Radiology reading services: comparing the 3 models in 2026

You have imaging volume and you need signed reports; the question is who interprets the studies and under what arrangement. Radiology reading services in the US resolve into three models: a hospital-based radiology department, an independent radiology group under contract, and a teleradiology service reading remotely. Most facilities end up running some combination, and the combination is usually inherited history rather than a decision anyone made deliberately.

This page compares the three models on the axes that decide the choice: cost structure, turnaround, subspecialty depth, procedure coverage, quality oversight, and what happens when a radiologist leaves. The comparison is not close in every row, and where one model clearly wins, the table says so.

What counts as a radiology reading service?

Any arrangement that turns an acquired study into a signed diagnostic report is a reading service; the label covers three structurally different things.

Hospital-based reading means radiologists employed by, or exclusively contracted to, the facility. They work on site or in a home-and-hospital hybrid, attend tumor boards, perform image-guided procedures, and take call. The facility carries their salary, benefits, malpractice, and recruitment.

Independent radiology groups are physician-owned or investor-owned practices serving one or more facilities under professional services agreements. The group manages its own staffing and scheduling, and typically bills payers directly for the professional component, sometimes with a subsidy or stipend from the facility for coverage the professional fees do not support.

Teleradiology reading services interpret remotely. Studies route out by DICOM push, and reports return through HL7 into the RIS and EMR. Billing is usually per report, paid by the facility, which then bills payers itself. Coverage is defined by contract: which modalities, which hours, which turnaround tiers, preliminary or final reads.

The boundaries blur in practice. Many independent groups run internal teleradiology for nights. Many teleradiology companies now offer daytime subspecialty final reads, well beyond the overnight preliminary coverage the category started with. What follows compares the pure forms, because the pure forms are what a contract actually specifies.

Why is the comparison harder in 2026 than it was five years ago?

Because the labor market moved underneath all three models. US radiology residency positions grew 33 percent between 2010 and 2025 while the count of practicing radiologists grew only 12 percent from 2010 to 2022, per a 2026 JACR analysis of the radiologist pipeline, and demand keeps climbing on the other side of the gap: the ACR's Health Policy Institute workforce update projects imaging growth of 17 percent in MRI to 25 percent in CT by 2055. Recruiting benchmarks from the Association for Advancing Physician and Provider Recruitment put the average time to fill a radiology position at around 130 days, and rural and independent facilities routinely run past a year. Where the shortage bites hardest is mapped in radiologist shortage geography.

That scarcity repriced everything. Hospital-based recruitment now competes against remote work offers. Independent groups consolidated, and a facility whose group was acquired often finds the service model changed without the contract changing. Teleradiology absorbed much of the displaced demand, which is why the comparison below is no longer "your real radiologists versus the night vendor" but a genuine three-way sourcing decision.

AstraRad sits in the third category: a subspecialist teleradiology practice where board-certified, fellowship-trained radiologists sign final reports, billed per report with no minimums and no platform fees. That is the model this page's author operates, stated plainly so you can weight the analysis accordingly; the comparison rows still concede where the other two models win, because they do.

How the three reading service models compare

Criterion	Hospital-based	Independent group	Teleradiology service
Cost structure	Salary plus benefits, malpractice, recruiting; fixed regardless of volume	Professional fees to payers, often plus a facility subsidy	Per report, or per report with minimums; scales with volume
Turnaround control	High during staffed hours, thin nights and weekends	Set by group staffing; overnight often preliminary	Contractual tiers; STAT commonly under 1 hour by SLA
Subspecialty depth	Limited to who you employ	Depends on group size; strong in large metros	Broad panels routed by study type
Procedures and consults	Full: biopsies, drainages, fluoroscopy, curbside consults	Full at covered sites	None; interpretation only
Coverage continuity	One resignation can break call	Group absorbs individual departures	Panel absorbs departures invisibly
Quality oversight	Internal peer review, variable rigor	Group-run peer review	Contract-defined; ask for the discrepancy rate
Scaling for growth	Hire, at roughly 130 days per search	Renegotiate the agreement	Add volume on the existing contract

Two rows deserve emphasis because buyers consistently underweight them.

Procedures and consults are the row teleradiology loses outright. No remote service performs a paracentesis, protocols a difficult case at the scanner, or walks to the emergency department to look at a study with the treating physician. A facility with meaningful interventional or fluoroscopy volume needs an on-site radiologist for that work no matter how good its remote reading is. The honest framing is that teleradiology replaces interpretation capacity, never a radiology department.

Coverage continuity is the row hospital-based reading loses most often. When one of three staff radiologists resigns, the remaining two absorb call until a 130-day search ends, and burnout math takes over. An independent group softens this; a teleradiology panel eliminates it, because no single departure is visible to the client. What a resignation actually costs, and the 30-day response to one, is worked through in the 30-day plan for when a radiology group gives notice.

Where does each model win on cost?

Cost comparisons fail when they compare a salary to a per-read rate without a volume assumption, so fix the volume first.

A hospital-based radiologist FTE costs $350,000 to $500,000 or more per year fully loaded, varying with market and subspecialty. Against 12,000 reads a year of mixed work, the middle of that range is roughly $35 per read, before recruitment costs and before the fact that you pay it whether or not the volume arrives. Against 7,000 reads a year, the same salary is $60 per read.

Teleradiology pricing runs per study. Published market figures for final reads in 2026 sit around $10 to $15 for X-ray, $30 to $50 for single-region CT, and $45 to $75 for single-part MRI; the modality-by-modality detail, with its sources, is on teleradiology cost per read. The crossover is straightforward: below the volume that keeps an employed radiologist fully productive, per-study pricing wins; above it, employment wins, and a busy multi-radiologist department beats any per-read rate on marginal cost. If your volume is steady, high, and subspecialty-narrow, in-house is usually cheaper per read, and this page will not pretend otherwise.

The independent group model prices differently again: the group bills payers, and the facility pays a subsidy only where coverage exceeds what collections support. That can look nearly free on the facility's books until the subsidy conversation arrives, and subsidy requests have climbed steeply as group labor costs rose. A facility comparing bids should reduce all three models to expected annual cost at its own volume, then rerun the number at 80 percent of that volume, because the models diverge hardest in the slow scenario. The same arithmetic, applied to quote structures, is in per-report versus subscription pricing.

AstraRad's commercial terms are built for the variable-volume case: one price per signed report by study type, no monthly minimum, no subscription, no platform fee, and a written rate card within one business day of a request.

Quality: what the discrepancy literature actually shows

Every reading service claims quality; the models differ in how visible their quality is. The research base is worth knowing before you read any vendor's number.

Studies of emergency radiology consistently find measurable disagreement between a first interpretation and a later expert review. A study of after-hours CT reporting published in Healthcare found discrepancies in 17 percent of 7,209 preliminary reports, 6 percent clinically important, when compared against the final read (Storjohann et al., PMC). Emergency department series place major discrepancy rates in the low single digits, with the rate itself proposed as a standing performance indicator for any reading operation (Issa et al., PubMed). And an analysis of 5.9 million teleradiology examinations in the American Journal of Roentgenology found higher discrepancy rates when radiologists interpreted advanced examinations outside their fellowship training, with no such difference on common examinations, which is a precise argument for subspecialty-matched routing of exactly the complex studies.

Three practical conclusions follow. First, preliminary-read models carry a built-in rework tax: someone re-reads in the morning, and the two interpretations sometimes disagree; the operational fallout is detailed in STAT versus preliminary versus final reads. Second, a vendor quoting a discrepancy rate of zero, or refusing to quote one, is telling you it does not measure. Third, subspecialty match is a quality variable, not a luxury, and it belongs in your evaluation criteria whichever model you choose.

For scale: AstraRad double-reads 1 in 20 signed reports, blind, by a second subspecialist, and major discrepancies run under 0.3 percent of signed reports, reviewed case by case at a monthly discrepancy meeting. Clients receive the numbers monthly. Whatever service you evaluate, ask for the equivalent report; a service that produces one inside a week has a quality program, and a service that needs a month to assemble one has a marketing claim.

How do you evaluate radiology reading services before signing?

Run every candidate, in any model, through the same seven questions. The questions are model-neutral; the answers will not be.

1. Final or preliminary? Get it in writing per modality and per shift. A low rate that buys preliminary reads is not a low rate.
2. Who reads what? Ask for the subspecialty match policy: does a pediatric CT go to a pediatric radiologist at 2 a.m., or to whoever is on? Subspecialty reads by study type maps where the match changes outcomes.
3. What clock does the turnaround start on? Last-image arrival to signature is the honest measurement. Order time to preliminary is a different, softer number.
4. What is the discrepancy rate, and how is it measured? Ask for methodology and a sample quality report.
5. What happens at your volume plus 40 percent? A group hires in months; a panel should absorb it in days. Ask for the mechanism, and for evidence it has been done.
6. What are the exit terms? Termination notice, report archive handoff, and credentialing portability decide how painful a bad choice is to unwind.
7. What does the total cost look like at 80 percent of your volume? Minimums, subsidies, and fixed salaries all behave differently in a slow quarter.

The ACR's practice parameters and technical standards set the professional baseline any service should meet, including qualification requirements for interpreting physicians; a candidate unfamiliar with them is disqualifying itself. A longer version of this checklist, aimed at the teleradiology model specifically, is in how to choose a teleradiology company.

Mixing models without creating a gap

Most facilities land on a hybrid, and the failures live at the seams. Three rules keep the seams tight.

Write one routing rule per study, by modality and shift, and publish it to the technologists. "Neuro MRI goes to the group weekdays 8 to 5, to the teleradiology service otherwise" survives a busy night; "use judgment" does not.

Assign every quality event a single owner. When a referrer disputes a read, someone specific must own the callback, whichever service signed the report. Split arrangements without this rule produce discrepancy reviews that nobody convenes.

Keep one turnaround definition across services. If your in-house number starts at order time and your vendor's starts at image arrival, your dashboard is comparing two different clocks and will flatter whichever service chose the softer one. How to audit a vendor's stated numbers, clock definitions included, is covered in verifying teleradiology vendor claims.

The decision between models is rarely permanent, and the right split moves as your volume and staffing move. What stays constant is the evaluation discipline: final reads, named readers, a measured discrepancy rate, a turnaround clock you can audit, and commercial terms that survive your slowest month. If the teleradiology column of the table fits part of your operation, AstraRad will put its terms in writing first: SLA tiers of STAT under 1 hour, urgent under 4 hours, routine under 24 hours, and a complete per-report rate card within one business day of a request. The SLA page carries the measured numbers, updated monthly.