Construction And Comparison Of Models Of Prostate Health Index And Multiparametric Magnetic Resonance Imaging In Detecting Prostate Cancer

Background and ObjectiveDue to the characteristics of low specificity and high false positive rate of serum prostatespecific antigen(PSA)screening,it is difficult to diagnose prostate cancer(PCa)accurately.Therefore,the use of a new serum marker,prostate health index(PHI),in combination with multiparametric magnetic resonance imaging(mpMRI)is necessary to improve the diagnosis of PCa and clinically significant prostate cancer(CSPCa).To confirm the diagnostic efficacy of PHI and mpMRI for PCa in Asian populations,we developed different prediction models based on PHI,mpMRI and other PSA derivatives and compared their efficacy in diagnosing PCa and CSPCa.MethodsThis study was conducted from September 2020 to November 2021 and 128 patients were enrolled in this study who underwent mpMRI with PHI testing prior to biopsy and ultimately transperineal targeted prostate biopsy.We use the Prostate Imaging Reporting and Data System(PI-RADS)version 2.1 to interpret the mpMRI.Univariable and multivariable logistic analyses were performed to screen for statistically significant predictors of CSPCa and non-CSPCa.We finally conduct the base prediction model based on PSA derivatives and other prediction models combining the basic prediction model with PHI or PI-RADS.Finally,the final predictive model was built by adding PHI and PI-RADS to the base model.The receiver operating characteristic(ROC)curve and area under the receiver operating characteristic curve(AUC)were used to assess the predictive ability of different models.The diagnostic indicators of different predictive models were compared,such as specificity,positive predictive value,negative predictive value,overall diagnostic accuracy,positive likelihood ratio,and negative likelihood ratio.Subgroup analysis was also performed at the PSA level.Statistical analysis was performed using SPSS V.25.0 and R statistical software.P<0.05 was considered statistically significant.Delong test was used to compare differences in AUC.ResultsOf the 128 patients in this study,47(36.72%)patients were diagnosed with CSPCa and 81(63.28%)patients were diagnosed with non-CSPCa.In the analysis of ROC curve in PSA 4-20 ng/ml,the combined diagnostic multivariable prediction model for PCa was significantly greater in diagnostic accuracy than the prediction model based on the PI-RADS(p=0.004)and PHI(p=0.031).The combined diagnostic multivariable model for CSPCa was significantly higher in diagnostic accuracy than the model based on the PI-RADS(p=0.003).In the subgroup analysis of TPSA,for detection of PCa in the TPSA 4-10 ng/ml,the combined multivariable prediction model had a significantly higher AUC area than the PHI-based prediction model(p=0.025).In the case of detecting CSPCa in the TPSA 4-10 ng/ml,the AUC of the combined diagnostic prediction model was greater than that of the PHI-based prediction model or PIRADS-based prediction model.In the case of TPSA 10-20 ng/ml,the AUC of the combined diagnostic prediction model detecting PCa and CSPCa was significantly higher than that of the PI-RADS-based prediction model(p=0.029;p=0.016).ConclusionIn this study,it was found that the prediction model combining PHI and PI-RADS was more effective in detecting PCa and CSPCa in PSA of 4-20 ng/ml than the prediction model based on PHI or PI-RADS alone.It was found that adding PI-RADS to the PHI-based model significantly improved the efficacy of the detection of PCa in the PSA 4-10 ng/ml.The addition of PHI to the PI-RADS-based model was more useful for detecting PCa and CSPCa in total PSA 10-20 ng/ml.