Improving allergy documentation: A retrospective electronic health record system-wide patient safety initiative

Abstract

Objectives:

Documentation of allergies in a coded, non-free-text format in the electronic health record (EHR) triggers clinical decision support to prevent adverse events. Health system-wide patient safety initiatives to improve EHR allergy documentation by specifically decreasing free-text allergy entries have not been reported. The goal of this initiative was to systematically reduce free-text allergen entries in the EHR allergy module.

Methods:

We assessed free-text allergy entries in a commercial EHR used at a multi-hospital integrated healthcare system in the greater Boston area. Using both manual and automated methods, a multidisciplinary consensus group prioritized high risk and frequently used free-text allergens for conversion to coded entries, added new allergen entries, and deleted duplicate allergen entries. Environmental allergies were moved to the patient problem list.

Results:

We identified 242,330 free-text entries, which included a variety of environmental allergies (42%), food allergens (18%), contrast media allergies (13%), “no known allergy” (12%), drug allergies (2%), and “no contrast allergy” (2%). The majority of free-text entries were entered by medical assistants in ambulatory settings (34%) and registered nurses in peri-operative settings (20%). We remediated a total of 52,206 free-text entries with automated methods, and 79,578 free-text entries with manual methods.

Conclusion:

Through this multidisciplinary intervention, we identified and remediated 131,784 free-text entries in our EHR to improve clinical decision support and patient safety. Additional strategies are required to completely eliminate free-text allergy entry, and establish systematic, consistent and safe guidelines for documenting allergies.

Introduction

Accurate documentation of patient medical information in the electronic health record (EHR) is critical for patient safety in healthcare.^1,2 Documentation of reactions to medications, foods, and other potential healthcare exposures occurs in the EHR allergy module to inform future prescribing and prevent adverse events by triggering allergy alerts as part of clinical decision support (CDS).³

Allergy module information standards can include multiple data elements describing an allergy (e.g., allergen, reaction, reaction type, date noted, severity, criticality). Despite the level of detail afforded by EHR allergy modules, current allergy documentation is poor: many allergies are missing reactions, fields beyond allergen and reaction are rarely completed, and reactions are often listed as “unknown.”⁴ Commercial EHR vendors vary as to which data elements are included in their allergy modules. Patient safety and allergy experts consider that, at a minimum, allergy entries should include allergen, reaction, reaction type, and severity.^5,6 However, this level of allergy documentation can be challenging to accomplish given inadequate training in entering allergies and limited general allergy knowledge among prescribers.^7,8

Allergy module entries may include drugs, food and environmental allergens.⁹ Allergens can be in a structured, coded data format (e.g., data generated from drop-down lists or checkboxes) or in an unstructured format, typically entered as free-text either alone or alternatively as a string of allergens and/or reactions. While coded allergens trigger alerts intended to protect patients, allergens entered as free-text will not trigger CDS which may be a hazard to patient safety (Figure 1). In our healthcare system prior to EHR conversion, free-text allergy data represented 6% of all entries.¹⁰ Because allergy inactivation or deletion was infrequent, there was an accumulation of both coded and free-text entries in the allergy module, resulting in poor user accessibility and provider alert fatigue.¹¹

Figure 1.

Representative cases of allergy documentation-related adverse events likely preventable by allergen coded entry

While tools have been developed to improve EHR allergy record documentation, such as correcting entry misspellings¹² and converting free-text data into coded data,^13,14 multidisciplinary healthcare system-wide patient safety initiatives to improve EHR allergy documentation by focusing on free-text have not previously been reported. While transitioning to a new system-wide EHR, Partners HealthCare System (PHS) leadership recognized that addressing free-text entries was critical for triggering CDS important to patient safety.¹⁵ We therefore developed a patient safety initiative to systematically address free-text allergen entries and improve the quality of EHR allergy module documentation.

Methods

Setting and Stakeholders

This patient safety initiative was performed at PHS, an integrated health care system in Massachusetts that includes community and large academic teaching hospitals. Stakeholders representing each hospital site across a variety of clinical disciplines including physicians, nurses, nutritionists, pharmacists, quality and safety and information technology (IT), formed the PHS Allergy Clinical Consensus Group (ACCG) and regularly convened to provide content knowledge and clinical decision making for this project.

Data Source

PHS implemented Epic (Verona, WI, v.2014) as the universal EHR in 2015. Prior to Epic, PHS hospitals and clinics had varied commercial (e.g., MEDITECH) and homegrown (e.g., the Longitudinal Medical Record, OnCall) EHR systems. Allergies from all EHR systems at PHS were stored in the Partners Enterprise Allergy Repository (PEAR)¹⁶ and were brought into the Epic allergy module with the staged EHR conversion.

For the patient safety initiative, drug allergy data were obtained from the EHR allergy section.¹⁶ Raw data were extracted and included allergy information for structured and free-text allergens, the latter comprising over 242,000 total entries. We use numbers with frequencies to describe allergy entry and remediation.

Allergy Module Prioritization

To identify potential target allergens for free-text remediation, we generated lists of the most frequent allergens in the allergy module.^4,9 ACCG members identified allergens with high severity reactions based on drug and food allergy literature and allergy-related safety incident reports over the past 10 years.^17,18 Allergen searches also identified strings of free-text suspected to include more than one allergen by finding all entries with text length >25 characters. ACCG members subsequently refined the lists of target allergens based on clinical experience.

This cross-sectional analysis yielded lists that were then synthesized and grouped into allergen categories (e.g., drugs, foods, or environmental) and, when necessary, subcategories (e.g., penicillin within medication, nuts within food) to facilitate further ACCG review. The ACCG assigned a priority level to each task, reaching consensus on those that had highest potential impact on PHS quality improvement and patient safety for remediation.

Target allergens underwent automatic remediation using existing utility functions in Epic designed to address these issues. Options for system-wide automated remediation included mapping the entry to an existing coded entry already on the patient’s allergy list, adding a new coded entry to the patient’s allergy list, or adding an entry to the problem list with deletion of the free-text allergy entry. Of note, free-text entries that listed multiple allergens within one entry could not be remediated automatically. For these, a manual remediation process was necessary by which each allergen was entered individually followed by deletion of the multiple allergen entry. Each attempt at automated remediation also had some entries that failed correction (termed “automated fall out”), which required manual remediation. All manual remediation was performed by general medicine physicians overseen by an allergy specialist (KGB) and pharmacist (BH). Manual allergy remediation processes followed a rule book set by the ACCG, which initially contained 25 allergy entry rules, but expanded to 89 rules by the end of the remediation process. To reduce alert fatigue, duplicate entries involving both a specific drug and its corresponding drug class were remediated to one structured entry reflecting the specific drug. If no solution was identified for remediation, entries were left as free-text, or deleted if the entry was considered unhelpful to safe prescribing (e.g., “antibiotic,” “small pill”). Allergy remediation occurred by an iterative process from August 2017 through August 2018 (Figure 2).

Figure 2.

Allergy free-text remediation process

For discontinued drug classes no longer considered a drug class by the Systematized Nomenclature of Medicine -- Clinical Terms (SNOMED-CT), similar drugs were chosen and entered in allergy modules as markers to trigger the desired CDS.¹³ For example, “loop diuretic” was a discontinued drug class that was mapped to furosemide as a coded entry with a visible comment that indicated the entry was converted from “loop diuretic.” Free-text entries indicating no drug allergies (e.g., “no known allergy”, “none known”, “NKA”) were deleted and replaced with the Epic coded checkbox for “No Known Allergies.”

The ACCG recommended that environmental allergy entries be included in the patient’s problem list rather than the allergy module given that these allergens do not inform safe prescribing or trigger CDS. As such, all entries identified as corresponding to environmental allergies (e.g., dust, cat, dog, tree, pollen, environmental, mold, hay fever, seasonal allergy) were removed from the allergy module and entered in the problem list.

Results

The most commonly identified free-text allergens included seasonal, contrast media, cats, no known drug allergies, dairy products, and nuts (Table 1). Of the top 20 most common free-text allergens entered in 2017, 41.5% (18,504/44,560 total entries) were environmental allergies, 17.6% were food allergens, 13.0% were contrast media allergies, 12.2% indicated no known drug allergy, 2.2% involved a drug allergy, and 2.0% indicated no contrast allergy.

Table 1.

Top 20 free-text entries as captured in the allergy module of the electronic health record at project initiation (August 2017)

Top Free Text Allergies	Count
SEASONAL	6897
UNSPECIFIED Contrast Media	5807
cats	5381
GENERIC: NKDA – NO KNOWN DRUG ALLERGIES	2994
Dairy Products	2957
nuts	2763
INGREDIENT Allergy: NONE	2432
Shrimp	2120
Seasonal Allergies	1845
3254: adhesive tape	1709
HAYFEVER	1551
Cat	1025
DRUG ALLERGY	979
Environmental	964
CLASS: 28:08.92 – ANALGESICS AND ANTIPYRETICS	921
No Known Contrast Allergy	884
UNKNOWN ALLERGY	869
latex	846
Hay fever	841
ADHESIVE TAPE	775
Total Count for Top 20 free-text entries	44,560

Each free-text allergy category included a variety of free-text entries. For example, “nuts” had 196 different ways of being entered in free-text, including descriptors “some nuts” “shelled nuts” “edible nuts” as well as various misspellings (e.g., “tree nuits”, Table 2).

Table 2.

Variability in representative (top 20) free-text nut, dust mite, and latex allergen entries as directly captured by the electronic health record

Nuts (total = 3,270 entries with 196 distinct variations)	Dust and dust mites (total = 1,357 entries with 84 distinct variations)	Latex (total = 1,024 entries with 89 distinct variations)	Penicillin (total = 199 entries with 122 distinct variations)	No known drug allergy (total = 4,296 entries with 126 distinct variations)
nuts some nuts other nuts mixed nuts RAW NUTS Nuts and peanuts shelled nuts nut allergy ‘all types of nuts’ 3 nuts ALL NUT PROBLEMS Edible nuts Multiple nuts nut and nut oils Nuts Allergy. All kinds. Mother reported. nuts; unknown nuts; tropical nuts PEANUT, NUT; PEANUT, NUT Tree nuits; nuts Peanuts, other nuts	Dust Dust. DUSTS Dust Mites Mites Dustmite Dust mights Dust mitee DUST MYTES house mites Environmental allergy – dust ENVIR-DUST HIGH SENSITIVITY TO DUST Household dust Mild dust Old dust nuts; unknown Possible allergy to dust Sneezing to dust Dust mites = asthma NKDA; DUST,MITES	Latex 8920:gloves, latex LAYTEX Natural rubber Prophylactic condom Apparent latex allergy manifest to dental rubber band; ALLERGY BACK BRACE CONTAINING LATEX Latex containing tape Latex from newspaper print, latex condoms Latex on bandaids Latex on the leads Latex sensitivity Latex; MISC/Food **DO NOT USE** NKDA latex sensitivity Rubber/latex Latex and rubber NKDA; DUST,MITES Sensitivity to latex	!Pen-vee K A MEDICATION IN THE PENICILLIN FAMILY a type of pcn adverse reaction to penicillin all penicillians ALL PENICILLINS ANTIBIOTIC-PCN ANY RELATIVE OF PENICILLIN FAMILY AVOIDSPCN Concern re: PCN consider penicillin allergy Generic PCN only GENERIC: PENICILLIN - He is allergic to penicillin; DRUG ALLERGY Liquid Penicillin Medication PCN Medication PCN-?RXN Medication PENCILLIN Medication PENCILLIN Medication PENICLLIN	GENERIC: NKDA - NO KNOWN DRUG ALLERGIES Medication NKDA NKDA GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; KNDA GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; n KNDA No Known Drug Allergies GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; NO NNKDA no known drug allergies: She has no known drug allergies GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; NKA TO MEDS GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; Nkda GENERIC: NKDA - NO KNOWN DRUG ALLERGIES; NKDA NKDAA no known drug allergies: No known drug allergies no known drug allergies: None known NKDA per pt no known drug allergies: No allergies to medicines NKDANKDA NKDS

Free-text entries in the ambulatory setting were most often entered by medical assistants (33.5%) and in the peri-operative setting by registered nurses (19.7%). Across inpatient and outpatient settings, medical assistants and registered nurses had the highest frequencies of free-text allergy entries (37.6% and 43.3% respectively). Physicians including fellows and residents entered just 7.9% of free-text allergies.

In total, 52,206 free-text allergy entries underwent automated remediation, and an additional 79,578 underwent manual remediation (Table 3). Manually remediated entries included those whose reaction severity was considered as high, or involved intravenous contrast, or medication-related free-text, or long text length (>25 characters). Since free-text entry remains a documentation option in the allergy module, during the period of remediation of historic allergy free-text, there was ongoing new Epic free-text being entered by providers across our healthcare system.

Table 3.

Free-texted allergy entry counts before and after remediation initial remediation (August 2017- August 2018)

Allergy module entries	Count
Total free-text at project start	242,330
Remediated (total):  Automated methods  Manual methods	131,784 52,206 79,578

Drug allergens

Beta-lactams and sulfonamide antibiotics represented two of the most common classes of free-text medication allergens. We deleted 199 free-text penicillin entries and added the coded entry for “Penicillins.” These 199 entries included 122 distinct spellings, misspellings and descriptions of penicillin. We deleted 688 entries representing amoxicillin and 8 representing beta-lactams with free-text and misspellings; a coded entry for “Penicillins” was added to the patient record. We deleted 158 free-text entries involving “sulfas” and the coded entry “Sulfa-Sulfonamide Antibiotics” was added to the patient record.

No known allergies

There were 4,298 free-text entries indicating variations of “no known drug allergy,” and 2,778 with entries of “no known allergy” were identified and deleted from the chart. These included over 130 unique free-text descriptions of no known allergy. The structured “No Known Allergies” entry was entered in the records for these patients, provided there were no other allergens in their record.

Environmental allergens

Environmental allergy entries had 10,683 free-text entries with 84 distinct variations of dust allergy (Table 2), 209 variations of cat allergy, and 46 variations of dog allergy, and 375 variations of trees or pollen allergy. These were deleted from the allergy module of patients’ charts and “environmental allergies” was added to the patient’s problem list. Similarly, 1,553 free-text hay fever entries, 2,602 environmental allergy entries, and 11,164 “seasonal” allergy entries were deleted and added to the patient’s problem list.

Food allergens

A total of 3,270 free-text entries for “nuts,” including 196 distinct variations (Table 2) were deleted, and coded entries for both “peanut” and “tree nut” were added to patients’ records. However, for the 345 free-text entries involving only specified tree nuts and 255 entries involving only peanuts, mappings were made precisely to the coded “tree nut” and “peanut” entries, respectively. An additional 1,956 free-text entries for shrimp were deleted and mapped to a coded “shrimp” allergen entry.

Contrast

We deleted 5,253 free-text entries for “unspecified contrast allergy”, and a coded entry for “Iodinated contrast - oral and IV dye” added. An additional 933 patients with a free-text entry indicating “no known contrast allergy” were identified; these entries were deleted and if the patient had no other allergies in their record, the coded “No Known Allergies” was selected for the patient.

Latex

Latex had 1,024 free-text entries including 89 distinct variations (Table 2), with free-text ranging from specific products (“gloves,” “latex tape,” “back brace containing latex,” “condoms,” “latex sutures,” “latex from newspaper print”) to misspellings (“latix,” “laytex”), or simply “latex.” These entries were deleted, and a coded entry indicating “Latex- Natural Rubber” was added to the patient chart. Of note, for the additional 65 individuals who already had a coded entry for “Latex- Natural Rubber” listed in the chart, the duplicate free-text entry was deleted.

Discussion

We identified and remediated free-text allergens in the EHR as a health system-wide patient safety initiative to improve documentation and ensure consistent CDS, including allergy alerting. We identified that many free-text allergy entries were for environmental allergies; lack of any allergies (e.g., “NKDA”), contrast media, and food allergens were also common free-text allergy entries. We found that non-contrast drug allergens, such as antibiotics, were less commonly entered in free-text. Unexpectedly, an allergy entry consistent with “no known drug allergy” was more than three times as likely to be entered in free-text. This creates a significant safety risk since it can be present along with a coded allergy, which could then be overlooked. We also found that free-text allergen entry is variable, and that a single term can be free-texted with hundreds of different variations and misspellings by different providers. With this health system-wide approach, we remediated 131,784 free-text entries.

Free-texted allergies fundamentally undermine quality and safety goals due to diverse terminology, misspellings, and abbreviations that do not reliably protect patients across healthcare settings. Our iterative process with both automated and manual remediation methods reduced the allergy free-text considered most likely to result in adverse patient consequences. We mapped unstructured entries to coded allergens, transforming inaccurate allergy information into clinically relevant data that would trigger CDS. A recent study examining allergy labeling during EHR transition periods identified free-text entries as a particular area of risk for errors.¹⁵ Our system’s remediation of over 100,000 free-text entries presents a new potential method and targets for improving EHR allergy documentation. Indeed, most EHRs permit free-text allergy entries, and allergy entries increase over time and over the patient lifespan,¹⁹ suggesting that periodic remediation may be necessary to preserve the integrity of a patient’s medical record, from the perspective of CDS functionality.

The most common free-text allergies described hay fever, cat allergy, or seasonal allergies, allergens considered by our expert group to be best documented in the EHR problem list rather than the EHR allergy module. Environmental allergies cause allergic rhinitis and allergic asthma but are not potential healthcare exposures and do not cause anaphylaxis except when given by injection intentionally by allergists in the form of allergen immunotherapy.²⁰ Further, in the allergy module of our EHR, these allergens do not communicate with pharmacy or nutrition systems or trigger any relevant CDS. We removed over 25,000 free-text entries reflecting environmental allergies from the allergy module and re-entered them in the problem list. Removal of these allergens can improve the accessibility and usability of the allergy module by increasing the visibility of more relevant, high-risk allergens.

Food allergens are an increasingly important consideration for hospitalized patients²¹ receiving hospital meals¹⁸ and medications that may contain food-derived substances.²² We found that nut terminology in the EHR allergy module presented a unique remediation challenge. Colloquially, among patients and providers, peanut (a legume) is considered a “nut” in addition to tree nuts. Remediation of free-texted “nut” entries with their 196 variations did not assume the allergen was tree nuts or peanut but remediated to both structured allergens. As food allergies become increasingly important, structured entries must be enforced and communicated to both pharmacy and nutrition services. Further, since food allergy entries include dietary preferences (e.g., vegan) and intolerances (e.g., lactose or gluten), as well as true allergic reactions, documentation of the “reaction type” in the allergy module should be enforced for food allergens as well as drug allergens.

Allergy entry is not adequately taught in health care or EHR training. Most systems, including our system, permit allergy entry by non-prescribers. Historically this practice has increased documentation of allergies and has become standard, but it presents unique challenges as non-prescribers frequently have different training of EHR usage. We found that over two-thirds of free-text entries undergoing remediation were entered by medical assistants and registered nurses. This not only suggests an education gap in allergy entry for these providers but represents an opportunity for our system to target initiatives to “catch and correct” errant allergy entry practices. Understanding the difference between coded and free-text, and why these users opt for free-text is also critical to sustaining any successful remediation intervention. In the absence of any published curriculum or studies tracking the efficacy of education interventions in this area, we developed an allergy entry tip sheet for broad dissemination as a training tool based on our findings (Supplemental Figure 1). As other institutions increasingly adopt allergy remediation as a patient safety imperative, we anticipate that education will be a key starting point, but ultimately stronger measures will be needed for sustained, highly reliable success.²³ Future research could focus on evaluating the relative strengths of different interventions in this area.

Our approach to remediating the EHR allergy module has several limitations. There is no consensus as to the most important allergens in a healthcare setting. Therefore, we employed a broad, but somewhat subjective approach to selecting target allergens that reflected health system level patient safety concerns and provider and patient clinical experience. The final list of allergens and free-text entries may not be generalizable to the priorities in other healthcare systems. For example, inactive ingredients were not included in our final list of target allergens but could be an area of consideration for future work. These interventions were conducted at one large health system located in the northeastern United States operating with a commercial EHR that included all allergens listed from the diverse prior EHRs throughout our healthcare system for decades. As a system, PHS chose to import existing allergy entries into the new system, and this may limit the generalizability of our efforts to other health care systems. Still, our interventions are generalizable since most EHRs permit free-text entries, and our interventions focused on free-text content, rather than allergy module architecture. Our process was informed by a wide range of stakeholders including safety experts, senior pharmacists, hospitalists, allergists and contracted information technology support. This operational commitment required leadership prioritization of allergy documentation as a patient safety initiative with committed funding and 0.83 full-time equivalent support.

In summary, we performed a healthcare system-wide allergy free-text remediation patient safety initiative. While we made substantial progress with respect to free-text reduction in high-risk areas, we continue to have unmet needs in free-text remediation with over 100,000 free-text entries remaining and ongoing, new Epic free-text entry by health system providers. In addition to consistently assessing and addressing free-text allergy entries on an ongoing basis, systems should explore EHR design changes to improve allergy documentation. Safety events and patient harm are metrics that can then be measured before and after implementation. Force functions exist in some current designs and include requiring certain allergy module fields to be completed or not permitting any free-text entries. However, more clinically intuitive, workflow-guided design to support structured allergy data entry is needed. An example of this includes dynamic pick lists where natural language processing is used to assist providers in correctly selecting a structured entry.¹² While free-text allergens were the primary patient safety concern addressed with this initiative, comprehensive longitudinal allergy module documentation improvements must also consider adoption of unified reaction lists, complete documentation, and deletion of allergies that are outdated, inaccurate, or repeatedly overridden.

Abbreviations:

ADE

adverse drug event

ADR

adverse drug reaction

EHR

electronic health record

HSR

hypersensitivity reaction

ACCG

Allergy Clinical Consensus Group

NKDA

no known drug allergy

CDS

clinical decision support

SNOMED-CT

Systematized Nomenclature of Medicine -- Clinical Terms