Same procedure, different price: Medicare Part B charge variation, 2024

A HCPCS code fixes two things and leaves a third loose. Medicare assigns each outpatient and physician service a HCPCS code — the Healthcare Common Procedure Coding System identifier — and sets a single administratively determined allowed amount for it under the Physician Fee Schedule. The code is fixed; the allowed amount is fixed. What a provider charges for that same code is not. Each practice sets its own list price, and those prices diverge far more than the clinical sameness of the code would suggest.

This study measures that divergence for Part B. It aggregates the CMS Medicare Physician & Other Practitioners by Provider and Service public-use file — one row per provider per HCPCS code per care setting — and, for every high-volume procedure, computes the spread of submitted charges across providers: the 10th percentile, the median, the 90th percentile, and the ratio between them. It is the outpatient, provider-level companion to the inpatient DRG charge-variation study, which measures the same dispersion across hospitals for inpatient stays. Here the unit is narrower — one provider, one code, one setting — but the question is the same: for one procedure, how far apart do providers price it?

How wide is the spread

Across 997 high-volume procedure-and-setting groups in 2024, the 90th-percentile provider billed a median of 3.6× what the 10th-percentile provider billed for the identical code in the same setting. The "high-volume" universe is every code-and-setting cell reported by at least 1,000 providers — 997 cells covering 8.76 million of the 9.0 million non-drug billing records in the file — so each percentile rests on a large, stable set of providers rather than a handful of outliers.

The middle of that distribution is high and tight: the 25th-percentile group spreads 2.9×, the 75th-percentile group spreads 4.6×, and 715 of the 997 groups (72%) carry a spread of at least 3×. At the floor, 995 of 997 groups (99.8%) run at least 2×. Wide provider-to-provider variation is not confined to a few exotic procedures — it is the standing condition of nearly every common service Medicare covers, from a routine office visit to a major joint replacement.

The HCPCS code is fixed and so is what Medicare allows for it. What a provider charges for it is not — for the same procedure, in the same setting, the dearest bill a median 3.6× the cheapest.

What it looks like on the most common services

Take the services Medicare pays for most often, and the spread holds across all of them, tightening for routine visits and widening for procedures. Each row below is the same code in its dominant care setting, priced across thousands of providers: the 10th-percentile provider's average charge, the median, the 90th, the 90/10 ratio, and how far the median charge sits above Medicare's allowed amount.

Established office visits — the workhorse of Part B — are the tightest cluster, at 2.4× to 2.8×. The most common single service, an established-patient visit at a moderate level of decision-making (CPT 99214), is billed by 450,513 providers, who price it from $160 at the 10th percentile to $416 at the 90th — a 2.6× spread for what is, by definition of the code, the same encounter. Move from visits to discrete procedures and the spread widens: a routine ECG runs 3.4×, a head CT 4.0×, and a cataract removal 4.7×, each across thousands of providers billing the identical code.

The widest-variation procedures

Rank the same universe by spread instead of volume, and orthopedic and spine surgery rise to the top. These are facility-setting procedures with a stable reporting base; we hold the 10th percentile above $50 so that a handful of nominal component-only charges cannot manufacture the ratio.

A total hip replacement (CPT 27130) is the clearest case because it is both common and dear: 8,164 surgeons bill it, and they bill it from $1,007 at the 10th percentile to $16,370 at the 90th — a 16.3× spread for the same coded operation. The clinical content of the code is the same; the list price is not. Arthroscopic shoulder work clusters even higher, from 15× to 20×, on smaller but still substantial provider bases.

Charges, allowed amounts, and which gap is which

Two different gaps live in this file, and they are easy to confuse. The first is the charge-to-allowed gap — how far above Medicare's allowed amount a provider's charge sits. Service-weighted across every non-drug Part B line in 2024, providers submitted $352.2 billion in charges against $103.1 billion that Medicare allowed — a 3.4× multiple. (Medicare then paid roughly 78% of the allowed amount, with the remainder falling to beneficiary coinsurance and deductibles.) The second gap, this study's subject, is the provider-to-provider spread on the same code.

The charge-to-allowed multiple itself varies by service, because providers mark up some codes more aggressively than others — the right-hand column of the common-services table makes that visible. A routine office visit carries a 2.1× charge-to-allowed multiple; a 12-lead ECG a 4.2×; a head CT a 4.8×. Neither multiple is what a patient or Medicare pays — Medicare pays its allowed amount — but together with the provider-to-provider spread they show submitted charges behaving as a markup policy decoupled from both cost and reimbursement.

How to read these numbers

The percentile columns and the charge-to-allowed column answer different questions, so read them in order:

1. The percentile columns (10th / median / 90th) are computed across providers for one code in one setting. Each provider contributes one observation — its own average submitted charge for that code — and a provider that billed the code five times counts the same as one that billed it five thousand times. This is deliberate: the question is how providers differ in pricing, not how many services each delivered.
2. The 90/10 ratio is the spread in a single number. A ratio of 4.0× means the provider at the 90th percentile of charges billed four times what the provider at the 10th percentile billed for the identical code in the same setting.
3. The charge-to-allowed multiple is a different axis entirely — service-weighted, comparing submitted charges to Medicare's allowed amount — and is included only to separate it from the spread, not as this study's finding.

A high spread on a code does not name any provider as overpriced or underpriced. It says that the same federally defined unit of care carries radically different list prices depending only on who bills it.

Methodology

Every figure is a direct aggregation over one public, read-only Postgres table: medicare_physician_service, built from the CMS Medicare Physician & Other Practitioners — by Provider and Service public-use file. The table carries one row per provider (npi) × HCPCS code (hcpcs_code) × place of service (place_of_service) × program year (data_year); it is public, read-only. The snapshot used here was ingested 2026-06-14 and covers program year 2024, the most recent annual release — 9,781,673 rows across 1,207,473 providers and 6,471 codes.

The program year is resolved as max(data_year) at query time, never hard-coded, so the figures advance when CMS publishes the next file. Drug codes (hcpcs_drug_indicator = true) are excluded throughout, because a drug's charge tracks its acquisition cost rather than procedure-pricing discretion. Charge percentiles are computed with percentile_cont over avg_submitted_charge, grouped by code and place of service, across the providers reporting that cell — one provider, one observation, unweighted. The setting is pinned into the grain because facility-versus-office billing and global-versus-professional-component splits otherwise inflate the spread (cataract removal pools to roughly 35× across settings but is 4.7× within the facility setting). The high-volume universe is restricted to cells with at least 1,000 reporting providers (997 cells) so each percentile is stable. The charge-to-allowed multiple, by contrast, is service-weighted — sum(submitted × services) / sum(allowed × services) — to reproduce the national per-service ratio.

CMS suppresses any provider-code cell with fewer than 11 beneficiaries; those rows are absent from the file and are excluded rather than imputed. The exact SQL is in the reproducibility block below, and the source-provenance contract is documented in the provenance methodology. Methodology version: medicare-physician-service/v1.

Limitations

• Medicare fee-for-service only. The file covers traditional Medicare Part B services billed under the Physician Fee Schedule. It excludes Medicare Advantage, Medicaid, and commercial claims, so the spread describes Medicare-billed activity, not every payer.
• Charges are list prices, not transactions. Average submitted charges are what providers billed, never what Medicare or a patient paid. The spread is a spread in markup policy, not in what care cost or what was collected.
• Provider averages, not per-service percentiles. Each provider contributes one averaged observation per code-and-setting cell. The spread is across providers, so it understates the full range an individual claim might carry within a high-volume practice.
• Within-code clinical variation is not fully controlled. A single HCPCS code fixes the procedure, but residual differences in patient complexity, modifiers, and regional input costs contribute to the spread alongside fee-schedule markup. The figures bound the variation; they do not attribute all of it to pricing discretion.
• Setting pinned, but two settings only. The file distinguishes facility from office (non-facility) place of service. Pinning the setting removes the largest confound, but device-inclusive and component-split codes still carry wide bundled-cost tails, which is why implant codes are reported separately.
• Suppressed small cells. Provider-code combinations under 11 beneficiaries are withheld by CMS and excluded, which slightly narrows the observed spread among low-volume providers.
• One program year, one snapshot. The figures are point-in-time: program year 2024, snapshot 2026-06-14. The Physician Fee Schedule and HCPCS definitions are revised annually, so they are not directly comparable across years without re-basing.
• A pricing-transparency signal, not a quality signal. These figures describe how providers price a code, never how well any provider performs. No provider is ranked, scored, or named in this study.

Sources

• CMS — Medicare Physician & Other Practitioners by Provider and Service — the Part B physician public-use file, program year 2024, the sole source for this study.
• CMS — Medicare Physician & Other Practitioners methodology — CMS's own definition of submitted charges, allowed amounts, payment, and the small-cell suppression rule.
• CMS — Physician Fee Schedule — the official source for how Medicare's allowed amount per HCPCS code is set.
• CMS — HCPCS Release & Code Sets — the procedure-coding system that defines each billable code.

The inpatient counterpart sits in the DRG charge-variation study and the national per-code DRG cost reference; for how Medicare's drug spending varies, see the most expensive Part D drugs and the prescribers behind half of Medicare's drug bill.