The reverse fragility index (RFI) is a novel metric to appraise the results of studies reporting statistically non-significant results. The purpose of this study was to determine the statistical robustness of randomized controlled trials (RCTs) reporting non-significant differences in anterior cruciate ligament reconstruction (ACLR) graft failure rates, defined as re-rupture/revision ACLR rate, between hamstring tendon (HT) and bone-patellar tendon-bone (BTB) autografts by calculating RFIs. A systematic review was performed to identify RCTs that compared HT to BTB grafts for ACLR through January 2022. Studies reporting non-significant differences in graft re-rupture and revision ACLR rate (n.s.) were included. The RFI, defined as the fewest number of event reversals needed to change the non-significant graft re-rupture/revision outcome to statistically significant (P < 0.05), was recorded for each study. In addition, the number of studies in which the loss to follow-up exceeded the RFI was recorded. Among the 16 included RCTs, the median (interquartile range [IQR]) sample size was 71 (64-114), and the median (IQR) total number of graft re-rupture/revision ACLR events was 4 (4-6). The median (IQR) graft re-rupture/revision ACLR rate was 4.3% (3.0-6.4) overall, 4.1% (2.6-6.7) in the BTB group, and 5.4% (3.0-6.3) in the HT group. The median (IQR) RFI was 3 (3-4), signifying that a reversal of the outcome in 3 patients in one arm was needed to flip the studies' result from non-significant to statistically significant (P < 0.05). The median (IQR) number of participants lost to follow-up was 11 (3-13), and 13 (81.3%) of the included studies had a loss to follow-up greater than the studies' RFI. The results of RCTs reporting statistically non-significant re-rupture/revision ACLR rates between HT and BTB autografts would become significant if the outcome were reversed in a small number of patients-a number that was less than the loss to follow-up in the majority of studies. Thus, the neutrality of these studies is fragile, and a true statistically significant difference in re-rupture/revision rates may have been undetected. Level I.
Read full abstract