Hamajima N, Hirose K, Inoue M, Takezaki T, Kuroishi T, Tajima K
Division of Epidemiology, Aichi Cancer Center Research Institute, Nagoya, Japan.
J Clin Epidemiol. 1994 Sep;47(9):971-5. doi: 10.1016/0895-4356(94)90111-2.
In order to realize the variation in the estimate of the odds ratio from case-control studies, results from individually matched sampling were compared with those from the analysis based on a large number of controls. The subjects were selected from those who visited Aichi Cancer Center Hospital from 1988 to 1990. Cases consisted of 251 male lung cancer patients aged 40-79 years. Age and year of visit matched controls were sampled independently 100 times and 5000 times from non-cancer male outpatients (cases to controls ratio: 1:1, 1:2, 1:3, and 1:4). As unmatched controls, all male non-cancer outpatients aged 40-79 years (4100 patients) were used. The smoking habit was adopted as an exposure variable. As logically expected, analysis based on 4100 male controls gave a steadier estimate than the matched analyses examined here, indicating that a matched sampling is not recommended when a large number of controls are available.
为了了解病例对照研究中优势比估计值的变化情况，将个体匹配抽样的结果与基于大量对照的分析结果进行了比较。研究对象选自1988年至1990年期间就诊于爱知癌症中心医院的患者。病例组由251名年龄在40 - 79岁的男性肺癌患者组成。年龄和就诊年份匹配的对照组分别从非癌症男性门诊患者中独立抽样100次和5000次（病例与对照的比例为1:1、1:2、1:3和1:4）。作为非匹配对照，使用了所有年龄在40 - 79岁的男性非癌症门诊患者（4100例）。吸烟习惯被用作暴露变量。正如逻辑预期的那样，基于4100名男性对照的分析比此处所检验的匹配分析给出了更稳定的估计值，这表明当有大量对照可用时，不建议采用匹配抽样。
