Abstract
BACKGROUND: Prostate cancer is a common malignancy in men worldwide. Approximately 30% of patients experience biochemical recurrence after radical treatment, leading to disease progression, increased treatment difficulty, and reduced survival rates. An accurate biochemical recurrence prediction model is required to assist in treatment decision-making. Therefore, we developed an effective assessment model using public databases. METHODS: The clinical data of 404 prostate cancer cases from The Cancer Genome Atlas (TCGA) were divided into training and validation sets (7:3 ratio). We developed a biochemical recurrence risk assessment model for prostate cancer based on training sets. DESeq2 was used to analyze differences between RNAs of prostate cancer in TCGA database, with thresholds of P<0.05, and |log(2)(fold change)| >1. Least absolute shrinkage and selection operator (LASSO) Cox regression analysis identified biochemical recurrence-associated RNAs using the minimum λ. The survival receiver operating characteristic (ROC) curve and area under the curve (AUC) assessed the model's predictive accuracy in the validation set. RESULTS: From TCGA data, 2,961 differential RNAs were identified. Of these, 502 RNAs were significantly associated with biochemical recurrence, 40 RNAs were selected via LASSO regression, and 10 RNAs were obtained through multifactor risk regression to construct a risk assessment model. The model showed strong predictive ability, with AUCs of 0.853 (training set) and 0.769 (validation set). The integrated model combined with clinical parameters had a better AUC of 0.869 in training and 0.774 for validation. CONCLUSIONS: This assessment model is a relatively accurate biochemical recurrence prediction tool for prostate cancer and can guide the treatment of prostate cancer.