Key statistical assumptions and methods in one-arm meta-analyses with binary endpoints and low event rates, including a real-life example in the area of endoscopic colonic stenting

Abstract

There are relatively few publications on the methodology of one-arm meta-analyses, especially when the outcome is binary and has low probability of occurring. We will discuss a few of the important assumptions underlying one-arm meta-analyses, including publication bias, fixed effect versus random-effects models, and raw event incidence rate transformations required when the event frequency is low. Finally, we will provide a real-life example taken from the endoscopic colonic stenting literature to illustrate the consequences of failure to thoroughly investigate these assumptions. In this example, we find arcsine transformation provides more appropriate results than logit transformation.

Keywords:

• one-arm meta-analysis
• low event rates
• publication bias
• random-effects model
• colonic stenting
• endoscopy

Public Interest Statement

Meta-analyses are incredibly useful tools for researchers that can be utilized to explore what the overall literature is showing about a particular topic. However, similar to using any statistical analysis method, there are assumptions that need to be understood prior to performing the analysis. In this paper, we look at the key statistical assumptions underlying one-arm meta-analyses with binary endpoints and low event rates. We also provide an example from the literature of the consequences of not understanding these assumptions.

Competing Interests

The authors declare no competing interest.

Additional information

Notes on contributors

Matthew J. Rousseau

Matthew Rousseau is a biostatistician who has been researching medical devices for the past 8 years. Primarily his work has been in the field of gastrointestinal endoscopy and he has extensive experience in clinical trial design and data analysis, along with performing meta-analyses for both study design and publication purposes. This work represents the authors’ hope to improve knowledge and understanding of assumptions underlying statistical analyses that appear in the literature, especially among non-statisticians.