目的 开发1种预测模型用于选择适合行最大雄激素阻断（maximum androgen blocking，MAB）治疗的高危局限性前列腺癌（high-risk localized prostate cancer，HRLPC）患者。 方法 本研究收集了2021年1月1日至2024年6月30日，在重庆医科大学附属第一医院泌尿外科经穿刺活检病理诊断并接受根治性前列腺切除术的96例HRLPC患者，所有患者在术前均接受3个月的MAB治疗，术后随访6个月。收集患者的基线人口学特征、实验室数据、影像学表现、围手术期临床资料及随访资料等。首先探究MAB治疗后前列腺特异性抗原密度（prostate-specific antigen density，PSAD）分层与病理和生化获益的关系，以确定最佳PSAD组别；随后采用机器学习筛选重要变量，构建预测模型，并通过ROC曲线、校准曲线及临床适用性评估其表现。 结果 低PSAD组［PSAD<0.17 ng/（mL∙cm³）］在围手术期结果、病理降级、术后检测不到前列腺特异抗原（prostate specific antigen，PSA）率、尿失禁恢复率及PSA随访情况均明显优于对照组（P<0.05）。用于模型构建的变量包括前列腺体积、低密度脂蛋白胆固醇、PSAD、吸烟史、总胆固醇、PSA、体质量指数，该预测模型的效能较好（AUC=0.769），具备一定的临床适用性。 结论 低PSAD组患者具有更好的病理和生化获益。本研究提供了1个可靠的预测模型，以辅助HRLPC患者的个体化治疗。

Objective To develop a predictive model for selecting patients with high-risk localized prostate cancer（HRLPC） who are suitable for preoperative maximum androgen blocking（MAB） therapy. Methods This study was conducted among 96 patients with HRLPC who were diagnosed based on transrectal biopsy and underwent radical prostatectomy in Department of Urology，The First Affiliated Hospital of Chongqing Medical University，from January 1，2021 to June 30，2024，and all patients received MAB therapy for 3 months before surgery and were followed up for 6 months after surgery. Related data were collected from all patients，including baseline demographic features，laboratory data，imaging findings，perioperative data，and follow-up information. At first，the association between prostate-specific antigen density（PSAD） stratification after MAB therapy and pathological and biochemical benefits was analyzed to determine the optimal PSAD group，then machine learning was used to identify important variables and establish a predictive model，and finally，the model was evaluated using the ROC curve，the calibration curve，and clinical applicability assessment. Results The low PSAD group ［PSAD <0.17 ng/（mL∙cm³）］ showed the best results of perioperative outcomes，pathological downgrading，the rate of undetectable PSA after -surgery，recovery from urinary incontinence，and PSA follow-up（P<0.05）. The model was established based on prostate volume，low-density lipoprotein cholesterol，PSAD，smoking history，total cholesterol，PSA，and body mass index，and this predictive model had good performance （with an area under the ROC curve of 0.769） and showed a certain degree of clinical applicability. Conclusion Patients in the low PSAD group tend to have better pathological and biochemical benefits. This study provides a reliable predictive model to assist in the individualized treatment of patients with HRLPC.