Evaluation of an optimized context-specific drug-drug interaction alerting system

IntroductionA common approach to reduce unwanted drug-drug interactions (DDIs) in hospital practice is the use of clinical decision support systems (CDSS). However, most CDSS for DDI screening generate alerts with low specificity leading to alert fatigue.ObjectiveThe main objective was to investigate whether a newly developed context-specific DDI alerting system would improve alert acceptance.MethodsA controlled pre-post intervention study was conducted in 4 departments in a university hospital. After a 7-month pre-intervention period, the new system was activated in the intervention group, while the old system remained activated in the control group. Post-intervention data was collected for a 7-month period. The new CDSS included a follow-up system for real-time evaluation of the alerts by a clinical pharmacist.ResultsA significant increase of the overall acceptance rate was observed between the pre- and post-intervention period (2.2% versus 52.4%; p < 0.001) for the intervention group and between the intervention and control group (2.5% versus 52.4%; p < 0.001) in the post-intervention period. There were no significant differences in acceptance rates between the pre- and post-intervention period in the control group and also not between the control and intervention group in the pre-intervention period. Although the clinical pharmacist conducted only 10 interventions, the follow-up of alerts seemed important because interventions were conducted for high risk situations.ConclusionsThe improvement was probably related to several optimization strategies including the customization of the severity classification, the creation of individual screening intervals, the inclusion of context factors for risk assessment, the new alert design and the creation of a follow-up system. The marked increase in alert acceptance looks promising and should be further evaluated after hospital wide implementation. System aspects that require further optimization were identified and will be developed. Further research is warranted to develop context-aware algorithms for complex DDIs.