Skip to main content
Version: 3.0

Score Reference

Each rate column on each ROID row receives one of these scores. Scores are additive: the integer tier reflects data source and validation status; the decimal component breaks ties within the same tier.

Score Breakdown

7.x Validated — Counterparty Agreement

Both payer MRF and hospital MRF independently report similar rates for this ROID. Agreement is within ±5%.

  • Decimal: rate / 1e8 — among validated rates, slightly prefer the higher dollar amount
  • Canonical rate score: 5
  • Condition: ABS(payer_rate - hospital_rate) ≤ 0.05 × payer_rate AND rate within outlier bounds
Payer MRF: \$1,000 negotiated rate for CPT 99213
Hospital MRF array: [\$950, \$1,050, \$2,000]
Check: |\$1,000 − \$950| = \$50 ≤ 5% × \$1,000 (\$50) → MATCH
Also required: \$1,000 is within outlier bounds (0.5–30× Medicare for non-IP)
Score = 7 + (1000 / 1e8) = 7.00001 Source label: "payer_hospital" (both sources agreed)

6.x Raw, Not Outlier — Single Source

Rate comes from a single MRF (payer or hospital) and is within outlier bounds. No counterparty validation.

  • Decimal: CDF value (0–0.999) — prefer rates that are more "common" for this code
  • Canonical rate score: 4
  • Condition: rate within [medicare_lower, medicare_upper] bounds; no counterparty match
Most rates land here

The majority of payer MRF and hospital MRF rates score 6.x. Score 7 requires an independent source to corroborate — uncommon for rare codes or small providers.


5.x Benchmark-Validated

Transformed rate (e.g. pct-to-dollar) falls within 95–1000% of Medicare. Applies to specific transformation columns where Medicare comparison is most meaningful.

  • Decimal: CDF value
  • Canonical rate score: 3

4.x Not Outlier — No Validation

Rate is within bounds but has no counterparty validation. Used for certain raw and transformed columns that don't meet the counterparty match threshold.

  • Decimal: CDF value
  • Canonical rate score: 2

3.x Imputation — Benchmark-Backed

Imputed rate grounded in a benchmark source: hospital MRF gross charge with percentage-of-charge, or state-level Medicare benchmark.

  • Decimal: CDF value
  • Canonical rate score: 3 (same as 5.x — benchmark-backed imputations are treated as moderate confidence)

2.x Imputation — Not Outlier

Estimated rate from tier aggregation (provider + code + payer combinations). Within bounds but fully estimated, no MRF source.

  • Decimal: CDF value
  • Canonical rate score: 2
Imputed rates can't beat validated raw rates

Score 2 (or 3) is always below score 6 or 7. An imputed rate only becomes canonical if no raw or transformed rate exists for the ROID.


1.x Outlier

Rate falls outside the Medicare-anchored bounds for its code type. Almost always a data error — wrong units, typo, or mis-classified rate.

  • Decimal: CDF value (computed but rarely relevant — outliers almost never win selection)
  • Canonical rate score: 1
  • Common causes: APR-DRG averaging bias pushing above IP bound; drug rate in wrong units (per mg vs per vial); capitation rate mistakenly included
Score 1 rates still appear in output

A ROID whose only rate is an outlier will still have canonical_rate set and canonical_rate_score = 1. Filter with canonical_rate_score > 1 to exclude outliers from analysis.


0 No Rate

No non-NULL rate column exists for this ROID, or the column scored 0 by default (e.g. a drug code's CDF column, which is always set to 0).

  • Canonical rate score: 0
  • The ROID row still exists in the output — a 0-score ROID means a confirmed coverage gap, not a missing row

The Decimal Tiebreaker

When many rates share the same integer tier, the decimal component determines the winner:

Integer tierDecimal =Effect
7rate / 1e8Higher dollar amount wins — among validated rates, prefer the larger value
6, 5, 4, 3, 2, 1CDF score (0–0.999)More common rate wins — prefer rates that cluster near the center of the observed distribution for that code

The CDF is computed as the probability mass under a log-normal curve centered on the code's observed median rate. A rate of $150 for CPT 99213 (common value) might score 0.78; a rate of $500 for the same code might score 0.05.

Drug codes always get CDF = 0

is_drug_code = true → CDF is set to 0, not computed. Drug pricing distributions are highly bimodal (brand vs generic), making the log-normal model unreliable. Drug tiebreaking falls back to array position ordering.