In 2024, the Centers for Medicare & Medicaid Services (CMS) estimated Medicare fee-for-service (FFS) improper payments were $31.7 billion, or 7.66 percent of its budget. Although CMS undertakes numerous activities to combat these improper payments, most of them consider whether a service that was already paid is in error and how the resulting overpayment can be recovered. This retrospective “pay-and-chase” approach, while valuable, can be complemented by prepayment reviews that prevent improper services from being paid in the first place. CMS is well positioned to build on its past success in this area. Coupled with analytics and artificial intelligence (AI), these prepayment review programs have significant potential to realize substantial savings and reduce improper payments.
One of CMS’s most successful demonstration, or pilot, programs involved the implementation of a targeted prior authorization program—where approval is required before a service or other benefit will be reimbursed—in a handful of states. The program saved approximately $1 billion over four years. This program, the Repetitive, Scheduled Non-Emergent Ambulance Transport (RSNAT) model, examined the impact of prior authorization for certain non-emergency transportation services, an area that had been repeatedly identified as having high payment errors. RSNAT services that were not approved in advance were not reimbursed by CMS. In its work as the evaluation contractor for this demonstration, Mathematica determined that CMS was able to dramatically reduce expenditures and use without meaningfully impacting patient care. Based on this evidence and other considerations, CMS expanded the program nationwide.
Another type of prepayment review performed by government and commercial payers is prepayment claims review, which identifies records submitted to their claims processing system at high risk of error and stops the claims for human review prior to payment. These capabilities are in addition to internal or commercially available claims edits or rule-based algorithms that identify coding errors or logically impossible scenarios that warrant an outright rejection of the claim.
Both prepayment claims reviews and prior authorization programs can leverage AI and analytics, deep Medicare subject matter expertise, and an array of data sources to identify concerning scenarios that can be indicative of fraud, abuse, or errors. These prepayment review programs can use and integrate the following:
- Unsupervised learning initially when training data are limited
- More powerful supervised learning as these data mature
- Network models that examine relational data to identify organized fraud rings that other models may miss
- Text mining approaches to identify signals developed from call center transcript data and other text data sources
They can also consider patients’ clinical conditions and logical disease progression and identify services that appear to be inappropriate for the beneficiary at a given point in their healthcare journey. This process can identify suspect claims and provide human reviewers with the insight needed to enable an efficient review to determine whether the claim should be paid or rejected. Over time, this detection-review process, as it acquires a high volume of training data, continues to improve its ability to target high-risk claims and realize an even greater return on investment.
For all of the success of the RSNAT demonstration, it’s worth noting that CMS used a relatively limited set of criteria to determine which services required review before being authorized. The demonstration likely would have been even more successful if AI and advanced analytics—informed and guided by deeply analytical Medicare subject matter experts—were used to identify high-risk RSNAT providers and requests for services for review. Both prepayment claims review and prior authorization review programs can use and integrate these approaches to develop scoring systems that rank transactions by fraud risk and present these transactions for human review based on their expected return on investment.
There are several other types of services at significant risk for Medicare FFS improper payments. One example is home health, which has an estimated 6.7 percent FFS Medicare improper payment rate and has been a CMS focal area for additional prior authorization review. Although identifying these types of services is a useful next step, it may be even more important for CMS to use available data to (1) go deeper and identify the array of factors that are associated with high-risk claims for those types of service and (2) model and score all of the claims for that service type to identify the specific transactions that should be prioritized for prepayment claims review or prior authorization review.
To expand the use of prepayment reviews in Medicare FFS, CMS would first conduct demonstrations to determine whether the approaches generate savings without negatively impacting patient care. As an initial step, CMS could launch multiple high-potential demonstration programs while simultaneously conducting cost-effective rapid evaluations. These evaluations would allow for iterative testing and real-time improvements and help identify which demonstrations are effective and suitable for nationwide implementation.
Prepayment reviews hold significant potential to help CMS achieve savings more quickly. CMS has already had recent success with the RSNAT demonstration program. Now, the agency can build on that experience and expand the scope and scale of prepayment reviews, making them more common and accepted by stakeholders in the traditional Medicare program. Cost-effective and timely evaluations can play a key role as well, helping CMS quickly assess which demonstration programs of prepayment reviews are working, how they can be improved, and whether they should be expanded nationwide.
