KarvBill
← Back to Home

How KarvBill Works

Complete transparency about our analysis methodology, data sources, and limitations.

Our Approach

KarvBill uses automated rule-based analysis to identify potential billing errors in medical bills. Our system checks your bill against official Medicare billing rules, CMS guidelines, and typical pricing data to flag issues that warrant further review.

We analyze billing codes (CPT/HCPCS), diagnosis codes (ICD-10), pricing, quantities, and relationships between charges to detect common errors like duplicate charges, unbundling, pricing outliers, and potential upcoding.

Detection Methods

Duplicate Charge Detection

We identify line items that appear multiple times with matching or highly similar characteristics.

What We Check:

  • • CPT/HCPCS code matches on the same service date
  • • Identical amounts and quantities
  • • Similar descriptions (using medical terminology normalization)
  • • Cross-page duplicates with matching position (bounding boxes)

Confidence Factors:

  • • CPT match: 30% weight
  • • Amount match: 20% weight
  • • Date match: 20% weight
  • • Description similarity: 15% weight
Threshold: 70% confidence
Unbundling Detection

We flag when services that should be bundled together are billed separately.

Detection Methods:

  • • Hardcoded pairs of codes known to bundle (e.g., venipuncture with lab panels)
  • • Pattern-based detection (panel codes with component codes)
  • • NCCI (National Correct Coding Initiative) edit references

Example:

CPT 36415 (venipuncture) is typically included in comprehensive lab panels and shouldn't be billed separately.

Threshold: 75% confidence
Pricing Analysis

We compare charges against Medicare fee schedules and typical allowed amounts.

What We Check:

  • • Unit price vs. Medicare allowed amount
  • • Math errors (quantity × unit price ≠ total)
  • • Unusual quantities (e.g., 50 units of a typically single-use service)
  • • Zero unit prices with non-zero totals

Severity Levels:

  • • High: >75% above typical pricing
  • • Medium: 25-75% above typical pricing
  • • Low: 10-25% above typical pricing
Threshold: 65% confidence
Upcoding Detection (Conservative)

We flag potential upcoding with extreme caution, only when multiple E/M codes appear on the same date.

What We Check:

  • • Multiple evaluation & management (E/M) codes on same date
  • • E/M code level vs. description keywords (low confidence)

Note: Upcoding is complex and context-dependent. Multiple E/M codes may be legitimate if different providers saw the patient. We use a conservative threshold (80%) and encourage reviewing documentation.

Threshold: 80% confidence

Data Sources

Our analysis is based on official government data and industry-standard billing rules:

CMS Medicare Fee Schedules

Physician Fee Schedule (PFS) data for typical allowed amounts and relative value units (RVUs).

View CMS Fee Schedules

NCCI Coding Edits

National Correct Coding Initiative edits that define which code pairs can be billed together.

View NCCI Edits

CPT/HCPCS Code Database

Current Procedural Terminology codes maintained by the AMA and HCPCS codes from CMS.

View CMS Coding Resources

Medical Terminology Dictionary

Abbreviation and synonym mappings to normalize medical descriptions across different billing formats.

Confidence Scoring System

Each flag includes a confidence score (0-100%) indicating how certain we are about the potential issue. Our confidence calculation is multi-factor and category-specific.

Base Formula:

confidence = (base_score × 0.3) + (weighted_evidence × 0.7)

Evidence Weights (Category-Specific):

Different evidence factors have different weights depending on the flag category. For example:

Duplicate Detection: CPT match (30%), Amount match (20%), Date match (20%)
Unbundling: CPT match (35%), Date match (25%), Provider match (20%)
Pricing: Amount match (40%), CPT match (30%), Reference price (20%)

Confidence Thresholds:

High (80%+)
Medium (60-79%)
Low (<60%)

Minimum Thresholds: We only show flags that meet category-specific minimum confidence thresholds (duplicate: 70%, unbundling: 75%, pricing: 65%, upcoding: 80%). This reduces false positives.

Limitations & Disclaimers

What We Can't Do:

  • We can't verify medical necessity. A charge might be coded correctly but still inappropriate for your condition.
  • We can't interpret clinical documentation. Some apparent errors may be justified by medical circumstances we can't see.
  • We can't analyze insurance contracts. Your specific insurance policy may have different rules than Medicare.
  • We don't negotiate or file appeals. We identify issues but don't handle disputes with providers or insurers.
  • We may miss context-specific issues. Automated analysis can't replace human expert review for complex cases.

False Positives:

Our system may flag items that are actually correct. Medical billing is complex, and many factors we can't see (modifiers, medical necessity, special circumstances) may justify charges that appear unusual. Always review flags with your provider and consider consulting a medical billing professional for significant concerns.

Not Medical or Legal Advice:

KarvBill provides informational analysis only. We are not providing medical advice, legal advice, or professional billing advocacy services. For disputes, appeals, or complex issues, consult with qualified professionals.

Frequently Asked Questions

How accurate is KarvBill?
We're in beta and actively building our track record. Our detection rules are based on official CMS guidelines and Medicare billing standards. We use confidence thresholds to reduce false positives, but some flags may not represent actual errors due to factors we can't see. Your feedback (marking flags as correct/wrong) helps us improve accuracy.
What should I do if KarvBill flags something?

1. Review the flag details and evidence

2. Check the authoritative sources we link to (CMS, NCCI, etc.)

3. Contact your provider's billing department to ask about the charge

4. If the issue isn't resolved, consider consulting a medical billing advocate

5. Mark the flag as correct/wrong to help improve our system

Can I trust the confidence scores?
Confidence scores reflect how much evidence supports a flag, not absolute certainty. High confidence means multiple strong factors align, but it's not a guarantee. Use confidence as one factor in deciding whether to investigate further. We're transparent about our calculation methodology (see above).
Why don't you flag everything above Medicare rates?
Providers can legally charge more than Medicare rates. We flag significant outliers (typically 10%+ above typical), but higher charges aren't necessarily errors. Factors like facility type, geographic location, and service complexity affect pricing. We focus on potential errors, not just high prices.
How do you handle my bill data?
See our Security & Privacy page for complete details. In short: bank-level encryption, no selling of data, secure cloud storage, and you can delete your data anytime.

Help Us Improve

We're committed to continuous improvement. When you mark flags as correct or wrong, you're helping train our system and improve accuracy for everyone. We use this feedback to adjust confidence calculations and detection rules.

Questions or concerns about our methodology? Contact us — we're always open to feedback.