Abstract
OBJECTIVE: To systematically review the utilization of artificial intelligence (AI) in health literacy, highlighting limitations and future developments. METHODS: A systematic review, following PRISMA guidelines, was conducted searching 6 databases for studies published from January 1, 2014, through April 10, 2024. Data extracted included population characteristics, health literacy definitions and measurement, study objectives, AI techniques, and metrics. Risk of bias was assessed using an adapted checklist. RESULTS: From 1296 studies, 18 (1.4%) met inclusion criteria. These studies primarily evaluated text-based materials, including online articles, and electronic health records, with most materials in English, but also incorporated other languages. Artificial intelligence played various roles, including evaluating complexity, text simplification/readability enhancement, translation, and question-answering. Only 5 studies involved participant engagement. Seven studies provided a health literacy definition, consistently describing it as an individual's ability to obtain, understand, and use health information for informed decisions, often linking it to external factors. However, only 1 study incorporated an individual level health literacy measurement tool, whereas organizational level health literacy measurement remained largely overlooked. The AI techniques used included traditional machine learning, deep learning, and transformer-based models. Evaluation metrics were categorized into human evaluation, readability, and machine learning metrics. CONCLUSION: The review highlights AI's dynamic application in relation to health literacy; however, measurement of health literacy, at both an individual and organizational level, to evidence AI's effectiveness remains limited. In addition, future work should not only measure health literacy outcomes more rigorously but also pursue research on enhancing AI model performance, robust evaluation, and their practical implementation in real-world settings.