Akademik Veri Yönetim Sistemi
Future oncology (London, England), ss.1-8, 2025 (SCI-Expanded, Scopus)
Background: Adjuvant chemotherapy decision-making in stage II colon cancer remains challenging. Although multidisciplinary tumor boards (MDTs) guide treatment, their recommendations vary. Artificial intelligence (AI) tools such as ChatGPT may support decision-making, but direct comparative evidence with MDTs is limited.
Methods: We retrospectively analyzed 179 patients with stage II colon cancer who underwent surgery between 2019-2024. MDT recommendations (observation, fluoropyrimidine monotherapy, or oxaliplatin-based chemotherapy) were compared with ChatGPT-5 outputs. Clinical factors - including age, ECOG performance status (PS), tumor stage, minor risk factors, and mismatch repair (MMR) status - were incorporated. Agreement was evaluated using Cohen's kappa (κ) and McNemar's test.
Results: Across the three treatment categories, agreement between MDT and AI was moderate (70.4%, κ = 0.542, p < 0.001), while in the binary comparison of adjuvant therapy versus observation, concordance improved to substantial (91.1%, κ = 0.719, p < 0.001). Discordance mainly reflected AI's tendency to escalate therapy. Agreement decreased in patients ≥70 years, those with ECOG PS 2, and those with multiple risk factors.
Conclusions: AI showed moderate agreement with MDTs in detailed three-category recommendations but substantial concordance in binary adjuvant decisions. While AI may serve as a supportive tool, clinical judgment remains essential, particularly for elderly and frail patients.
Keywords: Adjuvant chemotherapy; artificial intelligence; large language models; multidisciplinary tumor board; stage II colon cancer.
Background: Adjuvant chemotherapy decision-making in stage II colon cancer remains challenging. Although multidisciplinary tumor boards (MDTs) guide treatment, their recommendations vary. Artificial intelligence (AI) tools such as ChatGPT may support decision-making, but direct comparative evidence with MDTs is limited.
Methods: We retrospectively analyzed 179 patients with stage II colon cancer who underwent surgery between 2019-2024. MDT recommendations (observation, fluoropyrimidine monotherapy, or oxaliplatin-based chemotherapy) were compared with ChatGPT-5 outputs. Clinical factors - including age, ECOG performance status (PS), tumor stage, minor risk factors, and mismatch repair (MMR) status - were incorporated. Agreement was evaluated using Cohen's kappa (κ) and McNemar's test.
Results: Across the three treatment categories, agreement between MDT and AI was moderate (70.4%, κ = 0.542, p < 0.001), while in the binary comparison of adjuvant therapy versus observation, concordance improved to substantial (91.1%, κ = 0.719, p < 0.001). Discordance mainly reflected AI's tendency to escalate therapy. Agreement decreased in patients ≥70 years, those with ECOG PS 2, and those with multiple risk factors.
Conclusions: AI showed moderate agreement with MDTs in detailed three-category recommendations but substantial concordance in binary adjuvant decisions. While AI may serve as a supportive tool, clinical judgment remains essential, particularly for elderly and frail patients.
Keywords: Adjuvant chemotherapy; artificial intelligence; large language models; multidisciplinary tumor board; stage II colon cancer.