Machine Learning Techniques for Predicting Medicare Claim Denials and Improving Claims Management

Authors: Veeravaraprasad Pindi

Country: India

Full-text Research PDF File:   View   |   Download

Abstract: This paper explores how machine learning techniques can be applied in optimizing provider claims and preventing claim denials. Each of these techniques was applied in a real-world operational setting, and we report specific steps, models, and unique code segments used in implementing these approaches. Unlike fraud detection, which typically focuses on very small numbers of serious intentional adversarial behaviors, these techniques are designed to address a broad set of operational issues which, when added together, cause a large number of claim denials. This often results in an industry "cat and mouse" game with providers, as new operational issues identified are remedied and gradually new reasons for denials chip away at the denial problem rate. Healthcare providers and facilities are constantly struggling to manage and monitor the status of their claims, as each denial represents a loss and increases their administrative burden. Predicting the likelihood of claim denials can help providers and facilities manage claims more effectively by intervening with problem claims early in the process [1]. However, traditional regression models that have been used for prediction do not efficiently evaluate the complex relationships embodied in healthcare claims. Utilizing advanced machine learning algorithms, such as the gradient boosting decision tree ensemble method, can solve these issues and increase prediction performance while accounting for the many interactions and the high-dimensional nature of healthcare claims. Fostering high-value care has increased due to the fee-for-service (FFS) payment system [1]. Medicare FFS claims contain a multitude of services for its beneficiaries and turnover revenue for healthcare facilities. Each claim represents a unique interaction, but facilities are often burdened by the large volume and difficulty associated with monitoring the status of claims. As a result, many claims are unresolved because of poor management and oversight, triggering claim denials that lead to lost revenue for the provider and increased administrative burden. A 2011 study on denied claims found that 4% to 9% of claims submitted by U.S. hospitals are initially denied, representing 3% to 5% of their revenue. More recently, the Medicare Payment Advisory Commission Data Book reported that in 2017, Medicare FFS had an estimated 89% accuracy rate for claims, but the remaining 11% were inaccurate and resulted in improper payments, accounting for $31 billion in the fee-for-service program [2]. In order to reduce the burden of claim denials on medical providers and help patients receive medical treatment in a more timely manner, it is necessary to develop a system for predicting claim denials that can help medical providers identify and correct potential issues with claims before they are submitted to Medicare.

Keywords: Medicare Claim Denials, Predictive Analytics, Claims Management, Machine Learning, Supervised Learning, Unsupervised Learning, Anomaly Detection, Natural Language Processing (NLP), Healthcare IT Systems, Predictive Models, Regression Analysis, Neural Networks, Model Performance Evaluation

Paper Id: 1778

Published On: 2016-05-09

Published In: Volume 4, Issue 3, May-June 2016

Cite This: Machine Learning Techniques for Predicting Medicare Claim Denials and Improving Claims Management - Veeravaraprasad Pindi - IJIRMPS Volume 4, Issue 3, May-June 2016.

Share this