Abstract

Correlated categorical data often arise from studies involving cluster randomized trials, a cluster sampling scheme, or repeated measurements. The intra-cluster correlation coefficient (ICC) is used to estimate the average correlation within clusters. There have been numerous methods proposed to estimate ICC for correlated binary data, the ANOVA method for continuous data, and several methods for time-to-event outcomes. However, no method currently exists to estimate ICC for nominal or ordinal categorical responses with more than two categories. We developed a method based on resampling principles to estimate the ICC and its 95% confidence interval for categorical variables. We conducted a simulation study to show how our method estimates the population ICC under varying event rates, numbers of clusters, and cluster sizes. We also used real study datasets to estimate the ICC for ordinal and nominal categorical variables. We observed that the resampling method estimates the population ICC well for moderate to large numbers of clusters and moderate to large cluster sizes.

摘要

相关的分类数据通常来自涉及整群随机试验、整群抽样方案或重复测量的研究。簇内相关系数 (ICC) 用于估计簇内的平均相关性。已经提出了许多方法来估计相关二进制数据的 ICC、连续数据的 ANOVA 方法以及事件发生时间结果的几种方法。但是，目前不存在用于估计具有两个以上类别的名义或有序分类响应的 ICC 的方法。我们开发了一种基于重采样原则的方法来估计分类变量的 ICC 及其 95% 置信区间。我们进行了一项模拟研究，以展示我们的方法如何在不同的事件发生率、集群数量和集群大小下估计人口 ICC。我们还使用真实研究数据集来估计有序和名义分类变量的 ICC。我们观察到，重采样方法可以很好地估计中到大量集群和中到大型集群大小的总体 ICC。