Applying the fragility index to randomized controlled trials evaluating total neoadjuvant therapy for rectal cancer.

Abstract

32 Background: A relatively novel summary measure that is not commonly reported in randomized controlled trials (RCTs) is the fragility index (FI). FI describes the number of additional events required for an outcome to lose statistical significance. It has been applied to a number of medical subspecialities such as critical care, orthopedic surgery and colorectal surgery. Recently, there has been significant interest in, and adoption of, total neoadjuvant therapy (TNT) for locally advanced rectal cancer (LARC). A number of RCTs have assessed TNT, but the robustness of these practice changing RCTs has never been evaluated. As such, we designed the present study to assess the robustness of the RCTs evaluating TNT for LARC using the FI. Methods: Relevant articles were identified through a recently published review article by Johnson et al. in the Canadian Journal of Surgery, that narratively reviewed all of the previously published RCTs evaluating TNT for LARC. We manually searched Google Scholar and PubMed to identify any other relevant RCTs. Outcomes within these RCTs that were either dichotomous outcomes or time to event outcomes were eligible for inclusion if the reported effect size had an associated p-value of less than 0.05. The main outcome was the FI for each statistically significant outcome. Walsh et al.’s method of calculating FI was utilized. A RCTs results were considered fragile if the FI was less than the loss to follow up for a given outcome. Correlations between FI and research characteristics were assessed using the Spearman’s rank correlation coefficients. Results: Ten RCTs were identified with 25 outcomes having statistically significant differences between groups (p-values < 0.05). Eleven outcomes were time-to-event outcomes, while the remainder were dichotomous outcomes. About half (n=13) were oncologic outcomes (i.e., survival, recurrence), while the rest (n=12) were short- and long-term complications. The median FI was 2 (interquartile range [IQR] 1-16). The number of patients lost to follow-up exceeded the FI in 17 outcomes (68.0%) and thus these results were considered “fragile”. Lower FI was associated with high risk of bias (rho=-0.5594) and higher loss to follow-up (i.e., greater than 5% vs. less than 5%) (rho=-0.4394), while higher FI was associated with large sample size (i.e., greater than 500 patients vs. less than 500 patients) (rho=0.5120). Conclusions: The robustness of outcomes from trials assessing TNT for LARC was found to be questionable. Most of these outcomes were fragile, as determined by the FI. In most cases, two or less additional events would have resulted in a loss of statistical significance of the reported results. Those using the results of these studies, including clinicians and health policy experts, should apply caution when interpreting these types of trials.