CMS recently gave a big hint at the future in quality improvement and measurement. The CMS QualTech competition solicitation targets AI to help CMS with four opportunity areas:
- Surveys of charts and providers
- Annual wellness visit participation
- Digital quality measures
- The various hospital dashboards for readmissions, hospital-acquired conditions, and value-based purchasing
CMS is looking for eight finalists to present their work at a November 2026 event.
CMS’s Use of AI in Quality Measurement
What’s interesting is CMS’s use of AI as the forcing function to likely greatly expand the scope of quality measurement. Historically, CMS quality measurement for all of the inpatient and outpatient programs has had small measures of highly curated variables – maybe most famously the HEDIS measures which play a key role in Medicare Advantage Star Ratings and payments. When viewed through the scope of AI, these measures are literally a tiny handful of parameters. Practicing clinicians have been concerned that the limited scope of quality measures does not reflect the complexities of real-world clinical care.
Modern large language models (LLMs) powered by AI are, in large part, impressive because of the massive number of internal variables and parameters used. While AI model parameter numbers are closely guarded secrets, industry estimates suggest Claude Opus 4.5 has on the order of 2 trillion parameters. There are on the order of 50 parameters measured for CMS MA Star Ratings. So, there may be a roughly 40 billion-fold difference in analytic scale between CMS’s quality measurement today and when CMS uses modern AI.
How might the application of great analytic granularity change the nature of quality measurement? Economics will play a key role here. Researchers have identified quality measurement compliance costs of $15 billion annually and massive stock price swings for Medicare Advantage payers based on Star Ratings, suggesting massive downstream impacts on ratings performance. Currently having so few variables determining Star Ratings statistically suggests potentially large randomness in the scores.
AI may totally change the nature of quality measurement by looking at the entire patient record. While measures can still be gamed, it will be much harder to do and the classic cut and paste note generation tools will become obsolete. Increasingly deep biologic variables such as the classic hemoglobin A1c over time and new technology-enabled data, such as adiposity scores from cross sectional imaging and physical functioning based on cell phone accelerometers, can factor into care quality measurement.
What CMS’s Use of AI in Quality Measurement Means for Payers
What are the implications for payers as CMS starts incorporating AI into quality measurement? More fundamental measures require a shift from process-focused measures to targeting the patient’s deeper biology. Payers will need to do far more to change patient behavior. Payers have increasingly better tools to manage body mass, blood pressure, lipids, and inflammation, but doing so will require a new level of targeted digital engagement with patients (and their providers). Payer enterprise software will need to support not just the claims pipeline, but thinking strategically about clinical data. As just discussed, thinking about clinical data is a true big data problem (true even without new AI workloads).
Payers will need to not just store larger amounts of clinical data but compute and communicate with that data. Massive data and high-performance APIs to support that compute shape the winning payer strategies in the next decade. It is worth noting that CMS is encouraging not just AI quality compute tools, but the APIs which will be used to communicate care management changes. AI provides the crucible to force these changes in the quality measurement world and to do so in the near future.
You can learn more about other changes to digital quality measures by watching our recent webinar, “Beyond Claims: How Clinical Data and FHIR APIs Are Reshaping Digital Quality Management.”
