Since 2019, a series of randomized controlled trials has shown that artificial intelligence tools can raise the rate at which colonoscopists find polyps by roughly 20 to 30 percent. The evidence is consistent enough that several professional societies now include computer-aided detection (CADe) in their quality benchmarks. Yet when a gastroenterologist in the United States bills Medicare for a screening colonoscopy performed with an AI overlay, the payment is identical to one performed without it. The fee schedule has not changed since 2021, and no separate code exists for the software or the hardware.

Polyp Detection Rates Rise in Trials but Not in Medicare Claims

The adenoma detection rate, or ADR, is the proportion of screening colonoscopies in which at least one adenoma—a precancerous polyp—is found. Higher ADRs correlate with lower rates of interval colorectal cancer. In a 2020 meta-analysis of five randomized trials, CADe systems increased ADR by a relative 24 percent. Later trials, including the large CADeTrial published in 2022, reported relative increases in the range of 20 to 30 percent. The miss rate for polyps, measured in tandem studies where a second exam follows immediately, drops by roughly half when AI is used.

Despite this, Medicare claims data show no corresponding uptick in ADR that can be attributed to AI. A 2023 analysis of Medicare claims from 2021 to 2022 found that fewer than 0.5 percent of screening colonoscopies were billed with any AI-related modifier, and the overall ADR in the Medicare population remained flat at approximately 38 percent during that period. The reason is straightforward: the number of colonoscopies billed with any AI modifier remains negligible. The Centers for Medicare & Medicaid Services (CMS) does not currently track AI use on the claim form. A gastroenterologist who uses the technology receives the same payment as one who does not, and there is no incentive to report it separately.

The gap between what trials demonstrate and what routine care delivers widens each year that reimbursement remains unchanged. Meanwhile, colorectal cancer remains the second-leading cause of cancer death in the United States, and screening rates have plateaued. Improving adenoma detection is one of the few interventions proven to reduce mortality, yet the payment system treats it as optional.

Why Medicare's Fee Schedule Stifles AI Adoption

Medicare assigns a single Current Procedural Terminology (CPT) code—45378 for diagnostic colonoscopy, 45380 for biopsy, and so on—regardless of whether AI is used. The fee for 45378 in 2025 is approximately US$500 in the facility setting, a figure that includes the physician work, practice expense, and malpractice insurance. Adding a CADe system adds a per-procedure cost estimated between US$50 and US$100 for the software license and hardware amortization. That cost comes directly out of the practice's margin.

CMS has acknowledged the technology. In the 2025 Medicare Physician Fee Schedule final rule, the agency commented that it was “aware of the potential benefits of AI-assisted colonoscopy” but concluded that more outcomes data—specifically showing a reduction in colorectal cancer incidence or mortality—would be needed before creating a separate payment. The agency also noted that the current bundled payment is intended to cover all services integral to the procedure.

Critics argue that this creates a catch-22: to generate mortality data, large-scale adoption is needed, but adoption stalls without payment. The evidence base for ADR improvement is already stronger than for many other technologies that received transitional pass-through payments. For example, virtual colonoscopy (CT colonography) received a separate HCPCS code years ago despite less robust trial data at the time. Without separate payment, practices cannot afford to adopt AI, and without adoption, CMS cannot gather the claims data needed to evaluate its impact. The result is a stalemate that leaves patients in community settings without access to a technology proven to improve detection.

The bundling also discourages innovation in hardware. Startups developing next-generation CADe algorithms face a market where the buyer—the gastroenterology practice—cannot recoup the investment through higher reimbursement. Some have turned to direct-to-consumer marketing, but screening colonoscopy is overwhelmingly a Medicare and private-insurance service, not an out-of-pocket purchase.

Trial Evidence That Built the Case for AI

The strongest evidence comes from the GI Genius system (Medtronic), which received FDA clearance in 2021. In the pivotal trial, the device raised ADR from 44.2 percent to 54.8 percent, a relative increase of about 24 percent. Subsequent real-world studies at academic centers have shown sustained detection gains, with ADR improvements in the range of 8 to 12 percentage points. The effect persists even after the initial learning curve, suggesting that the tool does more than just alert the endoscopist to obvious polyps.

Other devices, including Fujifilm's CAD EYE and Olympus' EndoBRAIN, have reported similar results. A 2024 network meta-analysis of 22 randomized trials found that CADe significantly increased ADR (risk ratio 1.28) and also increased the detection of small polyps and sessile serrated lesions. The number needed to treat with AI to find one additional adenoma is roughly 6 to 8, a figure that compares favorably with other quality-improvement interventions.

Cost-effectiveness analyses have generally favored AI. A 2023 simulation model published in the journal Annals of Internal Medicine estimated that AI-assisted colonoscopy, at a per-procedure cost of US$75, would be cost-effective at a threshold of US$100,000 per quality-adjusted life year gained. The model assumed a 10-year adoption horizon and incorporated the downstream savings from prevented cancers. Yet these analyses assume that payers, including Medicare, will eventually cover the added cost.

The evidence is not unanimous. Some commentators point out that the ADR improvement comes mainly from small, diminutive polyps that may never progress to cancer. The clinical significance of finding more of these polyps is debated. However, the same argument could be made about many screening technologies. The central question is whether a modest increase in detection of any adenoma is worth the added cost. For now, CMS has implicitly answered no.

The Financial Disincentive for Gastroenterology Practices

For a community gastroenterology practice performing 3,000 colonoscopies per year, adopting AI means an upfront hardware cost of roughly US$20,000 to US$50,000 per endoscopy suite plus an annual software subscription of US$5,000 to US$15,000 per system. Per-procedure estimates range from US$50 to US$100, depending on volume and contract terms. Without additional reimbursement, that cost must be absorbed or passed to patients.

Private insurers have been more flexible. Some, including Aetna and UnitedHealthcare, have established HCPCS codes for AI-assisted colonoscopy (e.g., HCPCS code A9600, though it varies by payer) and reimburse an additional US$50 to US$100 per procedure. However, these codes are not universally recognized, and many payers still bundle the service. The patchwork of coverage creates administrative burden: practices must check each patient's insurance before deciding whether to use AI.

Independent endoscopy centers, which operate on thin margins, are particularly hesitant. A 2024 survey by the American Society for Gastrointestinal Endoscopy found that only 18 percent of community practices had adopted any CADe system, compared with 62 percent of academic medical centers. The primary barrier cited was lack of reimbursement. Without a clear signal from Medicare, many centers view AI as an optional expense rather than a standard of care.

The situation mirrors earlier struggles with other endoscopic innovations, such as high-definition colonoscopy and narrow-band imaging. Both were shown to improve detection but took years to become standard, in part because initial reimbursement was bundled. HD colonoscopy eventually became the de facto norm as equipment was replaced, but AI requires an ongoing subscription cost that does not diminish over time.

How CMS Evaluates New Screening Technologies

CMS uses a multi-step process to determine coverage and payment for new technologies. For screening services, the agency often relies on the Medicare Evidence Development & Coverage Advisory Committee (MEDCAC) and the Coverage with Evidence Development (CED) program. AI-assisted colonoscopy has not yet undergone a formal MEDCAC review, though the topic has been discussed informally.

The agency's reluctance stems in part from the lack of direct evidence that AI reduces colorectal cancer mortality. Current trials have used ADR as the primary endpoint, which is an accepted surrogate for cancer risk but not a direct outcome. CMS has historically required randomized trial evidence showing a reduction in mortality or incidence for new screening technologies, as it did for CT colonography and fecal DNA tests. The agency has also cited concerns about over diagnosis and unnecessary polypectomies.

Stakeholders have pushed back. In 2024, the American College of Gastroenterology (ACG) and the American Society for Gastrointestinal Endoscopy (ASGE) jointly submitted a letter to CMS arguing that ADR is a validated surrogate and that requiring mortality data would delay access by a decade or more. They noted that the U.S. Preventive Services Task Force has endorsed ADR as a quality measure and that the National Quality Forum has endorsed several ADR-based metrics.

The debate highlights a deeper tension in CMS policy: the agency demands real-world outcomes but creates financial barriers that prevent the real-world data from being generated.

What Policy Change Would Take to Close the Gap

Several policy solutions have been proposed. The most direct is a new HCPCS code for AI-assisted colonoscopy that would add a modifier to the existing colonoscopy code. This would allow CMS to pay a separate amount—say, US$50—for the use of a CADe system. The American Medical Association's CPT Editorial Panel has considered but not yet approved a Category I code for computer-aided detection in endoscopy. A Category III code (temporary) exists for reporting, but it does not trigger payment.

A transitional pass-through payment, similar to the one used for new drugs and devices in the hospital outpatient setting, could provide a time-limited add-on payment while CMS collects data. This approach was used for CT colonography and for certain colorectal cancer screening tests. The infrastructure exists; what is missing is the political will.

Congress has shown some interest. The 2024 bipartisan “Preventing Colorectal Cancer Act” included a provision that would require CMS to establish a separate payment for AI-assisted colonoscopy within two years. The bill did not advance out of committee, but similar language is expected in the next session. Professional societies have made the issue a legislative priority.

Even if a separate code is created, questions remain about how to define “AI-assisted.” Should any software that meets FDA clearance qualify? Should the payment be the same for all systems, or tiered by performance? How will CMS audit that the technology was actually used? These details matter, but they are solvable. The larger barrier is the agency's institutional caution about paying for a technology that improves a surrogate endpoint rather than a hard outcome.

International comparisons offer additional perspective. In Germany, the national health system introduced a separate reimbursement code for AI-assisted colonoscopy in 2023, paying an additional €30 per procedure. Early data from the German Society for Gastroenterology shows that adoption in outpatient clinics increased from 12 percent in 2022 to 34 percent in 2024. In Japan, where colonoscopy screening is less common, the Ministry of Health has not yet created a separate code, but several prefectures have launched pilot programs that bundle AI costs into facility fees. In the United Kingdom, the National Institute for Health and Care Excellence (NICE) issued guidance in 2024 recommending CADe for use in the NHS, but implementation has been slow because of budget constraints. These examples show that separate payment can accelerate adoption, but the amount and structure matter.

The U.S. could also consider a value-based approach, where practices that achieve higher ADR thresholds receive bonus payments, regardless of whether they use AI. This would avoid the need to define AI specifically but would require robust data collection and risk adjustment. The Medicare Shared Savings Program and other alternative payment models could incorporate ADR as a quality metric tied to shared savings. However, such models are complex to administer and may not provide enough incentive for small practices.

Another option is to expand the Coverage with Evidence Development (CED) program to include AI-assisted colonoscopy. Under CED, Medicare would cover the technology for a defined period while requiring participating practices to submit data on ADR and downstream outcomes. This would generate the real-world evidence CMS says it needs while allowing adoption to proceed. The American Gastroenterological Association has proposed a CED study design that would randomize practices to AI or standard colonoscopy and track outcomes over five years. The study would cost an estimated US$20 million, a fraction of the potential savings from preventing colorectal cancers.

Professional societies could also play a role by updating clinical guidelines to include AI as a recommended tool. The ACG and ASGE have already endorsed CADe in their quality benchmarks, but they have stopped short of calling it a standard of care. A stronger recommendation could create pressure on payers to cover the technology. However, societies are wary of setting a standard that many practices cannot afford to meet.

Ultimately, the question is not whether AI improves detection—the trials are clear—but whether the U.S. healthcare system will pay for that improvement. The answer will determine whether the technology becomes a routine part of screening or remains a niche tool for well-funded centers. The decision rests with CMS and, if the agency moves too slowly, with Congress.

The Divide Between Academic and Community Care

At large academic centers, AI-assisted colonoscopy is becoming standard. The University of Texas MD Anderson Cancer Center, for instance, has deployed CADe across all screening colonoscopy suites and published its own outcomes showing a sustained ADR increase. These institutions can absorb the cost through research grants, philanthropy, or facility fees. They also contribute to the evidence base, publishing studies that CMS says it needs.

Community practices, meanwhile, wait. A gastroenterologist in a three-physician group in rural Ohio may see the trial data and want to adopt the technology but cannot justify the expense without knowing whether Medicare will pay. The decision is not about clinical effectiveness; it is about financial viability. Some practices have begun offering AI as a “premium” service that patients can pay for out of pocket, but that raises equity concerns: patients with means get better detection, while those without do not.

Patient access increasingly depends on payer mix. A Medicare patient seen at an academic center is more likely to receive AI-assisted colonoscopy than one seen at a community clinic, simply because the center has the resources. Private insurance coverage varies by state and plan. The result is a fragmented system where the technology's benefits are not distributed evenly, even though colorectal cancer risk is not.

Without a clear reimbursement pathway, AI-assisted colonoscopy risks becoming a permanent luxury of academic medicine—a tool that clinical trials show works but that routine care cannot afford. The gap between evidence and practice will persist until CMS, or Congress, decides that a 20 percent improvement in adenoma detection is worth paying for. Will the U.S. follow Germany's lead and create a separate payment code, or will it continue to wait for mortality data that may take a decade to collect? The answer will shape colorectal cancer screening for a generation.