The Centers for Medicare and Medicaid Services (CMS) is focused on addressing “fraud, waste and abuse” in health programs including Medicaid. Efforts span across different types of health coverage and across all provider and service types. In November 2025, CMS issued a letter to states describing opportunities for federal and state governments to work collaboratively. Efforts to address fraud, waste and abuse are not new. The Center for Program Integrity (CPI), within CMS, was established in 2010 to coordinate program integrity efforts and move from a “pay and chase” model to higher reliance on data analytics to detect and prevent fraud. CPI has worked with states to provide training through the Medicaid Integrity Institute and access to broad sets of complete data to help promote program integrity efforts.
On February 14, 2026, CMS released a dataset with provider-level spending data that the agency suggests could be used to identify unusual billing patterns for specific services, states, or providers. This policy watch describes what the data include, what they exclude, and how they could potentially lead to mistaken conclusions given the limitations of the data.
What do the data include and exclude?
The new dataset includes seven types of data:
- The national provider identifier (NPI) for the billing provider,
- The NPI of the servicing provider (which may be an individual or an organizational entity),
- The procedure code (also known as the healthcare common procedure coding system or HCPCS code),
- The month and year,
- The number of beneficiaries seen,
- The number of procedures delivered (the count of claims), and
- The total amount paid for the services.
The totals include records of outpatient services paid for by Medicaid directly (“fee-for-service”) and those paid for by Medicaid managed care organizations on behalf of enrollees between 2018 and 2024.
The data exclude all institutional records and all information about prescription drugs, which are significant shares of Medicaid spending, with hospital care accounting for 37% and being the single largest source of Medicaid spending. Beyond excluding entire categories of services, the data omit several types of information that are important in evaluating the reasonableness of service volume and spending:
- Enrollment. The amounts of services used depend on how many people are eligible to receive the services, which varies based on state policies, the economy, and people’s demographics. Differences in service use over time or between geographic locations may not be comparable without accounting for the number of Medicaid enrollees and their age and health status.
- Benefits and Coverage. The volume of services used also depends on what services states elect to offer and how they determine who is eligible to use those services, features that may change over time.
- Payment rates. Spending on services depends on how much states are paying for each service, which could depend on the local cost of living as well as state decisions about how much payment rates should be to ensure reasonable access to care.
- Diagnoses. The data do not include any information to indicate what condition the procedures are used to treat.
- Place of service and other modifiers. The data exclude information about where the services were performed (including whether they were provided in-person or remotely) and other modifiers that are used to identify characteristics of the services.
How might the data lead to mistaken conclusions?
Although data analytics can be a powerful tool to identify potentially problematic patterns, they could lead to mistaken conclusions if used in isolation. A few specific shortcomings of the new Medicaid data stand out.
- The reported procedures are not always comparable to each other. The data provide counts of patients, services, and Medicaid spending for each procedure, but the procedures are not in all cases comparable to each other. Some are very narrowly defined whereas others encompass a wide array of procedures. In CMS’ example of how the data could be used, spending on personal care appears to be a significant outlier and the single largest source of Medicaid spending. However, the “procedure” for personal care includes a range of possible services lasting from 15 minutes up to an entire day. In contrast, there are multiple separate procedures for psychotherapy that each correspond to visit length (30 minutes, 45 minutes, 60 minutes, etc.). Similarly, there are multiple procedures for emergency department visits and office visits that correspond to the complexity of the case on a scale from one to five. Personal care would not be the largest category of spending if institutional spending were also included.
- Providers are not always comparable to each other. The data provide counts of patients, services, and Medicaid spending for each provider, but some of the providers are individuals whereas others are group practices, clinics, or even entire county and state health departments. In CMS’ example of how the data could be used to look at spending by provider, 10 of the 20 largest “providers” are state or local government agencies that both administer and deliver Medicaid benefits rather than health care providers. States’ approaches to delivering Medicaid benefits vary broadly, but in most cases, state or local health departments contribute by paying providers to deliver services and by providing services directly (particularly in the case of services for people with behavioral health needs and developmental disabilities).
- There is limited information about the methods used to create the dataset or the underlying data quality. The data do not indicate how underlying Medicaid data were aggregated to create the summary file. They also do not address issues related to the quality of the underlying data, which come from the Transformed Medicaid Statistical Information System or T-MSIS. T-MSIS is a rich source of data, but sometimes there are issues with the data in specific states for specific topics. CMS maintains a rigorous “data quality atlas” that supports “insightful, methodologically sound analysis” of the data. The atlas provides a wealth of information about potential data problems. It’s unclear how CMS handled those issues when generating the recently-released data. For example, CMS reports that in the 2024 data, there were six states with unusable information about total spending for the services included in the data; and an additional 16 states where the data were of high concern. It is unknown whether CMS included the unusable data in the public file or whether the public data reflect a different version of T-MSIS.
Beyond those data issues, the data are missing context about how Medicaid spending and use of care changed between 2018 and 2024. The COVID-19 pandemic, which began in 2020, resulted in major changes to Medicaid spending that resulted from increased enrollment during the continuous enrollment period and increased awareness of unmet needs for behavioral health and long-term care. As states increased access to those services, use of those services and associated spending grew due to changes in states’ policies regarding coverage, eligibility, and provider payment rates.
