(294) Optimization of Commercial Clinical Decision Support System for Inpatient Pharmacy at a Medical Center in Taiwan

Background: Clinical Decision Support Systems (CDSS) reduce prescription errors via drug alerts; however, their full potential is impeded by alert fatigue and may lead to compromising patient safety.

Objectives: To assess the effects of optimizing commercial CDSS (Medi-Span®) on alert fatigue, while analyzing it’s availability in the Taiwanese healthcare system.

Methods: This pre-post study of optimization (March to May 2023 and March to May 2024) was conducted at National Cheng Kung University Hospital, where a commercial CDSS-Medi-Span® was introduced to the inpatient pharmacy for prescription evaluation. Optimization strategies include (1) modified and customized alert rules (2) standard alert actions (3) a user-friendly interface, and (4) developing a satisfaction questionnaire to identify factors related to the availability of Medi-Span®. Data collection involved level 1 (major severity) drug alerts detected by Medi-Span®, along with evaluation results of pharmacists. The primary outcome, alert fatigue, was measured by alert burden, pharmacist intervention alert rate, and pharmacist non-intervention alert rate. To assess the questionnaire's validity and reliability, context validity used S-CVI/Ave (Scale Content Validity Index/Average), internal consistency employed Cronbach’s alpha coefficient, and test-retest reliability utilized the intraclass correlation coefficient (ICC). Categorical and continuous variables were compared using the chi-squared test and t-test, respectively. Statistical significance was set at p<0.05.

Results: Compared to pre-optimization period, the number of alert burden and pharmacist non-intervention alert rate reduced from 11,318 to 8,972 (p < 0.001) and from 97.2% to 92.64% (p < 0.001) in post-optimization period, respectively. Concurrently, the pharmacist intervention alert rate increased from 1.4% to 3.0% (p=0.001), with dosage or frequency modification for 73.0 % of prescriptions. Regarding the questionnaire, the S-CVI/Ave was 0.98 for relevance and 0.90 for simplicity. For reliability, the overall Cronbach’s alpha was 0.947, ranging from 0.742 to 0.949 for four domains with good to excellent internal consistency. The median ICC was 0.761 (IQR 0.638 - 0.806), indicating moderate to good test-retest reliability. In the final questionnaire feedback from 15 clinical pharmacists, it was revealed that they generally expressed satisfaction (mean score ≥ 3) with the use of Medi-Span®. Qualitative and quantitative analysis showed that ease of use and learning, as well as improving patient safety, are facilitators for the availability of Medi-Span®. However, barriers such as the lack of relevance and completeness of alert content, along with the absence of features fitting user expectations, may impact the availability of Medi-Span®.

Conclusions: Our optimizations of Medi-Span® resulted in a significant reduction in alert burden, the pharmacist non-intervention alert rate, and an increase in the pharmacist intervention alert rate. This contributed to the reduction of alert fatigue and the prevention of prescription errors. Additionally, we developed a valid and reliable questionnaire to assess the availability of Medi-Span®, which can identify the factors users consider important for CDSS. This study also provides a reference for applicable drug alerts to improve patient safety and future optimization directions of commercial CDSS or medical institutions in Taiwan.