Exploring the Potential of GPT-4 in Creating Billing Codes from Clinic Notes

Creating standardized billing codes from clinic notes is challenging due to the complexity of over 22,000 codes and the unstructured nature of medical records. This paper investigates how well GPT-4, can automate CPT/HCPCS codes generation. To assess its understanding, we prompted GPT-4 to produce CPT/ HCPCS codes from their textual descriptions. Next, we prompted GPT-4 to generate billing codes, with confidence scores, from Vanderbilt University Medical Center clinic notes linked to billing codes by patient visit identifiers. GPT-4 achieved 20.8% accuracy in generating exact codes from code descriptions and 28.9% mean true positive rate in assigning correct codes for clinic notes. These results highlight the challenges LLMs face in understanding and generating accurate billing codes.