Evaluation of context-specific alerts for potassium-increasing drug-drug interactions: a pre-post study (Preprint)

BACKGROUND Clinical decision support systems (CDSS) for drug-drug interaction (DDI) screening are often overly inclusive following a “better safe than sorry” approach leading to an excessive number of alerts, alert fatigue and high override rates which jeopardizes the effectiveness of CDSS. Evidence on the effect of CDSS on patient-specific outcomes is scarce. OBJECTIVE We investigated whether context-specific alerts for potassium-increasing DDIs reduced the alert burden and had a higher alert acceptance compared with alerts without context-specific rules. Secondly, the effect of these context-specific alerts on the occurrence of hyperkalemia was investigated. METHODS The intervention was the hospital wide implementation of a context-specific DDI alerting system where a recent laboratory value of potassium determined the alert level of the DDI. In the pre-intervention period all alerts for potassium-increasing DDIs were level 1 alerts advising absolute contraindication, while in the post-intervention period the same drug combinations could trigger a level 1 alert (very serious, absolute contraindication), a level 2 alert (serious, monitor potassium values), or a level 3 alert (informative, not shown to physicians to improve specificity). The number of alerts, alert acceptance and the occurrence of hyperkalemia after a potassium-increasing DDI alert was triggered were compared between the old CDSS and the new context-specific CDSS. RESULTS In the pre-intervention period 1461 level 1 alerts were triggered. In the post-intervention period 3 level 1 alerts, 86 level 2 alerts and 1148 level 3 alerts were triggered. The number of alerts shown to the physicians was reduced by 92.8%. The difference in alert acceptance between level 1 alerts was significant (P = 0.010), but should be interpreted with caution because of the small number of level 1 alerts in the post-intervention period. Better documentation of patient factors and follow-up of the high-risk patients was observed in the post-intervention period. In the multivariable regression model for the risk of hyperkalemia the intervention was not significant (OR 1.421, 95% CI 0.917 – 2.201). Three significant confounders were retained: the pre-DDI alert potassium measurement (OR 1.972, 95% CI 1.421 – 2.735), the administration of systemic corticosteroids (OR 2.409, 95% CI 1.502 – 3.864), and the administration of ACE inhibitors (OR 1.811, 95%CI 1.172 – 2.800). CONCLUSIONS We were able to reduce the alert burden with 92.8% without compromising patient safety. The results of this study demonstrate the proposed strategy seems effective to get a grip on the delicate balance between over- and under alerting. Further research into the development, optimization and evaluation of context-specific prediction rules for complex DDIs is warranted.