Institute of Statistics, National Chiao Tung University, Hsinchu, Taiwan.
Stat Med. 2021 Apr;40(9):2155-2176. doi: 10.1002/sim.8895. Epub 2021 Feb 3.
The Bland-Altman method, which assesses agreement via an assessment set constructed by the difference of the measurement variables, has received great attention. Other assessment approaches have been proposed following the same difference-based framework. However, the exact assessment set constructed by the difference is achievable only for measurements with certain joint distributions. To provide a more general assessment framework, we propose two approaches. First, when the measurement distribution is known, we propose a parametric approach that constructs the assessment set through a measure of closeness corresponding to the distribution. Second, when the measurement distribution is unknown, we propose a nonparametric approach that constructs the assessment set through quantile regression. Both approaches quantify the degree of agreement with the presence of both systematic and random measurement errors, and enable one to go beyond the difference-based approach. Results of simulation and data analyses are presented to compare the two approaches.
Bland-Altman 方法通过评估变量差异构建的评估集来评估一致性，受到了广泛关注。其他评估方法也提出了遵循相同的基于差异的框架。然而，只有在具有某些联合分布的测量中才能实现通过差异构建的精确评估集。为了提供更通用的评估框架，我们提出了两种方法。首先，当测量分布已知时，我们提出了一种参数方法，通过对应于分布的接近度度量来构建评估集。其次，当测量分布未知时，我们提出了一种非参数方法，通过分位数回归来构建评估集。这两种方法都量化了存在系统和随机测量误差时的一致性程度，并使人们能够超越基于差异的方法。我们提出了两种方法。首先，当测量分布已知时，我们提出了一种参数方法，通过对应于分布的接近度度量来构建评估集。其次，当测量分布未知时，我们提出了一种非参数方法，通过分位数回归来构建评估集。这两种方法都量化了存在系统和随机测量误差时的一致性程度，并使人们能够超越基于差异的方法。结果的模拟和数据分析被提出来比较这两种方法。
