前列腺癌风险预测模型的构建与验证

Construction and validation of a prostatic cancer risk prediction model

摘要:

目的 筛选前列腺癌（prostatic cancer，PCa）发病的影响因素，构建PCa风险预测模型并进行验证。方法 利用国家临床医学科学数据中心《前列腺肿瘤预警数据集》，对数据处理后将数据按7∶3随机分为建模组和验证组；使用最小绝对值收敛和选择算子（least absolute shrink-age and selection operator，LASSO）对建模组进行筛选，得到PCa特征指标；对特征指标进行多因素logistic回归分析，并利用其分析结果对建模组数据构建PCa风险预测模型，同时利用建模组数据进行内部评价及验证组数据内部验证。结果 共纳入880例样本数据，其中建模组616例，验证组264例；通过LASSO回归分析对筛选得到的14个特征指标进行多因素logistic回归分析，结果显示球蛋白（OR=1.112，95%CI=1.044~1.185）、无机磷（OR=65.167，95%CI=20.437~207.796）、总前列腺特异性抗原（total prostate specific antigen，tPSA）（OR=1.026，95%CI=1.014~1.037）与血清尿酸（OR=0.997，95%CI=0.994~0.999）的差异具有统计学意义（P<0.05），是PCa发病的独立影响因素；利用其构建的PCa风险预测模型内部评价和内部验证的校准曲线准确度较高；模型内部评价的曲线下面积（area under curve，AUC）为0.766（95%CI=0.728~0.804），患者的决策曲线分析（decision curve analysis，DCA）净获益率为9%~72%；而模型内部验证的AUC为0.704（95%CI=0.639~0.768），患者的DCA净获益率为18%~59%及63%~64%。结论 球蛋白、无机磷、tPSA与血清尿酸是PCa的独立影响因素，通过其构建的风险预测模型具有良好预测作用。

Abstract:

Objective To screen the influencing factors of the incidence of prostatic cancer（PCa），build a risk prediction model for PCa and validate it.Methods Based on the Prostatic Cancer Early Warning Data Set of National Clinical Medical Science Data Center，and after the data processing， the data were randomly divided into a modeling group and a verification group according to 7∶3. Least absolute shrink-age and selection operator（LASSO） regression was used to screen the PCa characteristic indicators of the modeling group， multifactor logistic regression analysis was carried out to analyze the characteristic indicators，and the analysis results were used to build a PCa risk prediction model for the data of the modeling group. At the same time，the data of the modeling group were used for internal evaluation and the data of the validation group for internal verification.Results A total of 880 sample data were included，including 616 in the modeling group and 264 in the validation group. The 14 characteristic indexes screened by LASSO regression analysis were used for multivariate logistic regression analysis. The results showed that only globulin（OR=1.112，95%CI=1.044-1.185），inorganic phosphorus（OR=65.167，95%CI=20.437-207.796），total prostate specific antigen（tPSA）（OR=1.026，95%CI=1.014-1.037） and serum uric acid（OR=0.997，95%CI=0.994-0.999） had significant differences（P<0.05），and they were independent influencing factors for PCa. The calibration curve of the internal evaluation and internal verification of the PCa risk prediction model had high accuracy. The area under curve（AUC） of internal evaluation of the model was 0.766（95%CI=0.728-0.804），and the net benefit rate of decision curve analysis（DCA） of patients was 9%-72%；the AUC of internal validation of the model was 0.704（95%CI=0.639-0.768），and the net benefit rate of DCA of patients was 18%-59% and 63%-64%.Conclusion Globulin，inorganic phosphorus，tPSA and serum uric acid are independent influencing factors of PCa. The risk prediction model constructed by them has a good prediction effect.