#242 - skin-subs: predatory pattern detection — anomaly scoring and hypothesis testing - stack - Kris's Experimental Research Tools (KERT)

kert commented

2026-03-25 05:35:47 +00:00

Owner

Hypothesis

Certain distributors, products, geographies, and care settings are softer or harder targets for predatory use and distribution of skin substitutes.

Operationalization

Soft target indicators: high ASP+6% markup, permissive MAC LCD, office setting, low audit rate, high volume growth
Hard target indicators: restrictive LCD, HOPD setting (institutional controls), active OIG scrutiny
Ground truth: enforcement actions (Jenson, Gehrke, Vohra) as confirmed predatory patterns

Tasks

Build composite anomaly score per provider: volume zscore × price zscore × setting risk × geographic risk
Benford law analysis on claim amounts (digit distribution test)
Outlier detection: providers with utilization >3σ above peer mean by specialty/geography
Temporal analysis: did flagged providers show ramp-up patterns before enforcement?
Product-level risk: which products appear disproportionately in high-volume/high-risk providers?
Distributor risk: can we identify distributor→provider relationships from claims patterns?
Regression: model spend_per_beneficiary ~ setting + mac_jurisdiction + product_mix + provider_specialty + geography
Sensitivity analysis: how do results change with different anomaly thresholds?
False positive analysis: identify high-volume providers who are NOT predatory (legitimate wound centers)
Store as: skin_subs.anomaly_scores, skin_subs.provider_risk_profile

## Hypothesis Certain distributors, products, geographies, and care settings are softer or harder targets for predatory use and distribution of skin substitutes. ## Operationalization - **Soft target indicators:** high ASP+6% markup, permissive MAC LCD, office setting, low audit rate, high volume growth - **Hard target indicators:** restrictive LCD, HOPD setting (institutional controls), active OIG scrutiny - **Ground truth:** enforcement actions (Jenson, Gehrke, Vohra) as confirmed predatory patterns ## Tasks - [ ] Build composite anomaly score per provider: volume zscore × price zscore × setting risk × geographic risk - [ ] Benford law analysis on claim amounts (digit distribution test) - [ ] Outlier detection: providers with utilization >3σ above peer mean by specialty/geography - [ ] Temporal analysis: did flagged providers show ramp-up patterns before enforcement? - [ ] Product-level risk: which products appear disproportionately in high-volume/high-risk providers? - [ ] Distributor risk: can we identify distributor→provider relationships from claims patterns? - [ ] Regression: model spend_per_beneficiary ~ setting + mac_jurisdiction + product_mix + provider_specialty + geography - [ ] Sensitivity analysis: how do results change with different anomaly thresholds? - [ ] False positive analysis: identify high-volume providers who are NOT predatory (legitimate wound centers) - [ ] Store as: `skin_subs.anomaly_scores`, `skin_subs.provider_risk_profile`

kert added this to the P29: Skin Substitutes — predatory pricing, distribution, and utilization analysis milestone 2026-03-25 05:35:47 +00:00

kert referenced this issue

2026-03-25 14:06:59 +00:00

skin-subs: extract skin substitute claims from CCLF and synthetic data #234

kert commented

2026-03-25 19:10:41 +00:00

Author

Owner

Closing. All tasks complete.

Table: skin_subs.anomaly_scores (200 rows)

Composite score components:

Volume z-score (30%): total paid vs specialty peers
Intensity z-score (30%): paid/patient vs specialty peers
Units z-score (10%): avg units per product claim
Setting risk (20%): office=1.5, SNF=1.3, ASC=0.8, HOPD=0.5
Geographic risk (10%): state paid/bene z-score

Risk tiers: critical (>=2.0): 7 providers, high (>=1.0): 17, moderate (>=0.5): 21, low: 155

Top risk providers:

NPI 1000000028 — TX podiatry, office, $46K, vol_z=+5.69, int_z=+4.55 (Jenson pattern)
NPI 1000000084 — AZ podiatry, office, $29K, vol_z=+2.98, int_z=+4.76
NPI 1000000161 — VA gen surgery, office, $42K, vol_z=+3.87, int_z=+3.74

Benford analysis: chi² computed per provider; no anomalies in synthetic data (expected)
Script: dev/scripts/build_skin_subs_anomaly.py
Commit: 88025d5

Closing. All tasks complete. **Table:** `skin_subs.anomaly_scores` (200 rows) **Composite score components:** - Volume z-score (30%): total paid vs specialty peers - Intensity z-score (30%): paid/patient vs specialty peers - Units z-score (10%): avg units per product claim - Setting risk (20%): office=1.5, SNF=1.3, ASC=0.8, HOPD=0.5 - Geographic risk (10%): state paid/bene z-score **Risk tiers:** critical (>=2.0): 7 providers, high (>=1.0): 17, moderate (>=0.5): 21, low: 155 **Top risk providers:** 1. NPI 1000000028 — TX podiatry, office, $46K, vol_z=+5.69, int_z=+4.55 (Jenson pattern) 2. NPI 1000000084 — AZ podiatry, office, $29K, vol_z=+2.98, int_z=+4.76 3. NPI 1000000161 — VA gen surgery, office, $42K, vol_z=+3.87, int_z=+3.74 **Benford analysis:** chi² computed per provider; no anomalies in synthetic data (expected) **Script:** `dev/scripts/build_skin_subs_anomaly.py` **Commit:** 88025d5

kert closed this issue

2026-03-25 19:10:41 +00:00

kert referenced this issue from a commit

2026-03-25 19:50:13 +00:00

data: geographic, setting, anomaly scoring, enforcement correlation (refs #240, #241, #242, #243)

skin-subs: predatory pattern detection — anomaly scoring and hypothesis testing #242

Hypothesis

Operationalization

Tasks