Construction and Validation of a Clinical Predictive Nomogram for Improving the Cancer Detection of Prostate Naive Biopsy Based on Chinese Multicenter Clinical Data.

objectivesProstate biopsy is a common approach for the diagnosis of prostate cancer (PCa) in patients with suspicious PCa. In order to increase the detection rate of prostate naive biopsy, we constructed two effective nomograms for predicting the diagnosis of PCa and clinically significant PCa (csPCa) prior to biopsy. MATERIALS AND

methodsThe data of 1,428 patients who underwent prostate biopsy in three Chinese medical centers from January 2018 to June 2021 were used to conduct this retrospective study. The KD cohort, which consisted of 701 patients, was used for model construction and internal validation; the DF cohort, which consisted of 385 patients, and the ZD cohort, which consisted of 342 patients, were used for external validation. Independent predictors were selected by univariate and multivariate binary logistic regression analysis and adopted for establishing the predictive nomogram. The apparent performance of the model was evaluated

resultsThe results of univariate and multivariate logistic regression analysis showed prostate-specific antigen density (PSAD) (P<0.001, OR:2.102, 95%CI:1.687-2.620) and prostate imaging-reporting and data system (PI-RADS) grade (P<0.001, OR:4.528, 95%CI:2.752-7.453) were independent predictors of PCa before biopsy. Therefore, a nomogram composed of PSAD and PI-RADS grade was constructed. Internal validation in the developed cohort showed that the nomogram had good discrimination (AUC=0.804), and the calibration curve indicated that the predicted incidence was consistent with the observed incidence of PCa; the brier score was 0.172. External validation was performed in the DF and ZD cohorts. The AUC values were 0.884 and 0.882, in the DF and ZD cohorts, respectively. Calibration curves elucidated greatly predicted the accuracy of PCa in the two validation cohorts; the brier scores were 0.129 in the DF cohort and 0.131 in the ZD cohort. Decision curve analysis showed that our model can add net benefits for patients. A separated predicted model for csPCa was also established and validated. The apparent performance of our nomogram for PCa was also assessed in three different PSA groups, and the results were as good as we expected.

conclusionsIn this study, we put forward two simple and convenient clinical predictive models comprised of PSAD and PI-RADS grade with excellent reproducibility and generalizability. They provide a novel calculator for the prediction of the diagnosis of an individual patient with suspicious PCa.