4CPS-151 Validation of an algorithm for prioritising medication reconciliation at admission using an artificial intelligence method

<h3>Background and Importance</h3> Medication reconciliation at admission (MRA) is a clinical pharmacy practice that contributes to tackling the burden of medication discrepancies and subsequent patient harm at care transitions. However, it is a time-consuming activity. A collaborative work with the <i>‘Institut de l’Intelligence Artificielle en Santé’</i> led to the development of an algorithm to prioritise MRA targeting patients most at risk for unintentional discrepancies (UD). <h3>Aim and Objectives</h3> This study aims to evaluate the algorithm’s performance before considering its routine use. Main objective: To compare the number of UD identified before and after implementation of the algorithm. Secondary objective: To compare the proportion of patients with a UD that has a significant clinical impact (score &gt;1C according to the CLEO scale, developed by the <i>‘Société Française de Pharmacie Clinique’</i>) before and after implementation of the algorithm. <h3>Material and Methods</h3> A monocentric retrospective study was conducted in medicine-surgery-obstetrics departments. The number of MRA and the number of UD were collected from 2 April 2024 to 31 May 2024, period without algorithm (control group), and from 3 June 2024 to 31 July 2024, period with algorithm (test group). Patients hospitalised for less than 3 days and unable to understand the aims of the study were excluded. The proportion of UD per MRA and the clinical impact of UD according to the CLEO scale were compared between the two groups using a Z-test for comparison of proportions. <h3>Results</h3> A total of 255 MRA were performed in the control group and 395 in the test group, with respective rates of 0.18 and 0.39 UD per MRA (Z = 5.46; p &lt; 0.001). The clinical impact was significant for 48.9% (n = 22) of UD in the control group, and 48.3% (n = 72) of UD in the test group (Z = 0.07; p = 0.947). <h3>Conclusion and Relevance</h3> The algorithm allows the identification of more UD, however their clinical impact according to the CLEO scale is not significantly different between the two groups. To improve it, variables that allow for targeting patients with clinically significant UD must be identified. Furthermore, the algorithm must be tested on a larger scale. <h3>References and/or Acknowledgements</h3> <h3>Conflict of Interest</h3> No conflict of interest