Design, Analyze, Communicate: Working together to improve clinical study outcomes.
The study design of a randomized trial must balance the need to collect information about the potential benefits and risks of interventions with the need to protect participants from potential harms. The amount of information needed depends on the effect of the intervention, which is unknown, and characteristics of the study population, which may not be precisely known. Study design calculations based on inaccurate information can lead to studies that are overpowered (i.e. collect more information than necessary), or underpowered (i.e. collect too little information), both of which have ethical implications.
To protect participants from unnecessary risk, analyses can be conducted at pre-specified points in the study. These interim analyses can be used to stop the study early if the study’s goal has been met or stop for futility if the data suggest the trial is unlikely to achieve its aims. Study designs that use pre-specified interim analyses and stopping rules are called group sequential designs.
Before a participant in a study is randomized, information is collected to assess their eligibility for the study. This information about baseline characteristics, or covariates, can include the presence and severity of conditions, risk factors for an outcome, or demographic characteristics. When these factors are associated with the outcome of interest, an analysis that uses these characteristics (a covariate-adjusted analysis) can provide more information than an analysis that does not (an unadjusted analysis). The amount of information gained from covariate adjustment cannot be known in advance, which complicates study-design planning.
A web book on covariate adjustment contains material on covariate adjustment, including worked examples with data and code, explanations, and key references to the literature.
When investigators and statisticians are planning out a future trial, PICARD (planning information-monitored, covariate adjusted randomized designs) allows them to visualize how a trial may play out under different assumptions about the population under study and the amount of information gained from baseline covariates. This can be used online through Shinyapps.io or downloaded from GitHub.
The `impart` (information monitoring for precision adaptive randomized trials) package for R is software that allows investigators to plan, monitor, and analyze randomized trials using covariate adjustment. First, it makes most covariate-adjusted analyses compatible with group sequential designs. Second, it can be used to conduct a precision-adaptive randomized trial, which uses the data to determine when enough information has been collected instead of relying on pre-trial calculations.
Web book on covariate adjustment
PICARD GitHub Repo
PICARD ShinnyApp
impart R-package
