihood of case status increased with length of hospitalization. Applying the probability of case status to the 56 591 potential cases, we estimated 475-875 to be valid SJS/TEN cases.
Saff et al. (2019) 138	ICD-9 codes and E codes	Allergic drug reactions	409	Specific ICD-9 codes can identify patients with allergic drug reactions, with antibiotics accounting for almost half of true reactions. Most patients with codes 693.0, 995.1, 708, and 995.0 had allergic drug reactions, with 693.0 as the highest yield code. An aggregate of multiple specific codes consistently identifies a cohort of patients with confirmed allergic drug reactions.
Combination of structured and unstructured data (n=4)
Kim et al. (2012) 142	Procedure codes and International classification of nursing practice terms	contrast-media-induced hypersensitivity reactions	266	An EHR-based electronic search method was highly efficient and reduced the charts that needed to be reviewed by 96% (28/759)
Cahill et al. (2017) 140	ICD-9 codes and an informatics algorithm	Aspirin-exacerbated respiratory disease (AERD)	593	An informatics algorithm can successfully identify both known and previously undiagnosed cases of AERD with a high positive predictive value. Involvement of an allergist/immunologist significantly increases the likelihood of an AERD diagnosis.
Fukasawa et al. (2019) 141	ICD-10 codes and informatics algorithms using clinical course and medical encounters	SJS/TEN	N/A	One algorithm, consisting of a combination of clinical course for SJS/TEN, medical encounters for mucocutaneous lesions from SJS/TEN, and items to exclude paraneoplastic pemphigus, but not ICD-10 codes, showed a sensitivity of 76.9%, specificity of 99.0%, positive predictive value of 40.5%, negative predictive value of 99.8%, and diagnostic odd ratio of 330.00.
Banerji et al. (2020) 139	ICD-9 codes and NLP	Allergic drug reactions	335	Among the 335 confirmed positive cases, NLP identified 259 true cases, resulting in a recall/sensitivity of 77% (range: 26%-100%). Among the 390 negative cases, NLP achieved a specificity of 89% (range: 69%-100%).

Abbreviations: DRESS, Drug Reaction with Eosinophilia and Systemic Symptoms; SJS, Stevens Johnson Syndrome; TEN, Toxic Epidermal Necrolysis; AERD, Aspirin Exacerbated Respiratory Disease; NLP, Natural Language Processing

Most of the studies reviewed (n=4, 44.4%) used a combination of structured and unstructured data for case identification. Banerji et al. have used ICD-9 codes to identify DHRs followed by a rule-based NLP algorithm to search free-text clinical notes and discharge summaries.¹³⁹ They found that the use of ICD-9 codes alone resulted in a positive predictive value (PPV) of 46% (range: 18%-79%) compared to a PPV of 86% (range: 69% - 100%) with the combination of ICD-9 codes and NLP. Alternatively, Kim et al. developed an EHR-based surveillance system using signals from standardized search terms within the international classification of nursing terms, and order codes for procedures that used contrast media, antihistamine, and epinephrine with a sensitivity of 66.7%, a specificity of 99.6%, and negative predictive value of 99.7%.¹⁴²

Genetic Studies

The genetic studies included 3 (2.1%) articles focused on investigating the contribution of genetic variation to DHRs. Zheng et al. conducted genome-wide association studies (GWAS) for ADRs in fourteen common drug/drug groups for 81,739 patients.¹⁴⁴ They identified seven genetic loci significantly associated with ADRs, which were consistent with additional expression quantitative trait loci and phenome-wide association analyses. They applied an approach that included a combination of prospective identification of cases and identification of cases in the Vanderbilt University Medical Center (VUMC) BioVu repository that links the EHR to DNA. Konvinse et al. reported a strong association between the HLA-A*32:01 and vancomycin-induced DRESS syndrome in a population of predominantly European ancestry.¹⁴⁵ In their study, nineteen (82.6%) of 23 vancomycin-associated DRESS syndrome cases had HLA-A*32:01 compared to 0 (0%) of 46 vancomycin-tolerant comparators matched by sex, race, and age using their BioVU deidentified EHR database. Lastly, Krebs et al. uncovered an association between HLA-*55:01 and self-reported penicillin allergy by extracting EHR data from over 1.1 million individuals, which was also replicated in 23 and me and VUMC BioVu.¹⁴⁶

DISCUSSION

We examined the use of EHRs in drug hypersensitivity literature by systematically reviewing articles in this field from five scientific databases from 2000 to 2021. We found that relevant studies fall into six study categories: observational studies, allergy documentation, case management, CDS, case identification, and genetic studies. We discuss the major findings in each study category, identify gaps and challenges that the field is currently facing, and outline the future directions.

Observational Studies

Overall, EHR-dependent analyses promote the use of extensive clinical data to build an epidemiological profile for DHRs and culprit drugs. With access to millions of medical records, reviewing EHRs conveniently increases sample size as demonstrated by Macy et al.¹⁸, Wong et al.²⁰, and Alvarez-Arango et al.²⁴ Furthermore, the retrospective use of EHRs allows for rare conditions, such as DRESS syndrome, to be studied in greater detail which would otherwise be difficult for conditions that already have a low prevalence and incidence rate.⁵¹,⁵⁴ In a similar vein, EHRs are used to study periodic trends, measuring the prevalence of DHRs throughout many years.¹³,⁴⁶,⁵⁴ Nonetheless, the retrospective design is considered a limitation as it is difficult to establish a clear temporal association compared to prospective cohort studies.

Furthermore, it was not consistently communicated what sections of EHRs were analyzed or how they analyzed EHRs. This heterogeneity is a limitation that cannot be always identified. Of the studies that defined their search methods, several studies used ICD codes to identify patients with the ADR or comorbidity of interest while a few others developed algorithms^34,65 and searched key terms in EHRs.^22,54,59 The problem list and allergy module were frequently used to measure the prevalence of allergies. Only 10% of studies highlighted the use of automated processes, like NLP, to extract information from free-text entries.^{16,20,24,30,33} 13% of the studies used a combination of the problem list, ICD codes, and the allergy module. With the profound growth of EHR literature, observational EHR studies should clarify how medical information was extracted for the data extraction to be considered transparent, replicable, and reliable. Additionally, the fact that allergy information has been documented in different places and often unreliably in the EHR highlights that allergy documentation needs to be systematically improved with reconciliation mechanisms in place, which is also discussed below.

Clinical Documentation

There was a consensus among the articles reviewed that there is a high rate of incomplete and inaccurate drug hypersensitivity documentation. This is especially concerning in cases that involve severe morbidity and high mortality rates including SJS/TEN and DRESS cases. Goldblatt et al. found that in 26 cases of SJS/TEN, only 27% of patients had all implicated drug notes in the outpatient primary record.⁶⁸ Similarly, Rukasin et al. reported that 5% of 511 patients had inaccurate descriptions for confirmed severe reactions like SJS/TEN, DRESS, or anaphylaxis, yet these errors were not identified and corrected for patients with new encounters after a system-wide EHR transition.⁷⁰

Sources of inaccurate documentation include lack of training and misconceptions among clinical staff responsible for documentation. Foreman et al. found that only 45.1% of reactions consistent with intolerance (eg, “nausea,” “diarrhea”) were correctly categorized as such.⁶⁹ Likewise, Inglis et al. observed that of 5,023 ADRs to penicillins, 95% were labeled as allergy rather than intolerance (n = 250, 5.0%), which is consistent with the known over-diagnosis of penicillin allergy in the hospital and other populations. Errors also stem from the failure to delete inaccurate allergies and redocument deleted allergies. Blumenthal et al. found that allergies tend to accumulate over time with relatively few allergy deletions, and Topaz et al. reported that over 50% of drug allergy alerts were triggered for previously tolerated medications, which undermines the purpose of CDS alerts.¹¹⁷

Successful approaches to improving DHRs documentation include creating standard formats for allergy recording⁸⁶ and implementing interventions to improve training and communication between medical record technicians and pharmacovigilance teams.⁸⁹ Concerted efforts should be made to educate clinicians documenting DHRs to improve the accuracy and completeness of records. Delabeling inappropriate allergies of multiple drugs in a single visit and delabeling after a negative oral challenge test were safe and effective methods to correct allergy documentation.^91,92 Furthermore, the involvement of patients in updating drug allergy records including allowing patients to review and update their allergy records could reduce allergy record errors.

Case Management

A majority of patients with beta-lactam allergy documentation do not have ‘true’ beta-lactam allergies.¹¹⁰ This leads to the overuse of antibiotics with potential lower efficacy, such as fluoroquinolones and vancomycin, harming individual and public health.⁹⁵ Hence, a priority of antimicrobial stewardship is to delabel inaccurate antibiotic allergies to optimize future treatments. The EHR is an essential tool for identifying and stratifying patients who would benefit from the evaluation of antibiotics allergies through skin tests, oral challenges, or patient interviews. Additionally, misconceptions about cross-reactivity or shared hypersensitivity between beta-lactams also contribute to the use of less optimal antibiotic alternatives. This presents an opportunity for the use of CDS applications to determine appropriate antibiotic alternatives¹⁴⁷ and evaluate the safety for drug rechallenges¹¹³ and desensitization.¹¹⁴

Clinical Decision Support

Though the Medicare and Medicaid EHR Incentive Programs encouraged the installation of CPOE systems to promote EHR usage, it has become increasingly evident that the high prevalence of allergy alerts and override rates jeopardizes patient safety and increases alert fatigue.¹¹⁹ EHR literature centered on CDS critically evaluates the effectiveness of current CDS alerts and calls for only essential allergy alerting mechanisms in response to the frequency of alert overrides. It is important to recognize that approximately 80% of these overrides are considered appropriate, suggesting that many alerts are not clinically relevant or useful; the fact that both Macy et al., 2021 and Buffone et al., 2021 have demonstrated that the reduction of beta-lactam alerts does not result in serious ADR reflects that our current alert rules need to be modified.^125,126 Still, there should be a greater concern for the 20% of alerts that are overridden but are in fact clinically relevant. Due to alert fatigue, clinicians may not be sensitized to the alerts that highlight allergens that can trigger severe DHRs. More research must actively explore and modify CDS to minimize the overwhelming number of alert overrides and improve clinical workflow.

Case Identification

Optimizing methods for the identification of DHRs in the EHR is important for studying the epidemiology of these reactions and for participant recruitment for prospective clinical studies including those assessing genetic risk factors. This includes real-time monitoring of EHR surveillance and the retrospective review of structured and unstructured EHR data which is especially essential for rare reactions like SJS/TEN that carry high mortality risk. The use of such optimized identification methods greatly reduced the amount of time and cost-intensive manual chart reviews. Kim et al. reported a reduction in the number of charts that needed to be manually reviewed by 96% using their electronic search strategy for contrast media hypersensitivity reactions.¹⁴² Furthermore, using a combination of case identification methods with structured data like ICD-9 and ICD-10 codes and unstructured data with text can greatly increase PPV compared to the use of structured data alone.¹³⁹ Future studies could leverage advanced computational and informatics technologies (e.g., NLP and machine learning) and diverse data sources for more efficient and accurate case identification and phenotyping.

Genetic Studies

Understanding genetic associations could facilitate the adoption of preventative, predictive, and diagnostic strategies to improve drug safety. Such studies have been limited by small sample sizes, however, Zheng et al. have used “drug allergy” labels from EHRs to obtain a large sample size from their DNA Biobank for GWAS.¹⁴⁴ Their high-throughput framework can be replicated to accelerate the discovery of drug response genetic association and improve precision medicine. Similarly, the discovery of HLA associations holds substantial promise for the prevention of severe cutaneous adverse drug reactions like DRESS syndrome which to date have shown strong associations with HLA class I alleles.

Limitations

This review has several limitations. While we repeatedly refined our search query to broadly capture relevant articles, it is possible we did not include relevant studies in our systematic review because we only extracted literature from five databases. We also excluded research letters, literature without abstracts, non-English studies, which may contribute to selection bias and limit generalizability. It is also worth noting that due to the heterogeneity of the studies included, we were not able to evaluate bias consistently. Additionally, the research categories were subjectively defined after the initial screening, and each article was placed in one category based on the predominant research theme. Hence, there may be important themes that are not studied extensively in this review and articles that fall into multiple categories.

CONCLUSION

EHRs present a rich source of clinical information that has been increasingly used to study drug hypersensitivity across diverse research topics. By leveraging large EHR databases, studies have been able to elucidate the epidemiology and genetic associations of DHRs. A growing number of studies has innovatively applied advanced informatics methods, including NLP, to analyze unstructured allergy data for case identification. Nonetheless, poor standardization and quality of allergy documentation jeopardize patient safety and pose a significant challenge to case identification. Additionally, the overuse of CDS alerts counters their utility while also contributing to clinician fatigue. Improvement of clinical documentation, more selective use of CDS alerts, and strategies to harmonize data from different sources are critical future avenues for drug hypersensitivity research.

Table 4.

Summary of articles related to case management

Focus/ Intervention	Drug of interest	Articles	Summary of findings
Antimicrobial stewardship (n=19)
Evaluation of documented beta-lactam allergies	All beta-lactam antibiotics	Abrams et al. (2016)110	ICD-9 codes and billing codes were used to identify patients who had underdone assessment for suspected beta-lactam antibiotic allergies for retrospective chart review. Following intradermal testing and oral challenge, 96% of patients with prior history of beta-lactam allergy were advised they could safely re-introduce beta-lactam antibiotics.
		Blumenthal et al. (2019) 99	This retrospective cohort study identified 1,046 test doses challenges prompted by an electronic guideline for hospitalized patients with reported beta-lactam allergies. The antimicrobial stewardship intervention was safe with only 3.8% of patients with beta-lactam allergy histories had a hypersensitivity reaction. Cephalosporin allergy histories conferred a 3-fold risk. Allergies were updated for 474 patients (45%), with records specified (82%), deleted (16%), and added (8%).
		Kwon et al. (2019) 101	A retrospective review of patients with a history of beta-lactam allergies in the EHR found that 15% (6/40) showed a positive cefazolin skin test result compared to only 1.36% (178/13,113) of cases with no such history.
		Shaw et al. (2020) 109	A retrospective review of 589 eligible patients with beta-lactam allergies who underwent allergy service consult, found that changes in the allergy record were recommended for 62% (n = 371) of patients; however, the allergy record was updated after the consult in only 74.9% (n = 278) of patients.
	Penicillins	Chen et al. (2017) 96	A specialized algorithm was used to flag and prioritize eligible inpatients in the EHR for penicillin skin tests and challenges resulting in the removal of penicillin allergy labels in 228 subjects (90.5%) and the use of beta lactams in 85 (38%) of patients who tested negative.
		Sundquist et al. (2017)97	The EHR was screened to identify patients with a history of penicillin allergy and a total of 37 patients were recruited for penicillin skin testing and oral challenge. None of the patients had a positive skin test or oral challenge, however 2 patients (5%) experienced reactions within 24 h.
		Kuder et al. (2020) 104	A retrospective review of pregnant women who underwent penicillin allergy evaluation via skin test found that 44 patients (95.6%) received negative results and 18 patients (39%) completed oral challenge and did not experience adverse reactions.
		Wolfson et al. (2021) 100	A retrospective study of the pregnant patients with penicillin allergies in the EHR found that of 220 patients skin tested, 209 (95%) had their penicillin allergy label safely removed and penicillin allergy testing was associated with significantly reduced broad-spectrum antibiotic use and increased first-line beta-lactam antibiotic use.
	General	Iammatteo et al. (2017) 102	A 5-year retrospective chart review of patients who underwent at least 1 single-blind placebo-controlled graded drug challenge was conducted. The reaction rate to drug and placebo was similar during beta-lactam challenges (9.4% vs 8.2%; P = .9) and during nonsteroidal anti-inflammatory drug challenges (14% vs 7%, P = .5), respectively.
Evaluation of allergy for treatment optimization	All beta-lactam antibiotics	Sigona et al. (2016) 107	A pharmacist-driven beta-lactam allergy interview was effective in switching 65% of eligible patients to beta-lactam therapy and identifying discrepancies between confirmed allergies and those documented in the EHR in 87%. Medical providers accepted 87.5% of pharmacists’ antimicrobial recommendations.
	Penicillins	McDanel et al. (2017) 103	Beta-lactam allergy screening using an electronic best practice advisory and Drug Allergy Clinic referral in orthopedic surgery patients resulted in higher cefazolin use in patients evaluated in the clinic versus those not evaluated (90% vs 77%) and lower use of non-beta-lactam antibiotics (16% vs 27%).
		Blumenthal et al. (2017) 106	Both application with clinical decision support and penicillin skin test increased the use of penicillin and cephalosporin antibiotics among inpatients reporting penicillin allergy by nearly 2-fold and 6-fold respectively.
		Ramsey et al. (2018) 108	Inpatients with penicillin allergies receiving moxifloxacin, intravenous vancomycin, aztreonam, daptomycin or linezolid were identified through an EHR report and a penicillin allergy history algorithm was used to determine eligibility for penicillin skin testing. Forty-seven patients (94%) were skin-test negative and were subsequently transitioned to beta-lactam antibiotic and no patients experienced immediate adverse reaction.
		Englert et al. (2019) 94	Patients with documented penicillin allergy (type I, immunoglobulin E [IgE]-mediated) who were prescribed alternate antibiotics enrolled in a pharmacist-driven PST service to facilitate delabelling of inaccurate penicillin allergies in the EHR. Of 22 patients, all were negative, and 68.2% (15) were successfully transitioned to beta-lactam antibiotics reducing the use of fluoroquinolones and vancomycin.
		Kuruvilla et al. (2020) 105	Using an institutional algorithm for antibiotic selection in penicillin-allergic surgical patients (n = 2296), treatment with cephalosporin increased from 22% at baseline to 80% after algorithm implementation without severe adverse reactions.
	Other antibiotics	Lin et al. (2020) 98	Patients with antibiotic allergy labels which interfered with the preferred antibiotic treatment were identified through physician recommendation in the EHR and 42 patients received oral challenges. In 40 (95%) of these patients, no allergic reaction was observed, and the preferred antibiotic treatment was given. Two (5%) patients developed a non-severe skin reaction after a drug challenge and continued an alternative antibiotic regimen
Improvement of desensitization	All beta-lactam antibiotics	Pandya et al. (2021) 111	A clinician survey found that the creation of standardized electronic beta-lactam antibiotic desensitization order sets results in increased overall efficiency.
Assessment of alternate antibiotic use		Mancini et al. (2021) 112	Of 2,276 inpatients receiving antibiotics for pneumonia at 95 U.S. hospitals, 450 (20%) had a documented penicillin and/or cephalosporin allergy. Inpatients with this documented allergy and pneumonia were less likely to receive recommended beta-lactams and more likely to receive carbapenems and fluoroquinolones.
Referral for allergy delabelling	Penicillins	Wang et al. (2021) 95	Using an attending physician educational session and Best Practice Advisory in the EHR, referrals to the allergy clinic for increased for penicillin-class drug allergies from 1.9% to 13.7% after the educational session with a further increase to 27.8% after the Best Practice Advisory.
Non-antibiotic drug desensitization and rechallenge (n=3)
Assess outcomes of drug desensitization	General	Murray et al. (2016) 115	A retrospective review of EHR of patients undergoing drug desensitization found that of 69 patients, desensitization was completed with no cutaneous reaction in 85% of patients. Reported histories of urticaria and labored breathing during prior exposure were significant in identifying patients who might have a reaction during desensitization.
Risk stratification and evaluation of taxane re-exposure	Taxane	Picard et al. (2016) 114	EHR of patients who had been treated for taxane-related DHRs were retrospectively reviewed to identify patients for re-exposure to taxanes. Of 138 patients desensitized, 29 (21%) had an immediate and 20 (14%) has a delayed DHRs with the procedure. Of 49 patients challenged, 2 (4%) had mild immediate and 1 (2%) had a delayed DHRs with the procedure.
Assessing safety of chemotherapy drug rechallenge protocol	Chemotherapy drugs	Wu et al. (2019) 113	Patients who attempted rechallenge with paclitaxel, docetaxel, carboplatin and oxaliplatin were identified through retrospective chart review. The first rechallenge cycle was completed successfully in 43/46 patients (93.5%) and 42/46 patients (91.3%) were hypersensitivity reaction-free throughout the treatment course under the rechallenge protocol.

Key Points.

• The widespread adoption of electronic health records has greatly enhanced the ability to study the epidemiology and management of a broad range of drug hypersensitivity reactions.

• There is a clear need for accurate drug allergy labelling given that many patients are incorrectly labelled with a beta-lactam allergy, undermining antimicrobial stewardship efforts.

• Although intended to promote safe drug prescribing practices, allergy alerting mechanisms lack specificity and clinical relevance, contributing to excessive override rates and alert fatigue.

• Accurate and complete clinical documentation in electronic health records is necessary to optimize the use of clinical decision support for drug hypersensitivity prevention.

Clinical Care Points.

• Emphasis should be placed on the completeness and accuracy of drug hypersensitivity reaction documentation to prevent future reactions.

• Drug allergy records need to be regularly updated to avoid unnecessary reliance on sub-optimal substitutes.

• Evaluation of drug allergies through skin tests, oral challenges, and patient interviews are safe and effective methods for delabelling inappropriate drug allergies.

• To minimize alert fatigue which compromises patient safety, it is critical for hospital systems to increase the specificity of allergy alerts.