Scholarlink

<p>Interrupted Time Series (ITS) designs are commonly used to estimate intervention effects in public health, epidemiology, clinical psychology, and many other fields. A common type of ITS design analyzed in time-series studies is the AB design for which there is one baseline phase and one intervention phase. Segmented linear regression is frequently used to estimate the causal effect of an intervention, allowing for the assessment of both immediate and delayed post-treatment effects. Many methods are available to estimate such models in R, but little guidance is available for researchers. We summarized ten available methods in R for estimating intervention effects in ITS and evaluated the performance of these methods to recover lag-1 autoregressive process (AR(1)) and change estimates at three time points post-intervention via a Monte Carlo Simulation study. The methods that had the least biased estimates of the lag-1 autoregressive effect with adequate Type I error rates and the highest statistical power were Restricted Maximum Likelihood (REML) performed in the package nlme and doubly-robust ordinary least squares performed in the package Dbfit. We therefore recommend researchers use one of these methods when estimating intervention effects in AB ITS design, especially in relatively short time series.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of statistical methods in R for estimating intervention effects using segmented linear regression in the AB interrupted time series design

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月23日

查看全文
  • Jinyong Pang,
  • Henian Chen,
  • Matthew J. Valente

摘要

Interrupted Time Series (ITS) designs are commonly used to estimate intervention effects in public health, epidemiology, clinical psychology, and many other fields. A common type of ITS design analyzed in time-series studies is the AB design for which there is one baseline phase and one intervention phase. Segmented linear regression is frequently used to estimate the causal effect of an intervention, allowing for the assessment of both immediate and delayed post-treatment effects. Many methods are available to estimate such models in R, but little guidance is available for researchers. We summarized ten available methods in R for estimating intervention effects in ITS and evaluated the performance of these methods to recover lag-1 autoregressive process (AR(1)) and change estimates at three time points post-intervention via a Monte Carlo Simulation study. The methods that had the least biased estimates of the lag-1 autoregressive effect with adequate Type I error rates and the highest statistical power were Restricted Maximum Likelihood (REML) performed in the package nlme and doubly-robust ordinary least squares performed in the package Dbfit. We therefore recommend researchers use one of these methods when estimating intervention effects in AB ITS design, especially in relatively short time series.