Edwards S, Slattery M L, Mori M, Berry T D, Caan B J, Palmer P, Potter J D
Department of Family and Preventive Medicine, University of Utah School of Medicine, Salt Lake City.
Am J Epidemiol. 1994 Dec 1;140(11):1020-8. doi: 10.1093/oxfordjournals.aje.a117192.
In this paper, the authors present an objective system to evaluate interviewer performance for use in epidemiologic studies. With this quality control system, all study interviews are audio-taped, and a random sample of interviews are coded according to interviewer behaviors, such as whether the interviewer asked the questions exactly as written and used the probes appropriately. With the use of data obtained from a large case-control study of colon cancer, the authors observed that 94.2% of all questions were asked in the same manner by all interviewers and that 89.5% of all probing behaviors were appropriate. They show that questions that required additional interviewer behaviors were more likely to result in variation in response that can be attributed to the interviewer. These findings have implications for study design and interpretation of study results. From simulations, the authors have estimated the impact of uncorrected interviewer variability on study power and ability to detect disease associations. Uncorrected interviewer variability could decrease study power from 84% to 56%. From simulations, the authors observed that odds ratios could be biased downward from 1.8 to 1.3. These findings illustrate the importance of using a continuous quality control program in epidemiologic research.
在本文中，作者提出了一个用于评估访谈者在流行病学研究中表现的客观系统。通过这个质量控制系统，所有的研究访谈都进行了录音，并且从访谈中随机抽取样本，根据访谈者的行为进行编码，比如访谈者是否完全按照书面形式提问以及是否恰当地使用了追问技巧。利用从一项关于结肠癌的大型病例对照研究中获得的数据，作者观察到所有访谈者以相同方式提出的问题占全部问题的94.2%，并且所有追问行为中恰当的占89.5%。他们表明，那些需要访谈者采取额外行为的问题更有可能导致因访谈者而异的回答差异。这些发现对研究设计和研究结果的解释具有重要意义。通过模拟，作者估计了未校正的访谈者变异性对研究效能以及检测疾病关联能力的影响。未校正的访谈者变异性可能会使研究效能从84%降至56%。通过模拟，作者观察到优势比可能会从1.8偏向下降至1.3。这些发现说明了在流行病学研究中使用持续质量控制程序的重要性。
