Prediction of medication risks and drug-related problems – a novel pharmacoepidemiological approach using real-world data, decision support algorithms and machine learning

Research problem and specific questionsDrug-related problems (DRPs) occur frequently and is a common cause of hospitalizations and death. Many DRPs may be prevented by using clinical decision support systems in health care.

In Sweden we have Janusmed knowledge databases, developed as decision support to detect potential drug-drug interactions, side-effects, and inappropriate dosing.

There is a need for better knowledge about the prevalence of these potential DRPs, how well current knowledge databases can predict actual problems, and how they can be improved.

The research project has three work packages to answer the following research questions:What is the prevalence of potential DRPs identified using Janusmed algorithms in the region of Kalmar, including drug-drug interaction, additive pharmacological effect and renally inappropriate medicines?How well can Janusmed algorithms predict actual DRPs, measured as correlation between calculated risk levels and clinical outcomes?Can machine learning be used to combine algorithms from Janusmed knowledge databases with additional factors to improve the prediction of DRPs?Data and methodThe core of the project is two sets of data which will be the same for all three WPs:(1) The real-world data primarily from the electronic health record in region of Kalmar including data on a population of approximately 200,000 patients for 10-years, covering both hospitals and primary care. (2) The rules and algorithms from three Janusmed knowledge databases (Interactions, Risk Profile, and Renal function).

WP1 will be a descriptive cross-sectional study, WP2 a nested case-control study, and WP3 will develop and evaluate new machine learning models.Plan for project realisationThe research project is carried out in an interdisciplinary team with broad competence including clinical pharmacology, pharmacoepidemiology, computer science, statistics, and visual analytics among other things.

Most of the cost for this project is salary for involved researchers.RelevanceDRPs is a major problem for society, especially with an ageing population and increasing use of medicines.

Knowledge from the current project can be used to improve current decision support algorithms, as well as provide new insights about how advanced technology such as artificial intelligence can be used to improve predictions. Improved predictions can lead to improved patient safety and reduced societal costs.