Diagnostic performance and generalizability of preoperative prediction models for lymph node metastasis in intrahepatic cholangiocarcinoma: a multimodal evidence synthesis

Accurate preoperative prediction of lymph node metastasis (LNM) in intrahepatic cholangiocarcinoma (ICC) is crucial for clinical decision-making, but the performance of existing models varies. This study, through a systematic review and meta-analysis, aims to evaluate the comprehensive diagnostic efficacy of these models and explore the key methodological factors influencing their performance, thereby providing empirical evidence for the optimization of future models. We systematically searched PubMed, Embase, the Cochrane Library, and Web of Science up to November 4, 2025. Studies developing or validating preoperative LNM prediction models in ICC were eligible. Methodological quality was assessed using the PROBAST tool. A bivariate random‑effects model was used to pool sensitivity, specificity, and the area under the summary receiver operating characteristic curve (SROC-AUC). Meta‑regression and subgroup analyses were conducted to examine sources of heterogeneity. Fourteen retrospective studies comprising 31 prediction models were included. Based on 11 independent models prioritized for optimal validation rigor, the primary pooled sensitivity was 0.81 (95% confidence interval [CI]: 0.71–0.87) and specificity was 0.77 (95% CI: 0.72–0.82), with an overall SROC‑AUC of 0.84 (95% CI: 0.80–0.87). Secondary exploratory heterogeneity analyses indicated that model category, data source, and algorithm type were key moderators of performance. Specificity dropped markedly in external test sets (0.68) compared with internal validation sets (0.82). This analysis showed no significant publication bias (p > 0.05). Preoperative prediction models show promising diagnostic potential for LNM detection in ICC, but their performance is highly dependent on feature categories and validation rigor. The marked decline in specificity upon external testing highlights an important gap in model generalizability. Future work should prioritize prospective, multi‑center external validation and focus on developing robust, interpretable models with standardized reporting to enable clinical translation.