Prognostic Scores for Liver Resection in Colorectal Metastases: Performance, Limitations, and Methodological Pitfalls—A Systematic Review and Meta-Analysis

Background/Objectives: The prediction of survival after resection of colorectal liver metastases is crucial for planning treatment strategies. Several prognostic scores have been proposed, but their reliability is debated. The present study aims to review available prognostic scores, focusing on their performance and the methodological approaches adopted for their evaluation. Methods: A systematic literature review was conducted using PubMed, Embase, and the Cochrane Database, including studies published between January 2015 and June 2024. Only English-language studies reporting the external validation of prognostic models were included. A random-effects meta-analysis was performed. Results: Overall, 48 prognostic scores were externally validated across 48 studies (n = 33,602 patients). A total of 286 performance measurements were reported, utilizing 17 different metrics and considering four outcomes: overall survival (OS), cancer-specific survival, recurrence-free survival (RFS), and recurrence rate. For OS, the pooled C-index values for the Fong, GAME, and RAS mutation Clinical Risk scores were 0.578 (0.570–0.587), 0.609 (0.592–0.625), and 0.579 (0.471–0.688), respectively. For RFS, the pooled C-index for the Fong score was 0.616 (0.578–0.653). Scores incorporating genetic, immunological and radiomic data performed better than purely clinical ones (C-index = 0.610, 0.657 and 0.635, respectively, vs. 0.585, p < 0.05). Analogously, the scores including perioperative data outperformed preoperative ones (C-index = 0.671 vs. 0.600, p = 0.007). Conclusions: The current survival prediction relies on scores with low reliability (C-index ≤ 0.65). Despite the abundance of available data, their heterogeneity and variable quality have limited their usability. Future research should prioritize the development of new prognostic tools and the standardization of prognostic modeling and reporting.