Exploratory pharmacovigilance with machine learning in big patient data: A focused scoping review

Area of Science:

• Pharmacovigilance
• Health Informatics
• Machine Learning

Background:

• Electronic patient records offer rich data for exploratory pharmacovigilance.
• Machine learning (ML) can operationalize this complex data for drug safety.
• Despite aspirations, ML applications in this field are not yet widely realized.

Purpose of the Study:

• To review the applications of machine learning and big patient data in exploratory pharmacovigilance.
• To identify trends and methodologies in ML-driven drug safety research.

Main Methods:

• Systematic search of PubMed and Embase databases.
• Inclusion of original articles on exploratory pharmacovigilance using ML in electronic patient records (≥1000 patients, post-market entry).
• Analysis of study characteristics, ML methods employed, and data scale.

Main Results:

• Seven studies met the inclusion criteria, published between 2015-2021 across six countries.
• Common ML methods included random forests, logistic regression, and support vector machines; neural networks and naive Bayes were also used.
• Most datasets contained thousands of patient records; two utilized big data (>1 million records).

Conclusions:

• Few studies have successfully translated the potential of ML and clinical data into practical exploratory pharmacovigilance outcomes.
• The field faces challenges in realizing the full potential of these advanced analytical techniques for drug safety.
• Further research is needed to bridge the gap between ML capabilities and clinical data for effective drug safety monitoring.