Westerdahl J, Anderson H, Olsson H, Ingvar C
Department of Surgery, University Hospital, Lund, Sweden.
Int J Epidemiol. 1996 Apr;25(2):245-51. doi: 10.1093/ije/25.2.245.
The reproducibility of a self-administered questionnaire on different epidemiological variables was examined in a random sample of a cohort of healthy women from the South Swedish Health Care Region. An identical questionnaire has previously been used for assessment of melanoma risk, in a population-based, matched, case-control study from the same region.
Repeat questionnaires were completed on two occasions, 1-3 years apart, by 670 randomly selected subjects. In addition, exposure data from a case-control study were used to estimate the effect of misclassification.
Overall, there was a fair to good consistency between the answers on the two occasions for the various epidemiological variables. Exposure variables identified as risk factors for melanoma development were reasonably reproducible. The estimated proportion of agreement, A, ranged from 0.74 to 0.92, the average correct classification rate (assuming common sensitivity and specificity), pi, was > or = 0.85 and the kappa coefficient, kappa, ranged between 0.52 and 0.83. However, the question on number of raised naevi on the arm was an exception since it had a lower test-retest reliability (A = 0.62, pi = 0.77, kappa = 0.40). When using data from the case-control study, the effect of the estimated random misclassification was found to only slightly bias odds ratios toward one.
Our questionnaire, used for assessment for melanoma risk, provided information with fair to good test-retest reliability, and corrected odds ratios were found to be only slightly higher than observed odds ratios.
在瑞典南部医疗保健地区的一组健康女性队列的随机样本中，对一份关于不同流行病学变量的自填问卷的可重复性进行了研究。此前，在同一地区的一项基于人群的匹配病例对照研究中，使用了相同的问卷来评估黑色素瘤风险。
670名随机选择的受试者在两次间隔1至3年的时间里完成了重复问卷。此外，还使用了病例对照研究中的暴露数据来估计错误分类的影响。
总体而言，两次回答中各种流行病学变量之间的一致性良好。被确定为黑色素瘤发生风险因素的暴露变量具有合理的可重复性。估计的一致性比例A在0.74至0.92之间，平均正确分类率（假设具有共同的敏感性和特异性）π≥0.85，kappa系数kappa在0.52至0.83之间。然而，关于手臂上凸起痣数量的问题是个例外，因为它的重测信度较低（A = 0.62，π = 0.77，kappa = 0.40）。当使用病例对照研究的数据时，发现估计的随机错误分类的影响只会使优势比略微偏向一方。
我们用于评估黑色素瘤风险的问卷提供了具有良好至优秀重测信度的信息，并且校正后的优势比仅略高于观察到的优势比。
