Development and internal validation of a parsimonious logistic regression model integrating multiparametric clinical indicators for predicting prostate cancer in the PI-RADS 3 cohort.

Objective: To develop and internally validate a parsimonious logistic regression model for clinically significant prostate cancer (CSPCa) specifically in patients with PI-RADS 3 lesions, integrating multiparametric clinical indicators to guide biopsy decision-making. Methods: We retrospectively enrolled 193 patients with PI-RADS 3 lesions who underwent mpMRI and prostate biopsy. A nested five-fold cross-validation framework was used, with LASSO regression performed independently in each training fold to select predictors from nine clinical variables. Four modeling approaches were compared, and logistic regression was chosen as the final model based on performance, calibration, and interpretability. Model discrimination, calibration, and clinical utility were evaluated internally. Results: LASSO selected age, PHI, and PHID as core predictors. Logistic regression achieved the highest cross-validated AUROC (0.840, 95% CI: 0.754-0.916) with acceptable calibration (slope 0.917, intercept -0.122). The model outperformed PSA and PSAD alone, and showed numerically higher discrimination than PHI or PHID alone. However, the incremental AUROC gain over PHI or PHID alone did not reach statistical significance. Exploratory risk stratification identified low-, intermediate-, and high-risk groups with CSPCa detection rates of 8.1%, 38.5%, and 83.3%, respectively. Conclusion: A three-variable logistic model incorporating age, PHI, and PHID demonstrated stable internal performance for predicting CSPCa in PI-RADS 3 patients. External validation is needed before clinical implementation.