Intracluster Correlation Coefficients in Healthcare Research

Presenter's Name(s)

Brennan E. ParadeeFollow

Conference Year

January 2020

Abstract

In a cluster randomization trial, which are often used in healthcare research, randomization occurs at the group level not at the individual level. It is often more practical to randomize an entire practice to a treatment rather than each individual patient. Clustering is very common in behavioral health research because interventions are often meant to be taken on by an entire practice. However, when randomization occurs at the practice level and outcomes are measured on the patient level, it can lead to some problems in the statistical analysis. Most standard statistical methods assume independence between individuals, which is true for a typical randomized experiment. However, there are many factors that cause individuals who visit the same practice to not be independent. This lack of independence must be accounted for in the analysis using intracluster correlation coefficients (ICC). ICCs are important to use for calculating the sample size in a study and can impact the statistical power of the clinical trial if not used correctly. This presentation will show what ICCs are, how to calculate them, and how they were used in the IBHPC study at UVM.

Primary Faculty Mentor Name

Abigail Crocker

Faculty/Staff Collaborators

Abigail Crocker

Status

Undergraduate

Student College

College of Engineering and Mathematical Sciences

Second Student College

Graduate College

Program/Major

Statistics

Second Program/Major

Biostatistics

Primary Research Category

Health Sciences

Abstract only.

Share

COinS
 

Intracluster Correlation Coefficients in Healthcare Research

In a cluster randomization trial, which are often used in healthcare research, randomization occurs at the group level not at the individual level. It is often more practical to randomize an entire practice to a treatment rather than each individual patient. Clustering is very common in behavioral health research because interventions are often meant to be taken on by an entire practice. However, when randomization occurs at the practice level and outcomes are measured on the patient level, it can lead to some problems in the statistical analysis. Most standard statistical methods assume independence between individuals, which is true for a typical randomized experiment. However, there are many factors that cause individuals who visit the same practice to not be independent. This lack of independence must be accounted for in the analysis using intracluster correlation coefficients (ICC). ICCs are important to use for calculating the sample size in a study and can impact the statistical power of the clinical trial if not used correctly. This presentation will show what ICCs are, how to calculate them, and how they were used in the IBHPC study at UVM.