The statistical significance of randomized controlled trial results is frequently fragile: a case for a Fragility Index

This report has been verified by one or more authors of the original publication.

この研究はハイインパクトの可能性があります。 OEのAI駆動型ハイインパクト指標は、論文が掲載されたジャーナルと論文自体の科学的内容の両方からのシグナルを統合することで、論文が持つ可能性の高い影響力を推定します。 最先端の自然言語処理を用いて開発されたOEハイインパクトモデルは、ジャーナルのインパクトファクターのみよりも、研究の将来の引用実績をより正確に予測します。 これにより、臨床的に意義のある研究をより早く認識することが可能になり、読者は将来の診療を形成する可能性が最も高い論文に集中することができます。

While P-values in clinical studies are an easy method to denote statistical significance, the interpretation of significant findings can be somewhat skewed. The Fragility Index is a newly developed metric that complements p-values and confidence intervals. It is calculated by adding consecutive events to a significant dichotomous finding until the result is no longer significant. The number of events needed to be added thus represents the fragility of the finding. Analysis of 399 trials published within high-impact journals found a median Fragility Index of 9, with 25% of studies at a Fragility Index of 3 or lower. Linear regression indicated that findings were less fragile when the reported P-values were lower (ie. P<0.001), the number of events reported was high, or when a large number of participants were included in the study. Application of the Fragility Index should hopefully aid readers in their determination of the impact of significant findings within the literature.