Plausibility Filters

Without plausibility, the Cartesian product of all payers × providers × codes would be hundreds of billions of rows. Plausibility filters narrow it to ~2–6B valid ROIDs.

code_plausibility_flattened

The key pre-computed table is tmp_ref_code_plausibility_{sub_version}. It is built before the ROS runs and maps (provider_type, billing_code) → valid combinations.

Two row variants exist:

• Provider-type-level rows (provider_id IS NULL): apply to all providers of a given type
• Provider-id-level rows (provider_id IS NOT NULL): apply to a specific provider (used for ASC specialty matching)

ASC plausibility is a 3-part UNION:

1. AmSurg specialty match — codes associated with the ASC's specific surgical specialty
2. Spines service flags (fallback) — codes mapped from service-line flags in the ASC spine
3. Multispecialty/unknown — all service lines included for ASCs with no specialty classification

PG Plausibility

Physician Group plausibility requires core_rates evidence — actual historical billing records. Two strategies:

• EIN-based (primary): Match the provider group's TIN to a provider_group_id of type 'ein' in core_rates. This is the most reliable strategy when TIN data is available.
• NPI-based (fallback): Match individual NPIs associated with the PG to NPI records in core_rates. Used when TIN is not available or doesn't match.

warning

PG plausibility is the largest scope reducer. The PG codeset has ~10K codes, but most physician groups actually bill only ~200–500 of them. Without this filter, PG ROIDs would dominate the ROS. The filter reduces PG ROIDs by ~95%.

Infusion Center Special Case

Providers with provider_subtype='Infusion Center' within provider_type='Physician Group' receive both drug/infusion codes AND the standard PG codeset. This allows infusion centers to be priced across their full billing scope, which spans both product categories.

Filter Rules by Provider Type

Provider Type	Plausibility Source	What it restricts
Hospital	None (all codes)	No plausibility filter; all hospital codes in scope
ASC	AmSurg specialty + Spines flags + multispecialty fallback	Codes not matching ASC specialty are excluded
Physician Group	core_rates EIN/NPI evidence	Only codes the PG has billed historically
Laboratory	code_plausibility (type-level)	Lab codes only; no provider-level filter
Imaging	code_plausibility (type-level)	Imaging codes only
Dialysis	code_plausibility (type-level)	Dialysis codes only
DME	DME evidence filter (supplier type flags)	Codes not supported by supplier's license type excluded
Urgent Care	code_plausibility (type-level)	Urgent care codes only
Exchange networks	Hospital only (except NYC Essential Plan)	Non-hospital providers excluded from Exchange ROS
MS-DRG 018 (CAR-T)	Designated CAR-T center list	Only CAR-T-certified hospitals included

CAR-T Plausibility (MS-DRG 018)

CAR-T (Chimeric Antigen Receptor T-cell) therapy is an advanced cancer treatment where a patient's own T-cells are genetically modified to attack cancer cells. It is represented in the CLD by MS-DRG 018. Because only specialized, certified hospitals can administer CAR-T therapies, we apply a provider-level plausibility filter to prevent MS-DRG 018 from appearing in the ROS for non-certified facilities.

How it works

1. Curated center list: The list of CAR-T centers was built by scraping a public directory of certified cellular therapy programs. The scraped centers were then manually mapped to CLD provider_ids by matching on name, address, and NPI. The resulting CSV is maintained at spines/plausibility/data/car_t_centers_enriched.csv and contains the center name, CLD provider_id, program type (Adult/Pediatric), and approved CAR-T products (e.g., Kymriah, Yescarta, Breyanzi, Carvykti, Tecartus, Abecma). This currently covers ~215 providers.

2. Table load: The spines_plausibility DAG loads this CSV into the Trino table referenced by tq_dev.internal_dev_csong_sandbox.plausibility_car_t_centers. This runs independently of the CLD sub-DAG and only needs to be refreshed when the center list changes.

3. ROS filter: In rate_object_space.sql, the filter logic is:

   -- Keep all rows except MS-DRG 018 for providers not in CAR-T centers
   (billing_code != '018' OR billing_code_type != 'MS-DRG')
   OR
   -- Allow MS-DRG 018 only for CAR-T centers
   (billing_code = '018' AND billing_code_type = 'MS-DRG'
    AND provider_id IN (SELECT provider_id FROM car_t_plausibility))

   This means every hospital keeps all its other DRGs, but MS-DRG 018 is restricted to only the ~215 certified CAR-T centers.

Future scope

The current filter only covers MS-DRG 018 (inpatient CAR-T). CAR-T drugs administered in outpatient settings use specific HCPCS codes (e.g., J-codes) and are not yet covered by this plausibility filter.

info

CAR-T plausibility only applies to Hospital provider types. ASCs, PGs, and other provider types do not have MS-DRG codes in their codesets, so they are unaffected.

Scale Impact

From 650 billion rows to ~4 billion

Theoretical max (no filters): ~90 payers × 3,000 networks × 60,000 providers × 20,000 codes × 2 bill_types ≈ 650 billion rows

After geo matching: ~80% reduction — most payer-provider pairs are geographically incompatible

After code_plausibility (type-level): e.g., Labs get ~1,500 applicable codes, not 20K; Hospitals get ~2,000 codes, not 20K

After PG plausibility (evidence-based): ~95% reduction on PG rows — most PGs bill only ~200–500 of the 10K PG codes

After DME plausibility: ~90% reduction on DME rows based on supplier license type

Final ROS: ~2–6B ROIDs depending on Clear Rates version scope. This is less than 0.001% of the theoretical Cartesian maximum.