(188) How Has Machine Learning Been Used in Patient Safety?

Publication Number: 519

Background: In patient safety, the use of machine learning is expanding in multiple areas including safety operations, signal management, and identification of target populations. However, the information has not been consolidated, which limits leaders’ ability to identify gaps in patient safety.

Objectives: To describe the use and applications of machine learning in patient safety through a systematic literature review.

Methods: A non-quantitative systematic literature review was performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The inclusion criteria were articles on machine learning used in patient safety, articles on modeling algorithms developed and used for safety signal identification, characterization, assessment, or management, and articles in the English language. Articles with incomplete information, letters to editors, notes, commentaries, and duplicated articles were excluded. The screening was completed in two levels: title/abstract review and full text review. Each level of screening was completed independently by separate researchers.

Results: Sixty-six articles were included in the final review. The sources of articles included MEDLINE (n = 8), the FDA Adverse Event Reporting System (n = 6), Side Effect Resource (n = 6), and social media platforms (n = 6). The main uses of machine learning in patient safety were in the identification of adverse drug events and reactions (57.6%), processing of safety reports or clinical narratives (21.2%), extraction or prediction of the effects of drug-drug interactions (7.6%), identification of populations at high risk of experiencing drug toxicity or guidance for personalized care (7.6%), prediction of drug side effects (3.0%), simulation of clinical trials (1.5%), and integration of prediction uncertainties into diagnostic classifiers to increase patient safety (1.5%). Some of the machine learning methods identified were the use of a natural language processing system to detect potential adverse events and the creation of a model based off laboratory tests to detect patients that may have had an adverse drug event linked to a drug-drug interaction.

Conclusions: The most frequent use of machine learning in patient safety is to gather information on adverse drug reactions or adverse drug events. The natural language processing systems and models for adverse drug events were able to capture adverse events or drug-drug interactions earlier than traditional methods. As machine learning is evaluated more in various databases, its application and scalability may prove to be paramount to patient safety.