Answers are grounded in this paper and cite their sources.
Objectives The differential diagnosis of prostate cancer (PCa) in men with serum prostate-specific antigen (PSA) levels in the 4–10 ng/mL range remains a clinical challenge, and there is currently no reliable quantitative tool to guide biopsy decisions. We sought to build a machine-learning-based model using routine clinical indicators to assist individualized prostate biopsy decisions in this population. Methods A total of 100 patients who underwent prostate biopsy at our center between July 2024 and February 2026 with PSA levels in the 4–10 ng/mL range were retrospectively enrolled, including 18 pathologically confirmed PCa cases and 82 benign prostatic hyperplasia (BPH) cases. Five routine indicators were included: free-to-total PSA ratio (f/t PSA), prostate health index (PHI), p2PSA, PSA density (PSAD), and prostate volume. A soft voting ensemble model was constructed via Optuna hyperparameter tuning and class imbalance correction, and internal validation was performed using repeated 5-fold cross-validation (20 repetitions). Results The constructed ensemble model achieved an area under the receiver operating characteristic curve (AUC) of 0.904 (95% CI: 0.836–0.972), with diagnostic efficacy significantly superior to all single indicators (P < 0.05). The optimal cutoff value determined by maximizing Youden’s index was 0.1726, yielding a negative predictive value (NPV) of 98.46%, sensitivity of 94.44%, specificity of 78.05%, and overall accuracy of 81.00%. Calibration curve and decision curve analysis confirmed good predictive consistency and significantly higher clinical net benefit than conventional strategies. Conclusions The ensemble model based on five routine clinical indicators has good predictive performance for PCa in the PSA 4–10 ng/mL range among Chinese men. Its high NPV may help identify low-risk patients in whom biopsy could be deferred, potentially reducing unnecessary invasive procedures, although the results need confirmation in larger cohorts.
No takes yet. Share an insight, caveat, or question.
Hua et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: