Patients diagnosed with metastatic breast cancer, colorectal cancer, and renal cell carcinoma between 2015 and 2023 at two hospitals affiliated with a single university were retrospectively reviewed. The exclusion criteria included metastatic lesions that developed during treatment, CT scans with motion artifact, and non-contrast CT scans. While patients from one hospital constituted the training cohort, those from the other served as the external validation cohort. The training cohort included 50 breast cancer, 55 colorectal cancer, and 55 renal cell carcinoma patients. The external validation cohort consisted of 20 breast cancer, 30 colorectal cancer, and 19 renal cell carcinoma patients. Clinical and radiological parameters, such as age, number and laterality of metastases, mediastinal lymphadenopathy, lymphangitic carcinomatosis, pleural effusion and pleural metastasis, as well as bone and liver metastases were recorded. The clinical-radiological model was developed based on five parameters that showed statistically significant difference among the three groups (p < 0.05). Radiomics analysis was conducted using dedicated software. Three-dimensional manual segmentation was performed (Figure 1), yielding 110 original features and 744 high-order features derived from wavelet filters, resulting in a total of 854 radiomics features. For the radiomics model, LASSO regression identified eight features with nonzero coefficients. Models were developed using the Random Forest algorithm. The combined model integrating both clinical-radiological parameters and radiomics features was constructed. Model performances were evaluated using macro-averaged and micro-averaged area under the curve (AUC) values.