A Primer on the Use of Frequentist Confidence Analysis in Neurology Clinical Trials

Clinical researchers are drawn toward Bayesian methods that facilitate probability statements quantifying confidence in particular treatment effects of interest. Although traditional frequentist statistics are not typically used to make such statements, frequentist confidence analysis is an alternative to Bayesian methods, which fully supports confidence statements about observed treatment effects. Despite frequentist confidence analysis dating back to the 1950s and being sometimes used in clinical studies, clinical researchers may not be aware of how to use frequentist confidence to design and analyze clinical trials. Therefore, this Research Methods in Neurology primer describes the application of frequentist confidence to address the practical topic of design and analysis of clinical trials. The primer expands on well-understood ideas around frequentist CIs to describe the notion of confidence distributions, which can be used to flexibly quantify the strength of evidence about any treatment effect of interest. Then, we describe how to use frequentist confidence in a specific treatment effect (e.g., benefit or lack of meaningful benefit) to monitor data in an adaptive trial setting. We illustrate how frequentist confidence-adaptive trials can achieve tight control of Type I error rate. Finally, we provide 2 R software packages ("confidenceCurves" and "confidenceSim"), with a clinician-friendly user interface ("confidenceCurvesApp"), to support frequentist confidence-based reasoning and confidence-based trial design for neurology research.