Survival-Inferred Fragility Index in Phase III Colorectal Cancer Trials: A Systematic Review

Conventional measures such as p-values and hazard ratios (HRs) do not directly quantify the sensitivity of time-to-event outcomes to small changes in individual outcomes. The Survival-Inferred Fragility Index (SIFI) evaluates the minimum number of survival-outcome changes required to reverse trial conclusions and provides complementary information on result robustness.

This systematic review (PROSPERO: CRD420251182325) identified phase III colorectal cancer randomized controlled trials (RCTs) published from 2015 to 2025 via PubMed, Embase, and Cochrane. SIFI was calculated using Bomze's framework with the "flip best" strategy. The fragility quotient (FQ) and the relationship between SIFI and withdrawals or losses to follow-up (WCLFU) were assessed. Analyses used R version 4.5.1.

SIFI was calculated for 50 phase III colorectal cancer RCTs, including 9 three-arm trials, resulting in 59 pairwise comparisons. Among positive results, median SIFI values for colorectal, colon, and rectal cancer were 6.0 (interquartile range [IQR]: 2.0 to 11.0), 19.0 (IQR: 15.5 to 22.5), and 4.0 (IQR: 4.0 to 4.0). Higher SIFI was observed in trials with HR < 0.75 than in those with HR 0.75-0.94 (median SIFI: 9.0 vs. 1.0; p = 0.035) and in trials with power > 85% than in those with power ≤ 85% (median SIFI: 8.5 vs. 2.0; p = 0.028). For negative results, median SIFI values were -4.5 (IQR: -7.2 to -3.0), -30.5 (IQR: -43.5 to -18.0), and -10.5 (IQR: -11.0 to -5.5). Greater absolute SIFI values were associated with adjuvant therapy, sample size ≥ 500, disease-free survival as the primary endpoint, and HR ≥ 0.95 (all p ≤ 0.005). Overall, 44.07% of pairwise comparisons had an absolute SIFI < 5.

SIFI shows some phase III colorectal cancer RCT conclusions were sensitive to small outcome changes. It complements conventional statistics by illuminating result robustness.