Radiology outsourcing: what to send out and how to govern it

Your reading capacity no longer covers your volume, and the choices are hire, cap the schedule, or send some of the work out. Radiology outsourcing is the third option, and it fails or succeeds on two decisions that get less attention than the vendor's name: which work leaves the building, and what governance follows it out the door.

This is a how-to for both decisions. It assumes you have read the case for and against in in-house versus outsourced radiology and have decided some interpretation is leaving; what follows is how to draw the line and how to stay accountable for what crosses it.

What should radiology outsourcing cover, and what should stay?

Draw the line by workload, not by percentage. Interpretation travels; presence does not. Four workloads outsource well, and four should not leave.

Send out: nights and weekends. Overnight coverage is the original outsourcing case and still the strongest. A facility running 8 to 15 overnight studies cannot keep a radiologist awake for them at any defensible cost, and a night vendor reading on scheduled shifts produces better reads than a solo radiologist on call after a full day. Insist on final reads; a preliminary-only night service leaves your own group re-reading every study in the morning, and the arithmetic of that rework is laid out in STAT versus preliminary versus final reads.

Send out: subspecialty studies outside your bench. A general radiologist can read a temporal bone CT; the question is whether they should when a neuroradiologist is a routing rule away. Discrepancy research on 5.9 million teleradiology examinations shows rates rise when radiologists interpret advanced examinations outside their fellowship training, with no such difference on common studies (AJR analysis), which makes subspecialty overflow the quality-driven outsourcing case rather than the cost-driven one.

Send out: overflow above daily capacity. Set a numeric trigger, for example any unread routine study older than 20 hours, and route the excess automatically. Overflow that waits for a manager to notice a backlog is not an overflow program.

Send out: backlog clearance. A one-time project with a defined study list, a schedule, and a weekly count. Do not blend it into the ongoing contract; blended backlogs never finish.

Keep: procedures, fluoroscopy, and contrast supervision. No remote service performs a drainage or supervises a reaction. If you have this work, you need on-site physician presence regardless of how much interpretation you outsource.

Keep: scanner-side protocoling and consults. The five-minute conversation at the console that turns a wrong study into a right one is invisible in any cost model and disappears when everything is remote. Facilities that outsource heavily should route protocol questions to a named vendor contact and measure the response time, because this is the quality seam most contracts ignore.

Keep: tumor boards and multidisciplinary conferences. These run on relationships and preparation. Some teleradiology practices join remotely; treat that as a negotiated add-on, never an assumption.

Keep: the quality program itself. You can outsource reads. You cannot outsource accountability for them, and the rest of this page is about what that means in practice.

Who is accountable for an outsourced read stream?

The facility is, and the governance structure should make that explicit before the first study routes out. Reporting in Radiology Business has warned hospitals for years that outsourcing arrangements erode trust when clinicians cannot see who stands behind a report; the antidote is named accountability at three levels.

A named clinical owner on your side. A medical director, chief of radiology, or supervising physician who owns the vendor relationship, chairs discrepancy review, and can suspend routing when quality slips. In an imaging center without a radiologist, this is the medical director of the center; the role cannot be vacant.

A named counterpart on the vendor side. Not a support queue. A specific person who receives discrepancy cases, answers protocol questions, and appears at the quarterly review with data. AstraRad structures this deliberately: every report carries the signing subspecialist's name, a named liaison owns each client account, and a monthly quality report goes to the client with turnaround by tier, SLA compliance, and the discrepancy log. Whoever you contract with, require the equivalent; a vendor whose quality reporting goes into a drawer has told you its priorities.

A written routing rule. One document stating which studies go where, by modality, shift, and priority, published to technologists and reviewed when volumes change. Every study's destination should be decidable in two seconds without judgment.

The professional baseline underneath all of this is set by the ACR practice parameters and technical standards, which define the qualifications and communication standards any interpreting physician should meet, on site or remote. Write conformance with the relevant parameters into the contract.

Credentialing: the governance step that sets your timeline

Nothing routes anywhere until the readers are credentialed, and this step decides whether your transition takes three weeks or five months.

Hospitals have two paths. The traditional one runs every remote radiologist through full medical staff credentialing, which is rigorous and slow, and painful with a vendor panel of 40 readers. The faster path is credentialing by proxy under 42 CFR 482.22 of the Medicare Conditions of Participation, which lets your governing body rely on the distant-site entity's credentialing and privileging decisions under a written agreement meeting the regulation's requirements, including a current privileges list for each physician and licensure in your state. Your bylaws must permit it, and your accreditor's parallel requirements apply; the mechanics and the agreement checklist are in credentialing by proxy for teleradiology, and the distinction between the two underlying processes is in privileging versus credentialing.

Imaging centers and urgent care groups without medical staff structures verify directly: state licensure for each reading state, board certification, malpractice coverage, and exclusion screening. Ask the vendor for its credential files during procurement, before signature. A vendor with organized files produces them in a day; disorder here predicts disorder everywhere.

Whichever path applies, put licensing responsibility in the contract: the vendor warrants that every reader interpreting your studies holds an active license in your state, and reports panel changes to you monthly.

How do you run quality oversight of radiology outsourcing?

Measure the outsourced stream at least as hard as you measure your own radiologists, and put the measurement on a calendar. A working program has four instruments.

Discrepancy tracking, with a defined path. When a treating physician disagrees with an outsourced read, that event needs a form, an owner, and a clock. Emergency radiology research treats the discrepancy rate between initial and reviewed interpretations as a standing performance indicator for a reading operation (Issa et al., PubMed); your program should produce that number monthly, split by vendor and by modality. Expect a real number: published major discrepancy rates in the literature run from a fraction of a percent to the low single digits depending on setting and definition. A vendor reporting zero forever is not measuring, and a facility that only counts complaints is measuring referrer patience instead of accuracy.

Peer review sampling. Pull a random sample of vendor reads monthly, 2 to 5 percent is a common range, and have your own or an independent radiologist score them using a structured scale. RADPEER-style scoring gives you trendable data instead of anecdotes; the scale and its mechanics are explained in RADPEER scoring. The sample must be random. A review queue built from complaints oversamples known problems and proves nothing about the stream.

Turnaround compliance, on one clock. Measure vendor turnaround from last-image arrival to final signature, monthly, against the contracted tiers, and reconcile the vendor's reported numbers against your own PACS timestamps quarterly. Vendors measure themselves generously when nobody checks; the audit method is in verifying teleradiology vendor claims.

A standing review meeting. Monthly for the first two quarters, quarterly after that: discrepancy log, peer review scores, turnaround compliance, invoice reconciliation, and any protocol or critical-results issues, with both named owners present. The meeting is the governance; everything else is data collection for it.

For calibration, the numbers AstraRad publishes to its own clients: 1 in 20 signed reports independently double-read by a second subspecialist, blind to the first read; major discrepancies under 0.3 percent of signed reports, each reviewed at a monthly discrepancy meeting and closed with the reader; 99.4 percent of reports inside their SLA tier over the trailing 12 months, measured from last-image arrival to signature. Those figures sit on the SLA page, dated. Ask every bidder for their equivalents in writing, and treat a refusal as data.

The commercial governance: fees, minimums, and exit

Three contract terms decide what outsourcing actually costs, and all three are cheaper to negotiate before signature than after.

Itemized pricing. Per-read rates are comparable only when every fee is on the table: platform fees, after-hours surcharges, subspecialty routing charges, critical-results call fees, addendum charges. The full catalog of quiet line items is in teleradiology hidden fees. AstraRad's structure, for scale: one price per signed report by study type, a printed STAT and urgent multiplier, and nothing else, with no minimums, subscriptions, or platform fees, and a written rate card within one business day of a request.

Volume terms that match your reality. A monthly minimum converts a slow month into a rate increase. If your volume swings, price the swing before you sign, and compare structures with the arithmetic in per-report versus subscription pricing.

Exit terms. Notice period, report archive handoff in a usable format, cooperation during transition, and no fee to leave. Outsourcing you cannot unwind is not a vendor relationship; it is a dependency. The time to read the exit clause is before the relationship needs one.

Four ways outsourcing programs fail, and the guard for each

The failure modes below come up repeatedly in facilities that outsourced first and governed later. Each has a cheap structural guard.

Scope creep without a decision. The contract covered nights; eighteen months later the vendor is reading 60 percent of daytime volume because routing "just grew." Nothing is wrong with a larger scope, but an unmanaged one means nobody re-ran the cost model, re-checked subspecialty match, or asked whether referrers noticed. Guard: the routing rule is versioned, and any change to it goes through the clinical owner.

The prelim trap. A preliminary-read night contract quietly doubles the true cost per study, because your own radiologists overread every case in the morning, and it doubles the discrepancy surface too, since two interpretations of the same study can disagree. Guard: final reads by contract wherever your group does not intend a full overread, and an explicit count of overread hours wherever it does.

Quality review that only fires on complaints. A referrer-complaint pipeline finds the loud discrepancies and misses the silent ones, and it finds them months late. The facilities burned worst by outsourcing are usually the ones that had no random sampling running when a pattern problem started. Guard: the monthly random peer review sample, sized in advance, reviewed on schedule even when it is boring.

Turnaround numbers on two clocks. The vendor reports 98 percent SLA compliance measured from image receipt; your emergency department measures from order time and experiences something much slower, and the gap between the two clocks is your own transmission delay that nobody owns. Guard: reconcile vendor-reported turnaround against your PACS timestamps quarterly, and assign the transmission leg to a named owner on your side.

None of these is an argument against outsourcing. Each is an argument for the governance being in place on day one, because every one of them is invisible in month one and expensive by month twelve.

A 60-day governance rollout

Sequenced, for a facility outsourcing a defined slice for the first time.

Days	Work
1 to 10	Define the outsourced slice by modality, shift, and trigger. Write the routing rule. Name both clinical owners.
5 to 30	Credentialing: proxy agreement or direct verification. Contract review against the seven-item checklist in the FAQ below.
15 to 30	Technical integration and test studies; confirm report format with referring physicians before go-live, and pick which reading service model covers each slice using the comparison of reading service models.
30 to 45	Go live on the defined slice only. Daily turnaround check for the first two weeks.
45 to 60	First monthly quality review: discrepancy log, peer review sample, turnaround reconciliation. Adjust the routing rule with data.

Facilities that skip straight to go-live usually retrofit the governance after the first disputed read, which is the most expensive time to build it. The program above costs a few hours a month once running, and it is what makes the difference between outsourcing you control and outsourcing that happened to you. When you are ready to price a defined slice, AstraRad will put SLA tiers and a complete per-report rate card in writing within one business day of a request.