Evidence
How each result was measured.
Every number on this site is listed here with its population, sample, period, comparison, method, and main limitation. Results describe the workflow they measured and do not carry over to other workflows. Availability of a workflow is stated separately on its page.
In production
9,067 · Imaging-report drafts generated by RadiologiQ
- Population
- Imaging studies at a participating orthopedic practice
- Sample
- Every draft generated since April 2026
- Period
- As of October 26, 2026, since April 2026
- Compared with
- None. This is a count, not a comparison.
Method. A running count of report drafts RadiologiQ has created in production. A physician reviews each draft before it can be saved to the chart. The count does not separate signed drafts from drafts still awaiting review. When the current count cannot be retrieved, the site shows the last recorded count with its date.
Limitation. Counts drafts generated, including drafts a physician has not yet signed. It does not measure payment.
Measured
97.5% · Agreement between CodeIQ OR's proposed CPT codes and the codes the operating surgeon billed
- Population
- Operative cases from one high-volume arthroplasty and trauma practice, five procedure families
- Sample
- 359 held-out cases
- Period
- Cases from 2018 to 2026, evaluated April 2026
- Compared with
- The operating surgeon's billed CPT codes
Method. The dataset is 3,951 de-identified operative cases from 2018 to 2026, held as a Limited Data Set. Cases were split with a fixed seed into example, evaluation, and held-out test sets; the held-out set was not viewed while the prompts were developed, and prompt changes were kept only when the evaluation set improved. Agreement is calculated for each procedure family and combined, weighted by the number of cases in each family. 4.2% of cases were flagged by the system as ambiguous and routed to human review.
Limitation. Measures agreement with how the surgeon billed, not clinical correctness. It applies to surgical CPT coding only, not to other OpAgent workflows.
| Procedure family | Cases | Agreement |
|---|---|---|
| Hip arthroplasty | 196 | 98.5% |
| Knee arthroplasty | 140 | 96.4% |
| Hip trauma | 11 | 90.9% |
| Knee non-arthroplasty | 10 | 100.0% |
| Hip infection | 2 | 100.0% |
| Weighted | 359 | 97.5% |
Measured
96.1% · Billed, fully adjudicated lines suggested by CodeIQ OR that the payer paid
- Population
- Surgical claim lines at one participating orthopedic practice
- Sample
- 1,436 claim lines
- Period
- Historical claims reviewed in the August 2026 benchmark
- Compared with
- The payer's adjudication of each line
Method. Among surgical claim lines that CodeIQ OR suggested and that were billed and fully adjudicated, the share the payer paid. Most cases were reviewed retrospectively, after billing.
Limitation. Most lines were billed before CodeIQ OR reviewed them, so payment does not show that the suggestion caused it.
Measured
201 · Unique patients flagged for bone-health review
- Population
- Patients on the clinic schedule at one participating orthopedic practice
- Sample
- 201 unique patients
- Period
- Four clinic days, August 2026 benchmark
- Compared with
- None. This is a count of flags, not a comparison.
Method. The number of distinct scheduled patients whose chart met the practice's criteria for a bone-health review flag over four clinic days.
Limitation. A flag asks the physician to consider a bone-health workup. It is not an order, a referral, or a completed visit.
Measured
39 of 40 · Bone-health flags supported by the chart on review
- Population
- Bone-health flags at one participating orthopedic practice
- Sample
- 40 flags in a fixed review sample: 39 supported, 0 unsupported, 1 with too little information to decide
- Period
- August 2026 benchmark
- Compared with
- Chart review of each flagged patient
Method. A fixed sample of 40 flags was selected before review. Each was judged against the chart as supported, unsupported, or not enough information to decide.
Limitation. A small sample. It shows whether each flag was supported by the chart, not whether care followed.
Measured
52.3% · Level-3 established visits whose signed note supported level 4
- Population
- Established-patient visits billed as 99213 at one participating orthopedic practice
- Sample
- 3,985 graded visits
- Period
- Retrospective sample, August 2026 benchmark
- Compared with
- The visit level that was billed
Method. Signed notes for visits billed as 99213 were graded against E/M documentation requirements. The figure is the share whose documentation supported 99214.
Limitation. Shows where to audit, not that every visit should have been billed higher. No encounter was re-billed.
Estimated exposure
$116,575 · Paid X-ray claims with no written report found in the chart
- Population
- X-ray claims at one participating orthopedic practice
- Sample
- Historical paid X-ray claims matched against chart documents
- Period
- Historical claims reviewed in the August 2026 benchmark
- Compared with
- Report documents on file in the chart
Method. Paid X-ray claims were matched to imaging report documents in the chart. The figure totals payments on claims where no written report was found.
Limitation. The amount already paid on claims missing a report. It identifies a documentation gap to review, not a confirmed recoupment or refund liability.
Measured
52.1% · Sampled Medicare epidural steroid injection records missing a core documentation item
- Population
- Traditional Medicare epidural steroid injection records at one participating practice
- Sample
- 190 sampled records that could be retrieved
- Period
- August 2026 benchmark
- Compared with
- Five core documentation items for repeat injections
Method. Each sampled record was checked for five items: a baseline pain scale, duration of conservative care, follow-up on the same scale, at least 50% relief before a repeat injection, and primary-care notification for long treatment courses. A record fails if any core item is missing.
Limitation. A missing item is a review finding, not a denied claim or a confirmed recoupment.
Measured
91 of 93 · Downcodes logged by billing staff that the detector also found
- Population
- Payer downcodes logged by one participating practice's billing team
- Sample
- 93 logged rows. One row had an invalid claim ID, so the detector found 91 of the 92 valid rows.
- Period
- Review log measured August 2026
- Compared with
- The billing team's downcode review log
Method. The detector was run over the claims in the billing team's downcode review log, and each logged downcode was checked for a matching detection.
Limitation. Finding a downcode is not recovering payment.
Measured
46 / 47 · Logged downcodes sorted into appeal work versus documentation or education follow-up
- Population
- Payer downcodes logged by one participating practice's billing team
- Sample
- 93 logged rows
- Period
- Review log measured August 2026
- Compared with
- Billing staff review of each logged downcode
Method. Each logged downcode was assigned to an appeal worklist or to documentation and education follow-up.
Limitation. Not every downcode can be appealed or recovered.