Statistical fragility of significant outcomes in spine surgery randomized controlled trials: a systematic review of binary and continuous outcomes.

Abstract

Purpose: Spinal disorders are a major cause of disability worldwide, and the increasing use of spine surgery has been accompanied by a growing body of randomized controlled trials (RCTs). However, statistical significance based on P-values alone may not adequately capture the robustness of trial findings. The Fragility Index (FI) is a complementary measure of evidence stability, quantifying how easily statistical significance can be reversed. Previous spine-specific fragility analyses have demonstrated that statistically significant dichotomous outcomes are frequently fragile, but continuous outcomes have not been comprehensively evaluated. Accordingly, a systematic review and fragility analysis was performed to evaluate the robustness of statistically significant outcomes in spine surgery RCTs across both dichotomous and continuous outcomes.

Methods: Embase, MEDLINE, Web of Science, and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched from inception to January 14, 2026 for RCTs in spine surgery reporting significant primary outcomes. Studies were screened and extracted in duplicate, and the risk of bias was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool; fragility was evaluated using the FI for binary outcomes and the Continuous Fragility Index (CFI) for continuous outcomes, where lower values indicate that statistical significance would be lost with fewer outcome changes, with results summarized descriptively and across predefined subgroup analyses.

Results: A total of 265 RCTs comprising 24,083 patients were included; 43 (16.2%) reported binary outcomes and 222 (83.8%) reported continuous outcomes. Overall median fragility was 10.5 (interquartile range (IQR) 5.0 to 18.7), with median values of 4.0 for binary outcomes (IQR 2.0 to 5.0) and 12.2 for continuous outcomes (IQR 7.4 to 21.0). Notably, 76.7% of binary outcomes had an FI ≤ 5, and fragility did not differ significantly by risk-of-bias category. Fragility also differed significantly by reported P-value category, with more strongly significant outcomes generally showing greater robustness.

Conclusion: In recent spine surgery RCTs with statistically significant primary outcomes, findings were often statistically fragile. For binary outcomes, reversal of statistical significance required a change in outcome status for a median of only 4 patients, with more than three-quarters of outcomes having an FI ≤ 5. Continuous outcomes also showed measurable fragility, with reversal requiring reallocation of a median of 12.2 patients. These results suggest that a substantial proportion of statistically significant findings in contemporary spine surgery RCTs may be less robust than conventional P-values alone imply. Fragility analysis may therefore serve as a useful complement to conventional significance testing when interpreting statistically significant findings in spine surgery trials, particularly for clinically important binary endpoints.