Diagnostic-agent reimbursement — first CPT code from CMS.

The AMA CPT Editorial Panel deliberations

The American Medical Association's CPT Editorial Panel is the body responsible for developing and maintaining the Current Procedural Terminology code set, the standardised vocabulary that defines what physicians and other clinicians can bill for in the US healthcare system. The Panel comprises seventeen members drawn from the AMA's House of Delegates, the major medical specialty societies, and the relevant healthcare delivery organisations, and it meets quarterly to consider Coding Change Applications submitted by specialty societies, manufacturers, or other stakeholders. The Panel's deliberations are technical, structured, and consequential: a CPT code is the gateway to reimbursement, and the absence of a CPT code is the structural barrier that has, until now, prevented clinical AI deployment from being a distinctly billable activity. The five-cycle deliberation on the diagnostic-agent code is the longest the Panel has ever held on a Category III code application and reflects the structural complexity of defining a billable activity whose intellectual content varies across specialties and across the range of FDA-cleared diagnostic agent capabilities available in operational practice.

The Coding Change Application that became 0X945T was submitted by the American College of Radiology on 14 January 2025 under the leadership of Dr. Geraldine McGinty, ACR's Chair of the Commission on Economics. The original application defined a code specifically for radiology applications — pulmonary embolism triage, incidental finding flagging, structured report prefill — and would have produced a code limited in scope to radiology billing. The Panel's October 2025 meeting expanded the application's scope after the American Society of Clinical Oncology submitted a parallel application requesting coverage for tumour board summarisation and clinical trial eligibility screening, and after the American Academy of Family Physicians submitted a third application covering ambient documentation in primary care. The Panel's January 2026 meeting consolidated all three applications into a single code application covering diagnostic-agent augmented clinical reasoning across specialties, with eight additional specialty societies joining as named co-sponsors by the March 2026 meeting: the American College of Physicians, the American Academy of Emergency Medicine, the American College of Cardiology, the American Academy of Neurology, the American Academy of Dermatology, the American Society of Hematology, the American Urological Association, and the American Gastroenterological Association.

The Panel's structural challenge in finalising the descriptor language was producing a definition that captured the clinical activity precisely enough to be billable while remaining broad enough to accommodate the eleven specialties' divergent use cases. The final descriptor, as published on 15 May 2026, reads: "Diagnostic-agent augmented clinical reasoning in the context of evaluation and management service, using an FDA-cleared software-as-a-medical-device (SaMD) capability that produces structured clinical reasoning outputs attributable to specific input modalities, documented within the patient's clinical record and reviewed by the responsible clinician prior to incorporation into the encounter documentation." The descriptor's key structural elements are the FDA-cleared requirement (which excludes non-cleared capabilities including most off-label deployments and most research-use deployments), the SaMD requirement (which excludes general-purpose AI capabilities not classified as medical devices), the attribution requirement (which requires the reasoning-attribution layer that the FDA's Pre-Sub feedback to Aidoc endorsed as a regulatory-quality structural feature), the documentation requirement (which requires the output to be incorporated into the patient's clinical record), and the clinician-review requirement (which excludes autonomous deployment without responsible-clinician override). The descriptor is structurally aligned to the FDA's emerging regulatory posture on multi-modal clinical AI and explicitly requires the kind of reasoning attribution that the Aidoc Pre-Sub feedback letter from earlier in May 2026 endorsed.

The Panel's deliberation also produced a Category III rather than Category I classification, which is structurally important. Category I codes are the principal CPT code set for established medical services with substantial documented clinical evidence; Category III codes are for emerging technologies and services that have not yet accumulated sufficient evidence to justify Category I status. The Category III classification means that 0X945T is provisional in important ways: it sunsets after five years unless the Panel converts it to Category I status during a renewal review, and it does not receive an Relative Value Unit (RVU) assignment under the Resource-Based Relative Value Scale (RBRVS) that drives Medicare physician fee schedule payments. The absence of RVU assignment means that CMS reimbursement for Category III codes is set by carrier-level determination rather than by the national fee schedule, which produces the variable coverage and reimbursement patterns that characterise the early period of any Category III code. The Category III classification was the Panel's deliberate choice and reflects the Panel's judgement that diagnostic-agent augmented clinical reasoning has accumulated sufficient evidence to be billable but not yet sufficient evidence to be folded into the established Category I code set. The judgement is defensible and is consistent with how the Panel has historically handled emerging technology services.

The CMS coverage determination and the reimbursement amount

The Centers for Medicare and Medicaid Services published its proposed coverage determination for 0X945T on 8 May 2026 under the formal CMS coverage analysis process, with public comment period closing on 7 June 2026 and final determination expected by 30 September 2026. The proposed determination is operationally consequential: it establishes the Medicare coverage pattern, sets the precedent that commercial payors will reference in their own coverage policy development, and defines the proposed reimbursement amount that will anchor pricing negotiations across the rest of the payor landscape. The proposed determination covers diagnostic-agent augmented clinical reasoning across the same eleven specialties that the CPT Editorial Panel's deliberation incorporated, with named coverage for radiology (pulmonary embolism triage, incidental finding flagging, structured report prefill), oncology (tumour board summarisation, clinical trial eligibility screening), primary care (ambient documentation), and the seven additional specialties through similar named-application clauses.

The proposed CMS reimbursement amount, documented in the coverage determination, is $4.21 on the national fee schedule for the new code 0X945T when billed in connection with a qualifying E/M service. The amount is approximately 3.5 per cent of the average Medicare reimbursement for the level 3 outpatient E/M code (99213) under which most of the diagnostic-agent applications would be billed in connection. The framing that industry reporting has characterised as "cents on the dollar" is empirically accurate: the diagnostic-agent code's reimbursement is roughly one-thirtieth of the underlying E/M code with which it must be billed in conjunction. The structural rationale CMS provides in the coverage determination is that the diagnostic-agent activity is a discrete component of clinical reasoning that supplements but does not replace the E/M service the clinician is performing, and that the reimbursement should reflect the marginal value of the supplemental activity rather than the totality of the clinical reasoning activity. The framing is defensible on principle but is producing substantive concern among the specialty societies that championed the code's adoption, several of which have signalled in published commentary that they will submit comments to CMS during the public comment period arguing for higher reimbursement.

The reimbursement amount also has structural implications for vendor pricing. The current vendor pricing models for clinical AI deployment vary substantially: Microsoft Nuance's DAX Copilot is priced as a subscription model with per-clinician licence fees ranging from approximately $400 to $800 per month depending on capabilities; Aidoc's BriefCase platform is priced as an enterprise contract with annual licence fees ranging from approximately $200,000 to $1.2 million per site depending on deployment scope; Abridge is priced on a hybrid model combining per-clinician subscription fees and per-encounter usage fees. The new CPT code creates a distinct economic signal for hospitals: the $4.21 per encounter that the deployment can recover through Medicare reimbursement is the floor of the revenue that justifies the deployment's cost. For high-volume specialties — emergency department triage at 60-120 encounters per shift, primary care at 25-40 encounters per shift — the per-encounter reimbursement aggregates to material revenue, with a single high-volume primary care practice potentially recovering $30,000 to $50,000 per clinician annually from the new code. For lower-volume specialties or for capabilities deployed in lower-volume contexts, the per-encounter reimbursement may not aggregate to cost-recovery levels, and the deployment economics depend on other revenue mechanisms (cost savings from documentation efficiency, improved patient outcomes affecting value-based payment, indirect quality-bonus payments) to remain net-positive.

CMS's coverage determination also imposes specific documentation requirements that the billing clinician must satisfy to bill 0X945T legitimately. The documentation requirements include: explicit naming of the FDA-cleared SaMD capability used in the encounter; documentation of the reasoning-attribution output produced by the capability; documentation of the clinician's review of the reasoning attribution prior to incorporating it into the encounter documentation; and documentation of any clinician modification or override of the capability's output. These documentation requirements are structurally similar to what the ACIS and CARS governance frameworks already require for institutional audit purposes, which means that hospitals operating under either framework will already have most of the documentation infrastructure in place. For hospitals not operating under formal governance frameworks, the documentation requirements impose a non-trivial workflow obligation that will need to be operationalised before the 1 January 2027 effective date. The documentation burden is a real cost that needs to be set against the reimbursement revenue when evaluating the code's net economic impact.

The commercial payor coverage policies — and the differential pattern

Commercial payors do not automatically inherit Medicare coverage determinations, but they typically reference Medicare policy as the starting point for their own coverage policy development. The four largest commercial payor networks — Anthem (operating under the Blue Cross Blue Shield licensing framework across fourteen states), UnitedHealthcare (the largest commercial payor by membership), Aetna (a CVS Health subsidiary since 2018), and the Blue Cross Blue Shield national network (covering the thirty-six independent BCBS plans not licensed through Anthem) — have begun publishing their coverage policies for the new CPT code in advance of the 1 January 2027 effective date. The four policies are not identical and produce a differential coverage pattern that will materially shape provider billing strategy and vendor pricing negotiations.

Anthem's coverage policy, published on 18 May 2026, follows the proposed CMS coverage determination closely. The Anthem policy covers 0X945T across the same eleven specialties CMS proposes, applies similar documentation requirements, and proposes a reimbursement amount of $4.50 per encounter on the Anthem national fee schedule — approximately 7 per cent higher than the proposed CMS amount, consistent with Anthem's general pattern of slightly higher reimbursement than Medicare for emerging technology codes. The Anthem policy is the most CMS-aligned of the four commercial payor policies and is being treated by industry as the baseline against which the other commercial payors will be evaluated.

UnitedHealthcare's coverage policy, published on 22 May 2026, is materially more restrictive than the Anthem and CMS posture. The UnitedHealthcare policy covers 0X945T only in connection with three specific clinical contexts — radiology pulmonary embolism triage, oncology tumour board summarisation, and ambient documentation in primary care — and explicitly excludes coverage for the other eight specialties that the CPT code's descriptor language accommodates. The UnitedHealthcare reasoning, published in the policy's coverage rationale section, is that "the clinical evidence base for diagnostic-agent augmented clinical reasoning in specialties beyond radiology, oncology, and primary care is at present insufficient to support routine coverage determination." The restrictive scope is producing substantive concern at multiple specialty societies, with the American College of Cardiology and the American Academy of Neurology publishing formal letters of disagreement with the UnitedHealthcare policy within seven days of its publication. UnitedHealthcare's proposed reimbursement amount for the covered specialties is $3.80 per encounter, approximately 10 per cent below the proposed CMS amount, consistent with UnitedHealthcare's general pattern of slightly lower reimbursement than Medicare for emerging technology codes.

Aetna's coverage policy, published on 26 May 2026, takes an intermediate position. Aetna covers 0X945T across nine of the eleven specialties (excluding the American Academy of Dermatology and American Urological Association applications that Aetna characterises as "insufficiently mature in clinical evidence"), applies the same documentation requirements as the CMS proposed determination, and proposes reimbursement of $4.15 per encounter — essentially at parity with the proposed CMS amount. Aetna's policy is positioned as a middle path between Anthem's CMS-aligned breadth and UnitedHealthcare's restrictive scope, and it is the policy that several specialty societies have signalled as their preferred reference template for negotiations with the smaller regional commercial payors that will follow the major networks' lead. The Blue Cross Blue Shield national network's coverage guidance, published on 28 May 2026, is the most flexible: it covers 0X945T across all eleven specialties and provides reimbursement amount guidance to the thirty-six independent BCBS plans that ranges from $4.00 to $5.00 per encounter depending on the plan's local market dynamics. The BCBS guidance is not binding on the independent plans, but it sets the framework within which the local plan coverage policies will be developed.

The differential coverage pattern across the four commercial payors produces operational complexity for providers and for vendors. A radiology practice billing across multiple commercial payors will need to track which payors cover which capabilities, what documentation each payor requires, and what reimbursement amount each payor pays. A vendor selling into the institutional market will need to support multiple coverage scenarios in its product positioning. The complexity is non-trivial but is structurally consistent with the differential coverage patterns that have characterised every prior emerging technology code's early payor uptake. The pattern typically converges over twelve to eighteen months as the major payors observe each other's experience with the code and adjust their policies toward a common posture. The convergence trajectory for 0X945T over 2027 will be a leading indicator of how the broader commercial payor market views the diagnostic-agent reimbursement category.

Vendor pricing implications and the long-term economic model

The new CPT code creates a distinct economic signal for clinical AI vendors that is more consequential than the per-encounter reimbursement amount alone suggests. The structural change is that diagnostic-agent deployment now has a documented, billable revenue stream, which means that hospital procurement officers can build cost-justification analyses that include the new reimbursement as a revenue offset against the deployment cost. This shifts the negotiation dynamic between hospitals and vendors. Previously, vendors had to argue that their deployments produced cost savings, improved patient outcomes, or other benefits that justified the deployment cost. Now, vendors can point to a specific dollar-amount revenue per encounter that hospitals can use to recover deployment cost directly. The argument becomes simpler, and the procurement decision becomes more tractable.

The vendor pricing implications, however, are not straightforwardly positive for vendors. The per-encounter reimbursement amount is fixed (at approximately $4.20 on the CMS schedule and $3.80 to $5.00 on the commercial payor schedules), which means that vendors cannot capture the reimbursement value directly through their own pricing. Any per-encounter vendor pricing that approaches the reimbursement amount will be vigorously resisted by hospital procurement officers, because the hospital captures none of the value when the vendor captures it all. The natural vendor pricing posture is therefore to price below the per-encounter reimbursement amount in a way that allows the hospital to retain meaningful net revenue from the deployment. Industry pricing analysis suggests that vendors will likely price per-encounter usage fees in the range of $1.50 to $2.50 per encounter, leaving the hospital with $1.70 to $2.70 per encounter in net reimbursement after vendor fees are paid. The pricing trajectory will be tested as the code becomes operational over 2027 and as hospitals build empirical data on actual reimbursement realisation and vendor cost recovery.

The longer-term economic model is more complex. The Category III classification of 0X945T means that the code sunsets after five years unless converted to Category I status, and the conversion review will require substantial accumulated clinical evidence. The CPT Editorial Panel's conversion criteria typically include published clinical evidence demonstrating the service's medical necessity, accumulated billing volume demonstrating the service's market acceptance, and specialty society support for the conversion. Diagnostic-agent augmented clinical reasoning will likely accumulate the necessary evidence over the five-year window, but the conversion is not assured. If the conversion occurs on schedule (likely target window of late 2030 to mid-2032), the code will receive an RVU assignment, will be folded into the Resource-Based Relative Value Scale, and will be subject to the same fee schedule update mechanics as other established CPT codes. The Category I conversion would also likely produce some adjustment to the reimbursement amount as the RVU assignment is calibrated against the broader CPT code set's relative value distribution. The conversion trajectory is consequential and will be a leading indicator of how the broader healthcare reimbursement system views diagnostic-agent deployment over the longer term.

The cross-jurisdictional implications are also worth noting. The US reimbursement framework is structurally distinct from the reimbursement frameworks that operate in the European member states, the UK, Canada, Australia, and other major healthcare markets. The new CPT code creates a US-specific economic signal that does not translate directly to other healthcare systems. European hospitals deploying clinical AI under the AI Act regulatory framework operate under reimbursement systems that vary by country — Germany's G-DRG system, France's T2A system, the UK NHS payment systems, and so forth — and that do not necessarily produce equivalent per-encounter reimbursement for diagnostic-agent activity. The cross-jurisdictional vendor strategy question is whether to price uniformly across markets (which would mean either lower US pricing or higher European pricing relative to local reimbursement) or to price differentially by market (which adds operational complexity but allows pricing to be calibrated to each jurisdiction's reimbursement framework). The choice will be made vendor by vendor over the next eighteen months, and the differential pricing posture is the more likely outcome on industry analytical grounds. The US CPT code is, in this sense, both a domestic reimbursement event and a structural signal in the broader global clinical AI economic landscape.

What to watch

The new CPT code becomes effective on 1 January 2027 and the first twelve months of operation will determine whether the reimbursement framework produces the institutional adoption acceleration its proponents anticipate. Five signals are the leading indicators.

• Whether the CMS public comment period closing 7 June 2026 produces material modifications to the proposed coverage determination — particularly to the proposed $4.21 reimbursement amount, which several specialty societies have signalled will be the focus of their comments arguing for higher reimbursement; the final reimbursement amount will anchor the commercial payor coverage policies and will define the economic signal the entire downstream market responds to.
• Whether UnitedHealthcare moderates its restrictive coverage scope — limited to radiology pulmonary embolism triage, oncology tumour board summarisation, and primary care ambient documentation — over the second half of 2026 as the formal letters of disagreement from the specialty societies accumulate; if UnitedHealthcare retains the restrictive scope through the 1 January 2027 effective date, the commercial payor coverage pattern will be materially more differentiated than the Medicare-anchored alignment that industry was expecting, and providers will face substantial operational complexity in tracking payor-specific coverage.
• Whether the documentation requirements the CMS coverage determination imposes — explicit naming of the FDA-cleared SaMD capability, documentation of reasoning attribution, documentation of clinician review and any modifications — produce significant workflow friction at hospitals not already operating under ACIS or CARS governance frameworks; the documentation burden is real and will be the test of whether the new code's economic benefits exceed its compliance overhead at hospitals that have not built formal governance infrastructure.
• Whether vendor per-encounter pricing settles in the $1.50 to $2.50 range that industry pricing analysis projects, or whether competitive dynamics push pricing toward either the lower bound (creating pricing pressure on smaller vendors who cannot operate profitably at sub-$1.50 per encounter pricing) or the upper bound (creating pricing pressure on hospital procurement officers seeking to retain meaningful net reimbursement after vendor fees); the pricing trajectory over the first half of 2027 will define the long-term economic model for the diagnostic-agent vendor market.
• Whether the Category III to Category I conversion trajectory remains on schedule for the late 2030-to-mid-2032 window or experiences material delays; the conversion review will require accumulated clinical evidence that the deployment population must produce over the five-year operational window, and any structural failures or material safety events during that window could delay the conversion or produce regulatory headwinds that affect the reimbursement framework's longer-term stability.

Frequently asked

What is a Category III CPT code, and how does it differ from a Category I code?

Category I CPT codes are the principal code set for established medical services with substantial documented clinical evidence; they receive RVU assignments under the Resource-Based Relative Value Scale and are subject to standard Medicare fee schedule update mechanics. Category III codes are for emerging technologies and services that have not yet accumulated sufficient evidence to justify Category I status; they sunset after five years unless converted to Category I, do not receive RVU assignments, and have reimbursement amounts set by carrier-level determination rather than by the national fee schedule. The Category III classification of 0X945T reflects the CPT Editorial Panel's judgement that diagnostic-agent augmented clinical reasoning has accumulated sufficient evidence to be billable but not yet sufficient evidence for Category I status. The conversion review is expected in the late 2030-to-mid-2032 window depending on accumulated clinical evidence and billing volume.

Why is the CMS proposed reimbursement amount so low — $4.21 per encounter?

CMS's structural rationale, documented in the proposed coverage determination, is that diagnostic-agent activity is a discrete component of clinical reasoning that supplements but does not replace the E/M service the clinician is performing. The reimbursement reflects the marginal value of the supplemental activity rather than the totality of the clinical reasoning activity. The $4.21 amount is approximately 3.5 per cent of the average Medicare reimbursement for the level 3 outpatient E/M code (99213) under which most diagnostic-agent applications would be billed in connection. The framing as "cents on the dollar" of the underlying E/M code is empirically accurate. Several specialty societies have signalled they will submit comments to CMS during the public comment period arguing for higher reimbursement, and the final determination after the 7 June 2026 close of comments may differ from the proposed amount.

Why is UnitedHealthcare's coverage policy more restrictive than the other commercial payors?

UnitedHealthcare's coverage policy, published 22 May 2026, covers 0X945T only in connection with three clinical contexts — radiology pulmonary embolism triage, oncology tumour board summarisation, and primary care ambient documentation — and explicitly excludes coverage for the other eight specialties the CPT code's descriptor language accommodates. UnitedHealthcare's published rationale is that the clinical evidence base for diagnostic-agent augmented clinical reasoning in specialties beyond the three named contexts is at present insufficient to support routine coverage determination. The American College of Cardiology and the American Academy of Neurology have published formal letters of disagreement within seven days of the policy's publication, and the restrictive scope is producing concern across the specialty society landscape. Whether UnitedHealthcare moderates the scope over the second half of 2026 is one of the principal open questions for the new code's commercial payor coverage trajectory.

How will the new code affect vendor pricing for clinical AI deployments?

The new code creates a distinct billable revenue stream that hospital procurement officers can build cost-justification analyses around, shifting the negotiation dynamic between hospitals and vendors. The per-encounter reimbursement amount is fixed (approximately $4.20 on CMS and $3.80 to $5.00 on commercial payor schedules), which means vendors cannot capture the reimbursement value directly through their own pricing. Industry pricing analysis suggests vendors will price per-encounter usage fees in the range of $1.50 to $2.50 per encounter, leaving the hospital with $1.70 to $2.70 per encounter in net reimbursement after vendor fees. The pricing trajectory over the first half of 2027 will define the long-term economic model for the diagnostic-agent vendor market and will be a leading indicator of competitive dynamics across the vendor landscape.

What documentation does a clinician need to maintain to bill 0X945T legitimately?

The CMS coverage determination requires documentation of four specific elements: explicit naming of the FDA-cleared SaMD capability used in the encounter; documentation of the reasoning-attribution output produced by the capability; documentation of the clinician's review of the reasoning attribution prior to incorporating it into the encounter documentation; and documentation of any clinician modification or override of the capability's output. These documentation requirements are structurally similar to what the ACIS and CARS governance frameworks already require for institutional audit purposes, which means hospitals operating under either framework will already have most of the documentation infrastructure in place. For hospitals without formal governance frameworks, the documentation requirements impose a workflow obligation that will need to be operationalised before the 1 January 2027 effective date. The documentation burden is a real cost that needs to be set against the reimbursement revenue when evaluating the code's net economic impact at the individual hospital level.

Does the new CPT code apply to clinical AI deployments outside the United States?

No. The CPT code system is a US-specific reimbursement framework operating under the American Medical Association's stewardship, and the new code 0X945T is enforceable only within the US healthcare system. European hospitals deploying clinical AI under the AI Act regulatory framework operate under reimbursement systems that vary by country — Germany's G-DRG system, France's T2A system, the UK NHS payment systems, and so forth — and that do not necessarily produce equivalent per-encounter reimbursement for diagnostic-agent activity. The cross-jurisdictional vendor strategy question is whether to price uniformly across markets or to price differentially by market; differential pricing by jurisdiction is the more likely outcome on industry analytical grounds, with the US CPT code functioning as both a domestic reimbursement event and a structural signal in the broader global clinical AI economic landscape.