四种预测前列腺穿刺阳性风险模型的准确性的外部验证与比较

An external validation and comparison of the predictive accuracy of four models designed to predict the probability of a positive prostate biopsy

目的验证与比较4种用于个体化预测经直肠超声引导下初次前列腺穿刺阳性风险模型的准确性。 方法回顾性分析2010年1月至2014年9月431例行经直肠超声引导下前列腺穿刺患者的临床资料。分别采用北美前列腺癌预防试验的前列腺癌风险计算（prostate cancer prevention trial derived cancer risk calculator，PCPT－CRC）模型、蒙特利尔模型以及文献报道的两个国内模型（国内模型1和国内模型2）进行个体化前列腺穿刺阳性风险计算，通过受试者工作特征（receiver operating characteristic，ROC）曲线下面积（area under the curve，AUC）评估各模型的预测准确性，并采用Z检验比较各模型AUC的差异。 结果本研究431例的穿刺病理结果中，123例（28.5%）为前列腺癌，308例（71.5%）为良性病变。前列腺癌组与非前列腺癌组的游离前列腺特异性抗原百分比（percentage of free prostate－specific antigen，%fPSA）比较差异无统计学意义（P＝0.242），年龄、PSA、直肠指检、前列腺体积以及超声结果比较差异均有统计学意义（P〈0.05）。PCPT－CRC模型、蒙特利尔模型、国内模型1、国内模型2和PSA的AUC分别为0.774（95% CI 0.726～0.822）、0.765 （95% CI 0.714～0.816）、0.813 （95% CI 0.767～0.858）、0.795（95% CI 0.749～0.842）和0.736 （95% CI 0.684～0.788），各模型间AUC比较差异无统计学意义（P〉0.05）。与PSA比较，当PSA范围无限制时，国内模型1的预测准确性提高了7.7%（P〈0.05）。当PSA为4～10 ng/ml时，4种模型和PSA的AUC分别是0.688（95% CI 0.560～0.816）、0.818 （95% CI 0.719～0.918）、0.830 （95% CI 0.740～0.919）、0.853（95% CI 0.771～0.935）和0.565（95% CI 0.419～0.710），国内模型2预测准确性最高，较PSA提高了28.8%（P〈0.05）。 结论4种模型均具有较高的预测准确性，国内两种模型与PCPT－CRC模型和蒙特利尔模型的预测准确性无差异，但与PSA比较，当PSA范围无限制时，国内模型1更有优势；当PSA为4～10 ng/ml时，国内模型2预测准确性最高。

Objective To validate and compare the predictive accuracy of four prostate cancer models designed to predict the likelihood of a positive initial transrectal biopsy. Methods Clinical data of 813 consecutive patients between January 2010 and September 2014 who had undergone a transreetal ultrasound （TRUS） guided prostate biopsy at our institution were reviewed ,431 patients fulfilling all criteria for four predictive models were enrolled for the final analysis. The risk of each individual positive biopsy was calculated using either of the four models. The predictive accuracy of each model was measured using area under the receiver operating characteristic curve （ AUC）, and the comparison of AUCs was performed by Z test. Results Of 431 participants, the statistical analysis of age, prostate-specific antigen （PSA） , digital rectal examination （ DRE）, prostate volume and TRUS findings were all significantly different （ P 〈 0. 05 ） , except percentage of free prostate-specific antigen （%v fPSA） （P = 0. 242 ） . AUCs were 0. 774 （95% CI 0.726-0.822）, 0.765 （95% CI0.714-0.816）, 0.813 （95% CI0.767-0.858）, 0.795 （95% CI0.749- 0. 842） and 0. 736 （95% CI O. 684-0. 788 ） for the North-American prostate cancer prevention trial derived cancer risk calculator （PCPT-CRC） model, Montreal model, domestic model 1, domestic model 2 and PSAalone,respectively. There was no significant difference among AUCs of the four models, and a 7.7% increased predictive accuracy was observed for the domestic model 1 compared to unlimited PSA alone（ P 〈 0. 05）. When serum PSA ranging from 4 to 10 ng/ml,AUCs were 0. 688（95% C10. 560-0. 816） ,0. 818 （95% CIO. 719-0. 918） ,0. 830 （95% CIO. 740-0. 919）, 0. 853（95% CIO. 771-0. 935） and 0. 565（95% C10. 419-0. 710） for the four models and PSA alone, respectively. Domestic model 2 owned the highest predictive accuracy and a 28.8% increased predictive accuracy was observed for the domestic model 2 compared to PSA alone （ P 〈 0. 05 ）. Conclusions External validation and comparison of the four models reveals that all of the four models have acceptable predictive accuracy in our cohort. There is no difference of predictive accuracy between foreign and domestic models according to the AUC resuhs. However, domestic model 1 is superior to unlimited PSA alone, and domestic model 2 has the highest predictive accuracy when serum PSA ranging from 4 to 10 ng/ml.