DATA-DRIVEN GRAPH NEURAL NETWORK MODELS FOR DETECTING FRAUDULENT INSURANCE CLAIMS IN HEALTHCARE SYSTEMS

Abstract

Fraudulent insurance claims remain one of the most persistent challenges in healthcare systems worldwide, draining billions of dollars annually and undermining the sustainability of both public and private insurance frameworks. Traditional fraud detection approaches, such as manual audits and rule-based or classical machine learning models, have demonstrated limited effectiveness in identifying the complex, relational, and often collusive nature of fraudulent activities. In response, recent research has increasingly turned toward graph neural network (GNN) models, which are uniquely suited to represent healthcare claims as interconnected networks of patients, providers, institutions, and transactions. This systematic review and meta-analysis, conducted under the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, examined a total of 62 peer-reviewed studies that applied GNNs or closely related graph-based methodologies to healthcare fraud detection. Collectively, these studies reported consistent improvements in accuracy, precision, recall, and interpretability, with GNN models frequently outperforming traditional approaches by margins of 10–20 percentage points. The reviewed literature also revealed methodological innovations such as hybrid GNN architectures, temporal graph learning, and privacy-preserving designs, underscoring the adaptability of these models to diverse healthcare contexts. Additionally, the global distribution of research—from North America and Europe to Asia and emerging markets—demonstrated the broad applicability of GNNs across different healthcare financing structures. While challenges remain in terms of scalability, interpretability, and data quality, the evidence strongly suggests that graph neural networks have matured into a robust and reliable solution for detecting fraudulent healthcare claims. By synthesizing the insights of 62 studies and more than 3,800 citations, this review positions GNNs as a transformative advancement in healthcare fraud prevention, offering both economic protection and enhanced trust in healthcare insurance systems.