Artificial Intelligence Audits Are Happening Now

Healthcare providers are starting to see the first claim audits based on analysis and determinations made by artificial intelligence (AI).

Although the technology is new, many of the issues remain the same. Especially when the companies that develop AI-based audit tools sell these tools and services to commercial insurance companies, AI-driven audits increasingly resemble audits of Medicare providers and suppliers performed by the Recovery Audit Contractors (RACs) or Unified Program Integrity Contractors (UPICs).

RACs and UPICs are Medicare contractors charged by the Centers for Medicare & Medicaid Services (CMS) to identify overpayments and underpayments made to providers and to facilitate return of overpayments to the Medicare Trust Fund. Primarily, RACs accomplish this by conducting audits and issuing repayment demands.

RACs are different from other types of Medicare contractors that conduct audits because RACs are paid on a contingency fee. That is, RACs receive a percentage of any funds they extract from providers, potentially making them significantly incentivized to deny claims and demand repayment even where there is no clinical or legal basis to do so. Sometimes it is difficult to assess whether an audit is being conducted by a RAC or a UPIC, which is important to know because different rules and regulations apply. For example, a RAC audit can only go back for review three years, whereas a UPIC can go back longer.

Sometimes, CMS disperses multiple contracts to one entity. CMS has come out and said, for example, that services such as blood transfusions will be audited. CMS guidelines mandate that blood transfusions administered in the hospital outpatient or physician office setting be billed with a maximum of one unit per patient, per date of service.

CMS guidelines also mandate that intravenous hydration therapy administered in the hospital outpatient or physician office setting be billed with a maximum of one unit per patient, per date of service (excluding claims with modifier 59).

Similarly, because few insurance carriers have developed sophisticated AI tools in-house, they often contract outside technology companies to provide such tools (and often, to conduct the audits themselves). These outside contractors are motivated to deny claims and identify alleged overpayments in order to retain the business of the insurance carrier. This motivation is further enhanced where the outside contractor is paid a percentage of the alleged overpayments that their AI tool identifies. Therefore, any provider should carefully scrutinize any such audit findings, much as they would scrutinize the findings of a similarly motivated RAC.

Furthermore, AI-driven audits also can raise concerns about the competence of the reviewer. RAC audits are often criticized for utilizing under-qualified coders, nurses, or others to attempt to review the documentation and complex decisions of physicians and specialists. Any AI tool is only as good as the underlying data on which it is built and trained. It is difficult to know how any AI tool has been trained, because technology companies generally consider this information proprietary. However, an AI tool that reviews physician records may make the same mistake or misunderstanding over and over again, because it simply does not understand the content, context, or the decision-making it is attempting to review. Providers should carefully review all audit findings, especially where any questions exist regarding the qualifications or competence of the reviewer.

Plus, adding more companies into the Medicare and Medicaid provider audit arena adds more incompetence. Already, RACs and UPICS and Medicare Administrative Contractors (MACs) conduct subpar provider audits.

It’s like the game “Telephone:” the more people are added, the more distorted the message becomes.