What hospitals told the IRS vs. what they told Medicare

A raw "biggest gap" ranking over all 1,295 systems is misleading — the largest raw gaps come from multi-hospital systems reported under one EIN (a name like "Memorial Hermann Kingwood" can actually be eight hospitals) and from rows whose two filings cover different fiscal years. The index filters to the subset where a difference is genuinely apples-to-apples:

• Single reporting facility — one CCN under one EIN, so the hospital name is trustworthy and there's no system-level aggregation smoothing or amplifying the difference.
• Period-aligned — the HCRIS fiscal-year end is within one month of the 990 tax-period end, so the two filings cover the same year.
• Material — both the CMS and IRS figures are at least $1M.
• Same-scale entity — the two forms' total expenses agree within 5%, confirming both describe the same organization for the same period.

99 of the 1,295 computable systems clear all four tests. The published CSV holds everything that doesn't, plus a wider set: 1,334 rows — every EIN matched across HCRIS, Schedule H, and the crosswalk — of which 1,295 are computable (both figures present, so a difference can be calculated) and searchable here. The remaining 39 matched but have a blank Schedule H 7a. The ranked index is committed at charity_care_gap_index_2022.md, and the full funnel with every caveat is in the method summary.

The comparison is net-to-net. The IRS figure is Schedule H Part I line 7a column (e), net community benefit expense (financial assistance at cost, less direct offsetting revenue); the CMS figure is the S-10 cost of charity care, net of partial patient payments. Comparing net to net avoids the gross-vs-net illusion — though the two "net" definitions still aren't identical, so a residual difference is expected even for a hospital reporting both correctly. The % shown throughout is the difference as a share of the larger of the two figures.

• Definitional differences. The two forms bound "financial assistance" / "charity care" differently — for example, how uninsured discounts are classified.
• Cost-to-charge ratio (CCR) handling. HCRIS S-10 applies a hospital-wide CCR; Schedule H allows different cost-step-down approaches.
• Charity care policy boundaries. Hospitals can categorize the same patient case as charity care, bad debt, or contractual allowance — moving cost between lines.
• S-10 data quality. The reliability and completeness of Worksheet S-10 charity-care reporting has been repeatedly flagged (e.g., by MedPAC), so a gap can simply mean one figure — often the S-10 side — is mis- or under-reported. This is a leading reason to open both source filings rather than treat a gap as settled.
• Multi-facility aggregation. A single 990 EIN often covers several HCRIS-reporting facilities. (The index above removes these; the full search below includes them, flagged per row.)
• Non-Medicare hospitals. S-10 charity care is all-payer, so low Medicare volume does not shrink it — but a few children's and specialty hospitals participate in Medicare so little that they file no Medicare cost report at all, or an S-10 with little in it. Treat a near-zero S-10 as suspect rather than a real zero.

HCRIS uses the federal fiscal year as a reporting cycle, not as a fiscal period. The file labeled "FY2023" contains hospital reports submitted during federal fiscal year 2023 — but the periods those reports cover vary by hospital fiscal calendar. About 29% of the 1,295 computable matched systems are period-aligned within 1 month; about 65% are exactly 12 months off. The index above uses only aligned rows. When you look up any hospital in the full search, the detail card shows the period-alignment status as a per-row signal — don't use a misaligned row to infer a direct like-period difference.

Coming in v2: per-hospital matching across HCRIS reporting cycles 2022/2023/2024 to find the report whose period actually matches each hospital's TY2022 990. This will substantially expand the well-aligned set.

Each detail card shows a Social Vulnerability Index value for the hospital's home county, sourced from CDC/ATSDR's 2022 SVI release (overall summary rank RPL_THEMES, 0–100). Higher percentile = more vulnerable. This is a proxy, not a true service-area measure — actual catchment areas can extend well beyond the home county. Coverage: ~98% of matched systems. Source: CDC/ATSDR SVI.

• Validation: Tampa General, Boston Medical Center, Baptist Medical Center (Jacksonville), and Cook Children's were re-verified against their actual TY2022 Form 990 Schedule H filings — the "Told the IRS" figure matches to the dollar in every case. Yale New Haven was confirmed on identity and expense scale (its exact 7a figure sits behind ProPublica's rate limiter, but nothing contradicts it). Bug reports welcome at the GitHub issues link.
• Crosswalk: 3,079 of 6,042 HCRIS hospitals appear in CBI's December 2024 crosswalk; the unmatched ~50% are mostly for-profit and government hospitals (no 990) plus hospitals not in CBI's frozen vintage.
• Time window: HCRIS Fiscal Year 2023 paired with 990 Tax Year 2022, ingested across IRS release years 2024, 2025, and 2026.
• This is a reference tool. A difference in any given row does not by itself say whether either filing is wrong. The value is making the cross-form view computable at all, where previously every researcher had to reconstruct it from raw filings.