Medicine recommender system

Abstract

The advances in recommender systems have motivated researchers to explore these applications in the healthcare domain. Adopting electronic health record databases in hospitals has enabled the use of medicine recommender systems as a clinical decision support tool. These systems alleviate the complexity of drug-related choices faced by healthcare professionals by generating a list of personalised prescriptions based on a patient’s data. However, we have identified some technical limitations in the encountered literature, such as missing out on effective recommender system techniques and not evaluating their models on patients with limited data. To address these issues, we generate models with popular recommender system techniques, such as Collaborative Filtering and implement several machine learning techniques to improve personalisation and precision performance. We have also compiled a more realistic dataset from the publicly available MIMIC-III database, including more medical codes, patients with limited data, and a time-based split up to standard with other recommender system datasets. However, the encountered research faced some practical limitations in the literature regarding the applicability and has not thoroughly considered these systems for real-life implementation. Therefore, we have a discussion with a medical expert to understand what direction these systems need to take to be used and applicable in real-life scenarios. Overall, medicine recommender systems are an exciting field that could lead to better clinical outcomes and improved quality of care.