When NLP Gets the Context Wrong: A DVT Case Study

An NLP coding tool read a chart, saw the word "DVT," and suggested an active clot code. The patient's DVT had actually resolved a year earlier. Here's exactly where the algorithm went wrong, and why that two-word miss matters a lot more than just one chart. In this video, I walk through a real-world style case: a patient with a history of DVT, resolved on imaging, still on anticoagulation as prevention rather than treatment. An NLP-assisted coding tool read "DVT" sitting near the anticoagulant name and suggested an acute embolism code. What the note actually supported was Z86.718, personal history of venous thrombosis and embolism, a code that doesn't carry the same risk-adjustment weight as an active or chronic clot. We'll break down the three codes involved (acute I82.4-, chronic I82.5-, and personal history Z86.718), why the NLP tool weighted keyword proximity over the qualifier language that actually decides the code, and why this exact fact pattern, a historical condition submitted as if it were still being actively managed, is precisely what the HHS Office of Inspector General flagged in its 2026 audit of a Medicare Advantage plan's risk-adjustment diagnoses (84% of sampled enrollee-years unsupported, an estimated $4.4 million in overpayments). You'll also get four practical guardrails for reviewing any NLP-suggested code before it ships, and why "the vendor's tool suggested it" isn't a defense CMS's RADV audit program will accept. Want the one-page version? Reach out through the contact form at help4docs.com and ask for the "NLP-Suggested Code Verification Checklist," and I'll send it your way, free. About the creator: Toni M. Appelberg-Parks, MBA-HM/LM, CDEO, CDEI, CRCR, CPC, CSBI, CPPM is President and Principal Consultant of Help4Docs, where she helps providers, coders, and CDI teams document and code in a way that actually holds up to audit. Learn more at help4docs.com. Category: AI & the Profession Series: Audit-Defense #MedicalCoding #RiskAdjustment #HCCCoding #ClinicalDocumentation #CDI #AIinHealthcare #RADV #MedicareAdvantage #HealthcareCompliance #DVT