Background: To assess the performance of an in-house developed AI Large Language Model (LLM) in assisting lung cancer patient management and compare its efficiency and accuracy against manual clinical workflows. Methods:We conducted a retrospective study on 849 consecutive lung cancer patient MDT records form Feb. 2025 to Dec. 2025.Our proprietary AI "Lung Cancer LLM" performed several tasks: rapid document scanning, structured data extraction, treatment suggestion generation, recurrence risk prediction, clinical trial matching, and follow-up management.Its performance was benchmarked against a team of three senior oncologists completing the same tasks.We measured time per case and assessed output accuracy (data extraction correctness, guideline compliance of suggestions) by independent experts. Results:The AI model demonstrated significant efficiency gains.It processed a full case in 2.1 minutes on average, over 13 times faster than the manual team (28.5 minutes).In clinical trial matching, the AI screened relevant trials in under 30 seconds per patient versus 12 minutes manually.Accuracy was comparable: the AI achieved 96.7% accuracy in key data extraction (vs.98.1% manually, p>0.05) and 94.5% guideline compliance for treatment suggestions (vs.96.0%, p>0.05).The AI also provided practical utility by generating risk stratifications and automatically identifying disease progression in follow-up notes.Conclusions: This AI lung cancer LLM matches specialist-level accuracy in core clinical tasks while offering order-of-magnitude improvements in processing speed.It serves as a highly efficient tool for automating data abstraction and initial clinical support, potentially freeing clinician time for higher-value care.Its integration of prediction and trial matching supports proactive patient management, demonstrating significant promise for real-world oncology practice.
Umihanić et al. (Tue,) studied this question.