CMS says go faster on AI šļø, AI coding adds $2.3B in excess spend šø, Pancreatic cancer's AI baby step š¬
CMS Plans to Fast-Track New Health Technologies, Including AI
CMS Administrator Mehmet Oz told the U.S. News Healthcare of Tomorrow conference yesterday that the agency is searching for āmore innovative waysā to get health tech and AI to patients faster.
He said CMS wants to be ājudicious in how we restrict these programs from prospering.ā
Thatās the largest payer in the country telling builders the bureaucratic friction theyāve been complaining about is about to get smaller.
CMS has already been moving. The HealthTech Ecosystem initiative launched its first wave of digital health tools. The ACCESS Model is offering recurring payments for chronic disease tech. CMS and FDA have proposed faster Medicare coverage pathways for breakthrough devices.
This isnāt talk. CMS is restructuring around technology adoption ā and that creates a specific window for clinician-builders who can demonstrate clinical value.
š¤ āOz is a showman, not a regulator. This is rhetoric.ā The HealthTech Ecosystem has 600+ organizations signed up and tools going live. Whether you trust the messenger or not, the institutional machinery is moving.
š¤ āFast-tracking AI in healthcare means less safety.ā The alternative ā a three-year coverage determination process while AI tools proliferate outside Medicare ā isnāt safer. It just means the tools reaching patients arenāt the ones CMS evaluated.
š¤ āCMS fast-tracks what saves money, not what improves care.ā Sometimes. But the ACCESS Model specifically pays recurring fees for chronic disease tech based on outcomes. If your tool demonstrably helps manage diabetes or hypertension, thereās a CMS payment rail for it now.
š” Builderās Radar
Why AI Will Accelerate Health Care Inflation
A Health Affairs analysis drops the number that should be on every builderās wall: the Blue Cross Blue Shield Association found AI-enabled coding added $2.3 billion in excess spending at hospitals ā driven by diagnoses coded at higher intensity than the clinical record supported.
The question isnāt whether AI raises costs. It will. The question is whether the payment system can tell the difference between inflation that buys better care and inflation that doesnāt.
The builderās dilemma in one stat: the same technology that improves documentation accuracy also optimizes billing intensity.
š¤ āThatās a coding problem, not an AI problem.ā The upcoding incentive existed before AI. But AI scales it from human-speed to machine-speed. The BCBSA didnāt flag $2.3B from manual coders.
š¤ āSo we should slow down AI adoption?ā No. Build the eval that catches it.
The Pancreatic Cancer āMiracle Drugā Is Just a Baby Step
Andrea Califano and Gideon Bosker argue in STAT today that daraxonrasib ā which doubled pancreatic cancer survival from 6.6 to 13.2 months at ASCO ā is the opening move, not the finish. The next advances will require AI to navigate the diseaseās molecular heterogeneity at a scale no individual researcher can manage manually.
The pattern is bigger than oncology: the first-generation win proves the concept, but the next 10x requires computation.
š” 80/20: Watch the clinical trial matching infrastructure being built around conversational AI. The bottleneck in oncology trials isnāt finding drugs ā itās finding the right patients. Conversation-derived phenotyping may move faster than genomic matching for enrollment.
OffPlan Raises $2.5M Seed for DPC-Driven Health Plans
OffPlan is building direct primary care memberships plus transparent specialty pricing plus catastrophic coverage for employers tired of the traditional insurance model. David Hardin ā a physician and clinician-builder ā is CMO. The thesis: stop using insurance for routine and predictable care. Worth watching because the model is structurally different from both FFS and VBC.
Dexcom Wins First OTC CGM Clearance for Children
FDA cleared Dexcomās Stelo Glucose Biosensor for OTC use in children ages 2+ who donāt use insulin. First OTC continuous glucose monitor for pediatric use ā 15-day wearable sensor, smartphone app. For builders in metabolic health: the data exhaust from OTC pediatric CGM creates a new design surface for family-facing health tools.
šļø From the Pods
Nineteen states have Medicaid enrollment approved for people in jails and prisons. Washington Stateās model is the template. But H.R. 1 is complicating expansion ā Oregon, Rhode Island, and Michigan pausing; Louisiana trimming benefits; Californiaās renewal in doubt. The clinical reality: overdose death rates are 100x higher in the two weeks after release.
š Speaker Blindspot: Status quo bias ā the episode frames the policy debate as āexpand vs. donāt expandā but spends almost no time on the technical infrastructure question. Even in states that approve enrollment, how does health data follow the person from jail to community? The policy wins without the plumbing are incomplete.
š” BTW: Andrea Califano ā whose STAT piece on pancreatic cancerās AI future published today ā started his career as a physicist in Naples, then spent a decade building computational biology tools at IBMās Watson Research Center before founding Columbiaās Department of Systems Biology. His labās algorithms helped identify daraxonrasib, the drug that just doubled pancreatic cancer survival.
š Upcoming: CMS HL7 FHIR Connectathon, July 14-16 (free, virtual).
What are you building this week? Email and tell me (kevin@clinicians.build) ā I read every one.
ā Kevin


