A prediction model for urological tumor metastasis using liquid biopsy-derived biomarkers

To construct and validate a prediction model for tumor metastasis in patients with urological tumors based on liquid biopsy biomarkers and clinical characteristics, to facilitate early clinical identification of metastasis risk and formulation of individualized diagnosis and treatment plans.

A total of 360 patients with urological tumors admitted to our hospital from January 2021 to December 2024 were retrospectively included. They were randomly divided into a training set (n = 252) and a validation set (n = 108) at a ratio of 7:3. Demographic characteristics and liquid biopsy biomarker indicators of the patients were collected. In the training set, demographic characteristics (age, gender) and liquid biopsy biomarkers (C-reactive protein, neutrophil count, monocyte count, platelet count, mean platelet volume, platelet distribution width, large platelet percentage, hemoglobin, white blood cell count, and urine parameters) were assessed. Univariate analysis was used to screen metastasis-related indicators, followed by Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariate Logistic regression to identify independent predictors. Three machine learning models (random forest, support vector machine, gradient boosting) were then constructed. The efficacy of the models was evaluated by the area under the receiver operating characteristic curve (AUC). The optimal model was selected, and the importance of key prediction indicators was analyzed.

There was no significant difference in baseline data between the training set and the validation set (P > 0.05). Multivariate Logistic regression indicated that C-reactive protein, neutrophil count, platelet count, platelet distribution width, hemoglobin, white blood cell count, and mean platelet volume were independent influencing factors for tumor metastasis in patients with urological tumors (all P < 0.05). The AUC of the random forest model (0.891) was significantly higher than that of the support vector machine (0.885) and the gradient boosting model (0.739), making it the optimal model.

The random forest model constructed based on liquid biopsy predictive indicators can effectively predict tumor metastasis in patients with urological tumors. Neutrophil count, platelet count, and white blood cell count are key prediction indicators for urological tumor metastasis.