Computerised prescribing support still needs a human touch

Abstract

Sir, We wish to compliment O’Mahony et al. with the SENATOR trial [1]. Despite the non-significant results for the primary outcome, we believe the SENATOR findings are still relevant for clinical practice. Their statements on the clinical decision support system (CDSS) uptake align well with our own experiences regarding automated alerts. In the University Hospitals Leuven, patients have an electronic health record (EHR) and medication is prescribed using a computerised physician order entry system, which incorporates multiple CDSS modules [2]. Uptake of these CDSS modules into clinical practice is very low. For example, from June 2019 to May 2020, 21,990 CDSS alerts were generated for very severe drug–drug interactions, only 11.8% was accepted by attending physicians. In our opinion, major causes were the low specificity and low clinical relevance of the CDSS alerts, contributing to alert fatigue [3]. To better impact prescribing, a more specific and clinically relevant approach was needed. Hence, the ‘Check of Medication Appropriateness’ (CMA) service was developed in adjunct to the existing CDSS modules. The CMA concerns a centralised service comprising both a rule-based screening for potentially inappropriate prescriptions (PIP) as well as a medication review performed by trained clinical pharmacists. All clinical rules are derived from guideline recommendations and validated by a local multidisciplinary expert panel to further increase downstream uptake. Each clinical rule is provided with a dedicated flowchart to guarantee consistency when performing medication reviews. Finally, all pharmacist recommendations are provided directly in the EHR by adding notes addressing the treating physician [2]. In a first stand-alone (i.e. non-integrated) prototype of the CMA, EHRs were screened daily. Preliminary results showed a high number of recommendations for clinically relevant PIPs with an acceptance rate of 69.4% [2]. Optimization of the service then led to the development of the current CMA software, which is fully integrated into the hospital’s information system (HIS). The rule-based screening now runs continuously on real-time and dynamic patient data and is able to discern rapid changes, such as those of biochemical values. From June 2019 to May 2020, 6,137 recommendations were provided with an acceptance rate of 77.4%. In sum, we believe that the SENATOR software might surely be worth further study. We are convinced that two aspects are important to take into consideration for future investigations. First, the software should be integrated in the HIS, allowing a continuous screening using real-time patient data. Second, software-generated recommendations should be interpreted clinically for relevance and communicated to the attending physician by a specified trained person (e.g. a pharmacist). CDSS without a human interface might only be successful if it leads to a limited number of highly clinically relevant alerts for a well-defined patient population. Preferably, the implementation should also be supported by local consensus. None. None.