Effectiveness of AI-Assisted Screening Tools for Depression and Anxiety in Primary Care: A Systematic Review

Abstract

This systematic review evaluates the effectiveness, diagnostic accuracy, and implementation outcomes of artificial intelligence (AI)-assisted screening tools for depression and anxiety in primary care settings, with a focus on literature published since the advent of advanced large language models. A systematic search was conducted across PubMed/MEDLINE, Embase, PsycINFO, and Cochrane Central from January 2020 to September 2025. Ten studies met the inclusion criteria, comprising randomized controlled trials, pragmatic trials, and validation studies. Included studies demonstrated that AI-assisted tools (including conversational chatbots, machine learning models using voice and behavioral data, and digital decision-support systems) can achieve high diagnostic accuracy, with balanced accuracy up to 88.2% and Area Under the Curve (AUC) up to 0.82. Clinically, AI-assisted interventions were associated with higher depression remission (e.g., 27.3% vs. 12.0%) and better anxiety outcomes than treatment-as-usual in individual trials. Feasibility and acceptability were generally favorable, with completion rates often exceeding 60-70% and user satisfaction reaching 79%, even among older adults. Key challenges included limited external validation, demographic homogeneity, and insufficient reporting on ethical safeguards. AI-assisted screening tools show potential to improve the detection and management of depression and anxiety in primary care, with encouraging accuracy, clinical effectiveness, and user engagement in the available studies. However, the evidence base is small and heterogeneous, and future research needs to address gaps in generalizability, long-term efficacy, ethical oversight, and seamless workflow integration to ensure equitable and responsible implementation.