Allergy alerting and overrides for opioid analogues across two health systems

Discussion

In this study across two large US health systems, we identified that the frequency of opioid alerts overridden was 71.8%. Of these, about a quarter of reactions were documented to be severe. Thus, approximately three in four alerts were for reactions of medium or low severity. Among the most frequent reactions to generate an alert, approximately 30% were true immune-mediated reactions, and the most common reactions associated with overridden alerts were non-immune mediated reactions. The data suggest that reaction severity and immune and non-immune mediated reaction classifications permit opportunities to improve opioid-related decision support.

Over the past few decades, there has been a rise in opioid-related alerts that may be related to increased prescribing of opioids and improved documentation of listing patient allergies in the EHR. This increase in alerts parallels an increase in override rates, increasing from 85.1% to 89.7% in a decade-long study at MGB.6 The most common reasons documented by clinicians for overriding alerts are ‘will monitor or take precautions’, ‘not clinically significant’ and ‘benefit outweighs risk’.13 Prior studies have shown that providing contextual information alongside the alert and specific to the patient can influence a provider’s acceptance of the alert.14 Quality improvement studies for specific conditions (eg, hyperkalaemia) or alert refinement based on alert data, multidisciplinary panels and surveys have been successful at reducing the number of alerts.15 16 However, the appropriateness of alerts and methods to reduce opioid-related alerts is still lacking in the EHRs. Genco et al evaluated the outcomes of patients receiving opioid alerts and found most alerts (98.6%) were inconsequential.2

Analysis of reaction type and severity reveals this is a critical element to consider when tailoring alerts. Nevertheless, it is also important to keep in mind when creating algorithms to consider other patient-specific concerns that may be documented as an allergy, intolerance or non-life-threatening side effects such as a patient who gets opioid-induced itching (without urticaria or anaphylaxis) or codeine-induced chest pain (oesophagitis-like) reactions.17 There is a balance to consider alongside clinical judgement, and we focused on one potential area to reduce interruptive alerts with reaction type and severity levels.

Based on previous studies, most opioid alerts are triggered by non-immune-mediated reactions, including opioid side effects.18 Among the overridden alerts, approximately one-third of reactions are true immune-mediated reactions (29.3%) and 72.8% were associated with mild and medium reactions. Furthermore, 7% of all opioid overridden entries which were mild reactions at both institutions would still trigger an interruptive alert with the new algorithm. The high proportion of override rates associated with entries of low severity indicates that allergy entries with mild reactions may be appropriate to not trigger interruptive alerts. Our findings indicate that strategically assigning interruptive alerts for severe reactions has the potential to reduce interruptive alerts. Discriminating between opioid immune-mediated reactions may also reduce overrides notably, as non-immune-mediated side effects were overridden 70.7% of the time.

Our findings also uncovered that over a quarter of overridden alerts are triggered by severe allergies. The concerning implications of overriding alerts associated with severe allergic reactions warrant greater analysis to evaluate why they are overridden. According to previous studies, physician-reported override reasons include previous tolerance of the medication, absence of a true allergy and a lack of reasonable alternative.1 These high override rates may be reflective of the effects of alert fatigue. As with excessive, untargeted alerting, clinicians may be more likely to insensitively trigger alerts despite varying levels of severity, as they are conditioned to override alerts rather than accept alerts. Greater investigation of the reasons for overriding highly severe allergy alerts can uncover what factors play a role in overriding alerts for more nuanced scenarios and the extent to which alert fatigue plays a role.

Allergy alerting logic is managed within the EHR which uses third-party databases (eg, First Databank, MediSpan) to determine if and when an alert should be triggered. Alerting is typically based on drug ingredient (active or inactive), drug class or cross-sensitivity. Health systems have the ability to configure which alerts will fire for various users based on level of match (eg, level 1: exact drug match, level 2: base ingredient match, level 3: drug class match or level 4: cross-sensitivity). Alerts can be filtered out based on classification (contraindications, intolerances/adverse reactions or miscellaneous exceptions) or severity. These functions allow the type of alerts to be configured to the needs of users with a specific role (eg, physician, pharmacist). Some commercial CDS products and EHRs offer limited customisation options for allergy alerts, such as restricting alerts to specific providers and allowing users to suppress certain warnings for a defined period (eg, ‘do not alert me again for 24 hours’). However, the effectiveness of these configurations is often constrained, because they often require manual adjustments by users, which can be tedious and time-consuming. As a result, many providers either ignore alerts or disable them broadly rather than fine-tuning them, thereby reducing the intended safety benefits of the system.

Existing solutions remain inadequate in their ability to fully tailor alerts to clinical contexts and workflows. Many systems lack the ability to incorporate contextual factors, such as recent provider interactions with similar alerts, the clinical relevance of the alert based on the patient’s medical history, or differentiation between high-risk allergies and low-severity or non-immune-mediated reactions. Furthermore, alert accuracy and specificity remain significant challenges, as many alerts are generated based on broad rule sets rather than personalised patient data, leading to excessive and often unnecessary interruptions.

What is lacking in commercial allergy alerting systems/EHRs is the ability of the system to learn from past interactions with its users (eg, patterns of overrides).19 Additionally, if a patient has been administered a medication for which an alert was overridden and reason documented (eg, tolerance), the system should be updated in a registry. This can help then guide reconciliation of allergy information with either removal of the allergen, change in classification (eg, intolerance or adverse reaction that can be suppressed to an informational only alert rather than interruptive alert). Based on our results, our algorithm functionality would decrease 46.4% of interruptive alerts. By reducing less consequential alerts, users can pay more attention to critical alerts.20

Back to top