What Are ACR Appropriateness Criteria in Medical Imaging?

The ACR Appropriateness Criteria are a set of evidence-based guidelines published by the American College of Radiology that tell clinicians which imaging study, if any, is most appropriate for a given clinical situation. They cover hundreds of specific scenarios across radiology and radiation oncology, rating each potential exam on a scale from 1 (least appropriate) to 9 (most appropriate). The criteria exist to solve a real problem: ordering the wrong scan wastes money, exposes patients to unnecessary radiation, and can lead to incidental findings that trigger even more testing. Yet whether the guidelines get used in practice, and how well they translate into better care, depends on a tangle of factors involving technology, physician behavior, and legislation.

How the Criteria Are Built

Each set of criteria starts with a clinical scenario, not an imaging modality. A panel might begin with “adult presenting with acute, nonlocalized abdominal pain” or “uncomplicated low back pain lasting less than six weeks.” The panels themselves are multidisciplinary, drawing not just from radiologists but from specialists who order imaging in relevant fields. The development process involves a systematic review of peer-reviewed literature, followed by a structured consensus method known as a modified Delphi technique, combined with the RAND/UCLA Appropriateness Method and the GRADE framework for evaluating the quality of evidence.¹ Panels meet, vote, discuss disagreements, and re-vote until they converge on ratings for each imaging option within a scenario. The process also follows standards set out in the Institute of Medicine’s guidelines for trustworthy clinical practice guidelines.

The criteria are reviewed annually by their expert panels, and the ACR updates or retires topics as new evidence accumulates.² This ongoing revision cycle is one of the features that distinguishes them from static guideline documents that may go years without updates.

What the Ratings Look Like in Practice

The criteria are organized by clinical topic, each containing several “variants” representing slightly different patient presentations. Within each variant, imaging procedures receive a numerical appropriateness rating. A rating of 7 to 9 means “usually appropriate,” 4 to 6 means “may be appropriate,” and 1 to 3 means “usually not appropriate.” The language is deliberately not absolute: the criteria acknowledge that individual cases may warrant deviations.

Low back pain is one of the most heavily referenced topics. The criteria state that uncomplicated acute low back pain, including radiculopathy, is a self-limited condition that does not warrant imaging at all.³ Imaging becomes appropriate only after about six weeks of conservative management with no improvement, or when red flags such as signs of cancer, infection, fracture, or a neurological emergency like cauda equina syndrome are present. This is one of those areas where the criteria run headlong into both patient expectations and physician anxiety. People with bad back pain often expect an MRI immediately, and the criteria explicitly push back against that impulse.

For acute abdominal pain in adults, the criteria vary depending on where the pain is localized and what the clinical suspicion is.⁴ A separate pediatric version addresses children with abdominal pain, covering scenarios like suspected intussusception, bowel obstruction without prior surgery, and necrotizing enterocolitis, with imaging recommendations tailored to each.⁵

In the realm of stroke workup, the criteria have measurable consequences for how effectively imaging is used. A study examining CT angiography (CTA) orders in the emergency department found that when CTA was ordered in line with the ACR recommendation of “usually appropriate,” about 29% of scans revealed findings that explained the stroke. When CTA was ordered in situations rated “may be appropriate” or “usually not appropriate,” explanatory findings dropped to 3% and 1% respectively.⁶ That kind of gap illustrates exactly why the ratings matter: the same scan in different clinical contexts has a dramatically different chance of telling you something useful.

Whether Doctors Actually Follow Them

Here is where the picture gets less tidy. Research on clinician awareness and use of the criteria has consistently found low uptake. One study examining how often clinicians relied on the ACR criteria when ordering imaging reported low overall utilization, leading the authors to suggest the ACR should rethink how the criteria are disseminated.⁷ The problem is not that the guidelines are controversial. Most clinicians, when shown the criteria, agree with the ratings. The problem is workflow: the criteria sit in documents and databases that many ordering physicians never consult in the middle of a busy clinic day.

This gap between having good guidelines and getting people to use them is a well-known challenge in healthcare. A systematic review of interventions aimed at reducing low-value imaging found that single-component efforts, like simply distributing guidelines, reduced unnecessary imaging in about 74% of the studies reviewed. Multi-component interventions, combining guidelines with education, audit, feedback, or computerized prompts, worked even better, reducing low-value imaging in 94% of studies examined.⁸ Education was a particularly effective ingredient when bundled with other components. The takeaway is that publishing a guideline is necessary but far from sufficient.

Clinical Decision Support and Report Cards

The most promising mechanism for getting the criteria in front of clinicians at the right moment is embedding them in clinical decision support (CDS) software, which can be integrated into electronic health record systems. When a doctor places an imaging order, the software checks the clinical indication against the ACR ratings and displays a recommendation, sometimes flagging that a particular scan is rated “usually not appropriate” for the stated reason.⁹

But how CDS is designed matters enormously. A study compared two approaches aimed at reducing unnecessary lumbar spine MRI orders for low back pain. Real-time alerts, which popped up at the moment of ordering, had no measurable effect on MRI ordering rates. Report cards, which gave primary care providers periodic feedback on how their ordering patterns compared to guidelines, reduced the likelihood of an MRI being ordered by roughly 37 to 38%.¹⁰ In other words, nudging doctors over time worked better than interrupting them in the moment. Alert fatigue is a real phenomenon: when clinicians are bombarded with pop-ups for every decision, they learn to click past them without reading.

The Legislative Push Behind the Criteria

The United States took the unusual step of legislating the use of imaging appropriateness criteria through the Protecting Access to Medicare Act (PAMA) of 2014. The law requires ordering physicians to consult a “qualified” clinical decision support mechanism before ordering advanced imaging (CT, MRI, nuclear medicine, PET) for Medicare patients. The ACR Appropriateness Criteria are one of the qualified sets of criteria that CDS systems can use to fulfill this requirement.

The scope of the mandate is considerable. An analysis of national emergency department data estimated that the eight priority clinical areas targeted by PAMA are present in roughly one in four adult ED visits, and CT or MRI is performed during up to 23% of those visits.¹¹ Depending on how individual health systems implement their CDS tools, the actual volume of interactions could be higher or lower than those estimates. Some hospitals have found they need local analytics platforms just to monitor how PAMA is affecting their ordering workflows.¹²

PAMA has been subject to rolling implementation delays, and its penalty phase, in which Medicare could deny payment for imaging orders placed without CDS consultation, has been postponed multiple times. The consultation requirement itself, however, has been live in various stages. The practical result is that most large health systems in the U.S. have already invested in CDS infrastructure that draws on the ACR criteria or equivalent guideline sets, regardless of whether the penalty phase has been enforced.

Cost Implications of Getting Imaging Right

When the criteria are actively applied, the financial effects can be substantial. The Radiology Support, Communication and Alignment Network (R-SCAN), an ACR quality improvement program, was shown to increase the odds that an imaging study would be appropriate and higher in diagnostic value. A cost analysis found that R-SCAN participation was associated with a reduction of about $260,000 in lesser-value imaging costs over just three and a half months. When extrapolated to the entire Medicare population, the estimated yearly savings from reducing lesser-value imaging came to roughly $433 million.¹³

The savings are not just about skipping unnecessary scans. Inappropriate imaging cascades. A CT scan performed without a clear indication may reveal an incidental finding, a small lung nodule or a cyst, that then requires follow-up imaging, possible biopsy, specialist referral, and patient anxiety. Each link in that chain carries its own costs and risks. By filtering out low-value orders at the front end, the criteria can prevent these downstream spirals from starting in the first place.

How the ACR Criteria Compare to International Equivalents

The ACR Appropriateness Criteria are not the only imaging referral guidelines in existence. The Royal College of Radiologists in the United Kingdom publishes iRefer, and the European Society of Radiology offers iGuide. All three aim to match clinical indications to the right imaging modality, but they do not always agree.

A study that compared the three guidelines head-to-head using cervical spine imaging referrals found stark differences for X-ray recommendations. The appropriateness rate of cervical spine X-ray referrals ranged from about 13% to 75% depending on which guideline was applied. The ACR criteria and the ESR’s iGuide showed moderate agreement with each other for X-ray recommendations, but the RCR’s iRefer had no meaningful agreement with either. For CT of the cervical spine, however, all three guidelines were in near-perfect agreement.¹⁴ The divergence for X-ray likely reflects different assessments of a modality that has been gradually supplanted by CT and MRI for many cervical spine indications. Where the evidence is stronger and more uniform, the guidelines converge.

These discrepancies matter for anyone practicing or receiving care in a system that follows a different set of guidelines. A referral deemed appropriate under one framework may be flagged as unnecessary under another, and vice versa. For patients, the practical implication is that the “right” imaging workup can depend partly on where in the world you happen to be.

Gaps in Pediatric Coverage

One criticism of both the ACR criteria and their international counterparts is that they were developed primarily with adult patients in mind. A study examining pediatric MRI orders across multiple centers found that the ACR criteria addressed only about 36% of the clinical scenarios encountered in children. The European EURO-2000 guidelines covered a similar proportion, roughly 38%. Coverage varied by body region: the ACR criteria did better for abdominal, pelvic, and musculoskeletal scenarios, while EURO-2000 offered broader coverage for head and spine imaging in children.¹⁵ Encouragingly, when a scenario was covered by guidelines, compliance was excellent, exceeding 94% for justified MRI exams. The problem is less about clinician willingness to follow the rules and more about the rules not existing for many pediatric situations.

The ACR has been expanding its pediatric content, as seen with the dedicated criteria for abdominal pain in children. But the pace of development may not keep up with the breadth of pediatric imaging questions, particularly for less common conditions and for modalities like MRI that are increasingly preferred in children to avoid radiation exposure.

AI-Assisted Guideline Use and the Horizon Ahead

The next frontier for the criteria involves artificial intelligence. Researchers are exploring whether large language models can be trained to recommend imaging in alignment with ACR guidelines, effectively acting as a smart intermediary between the clinician’s clinical question and the sprawling criteria database. One line of research uses reinforcement learning to align language model outputs with ACR recommendations, with the goal of handling not just the clinical scenarios that the criteria explicitly cover but also out-of-distribution cases that fall between existing topics.¹⁶ If such systems work reliably, they could help scale the guideline development process itself, reducing the significant effort involved in creating and updating criteria.

Another prototype, called RADAR, uses a structured knowledge base drawn from ACR criteria to ground its recommendations, with built-in citation validation and the ability to abstain when it lacks sufficient evidence to make a recommendation. The authors note that clinical readiness would require licensed ACR content, larger reference standards, and prospective evaluation in real clinical settings.¹⁷ These tools are far from deployment, but they hint at a future in which interacting with imaging guidelines feels less like consulting a static document and more like asking a knowledgeable colleague.

The deeper challenge, though, is not technical. It is cultural. The criteria work best when radiologists and ordering clinicians communicate about imaging decisions as a shared responsibility rather than a transactional request. CDS tools and AI assistants can make the right recommendation more visible, but whether clinicians act on those recommendations depends on institutional culture, workflow design, and whether the incentives in a health system reward thoughtful imaging over volume. That is a problem no algorithm, no matter how well-trained, can solve by itself.