Scholarlink

Background <p>The clinical relevance of concordance between multidisciplinary tumor board (MDT) decisions and artificial intelligence (AI)-based treatment recommendations remains unclear. This study evaluated not only concordance but also the clinical impact of discordance direction on outcomes in endometrial cancer.</p> Methods <p>We retrospectively analyzed 150 patients with endometrial cancer discussed at a single-center MDT. AI-based recommendations were generated using a large language model according to ESGO/ESTRO/ESP guidelines, based on standardized clinicopathological inputs without molecular classification. Concordance between MDT and AI recommendations was assessed at the patient level. Discordant cases were classified as relative undertreatment or overtreatment. Agreement beyond chance was evaluated using Cohen’s kappa. Recurrence-free survival (RFS) was analyzed using Cox regression and Kaplan–Meier methods.</p> Results <p>Overall concordance was observed in 76.7% of patients, with fair agreement beyond chance (Cohen’s κ = 0.365, 95% CI 0.181–0.549). During follow-up, 68 events (45.3%) occurred. Concordance was not independently associated with RFS in multivariable analysis (adjusted HR 1.26, 95% CI 0.64–2.48, <i>p</i> = 0.50). In contrast, relative undertreatment was associated with worse RFS (HR 1.94, 95% CI 1.03–3.64, <i>p</i> = 0.040), whereas overtreatment was not. Event rates were higher in the undertreatment group compared to concordant patients (67.9% vs 40.0%), despite more favorable baseline characteristics.</p> Conclusions <p>Concordance alone may be insufficient to determine the clinical relevance of AI-based recommendations. Discordance direction appears more informative, with treatment de-escalation associated with inferior outcomes. AI-based systems may help identify potential undertreatment and support more consistent implementation of guideline-based care.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Concordance between multidisciplinary tumor board decisions and AI-based recommendations in endometrial cancer: impact of discordance direction on clinical outcomes

  • 期刊论文

  • 发表日期：

    2026年7月5日

查看全文
  • Vali Aliyev,
  • Selin Cebeci,
  • Zeliha Birsin,
  • Murat Günaltılı,
  • Mehmet Cem Fidan,
  • Emir Çerme,
  • Hamza Abbasov,
  • Şefika Arzu Ergen,
  • Nebi Serkan Demirci,
  • Özkan Alan

摘要

Background

The clinical relevance of concordance between multidisciplinary tumor board (MDT) decisions and artificial intelligence (AI)-based treatment recommendations remains unclear. This study evaluated not only concordance but also the clinical impact of discordance direction on outcomes in endometrial cancer.

Methods

We retrospectively analyzed 150 patients with endometrial cancer discussed at a single-center MDT. AI-based recommendations were generated using a large language model according to ESGO/ESTRO/ESP guidelines, based on standardized clinicopathological inputs without molecular classification. Concordance between MDT and AI recommendations was assessed at the patient level. Discordant cases were classified as relative undertreatment or overtreatment. Agreement beyond chance was evaluated using Cohen’s kappa. Recurrence-free survival (RFS) was analyzed using Cox regression and Kaplan–Meier methods.

Results

Overall concordance was observed in 76.7% of patients, with fair agreement beyond chance (Cohen’s κ = 0.365, 95% CI 0.181–0.549). During follow-up, 68 events (45.3%) occurred. Concordance was not independently associated with RFS in multivariable analysis (adjusted HR 1.26, 95% CI 0.64–2.48, p = 0.50). In contrast, relative undertreatment was associated with worse RFS (HR 1.94, 95% CI 1.03–3.64, p = 0.040), whereas overtreatment was not. Event rates were higher in the undertreatment group compared to concordant patients (67.9% vs 40.0%), despite more favorable baseline characteristics.

Conclusions

Concordance alone may be insufficient to determine the clinical relevance of AI-based recommendations. Discordance direction appears more informative, with treatment de-escalation associated with inferior outcomes. AI-based systems may help identify potential undertreatment and support more consistent implementation of guideline-based care.