Fraud, waste, and abuse (FWA) in healthcare claims pose challenges to the accessibility of healthcare, adds a burden to health expenditures, and diminishes the quality of care. There is a wide variety of fraudulent behavior among providers, which necessitates lengthy investigative efforts. Timely identification of providers who share traits with other high risk or potentially fraudulent providers is of great importance to help curb FWA and expedite investigations. In this paper, we experiment with a graph topology built on healthcare claim-level features and provider risk indicators. Graph neural networks are used to compute provider node embeddings from this graph. For each provider in the graph we compute pairwise similarity using vector similarity methods, and create a rank order of providers who are similar to a given provider. We compute rank-based performance metrics on the predicted rankings and compare them to subject matter specialist rankings to show the effectiveness of our models in discerning providers with similar risk profiles versus those who are dissimilar.