Back to 2026 Abstracts
Head To Head Comparison Of Vascular Surgery Coding And Billing In Surgeons Vs Artificial Intelligence
Andrew Lee, MD1, Kerry Drabish, PhD, APRN, FNP-BC
2, Catherine Go, MD
2, Mohammad Eslami, MD, MPH, MBA
1.
1Charleston Area Medical Center, WVU, Charleston, WV, USA,
2Charleston Area Medical Center, Charleston, WV, USA.
OBJECTIVE Accurate Current Procedural Terminology (CPT) coding is critical for reimbursement, quality reporting, and compliance with the Centers for Medicare and Medicaid Services (CMS). Manual coding is labor-intensive, susceptible to human error, and often lags behind evolving operative techniques. Artificial intelligence (AI) natural language processing (NLP) models such as OpenAI’s ChatGPT-4o demonstrate strong performance in complex language processing and specialized reasoning, suggesting potential utility in surgical billing. This study compared the accuracy of vascular surgeons (VS) and ChatGPT-4o in identifying CPT codes for vascular surgery procedures.
METHODS This observational study evaluated ChatGPT-4o against six VS in coding 20 sample vascular operative reports that included pre-/postoperative diagnoses, indications, findings, and procedure descriptions. Reports were processed by both VS and ChatGPT-4o (queried in learning and non-learning modes). Certified professional coders and Society for Vascular Surgery Coding Committee guidelines served as benchmarks. Primary outcome was exact-match CPT accuracy. Secondary outcomes included accuracy by procedure type (open vs endovascular); simple (straightforward) vs complex (non-bundled codes, treatment involving multiple anatomic areas) endovascular procedures, work relative value units (wRVUs) generated, and revenue impact. Independent samples t-tests were used, with significance set at p<0.05.
RESULTS Overall CPT accuracy, total wRVUs, and revenue generation did not differ significantly between AI and VS. Accuracy for overall endovascular procedures was also similar between groups. However, when comparing simple and complex endovascular procedures, AI showed a significantly higher accuracy for complex cases (59% vs. 32%, p<0.0001). For open procedures, VS had significantly higher accuracy (66% vs. 43%, p=0.02), while AI produced significantly greater wRVU (p=0.004) and monetary error rates (p=0.01).
CONCLUSIONS Vascular surgeons outperformed AI in coding open and straightforward procedures, whereas AI excelled in coding complex endovascular cases. AI NLP tools show promise in being utilized to assist in interpreting and identifying CPT codes from operative reports, with promising implications for revenue integrity and administrative burden reduction. Further validation, training, and compliance safeguards are necessary before integration into clinical billing workflows.
Back to 2026 Abstracts
