基于多参数建立前列腺癌列线图预测模型及验证的研究

Establishment and validation of a multi-parameter model for predicting prostate cancer

关键词:  前列腺癌  前列腺特异性抗原  穿刺  模型  列线图

Key words:  prostate cancer  prostate-specific antigen  biopsy  model  nomogram

摘要:

目的 基于单中心数据探讨前列腺癌相关预测因素，建立并验证前列腺癌列线图预测模型。方法 回顾性收集2014年1月至2020年1月邯郸市中心医院行前列腺穿刺活检患者的临床资料，包括年龄（Age）、总PSA （tPSA）、游离PSA （fPSA）和前列腺体积（PV）等。资料完整者纳入研究，共697例，中位年龄71岁（40~95岁），中位tPSA 13.6 ng/mL （0.2~100 ng/mL）。随机选取495例（70%）为建模组，余202例（30%）为验证组。在建模组中利用单因素和多因素logistic回归分析，构建多参数列线图预测模型，利用ROC曲线下面积（AUC）评估该模型对前列腺癌的诊断价值，与tPSA、% fPSA和PSAD相比较，并进行内部人群验证。结果 697例中非前列腺癌组504例，前列腺癌组193例。两组患者的Age、tPSA、fPSA、PV、% fPSA、PSAD、直肠指检（DRE）结节、TRUS低回声和体质指数（BMI）差异有统计学意义（P<0.05）。单因素和多因素logistic回归分析显示建模组的年龄（OR=1.043）、tPSA （OR=1.025）、fPSA （OR=1.198）、PV （OR=0.971）、DRE结节（OR=3.195）、TRUS低回声（OR=4.288）及BMI （OR=1.703）是预测前列腺癌的独立预测变量（P<0.05），据此建立列线图预测模型。建模组模型最佳临界值为0.36时ROC曲线下面积（AUC）为0.855，显著高于tPSA、% fPSA、PSAD；验证组模型AUC为0.810，显著高于国内模型。结论 本研究建立的前列腺癌列线图预测模型对前列腺癌具有较高的预测价值，预测概率>0.36时，建议行前列腺穿刺活检。

Abstract:

Objective To discuss the possible predicting factors related to prostate cancer and establish a validated nomogram that predicts prostate cancer based on a single-center date set. Methods Clinical data of patients who underwent prostate biopsy in Handan Central Hospital from January 2014 to January 2020 were retrospectively analyzed, including age, tPSA, fPSA and PV etc. A total of 697 cases were included with complete set of data. The median age was 71 (40-95), the median tPSA was 13.6 ng/mL(0.2-100 ng/mL). Out of all cases, 495 (70%) were randomly selected as the development group, and the rest 202 (30%) as the validation group. In the development group, univariate and multivariate logistic regression analysis were performed to establish a multivariate parameter nomogram predictive model for prostate cancer. The area under the ROC curve (AUC) was used to evaluate the diagnostic value of the model and compared with tPSA, %fPSA, and PSAD. The results were eventually verified by internal population. Results Out of the 697 randomized cases, 504 were diagnosed with non-prostate cancer, and 193 prostate cancer. The differences of age, tPSA,fPSA,PV,%fPSA, PSAD, DRE,TRUS and BMI between the two groups were statistical significant (P<0.05). Univariate and multiple logistic regression analysis demonstrated that patients' age (OR=1.043), tPSA (OR=1.025), fPSA (OR=1.198), PV (OR=0.971), DRE (OR=3.195), TRUS (OR=4.288) and BMI (OR=1.703) were independent predictive factors for prostate cancer (P<0.05); and a nomogram model was developed based on these factors. In the development group, AUC of the model was 0.855 when a cut-off value was 0.36, which was significant higher than those of tPSA, %fPSA, and PSAD (P<0.05). In the validation group, AUC of the model was 0.810, significantly higher than those of domestic models (P<0.05). Conclusions The prostate cancer nomogram prediction model established in this study has a high predictive value for prostate cancer. When the cancer prediction probability of a patient is above 0.36, prostate biopsy is recommended.