Excluded

An Exploration of the LEIE

Overview

  • What?
  • Why?
  • Explore
  • Places to Start
  • Heatmaps
  • LEIE Alternatives
  • Observations
  • Questions

What?

LEIE

Waivers

  • Waiver when
    • sole provider;
    • essential service; and
    • community hardship

Reinstatement

Reinstatement if

  • after exclusion period;
  • after license reinstated; and
  • no patient neglect or abuse

Mandatory Exclusions

SSA 1128:

  • section(a) Mandatory (4)
    • program-related crime
    • patient abuse
    • felony healthcare fraud
    • controlled substance

Permissive SSA 1128

  • section(b) permissive (17)

Permissive SSA 1156

  • services must be
    • economical;
    • medically necessary; and
    • of sufficient quality; and if
  • provider fails in
    • substantial number of cases substantially to comply; or
    • grossly and flagrantly violated then
  • Secretary may exclude

Exclusions by Statute

statute exclusion description n_count
1128b4 Permissive License revocation or suspension 31084
1128a1 Mandatory Conviction of program related crimes 23109
1128a2 Mandatory Conviction for patient abuse 7449
1128a3 Mandatory Felony conviction for healthcare fraud 4987
1128a4 Mandatory Felony conviction for controlled substance 3140

Consequences

  • No participation
  • No payments
  • No hiring

Variables


lastname specialty city reindate
firstname upin state waiverdate
midname npi zip wvrstate
busname dob excltype
general address excldate

Why care about the LEIE?

Machine Learning Pipeline

  1. Find data (predictors & outcome)
  2. Create features
  3. Partition and stratify
  4. Choose models
  5. Resample
  6. Tuning
  7. Measure performance
  8. Test

Machine Learning Healthcare Fraud




Part B data <– LEIE by NPI

ML Healthcare Fraud Network Graph

Challenges - Class Imbalance

  • Herland (2020)
  • 4 years Part B
  • 1.1 million physicans
  • 1,312 fraud
  • 10 + per 10,000 physicians

Strategies

  • Random undersampling (90/10 optimal)
  • Random oversampling
  • Hybrid
  • Match LEIE w/ NPI?

Corpus/Matrix Size

Corpus/Matrix Example

Explore

Missing Identifiers by Year

Individuals Excluded by Statutory Provision

Exclusions by Statutory Provision

Average Monthly Exclusion

Excluded Physicans by Category

Exclusions by Category

Exclusions by General Category

Exclusion by Type

Exclusion by Type and Category

Physicians Excluded by Specialty

Variable Importance Plot

Places to Start

California

CMS vs. California

New York

CMS vs New York

Texas

CMS vs. Texas

Initial Effort

Three years of internal medicine from TX, CA, NY, FL, yielded 89k observations, 2500+ columns, and 42 fraud cases for a 5 per 10,000 result.

Heatmaps

Exclusion by State and Statute

Exclusion by State and Year

Exclusion to General

Physician Specialty to Exclusion

LEIE Alternatives

CMS Exclusions

link

CMS Preclusions

link

NPDB

link

Observations

  • Total annual exclusions are declining
  • Percent of excluded with NPIs is increasing
  • UPIN converted to NPIs around 2009
  • Medical practice –> to Doctor in 2009

Questions

  • Can the exclusions be scaled?
  • What exclusion types are indicative of fraud?
  • Are there strategies to boost NPI matches?
  • Will the LEIE remain relevant in future fraud detection?

Bibliography

[1]
“Exclusions | Office of Inspector General | U.S. Department of Health and Human Services.” Accessed: Aug. 30, 2023. [Online]. Available: https://www.oig.hhs.gov/exclusions/index.asp