Construct a predictive model to avoid excessive biopsy screening in patients with PI-RADs v2 3

Abstract

Objective Assist in clinical decision making by building models to predict the probability of clinically significant prostate cancer (CSPCa) with prostate imaging reporting and data system version 2 (PI-RADs v2) 3 and avoid unnecessary biopsy. Methods It’s a retrospective study which maintained database of 218 consecutive men who received prostate biopsy and with PI-RADs v2 category 3 in Capital Medical University, Beijing Friendship Hospital between January 2012 to July 2018, the average age was 70.7 years, and the age range was 63-77 years. Among them, 137 patients with benign diseases, 30 patients with clinically insignificant prostate cancer (CIPCa), and 51 patients with CSPCa. Models were established based on clinical variables. The measurement data were expressed as the median (interquartile range) [M(P25, P75)], and the rank sum test was used for comparison between groups; the Chi-square test was used for comparison between the count data groups. The decision curve was used to determine the clinical net benefit unilateral factors generated by the application of the model, univariate and multivariate logistic regression analysis to determine the predictors of positive outcomes. The diagnostic performance of the predictive model was evaluated by the area under the curve (AUC) of receiver operating curve, which was used to assess the overestimation or underestimation of the model, and the decision curve was used to determine the clinical net gain from the application of the model. Results Detection of prostate caner (PCa) and CSPCa in the PI-RADs v2 cohort were 37.2% (81/218) and 23.4% (51/218). The median prostate specific antigen of CSPCa patients was 12.1 ng/ml, which was higher than CIPCa (9.5 ng/ml) and benign (10.5 ng/ml) patients. The median prostate volume of CSPCa patients was 41.2 ml, lower than CIPCa (45.8 ml) and benign (57.3 ml) patients. The median prostate special antigen density (PSAD) was 0.28 ng/ml2, higher than CIPCa (0.20 ng/ml2) and benign (0.15 ng/ml2) patients. The predictive power of the developed model, based on age, PSAD, lesion region and ADC value, showed a higher AUC than that of parameters alone. Internally validated calibration curves showed that the nomogram might overestimate the risk of PCa when the threshold was above 35%. As for CSPCa, the predicted risk was closer to actual probability when the threshold was above 60%. Decision curves showed that a better net benefit was met when the model was used to guide clinical practice. Conclusions The models based on age, PSAD, lesion region and ADC value showed internally validated high predictive value for both PCa and CSPCa. It could be used to improve the detection rate of CSPCa and avoid unnecessary biopsy. Key words: PI-RADs v2; Prostatic neoplasms; Forecasting; Models, structural