文献速递流行病学与生物统计临床试验与 GCP

基于部分线性变系数模型与变化平面的纵向数据亚组分析

Subgroup analysis for longitudinal data based on a partial linear varying coefficient model with a change plane.
PubMed 文献 · Stat Med · 2023-06-14

纵向研究中亚组分析仍有限,本文提出部分线性变系数模型与变化平面,亚组由分组变量线性组合定义,估计不同亚组时变效应以捕捉预测变量与反应的动态关联;变系数用基函数近似、组指示函数用核函数平滑并纳入广义估计方程,建立估计量的渐近性质,模拟展示灵活性、效率与稳健性,基于标准与新型抗癫痫药物研究识别出特定时期对新药敏感的患者亚组。

要点

  • 提出部分线性变系数模型与变化平面用于纵向数据亚组分析,亚组由分组变量线性组合定义。
  • 变系数用基函数近似、组指示函数用核函数平滑,纳入广义估计方程估计。
  • 基于标准与新型抗癫痫药物研究识别出特定时期对新药敏感的患者亚组。

核心结论

  • 所提方法能灵活、高效、稳健地进行纵向数据亚组分析,并在标准与新型抗癫痫药物研究中识别出特定时期对新药敏感的患者亚组。

研究设计与样本

  • 方法学研究,提出部分线性变系数模型与变化平面,亚组由分组变量线性组合定义,变系数用基函数近似、组指示函数用核函数平滑并纳入广义估计方程,建立估计量渐近性质并进行模拟。
  • 基于标准与新型抗癫痫药物研究进行应用分析。

主要发现

  • 模拟研究展示了所提方法的灵活性、效率与稳健性。
  • 在标准与新型抗癫痫药物研究中成功识别出特定时期对新药敏感的患者亚组。

局限性

  • 原文未报告局限性

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

Subgroup analysis has become an important tool to characterize the treatment effect heterogeneity, and finally towards precision medicine. On the other hand, longitudinal study is widespread in many fields, but subgroup analysis for this data type is still limited. In this article, we study a partial linear varying coefficient model with a change plane, in which the subgroups are defined based on linear combination of grouping variables, and the time-varying effects in different subgroups are estimated to capture the dynamic association between predictors and response. The varying coefficients are approximated by basis functions and the group indicator function is smoothed by kernel function, which are included in the generalized estimating equation for estimation. Asymptotic properties of the estimators for the varying coefficients, the constant coefficients and the change plane coefficients are established. Simulations are conducted to demonstrate the flexibility, efficiency and robustness of the proposed method. Based on the Standard and New Antiepileptic Drugs study, we successfully identify a subgroup in which patients are sensitive to the newer drug in a specific period of time.

来源:Stat Med,查看原文。版权归原作者所有。

引用

  1. Wei K, Qin G, Zhu Z, et al. Subgroup analysis for longitudinal data based on a partial linear varying coefficient model with a change plane. Stat Med. 2023. PMID: 37314008. doi:10.1002/sim.9827

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。