Denial Rate by Specialty in 2026

This is the part of a specialty breakdown that is worth having. It carries no percentages, because the only specialty-group rates we found, MGMA's 2019 and 2020 first-submission figures in the ledger below, carry no sample statement. Each row names the mechanism that generates the specialty's denials, what the governing rule actually says, and where to read it. Treat every row as the first place to look in your own remittances, not as a claim about national frequency.

What decides the mix is procedure mix, not specialty identity. Procedural specialties generate code pairs, so their denials cluster on bundling and modifier discipline. Visit-based specialties generate evaluation-and-management levelling and same-day-service questions, so theirs cluster on documentation of what made a service separately identifiable. Order-driven categories — DME, home health, imaging — cluster on whether the upstream document existed before the service did. Two practices in the same specialty with different procedure mixes will not have the same denial profile, which is the practical reason a specialty average could not have told you much even if one existed.

Every rule stated above was read from its source on 17 September 2026. Payer policies change without notice and vary by plan and state, so rows that name payer policy rather than federal rule are a place to start a check, not a rule to build a scrubber edit on.

Primary care denials divide cleanly into two families, and neither needs a national rate to work on.

Eligibility, created at scheduling. Coverage that ended before the date of service, and patients the payer cannot identify, are decided at registration rather than in billing — which is why this is the cheapest denial category to eliminate and the one that recurs monthly until the front-end workflow changes. The corresponding adjudication codes are CARC 27 and CARC 31; their published descriptions are copyrighted and should be read from the X12 list rather than from any guide, including this one.

The same-day E/M interaction, which is the underrated one. CMS made the G2211 visit-complexity add-on separately payable from 1 January 2024, and it is the add-on most relevant to longitudinal primary care: it "captures the inherent complexity of the visit that's derived from the longitudinal nature of the practitioner and patient relationship," and CMS says to bill it when "you're the continuing focal point for all needed services, like a primary care practitioner." The trap is the interaction with modifier 25. CMS is explicit: "G2211 isn't payable when you report the associated office and outpatient E/M visit with modifier 25, except in certain cases starting January 1, 2025." From 1 January 2025 the exception is narrow — G2211 is payable alongside modifier 25 "only when the service or other procedure requiring the reporting of modifier 25 is an allowed Part B service," which CMS lists as Part B preventive services, immunization administrations and Annual Wellness Visits. A practice that appends modifier 25 for any other same-day procedure and still bills G2211 will have that line denied, and it will look like a mysterious add-on denial rather than a rule. Read the current text in CMS MLN Matters MM13473, revised 1 May 2025.

What to measure instead of a benchmark. Split your denials into the two families above, then count each as a share of your own claims and watch the trend across months. That comparison is valid; a comparison to the one published primary-care figure we found, MGMA's 4% for 2020 (a pandemic year, sample not stated), tells you little about this year.

Behavioral health is the specialty where the absence of a published rate matters least, because the denial mechanism is unusually legible even without one.

Medical necessity is discretionary here in a way it is not elsewhere. There is no federal session-limit rule to point at. Session limits, frequency caps and continued-treatment criteria are set in individual plan documents, they differ between a commercial plan and its behavioral carve-out, and they change without notice. That is the finding, and it is why we publish no threshold numbers on this page: a session count or a visit cap quoted to the number reads as policy and would be acted on, and we could not tie any specific threshold to a retrievable payer policy document. Verify the current policy for your plan and state.

The recurring exposure is documentation of time, not code selection. Longer psychotherapy sessions are distinguished from shorter ones by duration, and duration is what a records request asks you to evidence. Notes that describe duration narratively rather than recording discrete start and stop times are the single most common reason a clinically appropriate session becomes a recoupment. The fix is a note template, not an appeal template.

Authorization and reauthorization are two separate failure points. Programmes with treatment plans — intensive outpatient, partial hospitalisation, applied behaviour analysis — carry an initial authorization and a recurring reauthorization on the plan's own cycle. Denials cluster at the second one, because the calendar for it sits with the clinical team while the claim sits with billing. Track authorization expiry against the schedule, not against the claim.

On concurrent billing in ABA. Where a protocol-modification service and a direct-treatment service overlap in time, the defensible position is time-stamped start and stop per rendering provider, recorded at the session. Retrofitting it at appeal is what makes these denials unwinnable. We publish no denial percentage for this specialty, and we removed the one this page used to carry.

Two mechanisms produce most surgical denials, and both are documented in CMS policy you can read.

The global surgical period. For major procedures the global package is longer than most schedulers assume. CMS: "Total global period is 92 days; count 1 day before surgery, the day of surgery, and the 90 days following the surgery day." Everything normally provided during that window is inside the surgical fee, which is why post-operative encounters get denied as included rather than as invalid. Three modifiers open it back up, and each has a distinct trigger:

• Modifier 24 — an unrelated E/M by the same provider during the post-operative period. CMS: "The provider must document the E/M service billed with modifier 24 and must send documentation supporting the unrelated service."
• Modifier 58 — a staged or related procedure that was "planned prospectively or at the original procedure's time," is "more extensive than the original procedure," or is "for therapy following a diagnostic surgical procedure." A new post-operative period starts when you bill the next procedure in the series.
• Modifier 79 — an unrelated procedure or service by the same provider during a post-operative period, which also starts a new post-operative period.

Modifier 78 is the separate case of an unplanned return to the operating or procedure room for a related complication. Read the current text in CMS MLN Booklet 907166, December 2025.

NCCI bundling. Procedure-to-Procedure edits exist, in CMS's words, "to prevent improper payment when incorrect code combinations are reported," and they apply automatically. A distinct-service modifier overrides an edit only where the record supports the distinction — a separate session, a separate anatomic site, a separate encounter — and modifier discipline created at the encounter is what makes that appeal winnable. Created at appeal time, it is not.

Authorization, the third mechanism, is where the specialty's dollars sit. The recurring pattern is not a missing authorization but a mismatched one: the authorization was obtained for the planned procedure and the procedure actually performed differs, or the authorization lapsed between approval and the scheduled date. Both are calendar and scheduling problems rather than billing problems. We publish no share-of-denials percentage for this category; the one this page previously carried was not traceable to a study and has been removed.

Global obstetric billing is a construction problem, not a coding problem. The global package bundles antepartum care, delivery and postpartum care into a single code. Every denial pattern in it comes from the episodes that do not fit that shape: a patient who transfers in or out mid-pregnancy, a payer change between trimesters, a delivery by a practice that did not provide the antepartum care. In those cases the components have to be billed separately rather than globally, and billing globally produces an inclusion denial while billing components against a global authorization produces a different one. Decide which construction applies at the first visit and record the decision, because reconstructing it after delivery from the chart is where the rework cost lands.

Pediatrics: the periodicity schedule is a state artefact. Medicaid's EPSDT benefit "provides comprehensive and preventive health care services for children under age 21 who are enrolled in Medicaid," and the screening schedule is not federal: schedules "must be provided at intervals that meet reasonable standards of medical practice," with states required to "consult with recognized medical organizations involved in child health care in developing their schedules." The operational consequence is specific. A well-child visit that is clinically appropriate but falls outside your state's schedule is a covered-service question decided by a document your state publishes, not by the payer's general policy — so the check is against the state schedule, per state, for multi-state practices.

Vaccine administration is a series-selection problem. Administration codes come in two families: a component-based family that counts each vaccine component and requires the counselling element to be documented for patients under 19, and a per-vaccine family that does not. Billing the per-vaccine family where the component-based family applies leaves money on the table; billing the component-based family without documenting counselling produces a denial on records request. The determining fact is the patient's age and whether counselling happened and was recorded.

We publish no denial rate for OB-GYN or for pediatrics. The bands this page previously carried for both were not traceable to a published source and have been removed.

A correction this page owes its readers. The previous version attributed imaging denials in part to Medicare's Appropriate Use Criteria requirement for advanced diagnostic imaging. That requirement is no longer in force. CMS states that, as announced in the CY 2024 Physician Fee Schedule Final Rule and "effective January 1, 2024, CMS has paused efforts to implement the AUC program for reevaluation and rescinded the AUC regulations at 42 CFR 414.94," and that "effective January 1, 2024, providers and suppliers should no longer include AUC consultation information on Medicare FFS claims." Any scrubber rule, appeal template or training deck still citing AUC as a Medicare denial driver is enforcing a rescinded rule. Read the current status on the CMS AUC program page.

What remains, and it is commercial rather than federal. Advanced imaging and the interventional cardiology procedure set are heavily prior-authorized by commercial plans, and the characteristic failure is not an absent authorization. It is an authorization attached to the procedure that was planned rather than the one performed — a study that extends, a diagnostic catheterisation that converts to an intervention in the same session, a protocol changed at the scanner. The authorization was valid for a procedure that did not happen. Because that is decided in the room rather than at scheduling, the only workable control is a same-day check between the performed procedure and the authorized one, before the claim drops.

NCCI is the second mechanism, and it is federal. Where a diagnostic study and a therapeutic intervention occur in one session, Procedure-to-Procedure edits will bundle them absent a distinct-service modifier, and CMS's stated purpose for those edits — "to prevent improper payment when incorrect code combinations are reported" — is why the override has to rest on the operative note showing the diagnostic component was clinically necessary in its own right rather than a roadmap for the planned intervention.

Third, the ordering-provider identifiers. Imaging claims carry both an ordering and a rendering identity, and a mismatch between them is an information defect corrected in scrubbing rather than a clinical dispute. It belongs on a different worklist from the two above, and routing it to the appeals team is a common and expensive misclassification.

We publish no denial rate for radiology or cardiology; the bands previously on this page were unsourced and have been removed.

Dermatology's denial profile is modifier 25 and almost nothing else. Derm visits routinely pair an evaluation and management service with a same-day procedure, and Medicare's global surgery policy is explicit about what that costs: minor surgery and endoscopy visits by the same provider on the same day are included in the global package "unless they do a significant, separately identifiable service," and "use modifier 25 to separately bill an identifiable Evaluation and Management (E/M) service by the same provider on the same procedure day." The word doing the work is identifiable. The defensible record shows a chief complaint, history and medical decision-making that would have existed had the procedure not been performed. Where it does not, the E/M is bundled and the appeal fails on the note, not on the argument.

The consequence most derm practices have not priced in. Appending modifier 25 also removes the G2211 visit-complexity add-on from the same claim in every case except the narrow Part B preventive, immunization and Annual Wellness Visit exception that began on 1 January 2025 — CMS: "G2211 isn't payable when you report the associated office and outpatient E/M visit with modifier 25, except in certain cases starting January 1, 2025." Any office-procedure specialty that bills longitudinal E/M alongside same-day procedures is making that trade on every such encounter, usually without knowing it. See CMS MM13473.

Urgent care has a structural eligibility exposure. Walk-in volume means there is no pre-visit window in which to run a verification, so coverage is confirmed at or after the point of service rather than before it. That is a workflow constraint rather than a billing error, and the practical controls are real-time verification at check-in and a same-day rework queue for what it catches. The second urgent-care exposure is place of service, which is mapped per contract and differs between Medicare and commercial plans — a per-payer fact to confirm against your own contracts rather than a general rule.

Podiatry, pain management, ophthalmology and the other office-procedure specialties share the shape. The procedure codes themselves adjudicate cleanly; the denials sit on the same-day E/M, on frequency edits, and on unit limits under Medically Unlikely Edits, which CMS describes as existing "to prevent improper payments when services are reported with incorrect units of service." No denial-rate figures or days-in-A/R figures appear in this section; the ones previously here were unsourced and have been removed.

These categories are governed by conditions of payment written into federal regulation, which makes them the most precisely verifiable section on this page — and it also means the widely repeated summaries of their requirements are frequently out of date.

DME: the face-to-face window is six months, and it is not universal. Under 42 CFR 410.38, for power mobility devices and other items selected for the Required Face-to-Face Encounter and Written Order Prior to Delivery List, "the treating practitioner must document and communicate to the DMEPOS supplier that the treating practitioner has had a face-to-face encounter with the beneficiary within the 6 months preceding the date of the written order/prescription." Two details that get lost: the six-month window runs backward from the order date, not from delivery; and the requirement attaches to items on that list, not to DMEPOS generally. Every DMEPOS item needs a written order with six specified elements, and for listed items that order must reach the supplier before delivery, while for everything else it must reach the supplier before claim submission.

A second correction. The previous version of this page listed a Certificate of Medical Necessity as a requirement for certain DME categories. For Medicare, it is not one. CMS announced it was "discontinuing the use of Certificates of Medical Necessity (CMNs) and Durable Medical Equipment (DME) Information Forms (DIFs)" for claims with dates of service on or after 1 January 2023; suppliers submit that information only for earlier dates of service where it was required. Read the CMS notice.

Home health: the encounter has a two-sided window. Under 42 CFR 424.22, certification requires that "a face-to-face patient encounter, which is related to the primary reason the patient requires home health services, occurred no more than 90 days prior to the home health start of care date or within 30 days of the start of the home health care," performed by a physician, nurse practitioner, clinical nurse specialist, physician assistant or certified nurse-midwife, and "the certifying physician or certifying allowed practitioner must also document the date of the encounter as part of the certification." The encounter may occur through telehealth. Recertification is required at least every 60 days where care continues.

Why these denials behave differently from every other specialty on this page. The defect is almost always upstream of the billing team and inside a referring practitioner's record — a missing encounter date, a narrative that does not evidence the clinical findings, an order that reached the supplier after delivery. That makes the appeal a document-retrieval exercise against a third party rather than a re-coding exercise, which is why these denials age. The operational answer is a pre-service completeness check against the regulation's own element list, run before the item ships or the episode starts, rather than a stronger appeal afterwards.

We publish no denial rate and no days-in-A/R figure for DME or home health. The ones previously on this page were unsourced and have been removed.

This is the whole ledger. Every quantitative claim on this page is in this table, or it is labelled as arithmetic in a worked example, or it is not on the page. Each row states the figure as its source states it, the population it was measured on, the denominator it rests on, whether it is a measurement or a published target, and where we read it on 17 September 2026. The MGMA 2019 and 2020 row was added and read on 1 October 2026.

Notice what is in the table: one row broken out by physician specialty group. It is data from 2019 and 2020, about six years old, with no sample stated, for three broad groups rather than for your specialty. Every other row is a practice-wide target, a payer-side measurement, a federal audit of denials, or a hospital population, and none of them is comparable to another, because the populations and denominators differ. A practice that measures itself against the 19% marketplace figure, or against a hospital figure, reaches a wrong conclusion in whichever direction the mismatch runs.

The number this page used to lead with. An 11.81% initial denial rate circulates widely as a 2024 industry average, and the previous version of this page attributed it to a Change Healthcare Revenue Cycle Denials Index for 2024. No such edition exists; the index that body published reported 11.1% of claims denied on initial submission in the third quarter of 2020, on hospital transactions. The 11.81% figure is Kodiak Solutions' proprietary Revenue Cycle Analytics data, which is not public and cannot be read in a retrievable report, measured across the hospitals and practice-based physicians on that one platform. Wrong publisher, wrong year, wrong population — and it was on this page. It has been removed rather than re-attributed.

Why a denial rate is only interpretable with its denominator stated. Reported rates differ depending on whether the denominator is total claims submitted, total claim lines submitted, unique encounters, or dollars — and the last one is easy to miss. The AAFP target above is a dollar-based rate; almost every rate a practice computes from its own system is a claim-count rate. They are not the same metric, and a high-dollar denial mix will move one without moving the other.

Claim-based initial denial rate. Numerator: claims with at least one service line denied on first adjudication. Denominator: claims adjudicated in the measurement period. Event: the first remittance (835) for that claim. Period: the month the remittance posts, not the month of service.

Line-based denial rate. Numerator: service lines denied on first adjudication. Denominator: service lines adjudicated in the same period. Same event, same period, different unit — which is why the two rates are not comparable and neither converts into the other without knowing the lines-per-claim mix.

The worked example, and which way the bias actually runs. Take a single claim with five lines, one of which is denied. At the claim level that is one denied claim out of one claim adjudicated — a 100% denial rate. At the line level it is one denied line out of five adjudicated lines — 20%. The claim-based rate is the higher of the two whenever a denied claim also carries paid lines, because a single denied line puts the whole claim into the numerator while its paid siblings stay in the line-level denominator. Some sources report at the line level and do not say so. Confirm both sides are on the same unit before comparing your number to any published figure.

Rejections are not denials, and a first-pass metric mixes the two. The complement of a first-pass metric is not all payer denials. Take 10,000 claims submitted in a month at a 92% first-pass rate: 800 did not resolve on the first attempt. If 600 of those were denied by the payer and 200 were rejected by the clearinghouse before they reached adjudication, the payer denial rate is 6% (600 ÷ 10,000) and the clearinghouse rejection rate is 2% (200 ÷ 10,000). Publishing the whole 8% as the denial rate overstates denials by a third and sends the fix to the wrong team — rejections are pre-adjudication data failures corrected in scrubbing and enrollment, denials are adjudicated decisions corrected in coding, authorization and documentation.

Also confirm whether 'denial' in a source includes clearinghouse rejections (it should not — those never reached a payment determination and carry no appeal rights) and whether it counts contractual and cost-sharing adjustments, which are write-offs rather than recoverable denials. Counting every adjustment line as a denial is the single classification error that produces the wildly divergent denial rates practices report to each other.

What we removed from this page, and why. Listed explicitly so a reader who remembers seeing a number here knows it was withdrawn rather than relocated:

• Every specialty denial-rate band. Primary care 7-10%, internal medicine 8-11%, mental health 12-16%, behavioral health 13-18%, ABA therapy 14-19%, orthopedics and surgical 13-17%, OB-GYN 10-13%, pediatrics 8-11%, radiology 9-12%, cardiology 11-14%, dermatology 8-11%, urgent care 9-12%, DME and home health 15-20%. None was traceable to a source; the only specialty-group rates we found are MGMA's 2019 and 2020 first-submission figures, in the ledger above.
• An 11.81% industry-wide initial denial rate credited to a Change Healthcare 2024 Revenue Cycle Denials Index, with a companion 10.15% for 2020. That edition does not exist, and the figure belongs to a vendor whose data is not public — see above.
• The denial root-cause distribution — registration and eligibility 27%, missing or invalid data 16%, authorization 12%, medical necessity 11%, service not covered 8%, provider eligibility 5%. Only the first is traceable at all, and only to a 2020 hospital-population index reported second-hand.
• A 10-13% cluster attributed to AHIP and AHA. Neither publishes it in a form we could retrieve.
• A 2023 workplace survey credited to AAPC as reporting a similar mean denial rate. AAPC does not publish a denial-rate survey, and we could not source the claim.
• Recovery and appeal rates by specialty and by denial category — 65-75% primary care, 45-55% behavioral health, and the category ranges of 80-90%, 50-70%, 40-60% and 50-65%. None was traceable to a source with a stated population.
• A claim that CARC 197 represents 25-35% of all denials in surgical practices. No payer publishes code-level adjudication frequencies, so no national figure for an individual code exists to cite.
• Days-in-A/R figures by specialty — 40-55 days for surgical, 28-35 for dermatology, 50-70 for DME and home health — and the operational assertions attached to them, including that structured eligibility verification moves a practice from 10-12% to 7-8%.
• A statement that denial rates rose 1-2 percentage points across 2020-2024. It rested on the two figures removed above.
• A Medicare Appropriate Use Criteria denial driver for advanced imaging, and a Certificate of Medical Necessity requirement for DME. Both described rules CMS has since rescinded or discontinued; the corrections are in the radiology and DME sections above.

The one comparison that is always valid. Your own prior period, on a fixed definition. Every external figure in the ledger carries a population that is almost certainly not yours; your last quarter does not.