Tuo J Y, Bi J H, Li Z Y, Shen Q M, Tan Y T, Li H L, Yuan H Y, Xiang Y B
School of Public Health, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China State Key Laboratory of Oncogene and Related Genes, Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200032, China Department of Epidemiology, Shanghai Cancer Institute, Shanghai 200032, China.
State Key Laboratory of Oncogene and Related Genes, Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200032, China Department of Epidemiology, Shanghai Cancer Institute, Shanghai 200032, China.
Zhonghua Liu Xing Bing Xue Za Zhi. 2022 Mar 10;43(3):392-396. doi: 10.3760/cma.j.cn112338-20210812-00638.
To systematically introduce the design of case-cohort study and the statistical methods of relative risk estimation and their application in the design. First, we introduced the basic principles of case-cohort study design. Secondly, Prentice's method, Self-Prentice method and Barlow method were described in the weighted Cox proportional hazard regression models in detail, finally, the data from the Shanghai Women's Health Study were used as an example to analyze the association between obesity and liver cancer incidence in the full cohort and case-cohort sample, and the results of parameters from each method were compared. Significant association was observed between obesity and risk for liver cancer incidence in women in both the full cohort and the case-cohort sample. In the Cox proportional hazard regression model, the partial regression coefficients of the full cohort and the case-cohort sample fluctuated with the adjustment of confounding factors, but the hazard ratio estimates of them were close. There was a difference in the standard error of the partial regression coefficient between the full cohort and the case-cohort sample. The standard error of the partial regression coefficient of the case-cohort sample was larger than that of the full cohort, resulting in a wider 95% confidence interval of the relative risk. In the weighted Cox proportional hazard regression model, the standard error of the partial regression coefficient of Prentice's method was closer to the parameter estimates from full cohort than Self-Prentice method and Barlow method, and the 95% confidence interval of hazard ratio was closer to that of the full cohort. Case-cohort design could yield parameter results closer to the full cohort by collecting and analyzing data from sub-cohort members and patients with the disease, and reduce sample size and improve research efficiency. The results suggested that Prentice's method would be preferred in case-cohort design.
系统介绍病例队列研究的设计、相对风险估计的统计方法及其在设计中的应用。首先，介绍病例队列研究设计的基本原理。其次，在加权Cox比例风险回归模型中详细描述了普伦蒂斯方法、自我普伦蒂斯方法和巴洛方法，最后，以上海女性健康研究的数据为例，分析全队列和病例队列样本中肥胖与肝癌发病率之间的关联，并比较各方法的参数结果。在全队列和病例队列样本中均观察到肥胖与女性肝癌发病风险之间存在显著关联。在Cox比例风险回归模型中，全队列和病例队列样本的偏回归系数随混杂因素的调整而波动，但它们的风险比估计值相近。全队列和病例队列样本的偏回归系数标准误存在差异。病例队列样本的偏回归系数标准误大于全队列，导致相对风险的95%置信区间更宽。在加权Cox比例风险回归模型中，普伦蒂斯方法的偏回归系数标准误比自我普伦蒂斯方法和巴洛方法更接近全队列的参数估计值，风险比的95%置信区间也更接近全队列。病例队列设计通过收集和分析亚队列成员及患病患者的数据，可得出更接近全队列的参数结果，减少样本量并提高研究效率。结果表明，在病例队列设计中普伦蒂斯方法更可取。
