摘要目的:建立预测多参数MRI(mpMRI)前列腺影像报告和数据系统(PI-RADS)评分≥3分且前列腺特异性抗原(PSA) 4～20 ng/ml患者确诊为有临床意义前列腺癌(CsPCa)的列线图，并验证其预测价值。方法:回顾性分析2020年1月至2023年8月于天津医科大学第二医院首次接受超声引导下经会阴前列腺穿刺活检的865例患者的临床病理资料，患者的mpMRI PI-RADS评分≥3分且PSA 4～20 ng/ml。将865例作为队列A，年龄68(64，73)岁；f/tPSA 14.36(10.63，19.74)；PSAD 0.17(0.11，0.25)ng/ml 2；前列腺体积(PV)≤50 ml 375例(43.35%)，>50 ml 490例(56.65%)；PSA波动率[(PSA 首次－PSA 穿刺前最后一次)/PSA 首次×100]<－50% 84例(9.71%)，－50%～－20% 206例(23.82%)，>－20% 575例(66.47%)；PI-RADS评分3分546例(63.12%)，4分230例(23.59%)，5分89例(10.29%)；mpMRI上可疑病灶位于外周带619例(71.56%)，移行带181例(20.92%)，其他42例(4.86%)，外周带＋移行带23例(2.66%)。将865例中病例资料包含前列腺健康指数(PHI)的437例作为队列B，PSAD 0.17(0.12，0.25)ng/ml 2；D-二聚体310.00(230.00，411.48)ng/ml；PHI 49.75(35.90，73.27)；PV≤50 ml 198例(45.31%)，>50 ml 239例(54.69%)；PSA波动率<－50% 40例(9.15%)，－50%～－20% 107例(24.49%)，>－20% 290例(66.39%)；PI-RADS评分3分289例(66.13%)，4分103例(23.57%)，5分45例(10.30%)。将队列A、B患者分别使用R语言以"123"为随机数种子按7∶3比例随机分配到训练集和验证集，2组的训练集与验证集临床资料比较差异均无统计学意义( P>0.05)。基于队列A、B采用单因素和多因素后退法logistic回归筛选确诊CsPCa的独立危险因素，使用R软件分别构建列线图模型A、B。采用受试者工作特征(ROC)曲线、校准曲线和临床决策曲线分析(DCA)评估两个列线图模型的诊断性能。在验证集中对列线图模型进行外部验证。分析列线图模型A、B，以及不同阈值PSAD、PHI确诊CsPCa的敏感性、特异性、阳性预测值、阴性预测值、准确率、漏诊率。 结果:队列A训练集608例，验证集257例。多因素回归分析结果显示年龄( OR＝1.06， P＝0.001)、f/tPSA( OR＝0.96， P＝0.008)、PV>50 ml( OR＝0.36， P<0.01)、PSAD( OR＝145.19， P<0.01)、PSA波动率(-50%～－20%： OR＝1.97， P＝0.234；>－20%： OR＝6.81， P<0.01)、PI-RADS评分(4分： OR＝10.65， P<0.01；5分： OR＝21.20， P<0.01)、mp-MRI上可疑病灶定位(移行带： OR＝0.57， P＝0.074；其他： OR＝0.26， P＝0.022)为确诊CsPCa的危险因素。基于上述危险因素构建列线图模型A，其在训练集的ROC曲线下面积(AUC)为0.905(95% CI 0.881～0.928)，验证集的AUC为0.893(95% CI 0.854～0.931)。队列B训练集308例，验证集129例，多因素分析结果显示年龄( OR＝1.05， P＝0.053)、PV >50 ml( OR＝0.18， P<0.01)、PSAD( OR＝54.14， P＝0.021)、PSA波动率(－50%～－20%： OR＝4.78, P＝0.100；>－20%： OR＝20.37， P＝0.001)、PHI( OR＝1.02， P＝0.002)、D-二聚体( OR＝1.00， P＝0.031)、PI-RADS评分(4分： OR＝11.35， P<0.01；5分： OR＝57.61， P<0.01)为确诊CsPCa的危险因素。基于上述危险因素构建列线图模型B，训练集的ROC曲线AUC为0.933(95% CI 0.906～0.959)，验证集AUC为0.908(95% CI 0.859～0.958)。校准曲线显示2个列线图模型的校准度良好。DCA曲线显示，2个列线图模型在训练集和验证集均有较好的临床净获益。模型A的约登指数最大为0.657时，截断值为34%，特异性为81.50%，阴性预测值为89.64%，诊断CsPCa的准确率为82.51%；模型B约登指数最大为0.709时，截断值为41%，特异性为86.90%，阴性预测值为90.13%，诊断CsPCa的准确率为85.82%。 结论:本研究基于mpMRI及相关临床指标构建的2种列线图模型均对PI-RADS≥3分且PSA 4～20 ng/ml的患者前列腺穿刺活检诊断CsPCa有较好的预测价值。

abstractsObjective:Based on multi-parametric prostate magnetic resonance imaging (mpMRI) and related clinical indicators, a nomogram model for patients with PI-RADS ≥3 and PSA 4-20ng/ml was developed and validated, and the predictive value of the model in diagnosing clinically significant prostate cancer was evaluated.Methods:The clinical and pathological data of 865 patients who underwent ultrasound-guided transperineal prostate biopsy for the first time at the Department of Urology, Second Hospital of Tianjin Medical University from January 2020 to August 2023, with PI-RADS scores ≥3 and PSA levels between 4-20 ng/ml were retrospectively analyzed. These 865 patients were included in Cohort A, and from them, 437 patients with PHI were selected in Cohort B. In Cohort A, the median age was 68(64, 73); the median f/tPSA was 14.36 (10.63, 19.74); the median PSAD was 0.17(0.11, 0.25); 375 cases (43.35%) with PV≤50 ml and 490 cases (56.65%) with PV>50 ml; PSA fluctuation <-50% 84 cases (9.71%), -50%--20% in 206 cases (23.82%), and >-20% in 575 cases (66.47%); PI-RADS v2.1 3 scores 546 cases (63.12%), 4 in 230 cases (23.59%), and 5 in 89 cases (10.29%); localization of suspicious lesions on mpMRI in the peripheral zone in 619 cases (71.56%), transitional zone in 181 cases (20.92%), others in 42 cases (4.86%), and both peripheral and transitional zones in 23 cases (2.66%). In Cohort B, the median PSAD was 0.17 (0.12, 0.25); the median D-dimer was 310.00 (230.00, 411.48); the median PHI was 49.75 (35.90, 73.27); with 198 cases (45.31%) with PV≤50 ml and 239 cases (54.69%) with PV>50 ml; PSA fluctuation<-50% was in 40 cases (9.15%), -50%--20% in 107 cases (24.49%), and>-20% in 290 cases (66.39%); PI-RADS v2.1 scores 3 was in 289 cases (66.13%), 4 in 103 cases (23.57%), and 5 in 45 cases (10.30%).Patients in cohorts A and B were randomly assigned to the training set and validation set using R language with " 123" as the random number seed, at a ratio of 7∶3.There was no statistically significant difference between the clinical data of the training and validation sets for both groups ( P>0.05).Univariate and multivariate logistic regression analyses were used to identify independent risk factors for CsPCa, and a nomogram model was constructed using R. The diagnostic performance of the prediction model was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis(DCA).External validation of the model was conducted in the validation set. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and missed diagnosis rate analyses were performed on nomogram models A and B, as well as PSAD and PHI, under different thresholds. Results:Cohort A training set has 608 cases, and the validation set has 257 cases.The results of multivariate backward regression analysis in the training set show that age( OR=1.06, P=0.001), f/tPSA( OR=0.96, P=0.008), prostate volume (PV)>50ml( OR=0.36, P<0.01), prostate-specific antigen density(PSAD)( OR=145.19, P<0.01), PSA fluctuation(-50%--20%: OR=1.97, P=0.234; >-20%: OR=6.81, P<0.01), PI-RADS v2.1 score(4: OR=10.65, P<0.01; 5: OR=21.20, P<0.01), and localization of suspicious lesions on mpMRI(TZ: OR=0.57, P=0.074; Others: OR=0.26, P=0.022) were all risk factors for CsPCa. Nomogram A was developed based on these risk factors and had an area under the ROC curve (AUC) of 0.905 (95% CI 0.881-0.928) for the training set and 0.893 (95% CI 0.854-0.931) for the validation set. Cohort B training set developed based on age( OR=1.05, P=0.053), PV>50ml( OR=0.18, P<0.01), PSAD( OR=54.14, P=0.021), PSA fluctuation(-50%--20%: OR=4.78, P=0.100; >-20%: OR=20.37, P=0.001), PHI( OR=1.02, P=0.002), D-Dimer( OR=1.00, P=0.031), and PI-RADS scores(4: OR=11.35, P<0.01; 5: OR=57.61, P<0.01) as risk factors for CsPCa. Nomogram B had an AUC of 0.933(95% CI 0.906-0.959) for the training set and 0.908 (95% CI 0.859-0.958) for the validation set.The two nomogram models mentioned above both have excellent discrimination, and the calibration curves also indicated that the calibration of the two models were good.Moreover, both nomogram A and nomogram B demonstrate good clinical net benefits in the DCA curves of the training and validation sets, especially when applying nomogram B to predict CsPCa, with an accuracy rate of up to 85.82%. Conclusions:The two nomogram models developed in study, based on mpMRI and related clinical indicators, both have excellent predictive value for the diagnosis of clinically significant prostate cancer prior to prostate biopsy in patients with PI-RADS≥3 and PSA 4-20ng/ml.