Home Health AI-Powered Coding and OASIS QA: Accuracy Is the Product

AI is moving quickly into home health ICD-10 coding and OASIS QA. Agencies are looking for faster turnaround times, lower costs, and relief for overburdened clinicians and coding teams.

But here’s the reality many agencies are running into: AI without high accuracy creates more problems than it solves.

Recently, we spoke with the owner of a 1,000-ADC home health agency who had piloted multiple AI documentation and coding vendors. Their takeaway was simple:

“The quality was terrible.”

Unfortunately, that experience isn’t uncommon.

Low-accuracy AI increases risk in home health coding

In home health ICD-10 coding, 50–60% accuracy is effectively unusable. At that point it’s just coin flip coding. When agencies don’t know which codes are correct and which are wrong, they end up with:

• More QA rework

• Higher compliance and audit risk

• Missed diagnoses and lost acuity

• Increased clinician and coder frustration

At that point, AI adds overhead instead of efficiency, especially if agencies lack experienced in-house coders to catch errors.

AI that cannot be trusted at the chart level is not automation. It’s noise.

Most agencies don’t know how accurate their AI coding vendor really is

⚠️WARNING: Strap in for a quick trip down memory lane. Yes, back to high school Statistics 101. Don’t worry, no pop quiz. 🙂

Many AI coding vendors advertise “accuracy” without explaining what it means or how it’s measured. Some rely on disclaimers like “AI may make mistakes.” Others publish a single percentage without clarity on methodology and validity.

In home health ICD-10 coding and OASIS QA, accuracy breaks down in two critical ways:

• Low precision: Adding diagnoses that are not supported by documentation, creating compliance and audit risk

• Low recall: Missing valid diagnoses, leading to under-coding, lost revenue, and incorrect OASIS outcomes

This is why tracking F-score to measure accuracy matters. It measures both precision and recall together, providing a true view of coding accuracy on real production charts.

When evaluating AI vendors for home health coding and OASIS QA, agencies should ask:

• Do you have a QA process for measuring your coding and OASIS accuracy.

• Do you measure precision and recall separately?

• What is your F-score on real, production home health charts?

• How are certified coders involved in validating AI outputs?
  
  

A managed service approach to home health ICD-10 coding and OASIS QA

At Olli Health, we’ve taken a deliberate approach: accuracy is the product, not a marketing claim.

Home health documentation is complex, unstructured, and clinically nuanced. Traditional, human-only coding misses diagnoses buried in 100+ pages of referral and unstructured documentation.

Solving it requires proprietary AI models purpose-built for home health, paired with certified human coders and clinicians who validate and stand behind every output.

Our managed service model combines clinical AI with expert human review to deliver:

• Consistently Highly accurate ICD-10 coding

• Reliable, compliant OASIS QA

• Faster turnaround times without sacrificing quality

• Reduced compliance and denial risk

After years of working exclusively on home health ICD-10 coding and OASIS QA, one thing is clear: AI only works in home health if agencies can trust it.

That trust comes from measurable accuracy, transparent validation, and human accountability, not flashy demos or vague claims.

Want to learn more? Email us at sales@ollihomehealth.ai and we’re happy to show how Olli works.