Approximately 10% of people in the United States report a history of penicillin allergy. However, prior research shows that >90% of these patients subsequently tolerate penicillin without clinically significant hypersensitivity reactions. Reference Zhou, Dhopeshwarkar and Blumenthal1–Reference Pawlasty, Thompson and Chalmers4 The artificially high rate of documented penicillin allergies is due not only to inappropriate allergy assessment or documentation but also because ˜80% of patients with a positive penicillin skin-prick test actually lose their immune reaction after 10 years. Reference Rubin5,Reference Blanca, Romano and Torres6
Avoidance of penicillin among the large population of patients who report allergies but are actually tolerant presents a public health problem because it contributes to the emergence of drug-resistant bacteria. Reference Blumenthal, Lu, Zhang, Li, Walensky and Choi7 In addition, patients who report penicillin allergies are at increased risk for adverse drug events. Reference Macfadden, Ladelfa and Leen8 These are mediated by increased use of penicillin alternatives, which tend to be broader spectrum and more toxic than β-lactam antibiotics. Reference Blumenthal, Lu, Zhang, Li, Walensky and Choi7,Reference Macy and Contreras9 When hospitalized, patients with penicillin allergies have longer and more expensive hospitalizations compared to matched nonpenicillin allergic controls. Reference Mattingly, Fulton and Lumish10–Reference Picard, Bégin and Bouchard12
The goal of this study was to quantify the percentage and mean daily frequency of interruptive penicillin allergy alerts that are displayed in cases in which there is evidence in the electronic health record (EHR) of prior penicillin exposure, or any β-lactam exposure, after the allergy was noted to determine whether modification of these alerts with information about prior exposure may be beneficial.
Methods
Setting
We studied interruptive penicillin allergy alerts in CS-Link, Cedars-Sinai Health System’s (CSHS) branding of the EpicCare enterprise EHR product (Epic Systems Corporation, Verona, WI). CSHS comprises a large, nonprofit hospital, a smaller associated hospital, 2 emergency departments (EDs), and ˜250 outpatient clinics that are staffed by both employed and private community physicians. The system provides multidisciplinary care to a socioeconomically diverse population, with ˜794,000 outpatient visits per year. CS-Link is used throughout the health system.
Description of interruptive penicillin allergy alerts
At CSHS, allergies can be documented by any provider, but is most frequently done by nurses. Allergies are stored in the EHR indefinitely until they are modified or deleted by a user. No specific clinicians are responsible for removing inaccurate allergies. Interruptive alerts are displayed to prescribers attempting to order medications that are identical or related to medications to which patients have a documented allergy. For patients with penicillin allergies, these alerts, which contain information about reaction type and allergy severity, are displayed when any β-lactam is ordered, including penicillins, cephalosporins, and carbapenems (Supplementary Fig. 1 online). We refer to these interruptive penicillin allergy alerts as “alerts” for the sake of brevity.
We identified alerts that displayed for physicians, physician’s assistants, and nurse practitioners between January 1, 2013, and June 26, 2018. We did not examine alerts shown to pharmacists. To determine whether patients were exposed to β-lactams after their penicillin allergy was noted, we accessed medication administration record (MAR) and outpatient prescription data. We also obtained outpatient β-lactam prescription data from outside institutions participating in Care Everywhere (CE; Epic, Verona, WI), a health information exchange.
Prior β-lactam exposures of interest were those that occurred after the allergy had been documented but before the alert was displayed. Cases of interest are illustrated by the timeline shown in Figure 1. The time between each of these events was expected to vary across patients.

Fig. 1. Timeline representing alert of interest.
Analyses
First, we ran structured query language (SQL) queries to identify all alerts that were triggered by a penicillin-class antibiotic order and to determine the number of cases in which the patient had MAR data demonstrating prior penicillin exposure. Because not all alerts are clinically meaningful (eg, showing the same alert to the same provider regarding the same patient within a few minutes, before and after the provider talked to the patient about their allergy history), we restricted alerts to 1 per clinical encounter. In a secondary analysis, we included all alerts, without restricting to 1 per encounter.
In 2 additional analyses, we expanded our definition of “prior exposure” beyond MAR data to include outpatient prescriptions, first from CSHS and then from Care Everywhere, to measure the additional benefit that these data sources would provide.
In addition to examining alerts that displayed when penicillin-class antibiotics were ordered, we also ran SQL queries to identify alerts that displayed when any β-lactam antibiotic was ordered, and we identified the number of cases in which the patient had MAR evidence of any β-lactam exposure (including penicillins, cephalosporins, and carbapenems) after the allergy was noted. Again, we restricted alerts to 1 per encounter to capture those that were most clinically meaningful, but we performed an additional analysis that included all alerts.
In each situation, we calculated the mean daily number and the percentage of alerts for which EHR data suggested prior exposure. Additionally, because prior exposure does not necessarily indicate tolerance, we checked for diagnosis codes for allergic reactions documented in the 7-day period following each administration of a β-lactam. A list of the International Classification of Disease, Tenth Revision (ICD-10) codes we considered is listed in Supplementary Table 2 (online). If patients received a β-lactam at our institution and then presented elsewhere for allergy-related symptoms, these instances were not captured. However, in many cases (eg, receiving 5 days of a cephalosporin for pneumonia during a hospitalization), allergy-related symptoms would be unlikely to manifest after discharge.
Validation
In a subset of 100 randomly selected cases, we performed a manual chart review (1) to confirm that our data sources were accurate, (2) to identify scenarios in which the initial β-lactam was changed to a second-line antibiotic in response to the alert to gain qualitative insight, and (3) to ensure that prior exposure to β-lactam antibiotics did not result in any allergic reaction. We performed an additional chart review of 10 cases in which there was a diagnosis code for an allergic reaction in the 7-day period following the administration of a β-lactam to assess whether there was chart documentation of an allergic reaction.
Results
In total, 115,081 interruptive penicillin allergy alerts were displayed to prescribers between January 1, 2013, and June 26, 2018. Alerts were shown for 17,671 patients across 31,033 encounters: 8,811 outpatient encounters, 19,351 inpatient encounters, and 2,871 ED encounters. The median number of alerts per encounter was 7 (interquartile range, 3–25). Demographic information for these patients is listed in the Supplementary Materials (Table 1). Across the health system, there were 1,154,777 total encounters during the study period: 10,695,343 outpatient encounters, 307,427 inpatient encounters, and 545,000 ED encounters.
Table 1. Penicillin (PCN) Allergy Alerts for Patients With Records Showing Prior β-Lactam Administrations or Orders in our Health System

a The “alert order” is the order that a prescriber initially entered, which prompted the firing of the allergy alert.
Percentage of alerts associated with prior β-lactam exposure
In 8% of alerts displayed to prescribers ordering penicillin-class antibiotics, MAR evidence of prior penicillin exposure after the allergy was noted (Table 1). When the analysis was expanded to alerts triggered by orders for any β-lactam, including penicillins, 49% of alerts were displayed for patients with MAR evidence of prior exposure. When we removed the restriction to 1 alert per encounter, these percentages rose to 10% for prior penicillin exposure, and to 53% for any β-lactam exposure. Alerts per day are listed in Table 1. When alerts were stratified by encounter type, 61% of inpatient alerts were displayed for patients with MAR evidence of any β-lactam exposure, followed by 44% of ED alerts and 24% of outpatient alerts. In 96% of β-lactam administrations, there was no ICD-10 code for an allergic reaction within 7 days after the administration. Of the 2,666 cases (4% of administrations) in which there was an ICD-10 code, the plurality (37%) were codes for “allergy, unspecified.”
When we expanded the analysis to include outpatient prescriptions ordered in CS-Link as proxies for exposure, the percentage of alerts that were displayed for patients with prior penicillin exposure increased to 17%. After adding outpatient orders identified via Care Everywhere to administrations and prescriptions within our system, the percentage increased further to 19%. The percentage of outpatient orders identified using Care Everywhere data increased over time. Over the entire study period, Care Everywhere orders represented 2% of all prior orders. In 2018 alone, the percentage of orders from outside institutions was 6%.
Validation
Manual chart review of 100 penicillin allergy alerts revealed that our SQL-based analysis was accurate in 95 cases. Of the 5 misclassified cases, 4 had been classified as alerts without prior β-lactam exposure, but manual chart review found it to be present. In 1 instance, the case had been classified as an alert with prior β-lactam exposure, but manual chart review did not identify any such data.
Manual chart review also identified 18 cases in which, after the alert displayed, the original β-lactam order was discontinued and a new antibiotic was ordered. All cases had evidence of prior β-lactam exposure, either as an inpatient administration or outpatient prescription. We identified 7 scenarios in which the alert led to a change to a broader spectrum and/or more toxic antibiotic. Five of these, which we considered to be the most clinically significant antibiotic changes, are described in Supplementary Table 3 (online).
Among the 10 charts with an ICD-10 code for an allergic reaction within 7 days after administration of a β-lactam, there was no chart documentation of any allergic reaction to β-lactam antibiotics. There was 1 documented allergic reaction; however, the suspected culprit was intravenous immunoglobulin treatment, not a β-lactam.
Discussion
We aimed to quantify the extent to which interruptive penicillin allergy alerts are displayed for patients with EHR evidence of prior penicillin or other β-lactam exposure. We found that 8% of all penicillin allergy alerts displayed to prescribers were for patients with MAR evidence of prior exposure to penicillin-class antibiotics after their allergy was noted. This figure increased to 19% when we included outpatient prescriptions, both in our health system and systems participating in Care Everywhere, as proxies for prior exposure. Furthermore, nearly half (49%) of alerts were displayed for patients with MAR evidence of any prior β-lactam exposure. Based on national estimates of annual outpatient, emergency department, and hospital encounters, 13–15 along with our estimates of the rate at which alerts associated with prior exposure display across all encounters, we project that each year >7 million alerts display for patients who have previously been exposed to β-lactams: 325,597 outpatient alerts, 891,146 emergency department alerts, and 6,376,928 inpatient alerts. These alerts may contribute to alert fatigue, and a portion may even contribute to antimicrobial resistance and avoidable adverse drug events, if prescribers are unable to find information in the EHR about prior exposure.
Prior β-lactam exposure does not necessarily indicate that the drug was tolerated. However, we found that only a small minority (4%) of cases of prior β-lactam administration had a diagnosis code for an allergic reaction documented in the 7-day period following the administration. Furthermore, random chart review of 10 of these cases revealed that despite the diagnosis code, there was no chart documentation of a β-lactam-induced allergic reaction. In addition, the plurality of ICD codes were for an “allergy, unspecified,” which could relate to allergies for other medications, or simply patient-reported historical allergies as opposed to acute allergic reactions. Therefore, it is likely that most prior administrations were well tolerated. Still, because this may not be true in all cases, clinicians should ask the patient and review the chart for any evidence of adverse reactions to prior administrations before presuming tolerance.
Prescribers can currently find information about prior exposure in any EHR, but this task is often time-consuming, especially when it requires searching a health information exchange, such as Care Everywhere, which we identified as a rapidly maturing data source. Our finding that a high proportion of alerts display for patients who have previously received a β-lactam antibiotic suggests that modification to these alerts with information about prior exposure may be an efficient way to bring this information to prescribers at the point of care. Enhancing these alerts with information about prior β-lactam exposure could more efficiently alert prescribers to the possibility of prior tolerance, which could lead to higher rates of prescribing β-lactam antibiotics, and lower rates of prescribing second-line, broader-spectrum antibiotics. It may also lead prescribers to remove allergies from the chart completely, thus eliminating the possibility of future alerts. We have recently implemented such a modification to these alerts at our institution, with anecdotal success.
Many strategies have been explored for identifying inaccurately documented penicillin allergies. Detailed history about the timing of the reaction, and the reaction type, can help stratify patients into low-, moderate-, or high-risk for true allergy, and there is evidence that low-risk patients can be safely challenged with oral amoxicillin. Reference Tucker, Lomas, Ramchandar and Waldram16,Reference Iammatteo, Alvarez Arango and Ferastraoaru17 For cases of uncertainty and higher-risk patients, skin-prick testing has historically been the gold standard. Reference Shenoy, Macy, Rowe and Blumenthal2 Enhancement of allergy alerts thus represents a complementary, rather than an alternative, method to identify penicillin or β-lactam tolerance.
Our analysis had notable limitations. First, we were unable to exactly replicate our SQL-based findings with manual chart review. However, we detected only 5% misclassification, and 4% of this misclassification represented an underestimate. This finding gives us strong confidence in our SQL-based findings that considered every single alert during the study period. Second, our analysis was restricted to only 1 EHR instance serving providers in 1 health system. Even though the diversity of prescribers using this EHR, including both employed and private practice physicians, should contribute to good external generalizability, this analysis should be repeated at other organizations.
In conclusion, in our health system, nearly half of interruptive penicillin allergy alerts were displayed for patients who had EHR evidence of prior exposure to any β-lactam, and 8% had EHR evidence of prior exposure to a penicillin. Enhancing existing alerts with information about prior β-lactam exposure may provide valuable information to prescribers at the point-of-care, which could lead to improved antibiotic stewardship.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/ice.2021.329
Acknowledgments
Financial support
No financial support was provided relevant to this article.
Conflicts of interest
All authors report no conflicts of interest relevant to this article.