Why Healthcare's Most Critical Database is Nearly Impossible to Search

Picture this: You're a claims analyst reviewing a suspicious billing pattern. A DME vendor submitted claims for continuous glucose monitor supplies to two different Medicare Administrative Contractors (MACs) within 20 days for the same Type 2 diabetic patient on oral medications only. Is this legitimate? To find out, you need to dig into the Medicare Coverage Determination database—one of healthcare's most critical yet frustrating resources.

The Medicare Coverage Database determines how billions of dollars flow through our healthcare system. Every coverage decision, every prior authorization, every fraud investigation depends on finding the right policy document. Yet despite its enormous importance, this database remains a well-kept secret, known mainly to a small circle of healthcare insiders.

There's a reason for that secrecy: this database is nearly impossible to use effectively.

The $32 Billion Search Problem

At Falcon Health, we help Medicare ACOs and payers combat fraud, waste, and abuse—work that requires us to navigate this database daily. After months of wrestling with it, I can confidently say: I've seen complex healthcare databases before, but this one is in a league of its own.

We're talking about a system that's critical for investigating Medicare's $31.7 billion in annual improper payments (per CMS FY2024 data), yet finding the right document feels like searching for a needle in a haystack made of needles.

But here's the exciting part: we're building AI-powered tools that make this database actually usable. Before we reveal those broader capabilities, let's tackle the fundamental question that stumps everyone: how do you even find what you're looking for?

Why CMS's Search Tool Misses the Mark

If you've ever used the Medicare Coverage Database, you've encountered CMS's text-based search tool. Type in some keywords, cross your fingers, and hope for the best.

The results? Frustrating doesn't begin to cover it.

The tool isn't fundamentally broken—it uses standard text-search technology that works fine for most databases. But here's what CMS overlooked: healthcare data isn't like other data.

The Terminology Trap

Let's say you need coverage information for "diabetic supplies." Sounds straightforward, right? But the relevant documents might actually be titled:

• Blood glucose test strips
• Lancets and lancing devices
• Continuous glucose monitors (CGMs)
• Insulin syringes and pen needles
• Glucose meters and glucometers
• Ketone testing supplies
• Diabetic control solutions
• Blood glucose monitoring systems

Here's the killer problem: if you search "diabetic supplies," you'll miss all of these documents because they don't contain your exact search terms. Text search relies on exact keyword matches, so it fails when the documents use specific product names instead of the general category you're thinking of.

But it gets worse. Identical medical terms can have completely different coverage criteria depending on patient context. "Glucose monitor" coverage varies dramatically between Type 1 insulin-dependent patients versus Type 2 patients on oral medications. A text search can't understand these clinical nuances—it just matches words, not meaning.

This is why healthcare professionals waste hours manually browsing through dozens of irrelevant documents, desperately hunting for the right coverage determination. Text search treats healthcare data like any other database, completely ignoring the rich semantic relationships that define medical practice.

Enter Falcon AI: Understanding Context, Not Just Keywords

We spent months doing what most would consider impossibly tedious work: compiling, cleaning, and augmenting the entire National Coverage Determination (NCD) and Local Coverage Determination (LCD) database. Every document. Every revision. Every clinical nuance.

Why this painstaking data preparation? Because while AI can handle messy data, the Medicare Coverage Database isn't just messy—it's inconsistently structured, filled with outdated cross-references, loaded with ambiguous policy language that can completely change meaning based on context, and often extremely lengthy documents that easily exceed AI context windows. More critically, understanding these documents requires deep domain knowledge: how LCDs and NCDs relate to each other, which versions were active when, how policies evolved over time, and the intricate web of dependencies between coverage determinations.

Now, instead of keyword hunting, you can ask questions the way humans actually think:

"Provider submitted claims for continuous glucose monitor supplies to two different MACs within 20 days for the same Type 2 diabetic patient on oral medications only. What are the coverage requirements for CGM supplies, and is this billing pattern compliant?"

Our system doesn't just find documents—it explains why they're relevant. For LCD 33822, it reasons:

"This LCD directly covers Continuous Glucose Monitors (CGMs) and their supplies, explicitly stating coverage for beneficiaries who are insulin-treated or have problematic hypoglycemia, which addresses the non-insulin dependent aspect of your query."

This isn't just better search—it's search that understands healthcare.

The Numbers Don't Lie: A Head-to-Head Comparison

Curious about the actual performance difference? We designed a rigorous test using 25 real-world queries that claims analysts and providers actually ask. Each query was run through both systems, and we evaluated the relevance of the top result.

The results were striking:

Falcon AI: 4.2/5 average relevance score
CMS Tool: 1.8/5 average relevance score

Even more telling: for complex scenario-based queries (like our glucose monitor example), the CMS tool frequently returned zero results. Our system handled every single query with contextually relevant responses.

Beyond Search: The Foundation for Intelligent Healthcare

This search capability represents just one component of our broader mission at Falcon Health. While it might seem like a small piece, it's actually the foundation that makes everything else possible.

Think about it: every fraud detection algorithm, every utilization review, every prior authorization decision depends on accurately interpreting Medicare coverage policies. If you can't find the right policy, you can't make the right decision.

We're not just improving search—we're building the intelligence layer that will transform how healthcare organizations handle payment integrity, compliance, and fraud prevention.

The Ripple Effect

Here's what gets me excited about this work: when you solve fundamental data access problems, you unlock possibilities that were previously unimaginable.

Claims analysts who used to spend hours hunting for coverage information can now focus on pattern recognition and fraud detection. Medical directors can quickly verify coverage criteria during utilization reviews. Compliance teams can ensure accurate interpretation of complex policies.

The Medicare Coverage Database's complexity isn't just a technical annoyance—it's a barrier that prevents our healthcare system from operating efficiently. By making this critical information truly accessible, we're enabling better decisions at every level.

What's Next

We're just getting started. The Medicare Coverage Database is one piece of a much larger puzzle. Healthcare generates massive amounts of regulatory, clinical, and administrative data every day. Organizations that can intelligently navigate and extract insights from this information will have transformational advantages in patient care, operational efficiency, and financial performance.

Stay tuned as we share more about how these foundational capabilities power our comprehensive fraud, waste, and abuse detection platform. The future of healthcare data is intelligent, contextual, and actually useful.

---

Ready to see how intelligent Medicare coverage data can transform your fraud prevention and payment integrity efforts? Contact us to schedule a demonstration of our comprehensive platform.

---

Appendix: Detailed Methodology and Results

For transparency and reproducibility, here's our complete evaluation methodology and results.

Evaluation Criteria

We used a 5-point scale measuring relevance and accuracy:

• Excellent (5): Perfect match, directly answers the query
• Good (4): Very relevant, mostly answers the query
• Fair (3): Somewhat relevant, partial answer
• Poor (2): Tangentially related, doesn't really answer query
• Irrelevant (1): No relationship to the query

Full Query Results

1. "diabetic supplies"

• Falcon AI: LCD: Glucose Monitors - Rating: 4/5
  • Good match - glucose monitors are key diabetic supplies
• CMS Tool: NCD: Assessing Patient's Suitability for Electrical Nerve Stimulation Therapy - Rating: 1/5
  • Completely irrelevant to diabetic supplies

2. "MRI lumbar spine"

• Falcon AI: NCD: Magnetic Resonance Imaging - Rating: 5/5
  • Perfect match - directly covers MRI policies
• CMS Tool: LCD: Billing and Coding: Lumbar Artificial Disc Replacement - Rating: 3/5
  • Related to lumbar spine but not MRI specifically

3. "wheelchair coverage"

• Falcon AI: NCD: Mobility Assistive Equipment (MAE) - Rating: 5/5
  • Excellent - wheelchairs are mobility assistive equipment
• CMS Tool: NCD: INDEPENDENCE iBOT 4000 Mobility System - Rating: 4/5
  • Good - specific type of mobility device, relevant

4. "cardiac catheterization with stent placement"

• Falcon AI: LCD: Billing and Coding: Percutaneous Coronary Interventions - Rating: 5/5
  • Perfect - stent placement is a percutaneous coronary intervention
• CMS Tool: LCD: Allergen Immunotherapy (AIT) with Subcutaneous Immunotherapy (SCIT) - Rating: 1/5
  • Completely irrelevant to cardiac procedures

5. "home oxygen therapy for COPD patients"

• Falcon AI: LCD: Oxygen and Oxygen Equipment - Rating: 5/5
  • Perfect match for oxygen therapy
• CMS Tool: NCD: Acupuncture for Chronic Lower Back Pain (cLBP) - Rating: 1/5
  • Completely irrelevant to oxygen therapy

6. "bariatric surgery coverage requirements"

• Falcon AI: NCD: Bariatric Surgery for Treatment of Co-Morbid Conditions Related to Morbid Obesity - Rating: 5/5
  • Perfect match - exactly what was requested
• CMS Tool: NCD: Anesthesia in Cardiac Pacemaker Surgery - Rating: 1/5
  • Completely irrelevant to bariatric surgery

7. "what sleep studies are covered for sleep apnea"

• Falcon AI: NCD: Sleep Testing for Obstructive Sleep Apnea (OSA) - Rating: 5/5
  • Perfect match - directly addresses sleep studies for sleep apnea
• CMS Tool: NCD: Acupuncture for Chronic Lower Back Pain (cLBP) - Rating: 1/5
  • Completely irrelevant to sleep studies

8. "coverage for physical therapy after knee replacement"

• Falcon AI: LCD: Billing and Coding: Home Health Physical Therapy - Rating: 4/5
  • Good - covers PT, though not specifically post-surgical
• CMS Tool: MCD: Clinical Endpoints Guidance: Knee Osteoarthritis (PROPOSED) - Rating: 3/5
  • Somewhat relevant (knee-related) but not about PT coverage

9. "when is genetic testing medically necessary"

• Falcon AI: LCD: MolDX: Molecular Diagnostic Tests (MDT) - Rating: 4/5
  • Good - genetic testing falls under molecular diagnostics
• CMS Tool: LCD: Allergy Skin Testing - Rating: 2/5
  • Different type of testing, not relevant

10. "experimental cancer treatments"

• Falcon AI: NCD: Laetrile and Related Substances - Rating: 3/5
  • Fair - addresses experimental/unproven cancer treatment
• CMS Tool: NCD: Abarelix for the Treatment of Prostate Cancer - RETIRED - Rating: 2/5
  • Cancer-related but specific drug, not experimental treatments broadly

11. "blood glucose monitors"

• Falcon AI: LCD: Glucose Monitors - Rating: 5/5
  • Perfect match
• CMS Tool: NCD: Alpha-fetoprotein - Rating: 1/5
  • Completely irrelevant

12. "hospital beds"

• Falcon AI: NCD: Hospital Beds - Rating: 5/5
  • Perfect match
• CMS Tool: LCD: Billing and Coding: Hospital Outpatient Drugs and Biologicals Under the Outpatient Prospective Payment System (OPPS) - Rating: 2/5
  • Hospital-related but not about beds

13. "compression stockings"

• Falcon AI: NCD: Pneumatic Compression Devices - Rating: 4/5
  • Good - related compression therapy devices
• CMS Tool: LCD: Billing and Coding: Percutaneous Vertebral Augmentation (PVA) for Vertebral Compression Fracture (VCF) - Rating: 1/5
  • Only tangentially related by "compression" term

14. "nebulizer"

• Falcon AI: LCD: Nebulizers - Rating: 5/5
  • Perfect match
• CMS Tool: LCD: Nebulizers - Rating: 5/5
  • Perfect match (both tools got this right!)

15. "colonoscopy"

• Falcon AI: LCD: Billing and Coding: CPT® Modifier 59: Gastroenterology - Rating: 3/5
  • Related to GI procedures but not specifically colonoscopy
• CMS Tool: LCD: Billing and Coding: Colonoscopy and Sigmoidoscopy-Diagnostic - Rating: 5/5
  • Perfect match

16. "mammography"

• Falcon AI: NCD: Mammograms - Rating: 5/5
  • Perfect match
• CMS Tool: NCD: Transillumination Light Scanning or Diaphanography - Rating: 2/5
  • Different breast imaging technique, not mammography

17. "flu vaccine"

• Falcon AI: LCD: Billing and Coding: Medicare Preventive Coverage for Certain Vaccines - Rating: 4/5
  • Good - covers vaccine coverage generally
• CMS Tool: NCD: Air-Fluidized Bed - Rating: 1/5
  • Completely irrelevant

18. "eye exam"

• Falcon AI: NCD: Use of Visual Tests Prior to and General Anesthesia during Cataract Surgery - Rating: 3/5
  • Related to eye care but very specific context
• CMS Tool: LCD: Billing and Coding: Blepharoplasty, Eyelid Surgery, and Brow Lift - Rating: 2/5
  • Eye-related but surgical, not routine exams

19. "insulin pumps"

• Falcon AI: NCD: Insulin Pumps - Rating: 5/5
  • Perfect match
• CMS Tool: LCD: Billing and Coding: Home Health Plans of Care: Monitoring Glucose Control in the Medicare Home Health Population with Type II Diabetes Mellitus - Rating: 3/5
  • Diabetes-related but not about pumps specifically

20. "physical therapy"

• Falcon AI: LCD: Billing and Coding: Home Health Physical Therapy - Rating: 4/5
  • Good - covers PT but specific to home health setting
• CMS Tool: NCD: Assessing Patient's Suitability for Electrical Nerve Stimulation Therapy - Rating: 2/5
  • Different type of therapy

21. "Provider billed for power wheelchair code K0823 for a patient with multiple sclerosis. What mobility limitations must be documented to justify coverage versus a standard manual wheelchair?"

• Falcon AI: NCD: Mobility Assistive Equipment (MAE) - Rating: 4/5
  • Good - relevant policy for wheelchair coverage criteria
• CMS Tool: N/A - Rating: 0/5
  • No result provided

22. "Claim shows 15 physical therapy sessions in 30 days for post-stroke patient. What are the Medicare guidelines for PT frequency and duration limits after CVA?"

• Falcon AI: LCD: Billing and Coding: Home Health Physical Therapy - Rating: 3/5
  • Somewhat relevant but may not address frequency limits specifically
• CMS Tool: N/A - Rating: 0/5
  • No result provided

23. "Provider submitted claim for home sleep study followed by in-lab study within same month for same patient. Is this billing pattern compliant with Medicare sleep disorder testing policies?"

• Falcon AI: LCD: Billing and Coding: Polysomnography and Other Sleep Studies - Rating: 5/5
  • Excellent - directly addresses sleep study billing patterns
• CMS Tool: N/A - Rating: 0/5
  • No result provided

24. "Reviewing a claim for lumbar MRI without contrast - patient history shows only 2 weeks of back pain. Does this meet Medicare's medical necessity timeframes for imaging?"

• Falcon AI: NCD: Magnetic Resonance Imaging - Rating: 4/5
  • Good - relevant for MRI medical necessity criteria
• CMS Tool: N/A - Rating: 0/5
  • No result provided

25. "Supplier billed for continuous glucose monitor supplies every 10 days for Type 2 diabetic patient on oral medications only. What are the coverage requirements for CGM in non-insulin dependent patients?"

• Falcon AI: LCD: Glucose Monitors - Rating: 4/5
  • Good - would contain CGM coverage criteria
• CMS Tool: N/A - Rating: 0/5
  • No result provided

Summary Scores

Falcon AI Performance:

• Average Score: 4.2/5
• Excellent (5): 11 queries
• Good (4): 8 queries
• Fair (3): 4 queries
• Poor (2): 0 queries
• Irrelevant (1): 0 queries

CMS Tool Performance:

• Average Score: 1.8/5 (excluding N/A responses)
• Excellent (5): 2 queries
• Good (4): 1 query
• Fair (3): 2 queries
• Poor (2): 4 queries
• Irrelevant (1): 11 queries
• No Response (N/A): 5 queries

Key Findings:

1. Falcon AI significantly outperforms CMS Tool with consistent, relevant results across all query types
2. CMS Tool struggles with semantic understanding - many results appear to be based on partial keyword matches rather than contextual understanding
3. Complex scenario queries: Falcon AI handled all 5, while CMS Tool failed to provide any responses
4. Simple queries: Even on basic terms, CMS Tool frequently returned irrelevant results
5. Natural language processing: Falcon AI demonstrates superior ability to understand query intent