A recent study developed, validated and evaluated drug-drug interaction algorithms that alert clinicians to potentially harmful interactions, which take advantage of patient context information available in electronic health records data.
Drug-drug interactions are responsible for an estimated 5-14% of adverse events among hospitalised patients. To help manage drug-drug interactions, electronic prescribing and pharmacy systems have alerts for potential drug-drug interactions to warn prescribers and pharmacists of potentially harmful medication complications and provide documentation on how to avoid or mitigate the risk. However, clinicians often override these systems due to irrelevant alerts that have little relevance to the patient’s clinical situation, which has led to alert fatigue as clinicians can respond inappropriately to important alerts. Recent concerning research has suggested that less than half of the overrides were considered appropriate. Moreover, the rate of adverse drug events was higher with inappropriate versus appropriate overrides.
Reducing drug-drug interaction alert fatigue
Approaches to reduce alert fatigue include turning off entire categories of alerts and tiering drug-drug interaction alerts by clinical importance. However, most of these existing approaches trigger drug-drug interaction alerts based on the pair-wise combinations of the drugs. This therefore results in little or no consideration of the contextual factors, such as patient information, despite the fact that specificity of an alert to individual patient characteristics plays an important role in alert acceptance. Research has shown that when alerts failed to provide contextual patient information, prescribers bypassed the alert and then searched for the relevant data they needed. Therefore, for better management of drug-drug interactions new approaches are needed that are able to take into account patient context information.
Drug-drug interaction alert algorithms
The researchers behind this study developed and evaluated algorithms for alerting drug-drug interactions that were validated using both synthetic and real-world electronic health record (EHR) data. In total, the researchers created 8 algorithms for 8 high priority drug-drug interactions. The algorithms were created using data, taken over a 3-month period, on the rate at which drug-drug interactions were triggered but for which no action was taken. All the data was collected from a single tertiary care facility and was used to identify drug-drug interactions that were considered a high-priority for contextualised alerting.
The 8 drug-drug interactions that were selected for development of contextualized decision support algorithms were:
- Citalopram/QT prolonging agent
- Potassium/potassium-sparing diuretic
This data was then analysed by a panel of drug-drug interaction experts and the algorithms were developed, which incorporate drug and patient characteristics that affect the relevance of such warnings. Once implemented as computable artefacts, the algorithms were validated using synthetic health records data, and tested over retrospective real-world data from a single urban hospital. The results from the tests using retrospective real-world data showed that the 8 algorithms were able to reduce the alerts that interrupt clinician workflow by more than 50%. Moreover, two algorithms (citalopram/QT prolonging agents and fluconazole/opioid) demonstrated the ability to almost completely eliminate irrelevant results for these drug combinations.
This study addresses the need for contextual drug-drug interaction algorithms that are validated and shareable. The 8 algorithms all showed an impressive ability at reducing irrelevant alerts when tested using retrospective real-world data. The comprehensive evaluation in this study serves as an important step towards making drug-drug interaction altering more patient-specific across various healthcare organisations.
Image credit: jcomp – FreePik