Junto Analytics — Methodology

Version 1.0 · August 2026 Andrew T. Duncan, MBA, MS, PT, DPT, ATC Founder, Junto Analytics andrew@juntoanalytics.com

---

1. Purpose and scope

This document describes how Junto Analytics builds commercial reimbursement rate statistics from federally mandated price transparency disclosures, and how those statistics relate to physician compensation. It is written for readers evaluating the analytical approach — collaborators, reviewers, and anyone working with the same public sources.

It covers method, not implementation. Table structures, infrastructure, and code are omitted.

Every source described here is publicly available. No proprietary survey data, licensed benchmark, or institutionally supplied dataset is used at any stage. The approach is reproducible in principle by anyone willing to build the ingestion.

---

2. Data sources

2.1 Rate data

Hospital Price Transparency (HPT) filings. Required under 45 CFR §180. Hospitals publish machine-readable files disclosing gross charges, discounted cash prices, payer-specific negotiated rates, and de-identified minimum and maximum negotiated rates. This is the primary rate source.

Transparency in Coverage (TiC) filings. Required under 45 CFR §158. Payers publish in-network negotiated rates and out-of-network allowed amounts. Used for cross-validation and for professional-component rates where hospital filings are silent.

Hospital Ledger (hospitalledger.com). A CC0-licensed aggregation of HPT data. Used as a supplementary source where it covers hospitals not yet in our own ingestion. All rows from this source are tagged and can be isolated or excluded from any statistic.

2.2 Reference and benchmark data

2.3 Source authority

Every rate row carries an explicit authority classification:

Reference-only data is physically excluded from the analytical dataset rather than filtered at query time. Anything that must never be published should not be present where a query can reach it.

---

3. The central methodological problem

Professional-component and facility-component revenue are different streams. Comparing one against the other produces figures that appear meaningful and are not.

A total knee arthroplasty illustrates the scale. The surgeon's professional fee runs roughly $1,500–$2,000. The hospital's facility fee for the same case runs roughly $25,000–$40,000, covering operating room time, implants, nursing, supplies, and institutional overhead. These fund different things and are negotiated under different arrangements.

Early in this work, our model divided a facility contracted rate by a professional fee schedule and reported the result as a percentage of Medicare. It returned 1,408%. The arithmetic was correct. The comparison was not, and the figure described nothing anyone could act on.

The failure mode matters more than the specific error. An unlabeled ratio is not wrong so much as indefensible — and it collapses at the first informed challenge, taking the credibility of everything reported alongside it with it.

3.1 The rule

Every rate, ratio, and derived statistic declares its revenue basis:

A ratio may be computed only between a numerator and denominator sharing a basis. Cross-basis division is prohibited. Where cross-basis context is genuinely useful — and it often is, for site-of-service and service-line questions — it is expressed as a separately named metric, never as a ratio.

3.2 Practical consequence

Reporting a "percent of Medicare" without naming the payment system is insufficient. % of Medicare PFS, % of Medicare OPPS, and % of Medicare IPPS are three different statistics. A reader who compares an outpatient facility figure against a professional figure, both labeled "% of Medicare," reaches a conclusion the data does not support.

---

4. Medicare reference rates

Each lane is benchmarked against the payment system that actually governs it.

4.1 Professional component — PFS

Medicare professional rate =
  (work RVU × work GPCI + PE RVU × PE GPCI + MP RVU × MP GPCI)
  × conversion factor

The practice-expense component is basis-dependent: facility PE RVU applies where the service is performed in a facility setting, non-facility PE RVU otherwise. Using one where the other belongs shifts the result materially.

Two conversion factors now exist by statute — one for qualifying alternative payment model participants and one for all others. Both are carried; neither is ever written as a literal in code.

4.2 Hospital outpatient facility — OPPS

Medicare OPPS rate = APC relative weight × OPPS conversion factor

Codes are mapped to Ambulatory Payment Classifications through the CMS OPPS Addenda. Only codes with payable status indicators receive a reference rate. Packaged, non-covered, and excluded codes are left null rather than assigned a benchmark that does not apply to them.

The conversion factor is verified by division — payment rate divided by relative weight across all payable APCs — rather than taken from a secondary summary. Deriving it from the published data confirms the value rather than trusting a restatement of it.

4.3 Inpatient facility — IPPS

Medicare IPPS rate =
  DRG relative weight
  × [ operating standardized amount × (labor share × wage index + non-labor share)
      + capital federal rate × geographic adjustment factor ]

The labor-related share is tiered by wage index — a statutory lower share applies at or below 1.0, a rebased higher share above it. Non-labor share is derived as one minus the labor share rather than stored independently; two separately maintained numbers can drift out of agreement and silently stop summing to one.

Hospital-specific wage index and geographic adjustment are applied where the hospital can be resolved. Where it cannot, a national reference at wage index 1.0 is used and the row is labeled accordingly. The two are never presented under a single label.

This benchmark reflects base operating and capital payment. It excludes disproportionate share, indirect medical education, outlier, and other add-on payments, and is therefore a reference rate rather than an estimate of actual Medicare payment.

4.4 Clinical laboratory — CLFS

Laboratory services are paid under a separate national fee schedule and appear on neither the physician nor the outpatient schedule. A rate is published per code, with a small number of codes carrying a modifier-specific rate. A payment indicator distinguishes standard rates from those subject to a statutory inflation cap.

There is no geographic adjustment in the base file, so the CLFS reference is national.

Omitting this schedule was our most consequential reference gap. Matching only against the physician and outpatient schedules left roughly 87,000 five-digit codes unmatched, which appeared to be large-scale chargemaster contamination. The bucket was in fact dominated by venipuncture, complete blood count, hemoglobin A1c, and prothrombin time — among the most frequently billed codes in American medicine, present at more than 1,700 hospitals across 52 states. They were absent because no loaded file governed them.

4.5 Drugs and biologicals — ASP

Part B drugs are paid at the average sales price plus six percent, published quarterly as a payment limit per HCPCS billing unit. A minority of entries price through alternate methodologies — vaccines through an average wholesale price percentage, blood products and clotting factors through their own formulas — and those are flagged and analyzed separately, since the ASP-plus-six-percent floor does not apply to them.

Units are the binding constraint. The payment limit is per billing unit as the code defines it: one unit for some agents, a specific milligram quantity for others. Hospital filings frequently price a vial or a typical administration without stating the unit. A percentage against the ASP reference is therefore valid only where units align, and transparency filings rarely permit confirming that.

Every drug rate carries a unit-basis marker. Those rows are available for individual lookup and excluded from every published distribution, because a rate filed per vial against a per-milligram benchmark produces a ratio that describes nothing.

4.6 Equipment, orthotics, and supplies — DMEPOS

Durable medical equipment, orthotics, prosthetics, and supplies are paid under a fourth schedule that differs structurally from the others: rates vary by state and by rural designation, competitive bidding areas price differently again, and modifiers distinguish new purchase from rental from used equipment.

We take the national non-rural rate as the reference and record the variation as a limitation rather than modelling it. Actual state rates diverge materially from that figure. Like drug rates, DMEPOS rows carry a unit-basis marker and are excluded from published distributions — a purchase price compared against a rental benchmark is not a meaningful ratio.

Orthotic and prosthetic codes are the largest component by volume, which makes this schedule more relevant to musculoskeletal analysis than its overall share of the corpus suggests.

4.7 Discipline on constants

No CMS rate, conversion factor, or share is written as a literal anywhere in the pipeline. All are stored as versioned reference values with a year, a status of proposed or final, and a citation.

This is not a style preference. A constant with no recorded origin appeared in our IPPS calculation and survived for weeks before it was traced. It was neither the national standardized amount nor any published rate — it appeared to be a value copied from a hospital-specific rate applicable only to Sole Community and Medicare-Dependent hospitals, applied nationally. Every DRG benchmark computed during that period was wrong by roughly 11%.

An unattributed number in a pricing model is a defect regardless of whether it happens to be close.

---

5. Normalization

5.1 Payer identity

Payer names in filings are inconsistent across hospitals and often within a single file. A single national carrier may appear under dozens of string variants — case differences, abbreviations, legal entity names, product-line suffixes, and regional brands.

Names are normalized against a curated canonical mapping. Where no mapping exists, the raw name is retained and flagged as unmapped rather than nulled. A rate that exists should remain visible even when its payer cannot yet be resolved.

Blue Cross Blue Shield is deliberately not collapsed. BCBS comprises independent licensees with materially different rate structures — a bare "BCBS" string cannot be resolved to a licensee, and mapping it to a single canonical entity would produce payer comparisons that are wrong rather than incomplete. Bare BCBS strings remain unmapped by design.

5.2 Product line

A payer name alone does not identify the product. National carriers operate commercial, Medicare Advantage, Medicaid managed care, and exchange products under the same brand, and several carriers use distinct brands for Medicaid lines.

Product line is classified from plan-level detail where the filing provides it, falling back to payer-level inference where it does not. Rows classified at plan level are distinguished from rows classified by payer name alone.

This matters more than it appears. Pooling Medicaid managed care rates with commercial rates depresses every commercial statistic. Commercial rate reporting that has not separated product lines is reporting a blended figure under a commercial label.

5.3 Procedure codes

Filings contain a mixture of standard HCPCS and CPT codes, hospital-specific chargemaster identifiers, revenue codes, NDC drug codes, and compound multi-segment keys. In our corpus the raw code field held over 250,000 distinct values against an active code universe of roughly ten thousand.

Codes are normalized — case, whitespace, and leading-zero padding — then matched against the union of the current PFS and OPPS reference files. Non-matching identifiers are retained but excluded from rate statistics.

Leading zeros are handled in both directions. Some sources strip them, producing four-digit values from five-digit codes; others pad them, producing six characters. Both patterns appear, and normalizing only one direction discards real codes as junk.

The reference universe must span every applicable fee schedule. A validity flag built on an incomplete set of reference files measures the completeness of that set, not the validity of the code.

We learned this the expensive way. Matching only against the Physician Fee Schedule and the Outpatient Prospective Payment System left roughly 87,000 distinct five-digit codes unmatched, which looked like large-scale chargemaster contamination. Sampling showed otherwise: the bucket was dominated by venipuncture, complete blood count, hemoglobin A1c, and prothrombin time — among the most frequently billed codes in American medicine, appearing at more than 1,700 hospitals across 52 states.

They were absent because clinical laboratory services are paid under a separate schedule, the Clinical Laboratory Fee Schedule, which neither of the two loaded files contains. The same applies to most drug and biologic J-codes, which price through Part B average sales price rather than through outpatient surgical classifications.

Four distinct populations therefore sit in any "unmatched" bucket, and they require different treatment:

The first two are indistinguishable from the third by format alone. Reference matching against a complete schedule set is what separates them; a historical code master would be required to separate the second from the third.

Rather than a validity flag, each code carries the fee schedule that governs it — PFS, OPPS, IPPS, CLFS, ASP, or none. Rate statistics then gate on the schedule appropriate to the lane rather than on a boolean, which also handles the case correctly where a code is legitimately outside every schedule.

Aggregation across code type is never performed. Outpatient CPT rates and inpatient DRG rates differ by orders of magnitude; a single average across both is meaningless.

CPT code numbers are used as identifiers. AMA descriptor text is not displayed.

5.4 Provider identity

Hospital identifiers in filings are heterogeneous: valid NPIs, employer identification numbers used in an NPI field, CMS Certification Numbers, bracketed or delimited multi-NPI arrays where a system files under several campus identifiers, and synthetic keys from third-party sources.

Each identifier is typed rather than assumed. System-level filings attributed to a first campus identifier are labeled as an attribution limitation, not as resolved identity.

Identity resolution feeds geography, and geography gates market-level analysis. A hospital that cannot be resolved to a state cannot appear in any geographic statistic — which means identity quality determines analytical coverage more directly than raw row count does.

5.5 What a code represents, separately from what prices it

A procedure and the drug administered during it are billed as separate line items. An intra-articular injection is coded once for the administration and again for the substance injected. The administration code is identical whether the syringe contains a corticosteroid costing a few dollars or a hyaluronan preparation costing several hundred.

This has a consequence that is easy to miss and materially distorting.

Drug and device codes carry no physician work RVU. The physician's work is captured entirely in the administration code. The drug is an acquisition cost passed through with a markup. Any metric expressed per work RVU — revenue per wRVU, compensation per wRVU, yield — will therefore be inflated for specialties that administer expensive drugs, because the numerator grows while the denominator does not.

Rheumatology, oncology, and portions of orthopaedics are affected substantially. A clinician performing viscosupplementation or biologic injections generates the same work RVUs as one injecting a steroid, while the institution collects an order of magnitude more.

Every code therefore carries a service category independent of its fee schedule:

Category is derived from what the code represents rather than from its letter prefix, which is unreliable — the G series contains both pass-through drugs and genuine physician services, and the Q series spans drugs, supplies, and procedures. Outpatient status indicators encode the distinction directly and are the primary signal.

Drug rates are analyzed separately, and against a different question. For a drug, the meaningful statistic is not the rate but the multiple over Average Sales Price. Medicare pays ASP plus six percent; a commercial rate near that multiple represents cost recovery, while a substantially higher one represents margin on acquisition. That is a different economic question from what a payer pays for a procedure, and mixing the two produces a distribution that answers neither.

---

6. Exclusion rules

Filings contain values that are technically compliant disclosures but are not negotiated prices. Each is excluded from rate statistics and retained in the underlying data with a reason.

Gross charges filed as payer rows. Some hospitals disclose chargemaster amounts under a pseudo-payer label. These are list prices, not negotiated rates.

Per-diem and case-rate methodologies in the DRG lane. Where a single rate is applied uniformly across a large number of DRGs, the filing reflects a per-diem or case-rate arrangement rather than DRG-specific pricing. Treating a uniform daily rate as a DRG rate depresses inpatient statistics substantially. Detection is by uniformity across DRGs at a hospital-payer pair, corroborated by the filed methodology field.

Nominal sentinel values. Rates filed at values such as $0.01, $0.10, and $1.00 appear across a large number of hospitals, payers, and states. These satisfy the disclosure requirement without disclosing a price. They are flagged separately from data-quality outliers, because they appear to be a filing convention rather than an error — and the distinction is analytically meaningful.

Unit-dependent references. Drug and equipment rates are benchmarked per billing unit, and transparency filings seldom state the unit they priced. These rows are retained and available individually but excluded from every published distribution. A ratio that cannot be validated for unit alignment is not a statistic.

Stub values. A rate below one percent of its own Medicare reference is not a negotiated price under any schedule. This test uses the external benchmark rather than the code's own distribution, which is the more reliable anchor: a relative test fails precisely when a filing convention dominates the distribution it is measured against. Cell and gene therapies with references in the hundreds of thousands of dollars, filed at two cents, are the clearest instances.

Statistical outliers, bidirectional. Rates are flagged where they are extreme relative to the same code's distribution across all hospitals reporting it. A relative test is used rather than an absolute threshold: certain therapeutic agents legitimately carry rates in the millions, and a fixed dollar cap would silently delete correct data. Both high and low tails are flagged; a rate far below plausible is as defective as one far above.

Flagged rows are excluded from aggregates but retained and visible. Deletion destroys the evidence that a filing problem exists.

---

7. Observed distributions

Commercial rates as a percentage of the governing Medicare reference. Commercial product lines only; flagged outliers, placeholder values, and stub rates excluded; unit-dependent references omitted per Sections 4.5 and 4.6.

Reported by reference schedule rather than by billing class, because a laboratory rate benchmarked against the laboratory schedule and a procedure rate benchmarked against the outpatient schedule are different statistics and should not share a column.

Three features are worth naming.

The clearest finding is an inversion between two facility lanes. Laboratory rates sit at or above the fee schedule almost universally — the tenth percentile is approximately 95%, with a discrete cluster at exactly 95% that reflects a contractual convention rather than distribution density. Outpatient procedure rates, by contrast, have a tenth percentile of 36%: a substantial share of hospital outpatient procedures are contracted well below Medicare. Commercial payers hold laboratory pricing to the federal floor while negotiating outpatient facility rates beneath it for the least expensive procedures.

The outpatient procedure lane is the widest, a twelve-fold spread from p10 to p90, consistent with the least standardized of the payment environments.

The inpatient lane is the tightest, with almost no commercial rate below Medicare. That distribution emerged only after per-diem and case-rate arrangements were excluded; before that exclusion the lower quartile sat far below Medicare, because a uniform daily rate was being read as DRG-specific pricing.

7.1 Drug pricing, reported separately

Drug rates are expressed as a multiple of the ASP reference rather than as a percentage of Medicare, and are not included in the table above.

Across the corpus the median multiple is approximately 1.8 times ASP, against a Medicare payment of ASP plus six percent. A substantial share of drug rates fall at or below that cost-recovery threshold, concentrated in three economically coherent groups: generic injectables on declining price curves, where the two-quarter trailing average that defines ASP systematically overstates current acquisition cost; biosimilars, whose contracted prices run well below a reference blended across originator and biosimilar sales; and contracts written at exactly ASP with no markup, which appear as a discrete cluster at 1.00 rather than as distribution density.

No branded product without biosimilar competition appears in the below-threshold population, which is what distinguishes this as drug-pricing economics rather than a data artifact.

The upper tail is not interpretable. Multiples in the tens are unit mismatches, not negotiated margins, and no filtering resolves them because transparency filings do not state billing units. This metric therefore supports individual lookup where a user supplies the unit; it does not support a published distribution.

The exactly-at-schedule clusters — laboratory at 95% of CLFS, drugs at 1.00 times ASP — are the same phenomenon in two places: payers anchoring precisely to a federal rate rather than negotiating around it.

---

8. Interpreting these figures

These are not comparable to commonly cited commercial-to-Medicare ratios, and the difference is a difference of statistic rather than a discrepancy.

Published estimates in the range of 200% to 250% of Medicare derive predominantly from claims analyses. Those are volume-weighted: each rate is weighted by the services actually delivered under it, and the resulting figure describes what employer plans effectively pay.

Price transparency filings support a different statistic. A hospital files every negotiated rate line for every contracted plan — including narrow networks, exchange products, small regional plans, and legacy arrangements — without any indication of volume. A median computed across those lines is unweighted, and it necessarily includes many low-volume, low-rate arrangements that contribute little to a volume-weighted average.

An unweighted line-level median will therefore sit below a volume-weighted claims-based average, and should. The two answer different questions:

• What does a payer effectively pay for care delivered? — volume-weighted, claims-based
• What rates has this hospital agreed to, and how do they compare across payers and peers? — unweighted, transparency-based

The second question is the one price transparency data can answer, and the one that is directly actionable in contracting and compensation discussions.

Anyone computing percent-of-Medicare from HPT or TiC data should expect medians materially below published benchmarks. That result is not evidence of a pipeline defect.

---

9. Physician compensation

The reason to compute these rates carefully is to relate them to what physicians are paid.

9.1 The bridge

Compensation and reimbursement meet at the work relative value unit.

Compensation per wRVU derives from observed production — wRVU percentiles by specialty, year, and practice setting, computed from CMS Medicare claims — set against occupational wage distributions.

Reimbursement per wRVU derives from the transparency filings, normalized by the RVU values of the underlying procedures.

Both sides are then expressed in dollars per wRVU, which is what makes them comparable at all.

9.2 Two metrics, not one

Professional-fee yield — compensation per wRVU divided by professional-component reimbursement per wRVU. Both sides professional basis. This is the defensible figure for individual compensation discussion.

Facility revenue intensity — facility revenue per wRVU of physician work, reported on its own and never divided by compensation. This answers how revenue-dense a physician's work is to an institution, which is relevant to site-of-service and service-line questions. It is not a yield and is never labeled as one.

The distinction between them is Section 3 applied to compensation. A single ratio combining a professional numerator with a facility denominator was where this work began, and separating them made the analysis stronger rather than weaker.

Both metrics exclude drug, biological, device, laboratory, and supply codes, for the reason set out in Section 5.5: those codes carry no physician work RVU, so including their revenue inflates any per-wRVU figure without a corresponding increase in the denominator. A metric that rewards a specialty for the acquisition cost of what it injects is not measuring physician work.

9.3 A note on what facility revenue does and does not represent

Facility revenue associated with a physician's production is a top-line figure. It does not net implant and supply cost, staffing, space, or allocated overhead, and it is not a margin.

It is also, for surgical specialties, an understatement of the revenue a case originates: the procedure's facility fee does not capture the inpatient admission, pre-operative imaging and laboratory work, post-operative therapy, or downstream referrals arising from the same clinical decision.

Both of these should be stated wherever the figure appears. A number presented without its boundaries invites a conclusion it cannot support.

---

10. Known limitations

Single vintage. Our rate corpus reflects a single filing period. HPT files are point-in-time snapshots that hospitals replace; a vintage not captured when it was published cannot be retrieved later. Rate trend analysis requires an archive built forward from the present, not assembled retrospectively.

Commercial payers only, by design. Statistics reported here exclude Medicare, Medicare Advantage, and Medicaid managed care. Actual institutional payer mix includes all of them at materially different rates.

Medicare-derived production data. Physician wRVU percentiles come from Medicare fee-for-service claims and therefore understate total production for specialties with low traditional-Medicare volume. In obstetrics and pediatrics the understatement is severe enough that percentiles should not be reported at all. Every production percentile carries a Medicare-mix disclosure.

Professional-component coverage is thin. Most hospitals file facility rates only. Our professional-component statistics rest on 41 hospitals across 16 states — sufficient to compute the metric and to support specialty-level comparison, but far narrower than the facility lane. Professional-lane figures should be read as indicative rather than nationally representative, and expanding this coverage is the highest-value acquisition priority.

Filing quality varies. Compliance is uneven across hospitals in completeness, format, and accuracy. Coverage is reported alongside every result rather than assumed.

Identity resolution is incomplete. A meaningful share of filings use identifiers that cannot be resolved to a specific facility. Those rows are retained but cannot participate in geographic analysis.

Unit alignment cannot be verified for drug and equipment rates. Transparency filings do not state the billing unit priced. Rates for these categories are available individually but are excluded from every published distribution, and no filtering resolves the ambiguity.

DMEPOS rates are approximated nationally. Actual rates vary by state, rural designation, competitive bidding area, and purchase-versus-rental modifier. The national non-rural rate is used as a benchmark approximation only.

Facility reference rates are base rates. IPPS and OPPS references exclude add-on payments and reflect base operating and capital payment only.

---

11. Sources deliberately excluded

Junto uses no proprietary compensation survey data. MGMA, SullivanCotter, the AAMC Faculty Salary Report, Vizient, and comparable licensed benchmarks are excluded.

This is a methodological choice, not a licensing constraint.

Survey benchmarks are constructed from employer-reported compensation. Employers report what they pay; the survey publishes the distribution; employers then set compensation with reference to that distribution. The output of one cycle becomes the input to the next. That construction measures what employers pay accurately — and it structurally cannot measure what the work generates.

A benchmark built from claims data and mandated rate filings does not participate in that loop. It is an independent reference point, offered as a second opinion rather than a replacement. Both are worth having, and a physician holding an employer-cited survey percentile alongside an independently computed one is better equipped than one holding either alone.

NRMP match data is excluded separately: its terms prohibit use as input to AI models.

---

12. Reproducibility

Every source is public. Every reference rate carries a year, a status, and a citation. Every exclusion rule is stated. Every rate carries its revenue basis, its payment-system reference, its source authority, and its geography where resolved.

The methodology is auditable by anyone with access to the same federal files. That is the point: an analysis of prices that were made public by federal mandate should itself be transparent.

---

Junto Analytics · Methodology v1.0 · August 2026 Comments and challenges are welcome — andrew@juntoanalytics.com