Campbell Peter T, Sloan Margaret, Kreiger Nancy
Research Unit, Division of Preventive Oncology, Cancer Care Ontario, and Department of Public Health Sciences, University of Toronto, Toronto, Canada.
Ann Epidemiol. 2007 Apr;17(4):253-7. doi: 10.1016/j.annepidem.2006.07.012. Epub 2006 Dec 18.
This work provides an empirical assessment of the usefulness of obtaining exposure data from proxy respondents.
Two independent case groups were formed in data from a population-based case-control study. One case set was derived from proxy respondents. The second case series was derived from respondents who self-reported. The second case group was matched to the proxy case group for age, cancer site, and sex. These data were compared with a control group for completeness of information and identifying heterogeneity of risk estimates for a variety of variables. Index cases and controls were matched to the 829 available proxy respondents for sex, age (5-year groups), and cancer site.
Proxy respondents provided levels of complete information similar to index cases and controls for height and weight; occupational physical activity; consumption of coffee, alcohol, and cigarettes; and family history of cancer. Proxies had more missing responses for variables concerning recreational physical activity, clinical depression, age at first menstruation, use of oral contraceptives, and occupational exposure to pesticides. Polytomous logistic regression models found only a few examples of meaningful heterogeneity among all variables, limited to models for coffee consumption and cigarette smoking.
Our data suggest that proxy respondents, especially parents and spouses, provide adequately complete information for many, but not all, exposures common in epidemiologic studies.
本研究对通过代理受访者获取暴露数据的实用性进行了实证评估。
在一项基于人群的病例对照研究的数据中形成了两个独立的病例组。一组病例来自代理受访者。第二组病例系列来自自我报告的受访者。第二组病例与代理病例组在年龄、癌症部位和性别上进行匹配。将这些数据与对照组进行比较，以评估信息的完整性，并确定各种变量风险估计的异质性。索引病例和对照在性别、年龄（5年组）和癌症部位上与829名可用的代理受访者进行匹配。
代理受访者提供的关于身高、体重、职业体力活动、咖啡、酒精和香烟消费以及癌症家族史的完整信息水平与索引病例和对照相似。代理受访者在娱乐性体力活动、临床抑郁症、初潮年龄、口服避孕药使用以及职业接触杀虫剂等变量方面的回答缺失较多。多分类逻辑回归模型在所有变量中仅发现了少数有意义的异质性实例，仅限于咖啡消费和吸烟模型。
我们的数据表明，代理受访者，尤其是父母和配偶，为许多但并非所有在流行病学研究中常见的暴露提供了足够完整的信息。
