#1033

What can we learn from non significant primary endpoints? A reanalysis of non-significant primary endpoints in Urological RCT's

D. Noll^1,2, P. Stapleton^1,2, M. O'Callaghan^1,3,4

¹The University of Adelaide, School of Medicine, Adelaide, Australia
²NALHN, Urology Department, Adelaide, Australia
³Flinders University, Health and Medical Research Institute, Adelaide, Australia
⁴Flinders Medical Centre, Urology Department, Adelaide, Australia

Introduction:

Many randomised controlled trials yield non-significant results; such results are challenging to interpret using common statistical practices. An alternative approach, the likelihood ratio (LR), uses Bayesian statistical methods to compares the null and alternative hypotheses and provides quantitative strength of evidence for one compared to the other. A LR >1 supports the hypothesis of no effect, <1 supports the alternate hypothesis, the more extreme the value the greater the level of support. By quantifying the strength of support for one hypothesis over another, researchers and readers can determine the need, or lack thereof, for further research in an area with a non-significant result. We calculated a LR for non-significant primary endpoints in published randomised controlled trials (RCTs) to highlight its utility and provide further value for these otherwise difficult to interpret results.

Material and methods:

We performed a cross-sectional study of RCTs studying urological disease published in the top 15 relevant journals by impact factor between 2022 and 2024. Studies with a non-significant primary or co-primary outcome were included. For each non-significant primary or co-primary outcome, a LR was calculated. Journal, author(s) and article characteristics were also collected.

Results:

1638 articles were identified and screened. 41 articles that reported 49 non-significant primary endpoints met inclusion criteria. 4 results (8.2%) favoured the alternate hypothesis and 45 (92.8%) favoured the hypothesis of no effect. For 27 results (55.1%), the LR exceeded 10; for 14 (28.6%), it exceeded 100; and for 7 (14.3%), the ratio exceeded 1000. There was no correlation between p values and LR (p = 0.754, r = 0.05).