Development and validation of a prognostic model for predicting the persistence of prostate-specific antigen after radical prostatectomy
10.3760/cma.j.cn112330-20240701-00305
- VernacularTitle:根治性前列腺切除术后PSA持续的预测模型构建与跨中心验证
- Author:
Xianqi SHEN
1
;
Wenhui ZHANG
;
Jin JI
;
Yan WANG
;
Min QU
;
Zhenyang DONG
;
Jialun LI
;
Zenghui ZHOU
;
Jie WANG
;
Xu GAO
Author Information
1. 上海长海医院泌尿外科，上海 200433
- Publication Type:Journal Article
- Keywords:
Prostatic neoplasms;
Carcinoma;
Prostate specific antigen persistence;
Nomogram;
Prediction
- From:
Chinese Journal of Urology
2025;46(1):37-43
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the factors influencing the persistence of prostate specific antigen(PSA) following radical prostatectomy, and to develop and validate a predictive model for PSA persistence.Methods:Clinical data from 1 828 patients who underwent radical prostatectomy at Shanghai Changhai Hospital between January 2015 and December 2023 were retrospectively analyzed. Of these, 1 295 patients from January 2015 to April 2021 comprised the modeling group, while 533 patients from May 2021 to December 2023 formed the validation group. Additionally, 109 patients who underwent radical surgery at the Third Affiliated Hospital of Naval Medical University between March and December 2023 were included as an external validation group. Patients with incomplete clinical information, serum PSA levels exceeding 100 ng/ml, or those who received preoperative neoadjuvant therapy were excluded. Ultimately, 1 003, 369, and 86 patients were included in the modeling, validation, and external validation groups, respectively. The modeling group had serum PSA of 19.29 (8.43, 23.73) ng/ml; the clinical stages were distributed as T 1, T 2, T 3, and T 4 in 191, 673, 123, and 16 patients, respectively; the primary Gleason scores of biopsy were 3, 4, and 5 in 460, 466, and 77 patients, respectively; and the secondary Gleason scores were 3, 4, and 5 in 363, 486, and 154 patients, respectively. The validation group had serum PSA of 12.80 (6.82, 14.40) ng/ml; the clinical stages were distributed as T 1, T 2, T 3, and T 4 in 40, 289, 37, and 3 patients, respectively; the primary Gleason scores of biopsy were 3, 4, and 5 in 218, 145, and 6 patients, respectively; and the secondary Gleason scores were 3, 4, and 5 in 140, 184, and 45 patients, respectively. The external validation group had serum PSA of 12.84 (7.11, 12.97) ng/ml; the clinical stages were distributed as T 1, T 2 and T 3 in 9, 68, and 9 patients, respectively; the primary Gleason scores of biopsy were 3, 4, and 5 in 58, 27, and 1 patient, respectively; and the secondary Gleason scores were 3, 4, and 5 in 28, 50, and 8 patients, respectively. Logistic regression analysis was used to identify independent risk factors for PSA persistence after radical prostatectomy in the modeling group and a prediction model was constructed. The predictive performance of the model was analyzed using the area under the curve (AUC) of the receiver operating characteristics (ROC) curve, the calibration curve, and the clinical decision curve. The predictive performance of the model was verified by the ROC curve in the validation group and the external validation group. Results:The incidence of persistent PSA after surgery in the modeling group, validation group, and external validation group was 8.97% (90/1 003), 7.32% (27/369), and 17.4% (15/86), respectively. In the modeling group, univariate and multivariate logistic regression analysis revealed that serum PSA, percentage of positive needle cores, primary Gleason score on biopsy, and secondary Gleason score on biopsy were independent risk factors for PSA persistence ( P<0.05), and a prediction model was constructed based on these factors. The AUC value of this model was 0.790 (95% CI 0.745-0.835). Calibration curve and clinical decision curve analyses showed that the model's predicted probabilities aligned well with actual risks within the 0-40% prediction interval, providing clinical benefit. The AUC values of the ROC curves in the validation group and external validation group were 0.808 (95% CI 0.719-0.897) and 0.822 (95% CI 0.714-0.929), respectively, indicating that the model had good predictive performance. Conclusions:The predictive model for PSA persistence, constructed based on serum PSA, percentage of positive needle cores, primary and secondary Gleason score on biopsy, demonstrated good clinical predictive performance, exhibiting high accuracy in both internal and cross-center validation.
