Clinical decision-making using chatgpt-4 in pancreatic cancer

Introduction: With the advent of large language models, which are trained on extensive data sets of human-generated texts, different studies have demonstrated AI to be a useful tool in clinical-decision making. Since its launch in March 2024 ChatGPT-4 (OpenAI), has demonstrated superior capabilities compared to its predecessors. Whether ChatGPT-4 can assist clinical decision-making in patients with pancreatic cancer remains to be determined.

Objective: This study aims to assess the concordance between therapeutic decisions generated with ChatGPT-4 and those of a multidisciplinary tumor board (MDT) in patients with newly diagnosed pancreatic cancer.

Methods: Retrospectively collected patient data from tumor board referrals of newly diagnosed pancreatic cancer cases were transferred to a clinical data matrix and entered into ChatGPT-4, along with a request for a therapeutic recommendation. Following the initial output, 41 abstracts of key RCTs relevant to pancreatic cancer treatment were added, prompting ChatGPT to reassess its recommendation. Concordance between ChatGPT and the multidisciplinary tumor board (MDT) was evaluated both before and after literature input.

Results: Between 09/2024 and 03/2025, 45 patients with newly diagnosed pancreatic cancer were enrolled in this study. Concordance between ChatGPT and the MDT was evaluated both before and after additionally supplementing ChatGPT-4 with 41 high-ranking RCT abstracts. The initial concordance rate for treatment recommendations between ChatGPT-4 and MDT was 64.4%. Following literature-based prompting, the rate was 62.2%, with no significant difference between the pre- and post-prompting group. A comparison of patient characteristics between matched and unmatched groups also revealed no significant statistical differences.

Conclusion: In patients with pancreatic cancer, a structured clinical data matrix enables ChatGPT-4 to generate high-quality treatment recommendations. ChatGPT-4 demonstrated a concordance rate of 64.4% with MDT recommendations, which remained comparable (62.2%) even after supplementation with relevant literature. No significant differences were observed in concordance rates or patient characteristics between matched and unmatched groups. These findings suggest that ChatGPT-4 aligns substantially with guideline-based expert recommendations and may therefore serve as a valuable tool along the clinical decision-making process in patients with newly diagnosed pancreatic cancer.

 

Scroll to top

Publication History

Article published online:
04 September 2025