![\"Patient](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/10\/1793079133_379_image-0.jpg\")
<\/p>\n(UroToday.com) The 2026 European Society of Medical Oncology (ESMO) Annual Congress held poster 333P – Development and external validation of prostate-specific antigen (PSA)-based models combining blood markers for prostate cancer prediction in men with total PSA 4\u201310 ng\/mL.<\/p>\n
\n
<\/p>\n
Dr. Liang began by highlighting that although serum PSA remains widely used for prostate cancer screening, its specificity is suboptimal in men with PSA levels of 4\u201310 ng\/mL, frequently leading to unnecessary biopsies. While routine laboratory parameters may provide additional information reflecting the tissue microenvironment, their diagnostic accuracy remains limited when used individually. His study therefore evaluated whether combining PSA measures with routinely available blood test results could improve prostate cancer detection in men within this PSA \u201cgray zone.\u201d<\/p>\n
<\/p>\n
The investigators presented an observational study developing prostate cancer diagnostic models using electronic medical records from a tertiary hospital in China (XJTH) between 2020 and 2024, with external validation using the Japanese Medical Data Vision (MDV) database. The study included adult men with PSA levels of 4\u201310 ng\/mL who had hematologic testing within 7 days of PSA measurement. Patients with a prior prostate cancer diagnosis or without a clinical visit within 60 days were excluded. The primary outcome was a prostate cancer diagnosis within 2 months of the index PSA test.<\/p>\n
<\/p>\n
Candidate predictors included age, total PSA, percentage free PSA, and 54 routinely available hematologic and biochemical variables. Three diagnostic models were compared with total PSA alone: a knowledge-based model incorporating age, PSA measures, platelet-to-lymphocyte ratio, neutrophil-to-lymphocyte ratio, and absolute monocyte count; a data-driven model using variable selection across all candidate predictors; and an XGBoost machine-learning model.<\/p>\n
<\/p>\n
Model performance was assessed using optimism-corrected internal validation with 200 bootstrap resamples and external validation in the Japanese cohort. Discrimination, calibration, and specificity at 95% sensitivity were evaluated, while decision curve analysis assessed clinical utility, including the potential reduction in unnecessary biopsies across clinically relevant risk thresholds of 5%\u201320%.<\/p>\n
<\/p>\n
Patient selection began with 4,814 men undergoing PSA testing at XJTH and 183,490 in the Japanese MDV database between 2020 and 2024. An intermediate PSA of 4\u201310 ng\/mL was identified in 2,699 (56.1%) and 49,863 (27.2%) patients, respectively. After restricting the cohorts to adults with hematologic testing within 7 days of the PSA measurement and excluding patients with prior prostate cancer or insufficient follow-up, the final development cohort included 1,758 men and the external validation cohort included 9,524 men.<\/p>\n
<\/p>\n
<\/p>\n
<\/p>\n
The development cohort included 1,758 men with a median age of 70 years and a median PSA of 6.0 ng\/mL. Prostate cancer was diagnosed in 202 patients (11.5%). The external MDV validation cohort was substantially larger, including 9,524 men with a median age of 74 years and a similar median PSA of 6.0 ng\/mL; 1,248 patients (13.1%) were diagnosed with prostate cancer.![\"The](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/10\/1793079133_294_image-1.jpg\")
The combined models consistently improved discrimination compared with total PSA alone. In internal validation, tPSA alone achieved an AUC of 0.688, increasing to 0.757 with the knowledge-based model, 0.790 with the data-driven model, and 0.825 with the machine-learning model. Importantly, the improvement persisted in external validation. While tPSA alone had an AUC of 0.586, the knowledge-based model achieved 0.609, and both the data-driven and machine-learning models reached 0.698 (all P<0.01). Thus, incorporating routine laboratory parameters provided meaningful additional discriminatory value beyond PSA alone.![\"image-2.jpg\"](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/10\/1793079134_2_image-2.jpg\")
<\/p>\n
<\/p>\n
At 95% sensitivity, all three models substantially improved specificity compared with tPSA alone, increasing from 14.9% with tPSA to 31.2% with the knowledge-based model, 34.3% with the data-driven model, and 39.5% with the machine-learning model. This translated into progressively fewer unnecessary biopsies, with 27.6, 30.3, and 35.0 biopsies avoided per 100 men, respectively, compared with 13.2 using tPSA alone. Importantly, this improvement was achieved while maintaining fewer than 1 missed prostate cancer diagnosis per 100 men across all models.![\"image-3.jpg\"](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/10\/1793079134_769_image-3.jpg\")
<\/p>\n
<\/p>\n
Calibration analyses demonstrated good agreement between predicted and observed prostate cancer risk for both the knowledge-based and data-driven models. Decision curve analysis further supported their clinical utility, with the data-driven and machine-learning models providing greater net benefit than tPSA alone across clinically relevant risk thresholds of 5%\u201320%. Overall, these models could improve risk stratification while reducing unnecessary biopsies.<\/p>\n
<\/p>\n
![\"image-4.jpg\"](\"https:\/\/www.newsbeep.com\/il\/wp-content\/uploads\/2026\/10\/1793079135_402_image-4.jpg\")
<\/p>\n
<\/p>\n
Dr. Liang concluded:<\/p>\n
\n
\nKnowledge-based and data-driven models significantly improved prostate cancer discrimination compared with tPSA alone while maintaining good calibration.
\nRoutinely available blood tests provided incremental information beyond tPSA for men with PSA levels of 4\u201310 ng\/mL.
\nAll models improved specificity, increased the proportion of potentially avoidable biopsies at high sensitivity, and provided greater net clinical benefit than tPSA alone.
\nThe more compact data-driven model retained substantial discrimination and clinical utility, supporting further evaluation as a pragmatic tool for clinical implementation.\u00a0
\n
<\/p>\n
Presented by:\u00a0Liang Liang, Department of Urology, The First Affiliated Hospital of Xian Jiaotong University, Shaanxi, China<\/p>\n
<\/p>\n