This paper provides a comprehensive review of Graph Neural Networks (GNNs) in epidemic modeling, highlighting their growing importance in addressing the limitations of traditional mechanistic models. Traditional models often suffer from oversimplified assumptions, leading to sub-optimal predictive power and inefficiency in capturing complex relational information. GNNs, with their ability to aggregate diverse information through message-passing mechanisms, have emerged as a powerful tool for capturing relational dynamics within graphs, making them particularly suitable for epidemiological tasks. The paper categorizes epidemic tasks into four main categories: Detection, Surveillance, Prediction, and Projection, each with specific objectives and challenges. It also discusses the sources of data used in epidemic modeling, including demographic and health records, mobility information, online search and social media data, sensors, and simulated data. The construction of graphs for these tasks is further categorized into static and dynamic node features, static and dynamic graph structures. The methodologies of GNNs in epidemic modeling are divided into two broad categories: Neural Models and Hybrid Models. Neural Models focus on data-driven approaches, leveraging deep learning to uncover complex patterns in disease dynamics without explicit encoding of underlying epidemiological processes. Hybrid Models, on the other hand, integrate mechanistic epidemiological models with neural networks, combining the structured, theory-informed insights of mechanistic models with the flexible, data-driven nature of GNNs. The paper provides detailed illustrations of both categories, including spatial dynamics modeling, temporal dynamics modeling, and intervention modeling. It also discusses the integration of GNNs with mechanistic models for parameter estimation and mechanism-informed neural models, enhancing the accuracy and interpretability of disease forecasts. Overall, the paper aims to bridge literature gaps and promote the progression of the field by providing a systematic review of GNNs in epidemic modeling, highlighting current limitations and proposing future research directions.This paper provides a comprehensive review of Graph Neural Networks (GNNs) in epidemic modeling, highlighting their growing importance in addressing the limitations of traditional mechanistic models. Traditional models often suffer from oversimplified assumptions, leading to sub-optimal predictive power and inefficiency in capturing complex relational information. GNNs, with their ability to aggregate diverse information through message-passing mechanisms, have emerged as a powerful tool for capturing relational dynamics within graphs, making them particularly suitable for epidemiological tasks. The paper categorizes epidemic tasks into four main categories: Detection, Surveillance, Prediction, and Projection, each with specific objectives and challenges. It also discusses the sources of data used in epidemic modeling, including demographic and health records, mobility information, online search and social media data, sensors, and simulated data. The construction of graphs for these tasks is further categorized into static and dynamic node features, static and dynamic graph structures. The methodologies of GNNs in epidemic modeling are divided into two broad categories: Neural Models and Hybrid Models. Neural Models focus on data-driven approaches, leveraging deep learning to uncover complex patterns in disease dynamics without explicit encoding of underlying epidemiological processes. Hybrid Models, on the other hand, integrate mechanistic epidemiological models with neural networks, combining the structured, theory-informed insights of mechanistic models with the flexible, data-driven nature of GNNs. The paper provides detailed illustrations of both categories, including spatial dynamics modeling, temporal dynamics modeling, and intervention modeling. It also discusses the integration of GNNs with mechanistic models for parameter estimation and mechanism-informed neural models, enhancing the accuracy and interpretability of disease forecasts. Overall, the paper aims to bridge literature gaps and promote the progression of the field by providing a systematic review of GNNs in epidemic modeling, highlighting current limitations and proposing future research directions.