Biostatistics

Methods: Mind the Gap Webinar July 14: Overview of Statistical Models for the Design and Analysis of Stepped Wedge Cluster Randomized Trials

Speaker: 

Fan Li, PhD
Yale University School of Public Health

Description:

The stepped-wedge cluster randomized design has received increasing attention in pragmatic clinical trials (PCTs) and implementation science research. Since the design’s introduction, a variety of mixed-effects model extensions have been proposed for the design and analysis of PCTs. In this talk, Dr. Fan Li of Yale University will provide a general model representation and regard various model extensions as alternative ways to characterize secular trends, intervention effects, and sources of heterogeneity. He will also review key model ingredients and clarify their implications for the design and analysis of stepped-wedge trials.

Registration required: 

https://www.prevention.nih.gov/education-training/methods-mind-gap/overview-statistical-models-design-and-analysis-stepped-wedge-cluster-randomized-trials

October 1, 2018: Dr. Greg Simon Uses a Pie Eating Contest Analogy to Explain the Intraclass Correlation Coefficient

Posted on by Karen Staman

In a new video, Dr. Greg Simon explains the intraclass correlation coefficient (ICC) with an analogy to a pie eating contest. The ICC is a descriptive statistic that measures the correlations among members of a group, and it is an important tool for cluster-randomized pragmatic trials because this calculation helps determine the sample size needed to detect an effect.

Greg Simon from NIH Collaboratory on Vimeo.

“When we randomize treatments by doctors, clinics, or even whole health systems, we need to think about how things cluster, and the intraclass correlation coefficient is the measure of that clustering. When we think about sample sizes in pragmatic clinical trials, it’s important to understand what an intraclass correlation coefficient actually is.”

For most pragmatic trials, the ICC will be between 0 and 1. If the outcomes in a group are completely correlated (ICC=1), then all participants within the group are likely to have the same outcome. When ICC=1, sampling one participant from the cluster is as informative as sampling the whole cluster, and many clusters will be needed to detect an effect. If there is no correlation among members of the groups (ICC=0), then the available sample size for the study is essentially the number of participants.

For more on the ICC, see the Intraclass Correlation section in the Living Textbook or this working document from the Collaboratory’s Biostatistics and Study Design Core.

December 7, 2017: Dr. Greg Simon Explains Individual, Cluster, and Stepped-Wedge Randomization in a New Prop Video

Posted on by Karen Staman

In a new video in the Living Textbook, Dr. Greg Simon describes the differences between individual, cluster, and stepped-wedge randomization using props, including marbles, Play-Doh, and glassware.

“In the end, it’s all about randomly assigning who gets which treatment, or who gets which treatment when, so that we’re able to make some un-biased judgement about which treatment is really better.” —Greg Simon, MD