Abstract：
This letter argues how a previously published case-control (Horwitz et al.) study of estrogens and endometrial cancer illustrates how bias can arise in case-control studies when the proportion of interviews completed is related both to exposure and to disease. In the Horwitz study, 2 series were reported: 1 obtained from interview and 1 obtained from medical data. The odds ratios obtained from each of these series were similar, but a combined odds ratio produced very different results. Horwitz attributed this to a confounding factor not taken into account, but the letter writer argues that bias was introduced because of the incompleteness of the interview series where both exposure and disease were related. For example, in the conventional series, the interview rate in cases was higher for the exposed (91%) than the unexposed (61%), whereas in controls the reverse was true; in the alternative series, effect of modification was also present but to a much smaller degree, so the difference in odds ratio was correspondingly smaller. In sum, merely taking a subject as exposed if so reported in either data source should be avoided to prevent bias.
摘要：
这封信论证了以前发表的病例对照(Horwitz等人。）对雌激素和子宫内膜癌的研究说明，当完成的访谈比例与暴露和疾病有关时，在病例对照研究中会产生偏见。在Horwitz的研究中,报告了2个系列：1个来自访谈，1个来自医疗数据。从这些系列中获得的赔率比相似，但是综合比值比产生了非常不同的结果。霍维茨把这归因于一个没有考虑的混杂因素，但这封信的作者认为，偏见的引入是由于采访系列的不完整，其中暴露和疾病是相关的。例如,在传统系列中，接触者(91%)高于未接触者(61%)，而在对照组中，情况正好相反；在替代系列中，修饰的效果也存在，但程度要小得多，所以赔率比的差异相应较小。总之,如果在任一数据源中报告，则应避免仅将受试者视为暴露对象，以防止偏见。
