Clinical AI's week: a trial proves it works, a startup says aim it at regulation, Medicare sets the pay

The proof: a trial, not a demo

Clinical AI got its strongest evidence of the week from a trial rather than a product launch. The PARAMOUNT trial, which the company Cleerly describes as the first randomized trial of AI-enabled plaque-based care, met its primary endpoint, with results presented at the AHA Scientific Sessions 2026. It was a prospective, multicenter, open-label randomized trial of outpatients with suspected coronary artery disease, elevated LDL cholesterol and no prior history of the disease. Between January 2025 and February 2026, 530 participants at 21 sites were randomized to plaque-based care guided by AI-QCT or usual care guided by functional stress testing; 523 were analyzed. m:…7542

The headline number is a win ratio of 2.00 on the prespecified hierarchical composite endpoint at 180 days (95% CI 1.37–2.56, p<0.0001). The largest contributor was improved certainty about whether a patient's chest pain was caused by coronary artery disease, with a win difference of 16.6 percentage points. Women made up 57% of the participants analyzed — Cleerly notes this is a rare majority in cardiovascular research, where women have historically been underrepresented. m:…7542

The practical results concern treatment decisions, not just diagnosis. In the AI-QCT-guided arm, lipid-lowering therapy use rose from 31% at baseline to 75% after testing, compared with 32% to 47% in usual care. Among patients who had invasive coronary angiography, 38% in the AI-guided arm did not go on to revascularization with PCI or bypass surgery, versus 60% in usual care. Plaque-based care did not increase overall rates of catheterizations or revascularizations, and significantly fewer patients without obstructive coronary artery disease were among those referred to the catheterization lab. Full results are expected in early 2027, and peer review is still unresolved — so these numbers come from the company's announcement of late-breaking results, not a published paper. m:…7542

The redirect: aim AI at regulation

While Cleerly presented trial results, Weave Bio argued drug-development AI is pointed at the wrong problem. The company argues that current AI investment focuses on data generation rather than regulatory planning, the primary driver of project failure — the company puts it plainly: "Regulatory interactions define the success of a drug." Most programs, the company says, are planned with the molecule outward rather than starting from regulatory requirements, and early planning missteps often surface only once regulatory scrutiny is applied, causing significant delays and costs. m:…4330

Weave Bio's answer is a tool called Program Planner, which reviews a development plan against a curated library of more than 300 parsed guidance documents from the FDA, the CFR, ICH, EMA and similar agencies, flagging what is missing or misaligned, with every finding tied to its source. The company's cautionary example: some organizations it has worked with ran several times more nonclinical studies for an investigational new drug application than needed, turning work meant to take months into years, with wasted costs estimated in the tens of millions. The expensive failure, in this argument, is not a lack of data but a plan never checked against what regulators will ask for. m:…4330

The payment: Medicare's 2027 rules are final

The third part of the week is the one that decides what any of this gets paid. CMS released the calendar year 2027 Medicare Physician Fee Schedule and Hospital Outpatient final rules on November 4, 2026, with most provisions taking effect January 1, 2027. The headline change is the conversion factor: $32.63 for clinicians who are not qualifying APM participants, down 2.31% from $33.40, and $32.95 for qualifying APM participants, down 1.83% from $33.57. On the other side, the rules replace the G2211 add-on with Modifier B1, appended to the E/M base code, which raises its payment by 16%, while Modifier B2, for clinicians in a Medicare Shared Savings Program ACO or the LEAD Model, raises the visit payment by 32%. m:…5018

The rules also rewrite payment in adjacent areas. Payment for 340B drugs moves from ASP plus 6% to ASP minus 28%, based on CMS's drug acquisition cost survey. Same-day E/M visits billed with modifier 25 alongside a 0-day global procedure are paid in full for the higher-paid service and at 80% for the other. Site-neutral payment expands from drug administration to certain imaging services without contrast at grandfathered off-campus hospital outpatient departments, at 40% of the OPPS rate. Hospital outpatient prior authorization expands to eight more botulinum toxin injection codes starting July 1, 2027. The mandatory Ambulatory Specialty Model begins its first performance year on January 1, 2027, with 2027 performance setting a Part B payment adjustment of up to plus or minus 9% in 2029, while the rule paying for remote patient monitoring and remote therapeutic monitoring only when the practice's own employed clinical staff do the work was finalized with a two-year delay, taking effect January 1, 2029. m:…5018

What happens next

Read together, the three parts describe a sequence, not three separate stories. PARAMOUNT is the proof that AI-guided care can change treatment decisions in a randomized setting; Weave Bio's argument is that the same technology should be aimed earlier, at the regulatory plan that decides whether a drug programme survives; and the CMS rules are the payment reality that will decide whether clinics can afford to adopt either. Open questions remain: peer review of the PARAMOUNT results, with full publication expected in early 2027, and how revenue cycle teams handle the new modifiers, the 340B change and the Ambulatory Specialty Model from January 1, 2027. The trial proved the care model; the payment rules will decide how fast it spreads. m:…7542 m:…5018