Background: For prostate cancer, accurate prediction of the pathological stage before surgery is very important. Therefore, the aim of the present study was establishing the prostate-specific antigen (PSA) threshold nomogram to predict pathologically advanced prostate cancer using the novel method of area under the receiver operating characteristic curve boosting (AUCBoost).
Methods: The medical records of patients with clinically localized prostate cancer who underwent robot-assisted radical prostatectomy were retrospectively reviewed. Multivariate logistic regression analysis was performed to identify clinical covariates significantly associated with pathological tumor stage ≥3a. The best combination of the variables was determined by validated values of the area under the curve (AUC). The optimal individualized PSA threshold values were developed using AUCBoost.
Results: In the multivariate logistic regression analysis, PSA, prostate volume, clinical tumor stage, Gleason Grade Group, the number of positive cores, and the percentage of positive cores were independent predictive factors for pathological tumor stage ≥3a. A combination model comprising PSA, prostate volume, clinical tumor stage, percent positive core, and Gleason Grade Group produced the highest AUC for predicting pathological tumor stage ≥3a (AUC = 0.777). The PSA threshold values for detecting pathological tumor stage ≥3a were calculated and a table of individualized PSA threshold nomogram was developed using AUCBoost.
Conclusions: We developed a nomogram of the PSA threshold values for predicting adverse pathological tumor stages of prostate cancer using a novel statistical method. Further validation is necessary; however, the individualized PSA threshold nomogram may be useful in determining treatment strategies before surgery.
Keywords: Area under the receiver operating characteristic curve boosting; Extracapsular extension; Prostate cancer; Prostate-specific antigen; Threshold nomogram.
Copyright © 2021 Elsevier Inc. All rights reserved.