Patient Safety Risks Associated with Current Allergy-Related Clinical Decision Support

Background

Proper allergy documentation into an electronic health record (EHR) plays a key role in patient safety. Utilizing codified allergy information within clinical decision support allows for safety alerts notifying clinicians to the potential for harm with the administration of certain medications. Unfortunately, studies show that allergy information is commonly entered into the EHR as free text, which does not trigger important safety alerts.^1,2 In 2014, it was estimated that more than 80% of all available health care data existed as unstructured, or free-text, entries. ³ A 1997 study estimated that 12.1% of potentially harmful medication errors were due to incomplete or incorrect allergy documentation, and, despite advances in EHR technology, this has likely increased due to the high prevalence of free-text allergy entries. ³ This introduces one of the many information usability challenges that needs to be addressed. In this article we reveal certain gaps in current allergy alerting capability available in a widely used EHR software that could put patients at risk for harm. Finally, we outline what would be ideal functionality.

Current State

Allergies entered on a patient’s profile are checked against the drug information database embedded in the EHR. The primary time when allergy checking occurs is during computerized physician order entry (CPOE). The system can alert on many types of potential drug-related concerns (eg, allergies, side effects, intolerances, contraindications); however, concerns for alert fatigue have led many health care systems to only fire alerts for allergies and/or contraindications. Alerts are triggered only for active ingredients in drug products in the EHR. This introduces a clinically meaningful gap as clinical decision support (CDS) would also be valuable for specific inactive ingredients and unique terms. Customized configuration for select inactive ingredients was developed at a large integrated health system to fortify the EHR due to the lack of native functionality.

Existing allergy functionality includes 3 more important design elements that can be customized to optimize risk identification. Alerts in the standard design can be suppressed as a strategy to reduce minimally useful alerts. For example, many health care systems suppress alerts for patients with a documented intolerance for a specific medication, choosing instead to only see contraindication or anaphylactic allergies. If allergies have not been documented by the time a medication order is placed, then the clinician can be alerted to complete this task before being allowed to proceed. Lastly, allergy alert functionality can require an override reason before the user can move on to the next step.

Limitations

Limitations exist with current allergy alerting functionality that present numerous challenges. These shortcomings are generally either with how an allergy is documented or with the database of allergies and allergens. The search field works by searching terms that start with the entered term rather than searching for database items that may contain the search term, which inhibits the ability to effectively find appropriate allergies consistently. Additionally, the nomenclature does not allow for inaccurate spelling. When a clinician is unable to find the appropriate allergy to document, whether because it does not exist within the database or the appropriate search terms were not used, it is reasonable to assume that they will document the patient’s allergy as free-text. Allergies documented as free text are not able to be validated against the drug information database, and, therefore, will not generate an alert. After finding many free-text allergy entry issues with misspellings, and incomplete or inconsistent entries due to the strictness of the search terms, a large integrated health system has configured their EHR to only allow pharmacists to enter free-text allergies. This has not addressed all areas of vulnerability as external allergies can be downloaded as free-text content by any clinician.

Other problems with allergy documentation include the inappropriate entry of allergies as side effects or intolerances which may not trigger appropriate alerts, or the reverse of entering drugs with side effects and intolerances as allergies which would result in erroneous alerts that increase the risk of alert fatigue. Interconnectivity with other systems that house allergy information may be inconsistently used by clinicians, fragmenting allergy documentation in the EHR. If those interconnected systems utilize different nomenclature or have some other problem with translation, then an allergy may not be easily transferred into the EHR or, worse, may be documented only as free-text which gives the appearance of safety without actually providing it. Finally, the drug information and allergy databases do not contain information on all excipients or ingredients, disallowing the CDS system to check for allergies against those missing or inactive ingredients.

Ideal State

Configuration options to mitigate these limitations are themselves limited and require tradeoffs or choosing the better of 2 evils. As no complete technical solution exists, a foundational design change is warranted. Until further foundational changes can be made, providing safety education, updates, and training to the care team could promote the optimal usage and function of these tools, thereby improving appropriate allergy documentation.

The ideal allergy-alert functionality would allow for allergy checking at the product level. As drug products contain many inactive ingredients which are common allergens, the ability to check for allergies on the product level, combined with a more comprehensive database containing both active and inactive ingredients for each product, could aid in recognizing clinically relevant allergy interactions. Clinicians should be able to search products by their ingredients, including active and inactive ingredients, to avoid all products that could potentially cause a life-threatening reaction. It would also be beneficial to create a dynamic allergy search functionality to account for potential spelling errors and provide suggestions based on the letters typed at the time of entry. This type of search aid functionality is available in other technologies, such as the use in automated dispensing cabinets.

Allergy entries should be as specific as possible. They should not be considered complete until a reaction and severity are selected. Furthermore, alerts fired should look different based on the severity of reaction, allowing the most important alerts to stand out to clinicians. Reasons for an override should also be entered from a drop-down menu, and, when certain options are selected, such as “tolerated before,” clinicians should be prompted to update the patient’s allergy list. This would allow for the removal of false allergies to reduce the number of unnecessary alerts. The allergen and drug information databases should allow for custom updates or configurations when the need arises to incorporate new or unusual allergies or nomenclature.

Future Considerations

Many of the other challenges with allergy CDS could be potentially solved with natural language processing (NLP) as this technology matures. Developing NLP to provide structured, recognizable data association from free-text entries and synchronize with available data to trigger alerts would provide a stronger solution to these issues.^1,4 However, NLP is not easy to develop. A major barrier is the inconsistency in how free-text entries are formatted.^2,5 One study found that a single term could be entered in hundreds of different ways, including misspellings. ² This makes it difficult for NLP systems to properly match free-text to appropriate structured terms and CDS systems to trigger appropriate safety alerts. Another barrier is the time-consuming process of implementing an NLP system which may not convert all necessary data.^5,6 Also, NLP systems are often developed for specific tasks to meet the needs of a specific institution, making their widespread application difficult. ⁵ Some barriers to implementing allergy CDS involve the need for frequent updates to ensure information is relevant and the fine tuning of which alerts fire to avoid alert fatigue. ⁷

Functionally, the allergy documentation tools must be designed with the clinician in mind to make sure it is easy to enter and confirm the correct allergy information for a patient within the context of a normal patient encounter. Ideally, a patient’s allergy information would be documented only once and propagated to each relevant EHR through a Health Information Exchange interface. The systems providing allergy information must be fully interoperable to remove any systematic errors from being introduced.

Conclusion

The limitations in the existing allergy-alert functionalities pose risks for patient safety as triggered alerts are often clinically irrelevant, reinforcing dangerous alert-override practices. Understanding key limitations of allergy CDS is foundational to using these tools effectively. Innovations in NLP and CDS based on free-text allergies, the framework of allergy checking systems, and the usability of allergy search functions could enhance the clinical usefulness of the allergy checking process, improving patient safety.