The weakness of fragility index exposed in an analysis of the traumatic brain injury management guidelines: A meta-epidemiological and simulation study.

Abstract

OBJECTIVES:To perform fragility index (FI) analysis on the evidence that forms the basis of the guidelines for the management of severe traumatic brain injury (TBI), and develop a deeper understanding of the pitfalls associated with FI. DESIGN:Meta-epidemiological analysis and numerical simulations. METHODS:The Brain Trauma Foundation guidelines (4th edition) for management of severe TBI were used to identify relevant randomised controlled trials (RCTs). FI based on Fisher's exact test and relative risk was performed on eligible RCTs. The relationship between FI, event counts and P values was explored by exhaustively considering different combinations of outcomes for studies of total size ranging from 80 to 10000. Sample size calculations were also performed for a range of power, baseline risk and relative risk, to determine the influence of study design on FI. RESULTS:FI analysis of the severe TBI management guidelines revealed that most studies were associated with a low FI. In the majority of studies, FI was of a similar magnitude to the number lost to follow-up. The simulations revealed that while FI was inversely related to P value, a wide range of FI may be associated with a given P value. FI is also affected by sample size, baseline risk and effect size. Sample size calculations suggest that aside from very high-powered studies, most are likely to yield low FI values in the range typically encountered in the literature. CONCLUSIONS:Many studies are underpowered and are expected to be associated with a small FI. Furthermore, FI over-simplifies the complex, non-linear relationships between sample size, effect size and P value, which hinder comparisons of FI between studies. FI places undue importance on the "significance" of P values and accordingly should only be used sparingly.