Akademik Veri Yönetim Sistemi
Clinical and Translational Oncology, 2025 (SCI-Expanded)
Background: Liver is the most common metastatic site in colorectal cancer. This study aims to evaluate the effectiveness of different machine learning (ML) models in predicting liver metastasis in CRC patients using routine biochemical tests. Patients and methods: Cross-sectional study employed various ML algorithms for predictive modeling. The study was conducted at two academic reference centers in Ankara, Turkey: a total of 810 CRC patients diagnosed between January 2010 and December 2023 were included. The training and internal validation dataset comprised 710, and external validation dataset included 100 patients. Inclusion criteria were patients aged ≥ 18 years with a pathological CRC diagnosis, pre-treatment biochemical tests, and known initial staging. Exclusion criteria encompassed non-adenocarcinoma histologies, incomplete biochemical data, other malignancies. Results: Logistic regression achieved the highest internal validation AUC (0.956), accuracy (0.901), and F1 score (0.936), with a sensitivity of 0.971 and specificity of 0.703. ElasticNet and Lasso regression followed closely with AUCs of 0.958. In external validation, logistic regression maintained high performance (AUC 0.951, accuracy 0.900), while the K-nearest neighbors (KNN) model achieved perfect sensitivity (1.0) with an AUC of 0.891. The optimal predictor combination included ALP, LDH, CEA, and CA-19-9. Conclusion: Different ML models, can effectively predict liver metastasis in CRC patients using routine biochemical tests. Further refinement and prospective clinical trials are necessary to validate and implement these predictive tools in clinical practice.