Umeå universitets logga

The integration of Artificial Intelligence (AI) technology in mental healthcare addresses the widespread need for accessible and individualized care. However, the rapid evolution of AI tools presents research challenges, emphasizing the need for rigorous and adaptable methodologies. This systematic review provides an overview of experimental practices in AI and anxiety research, focusing on methodological considerations (e.g., design, procedures, ethical constraints) and general trends (e.g., constructs, limitations) of experiments including direct human participation. Following PRISMA guidelines and using PICOS to specify inclusion and exclusion criteria, two independent researchers conducted a search in Scopus in March 2024, resulting in 22 included articles. Data extraction was performed using a tested and adjusted protocol aligned with the research questions. The review found that ethical approval is uncommon, and resource requirements decrease with advanced AI and internet technology, enabling more remote administration. However, these interventions were frequently executed by third parties with no mention of standardized protocols. There was also an overrepresentation of situational anxiety, particularly learning anxiety. Additionally, the sensitivity of AI systems to anxiety is variable, often lacking integration of psychological theories. Future research should enhance ethical oversight, extend intervention durations, adopt more generalizable study designs, and evaluate the sensitivity of AI tools according to the operationalization of anxiety explored.