Fragility of Results in Ophthalmology Randomized Controlled Trials: A Systematic Review

Abstract

Evidence-based medicine is guided by our interpretation of randomized controlled trials (RCTs) that address important clinical questions. Evaluation of the robustness of statistically significant outcomes adds a crucial element to the global assessment of trial findings. The purpose of this systematic review was to determine the robustness of ophthalmology RCTs through application of the Fragility Index (FI), a novel metric of the robustness of statistically significant outcomes. Systematic review. A literature search (MEDLINE) was performed for all RCTs published in top ophthalmology journals and ophthalmology-related RCTs published in high-impact journals in the past 10 years. Two reviewers independently screened 1811 identified articles for inclusion if they (1) were a human ophthalmology-related trial, (2) had a 1:1 prospective study design, and (3) reported a statistically significant dichotomous outcome in the abstract. All relevant data, including outcome, P value, number of patients in each group, number of events in each group, number of patients lost to follow-up, and trial characteristics, were extracted. The FI of each RCT was calculated and multivariate regression applied to determine predictive factors. The 156 trials had a median sample size of 91.5 (range, 13-2593) patients/eyes, and a median of 28 (range, 4-2217) events. The median FI of the included trials was 2 (range, 0-48), meaning that if 2 non-events were switched to events in the treatment group, the result would lose its statistical significance. A quarter of all trials had an FI of 1 or less, and 75% of trials had an FI of 6 or less. The FI was less than the number of missing data points in 52.6% of trials. Predictive factors for FI by multivariate regression included smaller P value (P<0.001), larger sample size (P= 0.001), larger number of events (P= 0.011), and journal impact factor (P=0.029). In ophthalmology trials, statistically significant dichotomous results are often fragile, meaning that a difference of only a couple of events can change the statistical significance. An application of the FI in RCTs may aid in the interpretation of results and assessment of quality of evidence.