Radiology peer review and discrepancy rates: how reporting quality is measured

There is no single number that says whether a radiology service is good. Departments and teleradiology networks therefore measure quality with a small set of process measures: peer review of a random sample of reports, the discrepancy rate that results, compliance with turnaround targets, closed-loop communication of critical results, the addendum rate and report completeness. The most widely used peer review method is the American College of Radiology's RADPEER programme. An analysis of 41 million RADPEER reviews found that about 0.7 percent of reviewed cases were scored as discrepant and about 0.4 percent as clinically significant.¹ Those two figures are the benchmark most programmes compare themselves against.

What radiology peer review is

Peer review is a second radiologist re-reading a completed report and scoring whether they agree with it. Three features separate a real programme from an informal look over a colleague's shoulder.

• Random sampling. Cases are selected by the system, not by the reviewer or the author. A fixed share of each radiologist's volume is drawn across modalities, shifts and sites, so the sample reflects ordinary work rather than the interesting cases people remember.
• Blinding. The author does not learn who scored the case, and in many programmes the reviewer's identity is hidden from everyone except the quality lead. This keeps scoring honest in a small panel where everyone knows everyone.
• A fixed scale. The reviewer records a score on a short scale, plus a flag for whether any discrepancy was clinically significant. Programmes built into the reporting workstation, where the reviewer scores the prior report while reading the new study, are the model described in the peer review literature and remove the excuse of a separate log-in.³

The RADPEER-style scale used by most programmes runs as follows. The exact wording varies; the structure does not.

Discrepancy rates and what "significant" means

The discrepancy rate is the share of reviewed cases scored 2, 3 or 4. The clinically significant rate is the share flagged as likely to change management. In the 41-million-review analysis those rates were about 0.7 percent and 0.4 percent respectively.¹ Three cautions apply when reading these numbers.

1. They are averages across many practices. Case mix matters: a night service reading emergency CT of the head will not have the same profile as a daytime musculoskeletal MRI list. Compare like with like.
2. A very low rate is not proof of excellence. It can mean lenient reviewing, a sample skewed towards easy cases, or reviewers reluctant to score colleagues. A programme that never finds anything is not measuring anything.
3. Individual rates need volume. At a 0.7 percent base rate, a radiologist with 100 reviewed cases has, on average, less than one discrepancy. Trends per radiologist only become meaningful over hundreds of reviewed cases; before that, the group rate is the useful figure.

"Significant" is the flag that matters clinically. A missed incidental finding of no consequence and a missed pulmonary embolus are both discrepancies, but only one changes what happens to the patient. Every discrepancy scored 3 or 4 with a significant flag should trigger an addendum and a conversation with the referring clinician, whatever the wider statistics say.

Double reading and second reads

Double reading means two radiologists independently read every study before a single report is issued, and disagreements are resolved by discussion or a third reader. It is the most thorough form of quality control and roughly doubles reading cost per study, so it is used selectively: screening programmes, some high-stakes protocols and trainee reporting under supervision. It is not the norm for a general CT and MRI workload anywhere.

A second read, also called an overread or a second opinion, is different. It is requested for one specific study after the first report exists, usually because a clinician disagrees with the report, an unexpected finding needs confirmation before surgery, or a patient asks for another view. The ACR White Paper on Teleradiology Practice lists an overread path, a discrepancy process and addenda among the things a teleradiology provider is expected to offer.² On Raydiac a second read is a second assignment on the same order to a different credentialed radiologist, and the two reports sit side by side in the audit trail. See the second read service for how it is requested.

Other quality measures

Peer review captures interpretation. Four other measures capture whether the service around the interpretation worked.

• Turnaround compliance. The share of studies reported within the target for their priority. The ACR position is that teleradiology turnaround should be "not more or less stringent than for on-site radiology".² Measure breaches by priority, modality, site and shift, with the reason recorded. Our guide to radiology report turnaround time covers what the targets usually are.
• Critical results, closed loop. The ACR practice parameter on communicating imaging findings expects urgent findings to be communicated in a way that ensures receipt, and documented.⁴ A closed loop records who found what, who was called, on which channel, when they acknowledged it, and who was escalated to if nobody answered. The measure is the share of critical findings with a documented acknowledgement.
• Addendum rate. The share of signed reports later corrected or extended. A rate near zero can mean nobody looks back; a rising rate for one radiologist is a signal. Record the reason for every addendum.
• Report completeness. Whether each report contains the expected sections (technique, comparison, findings, impression, recommendations), answers the clinical question, and carries the radiologist's name, qualification, registration number and council. Structured templates make this measurable; our guide to the radiology report format explains what belongs in each section.

How a hospital should audit an outsourced reporting partner

A hospital that outsources reporting remains responsible for the quality of the reports its clinicians act on. NABH-accredited hospitals may use outsourced reads provided the partner meets the same documentation and quality standards as an in-house department.⁵ The practical way to enforce that is a monthly evidence pack, agreed in the contract, that the partner produces from its own system rather than compiles by hand.

If a partner cannot produce this pack, or produces it only as a typed summary with no underlying records, treat that as the finding. The evidence should come from an audit trail, not a memory.

How a teleradiology network should run QA

A distributed panel of radiologists working from different cities needs a stricter quality system than a single department, not a looser one, because nobody shares a corridor. Raydiac runs quality on the following principles.

• Random-sample peer review across the whole panel. The platform draws the sample; no radiologist chooses which of their cases are reviewed, and no reviewer chooses whose work they score.
• Discrepancy tracking per radiologist and per site. Scores, significant flags and addenda are recorded against the order in an append-only audit trail, and reported to each hospital for its own studies.
• Second reads as a second assignment. A hospital can request a second read on any order; it goes to a different radiologist credentialed for that organisation and modality.
• Service-level compliance, not turnaround claims. Each site has a service-level target per priority with a visible countdown, and compliance is reported per site, per priority and per shift.
• No star ratings. Ratings are opinions. Assignment and panel management use credentials, peer review results and service-level compliance instead.

The hospital-facing side of this is described on the for hospitals page. Our comparison of teleradiology companies in India notes which providers publish anything about their quality process at all.

A learning loop without blame

The purpose of measuring discrepancies is to reduce them, and a programme that feels punitive produces lenient scoring and defensive reporting. The practices that work are simple. The author of a discrepant report is told privately, sees the reviewer's reasoning and can respond, including disagreeing. Cases scored 3 or 4 are anonymised and discussed at a regular meeting as teaching cases, with the emphasis on the pattern rather than the person. Thresholds for escalation, such as a run of significant discrepancies in a defined period, are written down in advance so that intervention is a rule, not a reaction. And the same data is shown to the radiologist that is shown to management, so that nobody is judged on numbers they cannot see.