An alternative method for assessing the fragility of survival analysis results: a proof-of-concept study based on the log-rank test

Misused P-values and an excessive focus on significance have prompted calls for added robustness metrics. The Fragility Index (FI), which quantifies how many event status changes are needed to reverse statistical significance, serves as a useful complement. Although FI has been applied in various settings such as dose-finding trials and meta-analyses, its use in survival analysis is limited due to complexities like censoring, variable follow-up, and hazard assumptions. Existing FI adaptations for survival data often reassign individuals across intervention arms in randomized controlled trials (RCTs), diverging from FI's original philosophy and reducing clinical plausibility. We propose a modified fragility index for survival data, termed FIS, to assess the robustness of survival analysis results in RCTs. Rather than reassigning individuals between intervention and control groups, FIS preserves the foundational principles of the original FI by quantifying the minimum number of changes in outcome status, either events or censoring, needed to overturn statistical significance. To enhance flexibility and practical utility, we extend FIS to assess fragility in both directions: from statistically significant to non-significant results and vice versa. We demonstrate the performance of the proposed method through two real-world cases from RCTs.