Hospital surgical and coding departments within German DRG-based revenue system
Key Highlights
Professional coders achieved highest accuracy in both sufficient (99-100%) and optimal (94%) OPS coding.
GPT CodeMedic, a medically fine-tuned LLM, outperformed GPT-4o in coding accuracy by over 10%.
Surgeons showed high sufficient coding accuracy but lower optimal coding performance compared to coders and GPT CodeMedic.
Guideline-Based Recommendations
Diagnosis
Use ICD D32.0 to identify meningioma cases for OPS coding extraction.
Management
Initial OPS coding should be performed by the operating surgeon.
Professional coders must review and revise coding before case closure to ensure accuracy.
LLMs such as GPT CodeMedic can assist in coding but should not replace professional coders.
Monitoring & Follow-up
Cases are subject to review by the Medical Service of Health insurance providers (MDK) to verify coding accuracy.
Monitor for hallucination errors in LLM-generated codes to avoid incorrect billing.
Risks
Incorrect OPS coding may lead to revenue reduction and fines.
Overcoding or adding codes not explicitly documented in surgical reports is not permitted.
Patient & Prescribing Data
100 patients undergoing meningioma resection from January 2023 to December 2024
Accurate OPS coding is critical for appropriate DRG classification and hospital revenue; LLMs show promise but require validation and professional oversight.
Clinical Best Practices
Ensure OPS codes include correct procedure and approach codes; accessory codes are optional but must be documented.
Provide LLMs with OPS catalogue and internal code lists to improve coding accuracy.
Use retrospective analysis and statistical validation to compare coding performance among surgeons, coders, and LLMs.
Maintain anonymized databases for coding accuracy assessment and quality control.
