The Extraction Index: Methods & Data
How the Extraction Index timeline is built: what is measured, what is modeled, every source named, all code open.
The Extraction Index: Methods & Data
The Extraction Index chart shows cumulative dollars extracted from the US healthcare system, by company, climbing year by year. Because some of those totals are larger than any single source measured directly, this page documents exactly how each line is built: what is measured, what is modeled, and where every number comes from. All reconstruction code is open.
The one rule
A line only appears if a published issue, or the peer-reviewed or government source that issue is built on, attributes a specific dollar figure to that company. We never split an industry pool across companies without a source that does the splitting, and we never annualize a one-time settlement into the ranked tier. Where a number is the documented annual rate carried forward, the line is dashed and the card shows the measured rate next to the accumulated total.
How the timeline accumulates
Each company's documented figure is an annual rate. The line is the cumulative of that rate from the year a source first detected it to the present (cumulative = annual rate times years elapsed). Solid segments are years a source measured directly; dashed segments carry the documented rate forward through years the mechanism kept operating but no source independently re-measured. The vertical axis is logarithmic. This means the headline totals are larger than any single-year measurement, and that is the point of a cumulative view, but it also means the later years of most lines are the documented rate projected, not independently measured. We label that distinction everywhere.
Line by line
MA industry (all insurers)MEASURED, per year
Measured: MedPAC's annual Medicare Advantage coding-intensity overpayment, every year 2016-2026, accumulated to ~$218B.
Modeled / carried: Nothing modeled. This is real per-year data. It is a broader, different measure than the company lines (coding intensity above the statutory adjustment, all insurers), so the company lines do not sum into it; it is shown for scale.
Source: MedPAC March 2026 Report to Congress, Ch. 12, Fig. 12-6.
UnitedHealth, Humana, Aetna (MA coding)MEASURED anchor, projected forward
Measured: Each insurer's 2021 differential-coding revenue, measured at the insurer level by Kronick et al. 2025 (UnitedHealth $13.9B / 42%, Humana ~$6.27B / 19%, Aetna ~$1.98B / 6% of the $33B 2021 total).
Modeled / carried: The climb after 2021 carries that measured 2021 annual rate forward (dashed). The insurer-level figure for years other than 2021 is not public, so we do not claim it. The cumulative is the measured rate times years elapsed.
Source: Kronick R, Chua FM, Krauss RC, Johnson L, Waldo D. Annals of Internal Medicine 178(5):655-662 (2025), DOI 10.7326/ANNALS-24-01345; per-insurer split via STAT's reporting of the paper.
Big 3 PBMs (specialty markups)RECONSTRUCTED to published anchors
Measured: The FTC documented a $7.3B total (2017-2022) and a 42% compound annual growth rate (2017-2021) in PBM-affiliated specialty-generic dispensing revenue above acquisition cost.
Modeled / carried: The FTC did not publish a per-year data table, so we reconstruct the per-year curve to fit both published anchors (42% growth, $7.3B total), holding 2022 at the 2021 level as a partial-year proxy, and carry the 2021 full-year run-rate forward. The shape is faithful to the FTC's reported acceleration; the exact per-year split is our reconstruction.
Source: FTC, Specialty Generic Drugs (Second Interim Staff Report), Jan 14 2025.
CVS Caremark (PBM spread)MEASURED, single point
Measured: Ohio's 2018 Medicaid audit found $224.8M of spread pricing in one state in one year.
Modeled / carried: Carried forward at that documented annual rate (one state), labeled as such. Not a national figure.
Source: Ohio State Auditor Medicaid PBM audit (2018), via Issue #4.
Cigna / Express Scripts (settlement)PROJECTION, from a settlement
Measured: The FTC's February 2026 Express Scripts settlement projects $7B in patient savings over ten years (~$0.70B/yr).
Modeled / carried: An FTC forward projection, not a backward-looking measurement; begins 2026.
Source: FTC settlement with Express Scripts, Feb 4 2026, via Issue #4.
Reproducibility
The two reconstructed series are produced by short, self-contained scripts you can run and check:
scorecard/backfill/ftc_pbm_series.py reconstructs the Big 3 PBM per-year curve from the FTC's published 42% growth rate and $7.3B total.scorecard/backfill/medpac_industry_line.py builds the all-insurer MA industry line from MedPAC's annual coding-intensity figures.
The whole page renders from one data file, scorecard/extraction_index.json, via tools/build_scorecard.py.
What we deliberately did not do
We did not extend the company-level Medicare Advantage lines backward across the decade by scaling them to enrollment and coding-intensity growth. That model is buildable from public data and would push UnitedHealth's cumulative to roughly $125B, but those earlier-year figures would be mostly modeled, and we will not put a multi-billion-dollar modeled figure under a named company. The insurer-level per-year series that would settle it exists only in restricted CMS data (the Chronic Conditions Data Warehouse, accessible under a data-use agreement). That is the one piece we cannot reconstruct from public files, and it is an open invitation.
If you have the data
If you hold or have access to licensed or restricted claims data (CMS LDS or VRDC, a state all-payer claims database, MarketScan, Optum Clinformatics, IQVIA) that would let us measure these mechanisms at the company-year level instead of modeling them, we would value a conversation, on background and in confidence. Reach us at contact@ahcdata.fund.
Last updated 2026-06-10, through Issue #15.