基于第二版PI-RADS建立的预测模型对前列腺穿刺活检结果的预测价值

Predictive value of prostate biopsy results based on predictive model established by the PI-RADS version 2

目的探讨基于第二版前列腺影像报告和数据系统(PI-RADS v2)联合前列腺特异性抗原(PSA)及其相关参数建立的列线图模型对前列腺穿刺活检结果的预测能力.方法回顾性分析2014年1月至2018年12月厦门大学附属第一医院509例行B超引导下经直肠前列腺穿刺活检术患者的临床资料.年龄(68.1±7.2)岁.前列腺体积(PV)(55.8 ±30.7)ml.tPSA(19.86±18.94)ng/ml,fPSA (2.63±3.60) ng/ml,f/tPSA (0.14±0.08),PSAD(0.46±0.52) ng/ml^2.根据PI-RADS v2对前列腺主要病灶进行评分:1分37例,2分131例,3分152例,4分102例,5分87例.509例中选取407例(80%)为建模组,余102例(20%)为验证组.对建模组进行单因素及多因素logistic回归分析,筛选能预测前列腺癌的独立影响因素,建立前列腺穿刺活检结果的列线图模型.利用受试者工作特征(ROC)曲线、校准曲线及决策曲线对建模组、验证组及tPSA 4.1 ~ 20.0 ng/ml患者进行模型的验证与评估,并与tPSA、fPSA、f/tPSA、PSAD及PI-RADS v2的预测能力进行比较.结果本研究509例的前列腺癌检出率为43.0%(219/509).logistic回归分析结果显示,建模组的年龄(OR =1.113)、f/tPSA(OR=0.004)、PV(OR=0.986)、PSAD(OR=11.023)、直肠指检质地(OR=2.295)、经腹超声检查有无低回声(OR=2.089)及PI-RADS v2(OR=1.920)是预测前列腺癌的独立影响因素(P<0.05),利用上述指标建立列线图模型.建模组中,列线图预测前列腺癌的ROC曲线下面积(AUC)为0.883,高于tPSA (0.686)、fPSA(0.593)、f/tPSA(0.626)、PSAD(0.777)及PI-RADS v2(0.761);验证组中,列线图预测前列腺癌的ROC曲线下面积为0.839,高于tPSA(0.758)、fPSA(0.666)、f/tPSA(0.648)、PSAD (0.832)及PI-RADS v2(0.803);tPSA 4.1 ~ 20.0 ng/ml患者中,列线图预测前列腺癌的ROC曲线下面积为0.801,高于tPSA(0.570)、fPSA(0.426)、f/tPSA(0.657)、PSAD(0.707)及PI-RADS v2(0.701).列线图模型的校准曲线显示预测曲线与标准曲线基本拟合,霍斯默-莱梅肖拟合优度检验结果为(χ^2=5.434,P=0.710),均提示列线图模型有较好的校准能力;决策曲线显示基于PI-RADS v2建立的列线图模型有较高的临床应用价值.结论基于PI-RADS v2建立的列线图模型对前列腺癌有较高的预测价值,能显著提高前列腺癌的诊断效能,较单独应用PSA及其相关参数具有更高的诊断价值.

Objective To explore a predictive nomogram for the result of prostate biopsy based on Prostate Imaging Reporting and Data System version 2(PI-RADS v2)combined with prostate specific antigen (PSA) and its related parameters, and to assess its ability to diagnose prostate cancer by internal validation. Methods We retrospectively analyzed the clinical data of 509 patients who underwent transrectal prostate biopsy guided by ultrasound during the period from January 2014 to December 2018 in the Department of Urology, First Affiliated Hospital of Xiamen University. In 509 cases, the mean age was (68.1±7.2) years. The mean prostate volume(PV) was (55.8±30.7) ml. The mean tPSA value was (19.86±18.94) ng/ml. The mean value of fPSA was (2.63±3.60) ng/ml and the mean f/tPSA was 0.14±0.08. The mean PSAD was (0.46±0.52) ng/ml^2. Based on the PI-RADS v2, score 1 point have 37 cases, score 2 point have 131 cases, score 3 point have 152 cases, score 4 point have 102 cases, score 5 point have 87 cases. Of these patients, we randomly selected 80%(407 cases) as development group, and the other 20%(102 cases) as validation group. Univariate and multivariate logistic regression analysis of the development group was performed to identify the independent influence factors that can predict prostate cancer (PCa), thereby establishing a predictive model for the result of prostate biopsy. In the development group, validation group and tPSA was between 4.1-20.0 ng/ml, the model was evaluated by analyzing the receiver operating characteristic (ROC) curve, calibration curve and decision curve, and compared to PSA, fPSA, f/tPSA, PSAD, PI-RADS v2. Results Among the 509 patients enrolled in the study, the detection rate of PCa was 43.0%(219/509). In the development group, the logistic regression analysis demonstrated that patient age (OR=1.113), f/tPSA (OR=0.004), PV (OR=0.986), PSAD (OR=11.023), digital rectal examination (DRE) texture (OR=2.295), transabdominal ultrasound (TAUS) with or without hypoechoic (OR=2.089), and PI-RADS v2 (OR=1.920) were independent factors for PCa (P<0.05). The nomogram based on all variables was established. In the development group, the area under the curve (AUC) of the model (0.883) was greater than those of tPSA (0.686), fPSA (0.593), f/tPSA (0.626), PSAD (0.777), PI-RADS v2 (0.761). In the validation group, the area under the curve of the model (0.839) was greater than those of tPSA (0.758), fPSA (0.666), f/tPSA (0.648), PSAD (0.832), PI-RADS v2 (0.803). In patients whose tPSA was between 4.1-20.0 ng/ml, the area under the curve of the model (0.801) was greater than those of tPSA (0.570), fPSA (0.426), f/tPSA (0.657), PSAD (0.707), PI-RADS v2 (0.701). The calibration curve of the nomogram indicated that the prediction curve was basically fitted to the standard curve, and the Hosmer-Lemeshow showed that χ^2=5.434, P=0.710, both suggested that the prediction model had better calibration ability. The decision curve showed that the model based on PI-RADS v2 had high clinical application value. Conclusions The nomogram based on PI-RADS v2 had a high predictive value for prostate cancer and could significantly improve the diagnostic performance. It had better diagnostic value than PSA and its related parameters. It also provided important guidance for the prostate cancer on clinical treatment of patients to some extent.