Development of an Artificial Intelligence powered Medication Risk Score calculator application (AUGMERIS)

(Másodközlés)

Abstract

Introduction: Identifying high-risk therapies is a primary task for clinical pharmacists. The Danish Medication Risk Score (MERIS) is an effective tool for this purpose, but its automated application is restricted by unstructured therapy data in electronic health records (EHR). Our aim was to develop a web application (AUGMERIS) enabling risk assessment through hybrid text processing.

Methods: We developed a hybrid model in Python combining strict rule-based algorithms with GPT (Generative Pre-trained Transformer) technology for drug name recognition. The system was tested on low-quality domestic hospital data (N=2999) and data from a globally available web application (N=383), and compared with expert evaluations.

Results: Validation showed a moderate, significant correlation between system-calculated scores and clinical pharmacists’ evaluations (Spearman r=0.463 and 0.496; p<0.05). AI was needed for 25.8% of tests conducted worldwide, compared to 10.3% in the Hungarian sample.

Conclusions: The AUGMERIS application efficiently processes unstructured therapy data, supporting risk-based patient screening, but AI integration is essential for handling international and heterogeneous data sources.