The Fragility Index: a<i>P</i>-value in sheep’s clothing?

Carter, Rickey E.; McKie, Paul M.; Storlie, Curtis B.

doi:10.1093/eurheartj/ehw495

Cited by 74 publications

(97 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We find that the “ca” (# of major vessels colored by flourosopy) and “thal” (defect type) coefficients are the least fragile as heart disease predictors, while “thalach” (maximum heart rate achieved), “trestbps” (resting blood pressure), and “exang” (presence of exercised induced angina) are the most fragile. As previously indicated (6, 12), we do find a positive correlation between degree of significance and fragility (i.e. higher significance generally leads to more stable fragility scores) and, as such, both metrics should be considered when interpreting outcome associations.…”

Section: Resultssupporting

confidence: 80%

fragilityindex: An R Package for Statistical Fragility Estimates in Biomedicine

Johnson

Rappaport

Shameer

et al. 2019

Preprint

View full text Add to dashboard Cite

Poor reproducibility is a growing crisis in biomedical research. The fragility index was introduced as a convenient measure to estimate how fragile statistical results in clinical trials are to small perturbations in event outcome counts. There is currently no freely available R package to produce this calculation. Furthermore, the original definition of the method is applicable only to 2x2 contingency tables.As such, we developed an R package to calculate fragility index. We have also extended the concept of a statistical fragility index to two of the most commonly used methods in clinical research, survival analysis via weighted log-rank tests and logistic regression, and implemented these technique sin this R package. We describe example applications of these methods to existing publically available datasets. This R package is freely available under the AGPL license on CRAN (https://cran.r-project.org/web/packages/fragilityindex/index.html). The most recent versions may be downloaded and installed via Github (https://github.com/kippjohnson/fragilityindex).

show abstract

Section: Resultssupporting

confidence: 80%

fragilityindex: An R Package for Statistical Fragility Estimates in Biomedicine

Johnson

Rappaport

Shameer

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…While a low Fragility Index indicates that the P value hinges on only few events, it is important to remember that the Fragility Index, and indeed P values themselves, are not measures of clinical effect. Carter et al reported in a simulated data set how the Fragility Index repackages the P value to correlate inversely with it. When the Fragility Index is low, it provides value as an intuitive metric to contextualise a study's results with respect to discontinuation or lost‐to‐follow‐up rates.…”

Section: Discussionmentioning

confidence: 99%

The fragility of statistically significant findings from randomised controlled trials in the urological literature

et al. 2018

View full text Add to dashboard Cite

The results of urology RCTs that study dichotomous outcomes and report statistically significant differences between groups are sometimes fragile and depend on few events. Urologists should interpret these RCTs cautiously, particularly when the number of participants lost to follow-up exceeds the Fragility Index. Routine reporting of Fragility Index values alongside P values may provide additional guidance about the robustness of statistically significant findings.

show abstract

“…In our review, we found that reported P values were moderately correlated with FI scores, an association that has been repeatedly observed in the FI literature. 5,8,10,51 Carter et al recently demonstrated that the P value and FI score are strongly associated with one another through a series of in silico (simulation) studies. 51 They cautioned readers not to interpret the FI score as a measure of effect, a concept that is widely, but incorrectly, used for P values.…”

Section: Discussionmentioning

confidence: 99%