Depression, as a highly prevalent mental disorder, is traditionally diagnosed using subjective scale-based assessments, which suffer from limitations such as insufficient objectivity and delayed early warning. With the widespread adoption of electronic health records and the rapid advancement of artificial intelligence technologies, data-driven approaches for depression prediction have become a major research focus. This paper systematically reviews the evolution of depression prediction methods, spanning from traditional assessment scales to machine learning and further to deep learning techniques. The paper places particular emphasis on the current applications and unique advantages of graph neural networks (GNNs) in this domain. Existing studies indicate that GNNs can effectively model complex relationships among patient features, thereby improving predictive performance while enhancing model interpretability and offering a novel technical pathway for early detection of depression. However, several challenges remain, including the lack of standardized graph construction methods, issues related to data privacy and quality, and insufficient model generalizability. Future research is expected to focus on constructing multi-center graph datasets, developing interpretable GNN models, and promoting their practical application in public health screening, ultimately advancing mental health services toward a “data-driven and proactive prevention” paradigm.