More data, less information? Potential for nonmonotonic information growth using GEE.
More data, less information? Potential for nonmonotonic information growth using GEE.
复制标题
更多数据，更少信息？
DOI：
10.1080/10543406.2016.1167071
复制
发表时间：
2017
影响因子：
1.1
通讯作者：
Emerson,ScottS
中科院分区：
文献类型：
--
作者：
Shoben,AbigailB;Rudser,KyleD;Emerson,ScottS
统计学直觉表明，增加可用于分析的观测总数应该会增加参数估计的精度。统计信息的这种单调增长在连续分析数据时特别重要，例如在确证性临床试验中。然而，单调的信息增长并不总是得到保证，即使使用有效的，但效率低下的估计。在这篇文章中，我们证明了非单调信息增长的理论可能性时，使用广义估计方程（GEE）估计斜率，并提供直观的为什么这种可能性存在。我们使用理论和基于模拟的结果来描述可能导致非单调信息增长的情况。非单调信息增长最有可能发生在以下情况：（1）相对于每个个体的随访，累积速度较快;（2）同一个体的测量值之间的相关性较高;（3）测量值在随机化后变得更加多变。在可能导致非单调信息增长的情况下，研究设计者应计划中期分析，以避免最有可能导致非单调信息增长的情况。
Statistical intuition suggests that increasing the total number of observations available for analysis should increase the precision with which parameters can be estimated. Such monotonic growth of statistical information is of particular importance when data are analyzed sequentially, such as in confirmatory clinical trials. However, monotonic information growth is not always guaranteed, even when using a valid, but inefficient estimator. In this article, we demonstrate the theoretical possibility of nonmonotonic information growth when using generalized estimating equations (GEE) to estimate a slope and provide intuition for why this possibility exists. We use theoretical and simulation-based results to characterize situations that may result in nonmonotonic information growth. Nonmonotonic information growth is most likely to occur when (1) accrual is fast relative to follow-up on each individual, (2) correlation among measurements from the same individual is high, and (3) measurements are becoming more variable further from randomization. In situations that may lead to nonmonotonic information growth, study designers should plan interim analyses to avoid situations most likely to result in nonmonotonic information growth.