Rämsch C, Pfahlberg A B, Gefeller O
Department of Medical Informatics, Biometry and Epidemiology, University of Erlangen-Nuremberg, Germany.
Comput Methods Programs Biomed. 2009 Apr;94(1):88-95. doi: 10.1016/j.cmpb.2008.10.006. Epub 2008 Dec 4.
The attributable risk (AR) is an epidemiologic measure quantifying the relationship between an exposure factor and a disease at the population level. In addition to its original use as a one-dimensional parameter the AR is increasingly applied in multifactorial epidemiologic situations when the combined impact of multiple exposure factors has to be partitioned into factor-specific components. We discuss the point and interval estimation of the resulting multidimensional parameter termed partial attributable risk (PAR) and introduce the R-package 'pARccs', a comprehensive software enabling the application of the methods. 'pARccs' allows for point and interval estimation of PAR from case-control data utilizing the non-parametric bootstrap with stratified resampling in combination with the percentile or BC(a) method to compute confidence intervals. We illustrate the concept of partial attributable risks and the application of the software by an example from a recent case-control study on risk factors for melanoma. We also discuss practical aspects of the software application for epidemiologic purposes and its limitations.
归因风险（AR）是一种流行病学指标，用于量化人群层面暴露因素与疾病之间的关系。除了最初作为一维参数使用外，当必须将多个暴露因素的综合影响分解为特定因素的组成部分时，AR在多因素流行病学情况中的应用越来越多。我们讨论了所得多维参数（称为部分归因风险（PAR））的点估计和区间估计，并介绍了R包“pARccs”，这是一个全面的软件，可实现这些方法的应用。“pARccs”允许利用非参数自助法结合分层重采样以及百分位数或BC(a)方法从病例对照数据中对PAR进行点估计和区间估计，以计算置信区间。我们通过最近一项关于黑色素瘤危险因素的病例对照研究中的一个例子来说明部分归因风险的概念以及该软件的应用。我们还讨论了该软件在流行病学应用中的实际问题及其局限性。
