What the study found
Researchers linked physicians' scores on the American Board of Internal Medicine's Longitudinal Knowledge Assessment (LKA) to the care their patients actually received, and published the results in JAMA Internal Medicine. The analysis covered nearly 900,000 Medicare beneficiaries cared for by 7,089 outpatient general internists who first enrolled in the LKA in 2022 or 2023.
Patients of physicians in the top quartile of knowledge scores were nearly 8% less likely to receive any of 25 tracked low-value services — 28.6% versus 31% for patients of the lowest-scoring physicians. For low-value imaging specifically, the gap was about 4% (13.2% versus 13.8%), alongside a 16% lower rate of unnecessary diagnostic and preventive testing and an 11% lower rate of unnecessary cancer screening.
The authors — Jonathan L. Vandergrift, Bruce E. Landon, Weifeng Weng, and Bradley M. Gray — estimate that if every general internist ordered low-value services at the rate of top scorers, about 80,000 fewer Medicare fee-for-service beneficiaries would receive them each year. Low-value care overall is estimated to cost the U.S. health system roughly $100 billion annually.
Volume and value are two different problems
Most of the debate about imaging demand is about volume — more scanners, more orders, more studies stacking up in the reading queue. That capacity story is real, and it is covered separately in the overutilization and radiologist-capacity squeeze. But this study points at a different axis: appropriateness. A low-value scan isn't a problem because it adds one more study to the pile — it's a problem because it was unlikely to change the patient's management in the first place.
The two problems interact in an uncomfortable way. A low-value scan doesn't just consume a slot — it generates findings. Many are incidental and clinically irrelevant, but each one can seed a follow-up recommendation, another scan, a specialist referral, sometimes a biopsy. Inappropriate imaging is therefore a common on-ramp to more inappropriate imaging.
Volume problem
Too many studies for the available reading capacity. The lever is throughput — reading the queue faster without proportionally growing headcount.
Value problem
The wrong studies get ordered relative to guidelines. The lever is appropriateness at the point of ordering — knowledge, decision support, and incentives.
Where they meet
Low-value scans create incidental findings that trigger downstream low-value follow-up. Ambiguous reporting amplifies the cascade; clear, guideline-aligned reporting dampens it.
Knowledge helps — but it isn't the whole fix
It would be easy to read this study as "just train doctors better." The effect is real, but note its size: the imaging gap between the highest- and lowest-knowledge physicians was only a few percentage points. Ordering behavior is also driven by defensive medicine, patient expectations, time pressure, and habit — factors that a knowledge score doesn't capture.
That matches the broader evidence. Efforts to curb low-value ordering through education and clinical decision support alone have produced mixed and often disappointing results. Appropriate use is a system property, not a personal virtue — which is exactly why the reporting layer, downstream of the ordering decision, matters more than it first appears.
Where AI CT reporting fits — on the reads that do happen
AI CT reporting sits on the far side of the ordering decision. It doesn't — and shouldn't — decide whether a scan was appropriate; that call belongs to the referring clinician and the appropriateness criteria they follow, and it's the same gate that prior authorization reform is trying to make faster. What reporting can do is control the signal quality of the reads that make it through.
A vague report — an incidental finding described without size, characterization, or a guideline-anchored recommendation, capped with "clinical correlation advised" — is how a marginal scan turns into a follow-up cascade. A structured, comprehensive report that characterizes findings consistently and states clear, evidence-cited follow-up recommendations does the opposite: it reduces the ambiguity that manufactures downstream low-value imaging.
That is the model xAID is built on. The AI produces a structured, guideline-aligned draft; xAID's in-house radiologist reviews every preliminary; and the report is delivered ready-to-sign so your reading radiologist signs the final. The goal isn't more reads — it's cleaner signal on each one, which is the reporting-side complement to the appropriateness work happening at the point of order. You can see the workflow in detail on how AI CT reporting works.
Frequently asked questions
What did the study find about clinician knowledge and low-value imaging?
A 2026 JAMA Internal Medicine study of nearly 900,000 Medicare beneficiaries cared for by 7,089 general internists found that patients of physicians in the top quartile of the ABIM Longitudinal Knowledge Assessment were about 4% less likely to receive low-value imaging (13.2% vs 13.8%) and nearly 8% less likely to receive any of 25 low-value services (28.6% vs 31%). Higher measured clinical knowledge correlated with more appropriate ordering.
What is low-value imaging?
Low-value imaging is a scan ordered when guidelines and evidence suggest it is unlikely to change management or benefit the patient — for example, routine imaging for uncomplicated low back pain or headache. It is distinct from overutilization by volume: the issue is appropriateness of the individual order, not the total number of scans. Low-value care overall is estimated to cost the U.S. health system roughly $100 billion a year.
Does more clinical knowledge eliminate low-value imaging?
No. The knowledge effect is real but modest — a few percentage points for imaging — and separate research shows education and clinical decision support alone have produced mixed or limited reductions. Ordering behavior is shaped by defensive medicine, patient expectations, time pressure, and habit as well as knowledge, so appropriate use is a system problem, not only an individual one.
How does AI reporting relate to low-value imaging and appropriate use?
AI CT reporting does not decide whether a scan should be ordered — that is the referring clinician's call, guided by appropriateness criteria. Its role is on the reads that do happen: a structured, guideline-aligned, evidence-cited report with clear follow-up recommendations reduces the ambiguous incidental findings and vague 'clinical correlation advised' language that trigger downstream low-value follow-up imaging. At xAID, every report is radiologist-reviewed and delivered ready-to-sign, so your reading radiologist signs the final.
Source: Vandergrift JL, Landon BE, Weng W, Gray BM. "Low-Value Services and Longitudinal Knowledge Assessment Performance," JAMA Internal Medicine (2026), doi:10.1001/jamainternmed.2026.2889, as reported by Radiology Business and News-Medical. Figures are rounded as reported.