Assess AI’s Auditing Capabilities Before Deploying the Technology
Question: What best practices should we implement when using artificial intelligence (AI) to perform audits? Texas Subscriber Answer: One of the most important places to start when deploying AI to perform audits is by removing sensitive information from the claims data. This means that any claims data you upload into an AI system or a large language model (LLM) should not have any protected health information (PHI) included. “I would make sure that the platform you’re utilizing allows the data to be desensitized or anonymized,” explained Christopher Parrella, CPC, CPCO, of Parrella Health Law, during his “Fraud in the Age of AI: The New Threat to Medical Billing Integrity” AAPC webinar. Then you can use the AI platform to identify patterns or abnormalities in the data. You can test the analysis of the dataset or audit the records compared to the AI’s output to see how accurately the AI system performs. “You can also upload the coverage and reimbursement criteria for some of the codes that may be highly utilized within your organization,” Parrella explained. Finally, you can upload some anonymized patient records to test the AI platform’s capabilities to evaluate compliance, documentation, medical necessity, and frequency of the code usage. If the system identifies any issues, you’ll have a human coder or auditor reviewing the accuracy of the AI’s outputs. Mike Shaughnessy, BA, CPC, Production Editor, AAPC
