Analysing university hospitals’ clinical decision support system alerts for inpatient medication prescribing process: a cross-sectional register-based study

Abstract Background Computerized physician order entry (CPOE) with structured prescribing process and clinical decision support (CDS) system is a widely used systemic defence to enhance medication safety in hospitals. However, the implementation of these systems requires major changes to the workflows, which has been associated with an increased risk of prescribing errors. The aim of this study was to describe and analyse the CDS alerts and the responses of the physicians at prescribing process for the further development and optimisation. Methods This cross-sectional register-based study included CDS alerts of the Helsinki University Hospital’s adult inpatient prescriptions during year 2022. The alerts were studied with descriptive quantitative analysis to identify the prevalence of alerts and physicians' responses to them. The alerts displayed for somatic conservative and operative specialties were analysed separately to identify possible differences. The data was analysed using descriptive statistics with Microsoft Excel. Results In 2022 there were altogether 5,342,217 adult inpatient medication prescriptions. For 16% (n = 874,317) of the prescriptions, the CDS produced a soft-limit pop-up alert and 87% (n = 762,746) of these alerts were overridden without changing the original prescriptions. The alerts for somatic operative and conservative specialties covered almost 57% (n = 495,399/874,317) of all alerts and there were no major differences between the specialties. Interactions caused most alerts and had high override rates (93–94%). The highest override rate was for geriatric (97%), dose (96%) and pregnancy and lactation alerts (94–97%). The alerts related to drug allergies (55–65% overridden) and duplicate prescriptions (69–88% overridden) had the highest acceptance rates. Conclusions The highest override rates were connected to geriatric, dose, pregnancy and lactation alerts, which should be optimized to decrease the alert fatigue. The biggest volume of alerts was associated to drug interactions, but filtering of these alerts would be problematic due to large variety of medical specialties using the same EHR system. This study gave an overview to the current situation in one university hospital shortly after CDS implementation and provided a baseline for creating a collaborative model for system optimisation.