Quick overview of healthcare fraud, waste, and abuse

Defining FWA in the AI Era:

• Fraud: Intentional deception (e.g., phantom billing)
• Waste: Inefficiency (e.g., redundant testing)
• Abuse: Improper practices (e.g., upcoding)

How to look at Fraud vs. Waste vs. Abuse?

• AI Perspective: How algorithms distinguish "intent" (patterns)from "error" (randomness)

Data Ecosystem: What kind of data you need for detecting FWA?

Primer: AI vs. Gen

• Predictive AI (ML): "Classify this claim as risky" or "Predict the cost."
• Generative AI (GenAI): "Summarize this chart" or "Write a SQL query."

Medical record review (Analyzing Diagnosis (ICD-10) vs. Procedure (CPT/HCPCS) mismatches)

Identifying potential fraud schemes

AI assisted data wrangling (Investigator's Bottleneck: Cleaning messy data.

Hands-on Lab: Using a GenAI tool (e.g., ChatGPT Data Analyst, Julius.ai) to clean a dataset.

The Art of Persuasion: Moving from "Excel grids" to "Evidence." How to create visuals that can stand up in a summary judgment or medical review.

No-Code Graphics (using Natural Language to generate simple charts).

Hands-on exercise: Creating distributions, Visualizing "The Bell Curve" vs. The Outlier.

Using visualization to illustrate spatial relationship (mapping patient-to-provider distance,

Detecting "Teleportation” where providers bill for services in 2 distant locations simultaneously).

Network Analysis (Fraud Rings): Visualizing shared addresses, shared bank accounts, or patients shared between suspect providers/pharmacies.

Generating a simple network graph to identify collusive behavior between a prescriber and a pharmacy.

Demystifying ML Tasks: Supervised Learning vs. Unsupervised Learning

What are Machine Learning tasks?

• Regression: Predicting a number (e.g., Expected length of stay).
• Classification: Predicting a Label (Fraud vs. Not Fraud).
• Clustering: Grouping peers

How you build ML models for FWA detection?

How to use Gen AI tools to build supervised classification ML models for detecting FWA?

Building anomaly detection models for FWA detection (identify providers who are statistical outliers without writing a single line of code manually).

GenAI for Building ML Models Explainable AI: Generating SHAP plots to explain why the model flagged a specific provider or claim.

Interactive Session: Using GenAI as Your Virtual Data Scientist.

Dental/Vision: Pattern recognition in tooth numbers/surfaces (ML Clustering)

DME: Predicting "Need" vs. "Prescription"

Pharmacy: Opioid cocktail detection using Supervised Classification

How to improve LLMs with your own data? (Why ChatGPT doesn't know your specific medical policies?)

RAG (Retrieval-Augmented Generation) – How to build custom GPTs / Google Gems?

Use cases: GenAI for Medical Record review

Scenario Exercise: Designing Your First AI-Enabled FWA Use Case.

Agentic Workflows:

• Moving from "Chatting" to "Doing."
• Chaining tasks together (Analyze Data -> Review Record -> Draft Letter) Closing:
• Ethics & Bias: Ensuring AI doesn't profile based on patient demographics.
• Governance: The concept of "Human-in-the-Loop." AI is the investigator's assistant, not the judge.
• Final Q&A: Road mapping the implementation of these tools in your SIU.