Investigating providers suspected of billing for fraudulent procedures can be a costly and time consuming process. The goal for this project was to identify and quantify fraud, waste, and abuse indicators for a Medicare and Medicaid Dental Insurance client so that they could rank potentially fraudulent providers and target them for appropriate interventions. Having explainable scoring was a key component of success, since the model results would be used as evidence to warrant opening an investigation.
When developing analytical models, there is often a trade off between interpretability and accuracy of the results. For this client, the solution needed to be interpretable, and more importantly needed to incorporate relevant and actionable results that could be used to assist in investigations. Elder Research utilized ensembles to achieve explainable and actionable results while minimizing sacrifices in accuracy. Dr. Elder was one of the first to discover that combining different methods or algorithms into an ensemble usually outperforms a single algorithm, an approach that has become a best practice for predictive analytics. Our analytics solution enabled investigators to focus their time and effort by automatically generating a list of prioritized cases for review.
Based on the available resources, the client wanted a solution that would provide a ranked list of 50 service providers to target for fraud investigation. Our deployed solution increased the hit rate from 5% chosen at random to 48% for the top 50 riskiest providers.
Providers identified as the top 50 based on risk score were nearly ten times more likely to be fraudulent than a provider selected at random. These results enabled the client to provide targeted interventions, recoup fraudulent charges, educate providers on correct billing procedures, and ultimately reduce Medicare and Medicaid spending.