The radiology scan backlog is not a new problem. It was a documented operational challenge in academic medical centers before the pandemic, worsened significantly during COVID-era volume disruptions, and has continued growing in most settings since. The common explanation is simple: more imaging orders are being placed than radiologists can read. That explanation is correct but not complete, and incomplete diagnosis leads to incomplete solutions.
A useful root cause analysis of the scan backlog separates the supply-demand imbalance into its component parts, identifies which components are structural, which are cyclical, and which can be addressed by workflow changes. Those are three different intervention types, and conflating them produces programs that tackle the wrong thing.
The Demand Side: Why Order Volume Has Grown
Imaging order volume in the U.S. has increased for reasons that predate AI and telehealth and that will not reverse regardless of what happens in the reading room. The aging population drives disproportionate imaging demand: patients over 65 are imaged at two to three times the rate of younger adults, and the share of the population in that cohort has been growing steadily since the early 2000s. This is a structural driver, not a cyclical one.
Clinical pathway expansion also increases orders. As guidelines have been updated to include imaging in earlier stages of diagnostic workups for conditions including lung cancer screening (LDCT following USPSTF expansion), cardiac evaluation, and orthopedic pre-surgical assessment, the number of indicated studies has grown. Physicians ordering appropriately are still ordering more.
The third demand driver is ordering behavior drift. Research on imaging utilization suggests that a portion of imaging orders, estimates vary widely but published figures in the 20-30% range have been discussed in the literature for outpatient settings, may not meet high-acuity clinical necessity thresholds. This portion is harder to address through workflow tools because it requires upstream behavior change. But it's part of why volume growth outpaces population growth alone.
None of these three demand drivers are addressable by a radiology workflow tool. A pre-reading layer does not reduce order volume. Anyone claiming otherwise is misrepresenting what the technology does.
The Supply Side: Why Reading Capacity Has Not Kept Up
The radiologist supply problem is also structural, but for different reasons. Residency programs have relatively fixed throughput: graduating around 1,200 diagnostic radiology residents per year in the U.S. has been roughly stable for a decade while imaging volume has grown. The pipeline cannot rapidly expand even when the demand signal is clear.
Radiologist attrition compounds this. Radiology has seen elevated early retirement and career change rates post-pandemic, concentrated among mid-career practitioners who accumulated the combination of volume pressure, administrative burden, and remote reading isolation that shifted reading patterns during COVID. These practitioners left with experience that takes years to rebuild in the trainee pipeline.
Teleradiology has partially redistributed reading capacity by allowing radiologists to read for multiple sites simultaneously, particularly for overnight and weekend coverage. But teleradiology does not create new radiologists: it reallocates existing ones. And the subspecialty concentration that matters for complex reads (neuroradiology, musculoskeletal, interventional) is particularly constrained because fellowship pathways don't scale quickly.
The reading capacity constraint is structural and will not be resolved by hiring alone, at least not in the short term.
The Efficiency Dimension: Where Workflow Fits
Given that order volume will continue to grow and reading capacity will not grow proportionally in the near term, the third component of the backlog equation is efficiency: how much reading capacity is extracted from the radiologists who are available.
This is where workflow analysis yields actionable insights. A radiologist reading 40 studies in an 8-hour shift is not extracting the same capacity as one reading 60 studies in the same shift. The difference is almost never the radiologist's reading speed on individual studies. It's the time spent on activities that are not reading: worklist navigation, prior study retrieval, dictation overhead, interruptions, template loading, and report formatting.
Dictation time specifically is a documented source of variability. For a routine study where the report is essentially a template with findings documented: normal chest X-ray, no acute cardiopulmonary process, a radiologist dictating from scratch may spend 2-4 minutes on a report that, if pre-populated with an accurate AI draft, requires 30-60 seconds to review and sign. Across 30 routine studies in a day, that's a meaningful time savings. It doesn't address structural backlog in the way that adding 200 radiologists would, but it does extract more effective reading capacity from the panel that exists.
The Prioritization Problem Within the Backlog
Not all backlog is the same, and one of the least-discussed components of the backlog problem is the hidden cost of poor prioritization within it.
A worklist with 120 studies at 8am on a Monday morning is a backlog. But a worklist where 8 of those studies have clinically urgent indications and 112 are routine outpatient reads is not a uniform problem. The urgent studies have TAT implications that the routine studies don't. If the worklist sorts by arrival time, the first study opened is likely not the most urgent one. The urgent studies sit mid-list until a radiologist happens to scan far enough down to identify them.
Prioritization failure means that the backlog, even when it eventually clears, clears in the wrong order from a clinical standpoint. An urgent study read at hour four of a shift that should have been read at hour one is not a backlog metric failure: the study got read, and it will show up in the right column of the TAT dashboard. But it's a workflow failure that a flat TAT number conceals.
Pre-reading layers that include triage scoring address this by computing urgency indicators before the radiologist opens the study. The indicators are not diagnostic determinations: they're flagging signals that put likely-urgent studies near the top of the worklist. The radiologist still makes the clinical judgment. The computational layer just changes the order in which that judgment is applied.
What "Addressing" the Backlog Actually Means
The backlog can only be fundamentally reduced by reducing order volume, increasing reading capacity, or both. Neither is something a radiology workflow tool accomplishes.
What workflow tools address is the rate at which the backlog grows, not its existence. By extracting more effective reading capacity from the current radiologist panel, reducing TAT for routine studies, and improving prioritization within the queue, a pre-reading layer changes the dynamics of how the backlog accumulates and clears. In a steady state where order volume and reading capacity are matched, better efficiency could theoretically clear backlogs that would otherwise persist. In a steady state where volume meaningfully exceeds capacity, no workflow tool closes that gap independently.
We're not saying this to lower expectations. We're saying it because accurate framing of what the technology does matters for decisions about whether and how to deploy it. A department that believes a pre-reading tool will solve their backlog problem will be disappointed and will blame the tool. A department that understands it as one tool for improving efficiency within a structural supply-demand challenge will set it up correctly, measure the right things, and extract the value it can realistically provide.
The backlog problem requires a portfolio of responses: clinical decision support for ordering appropriateness, radiology training pipeline investment, teleradiology utilization, and workflow optimization. Pre-reading sits in the last category. It's a real lever. It's not the only one.