B0483
Title: Benefit-risk assessment via generalized pairwise comparisons
Authors: Esben Budtz-Joergensen - University of Copenhagen (Denmark)
Brice Ozenne - University of Copenhagen (Denmark) [presenting]
Julien Peron - Hospices Civils de Lyon (France)
Abstract: The benefit-risk balance is critical when evaluating a new treatment, especially in oncology, where side effects may outweight small gains in survival. Combining traditional approaches, e.g. using a hazard ratio to summarize gains in survival and a chi-squared test to compare toxicity, typically leads to an obscure estimand. Instead, the net benefit has been proposed, as an extension of the Mann-Whitney parameter to multiple outcomes. It can be interpreted as the probability for a random patient in the treatment group to have a better overall outcome than a random patient in the control group (e.g. difference of at least two months in survival, or, if equivalent survival, a lower grade adverse event), minus the probability of the opposite situation. We present how estimation and uncertainty quantification can be performed in the presence of right-censoring, leveraging results from the U-statistic theory. In particular, we derive the first-order H-decomposition of the statistic. This enables us to quantify the uncertainty in sensitivity analyses, e.g. when varying the threshold for survival. These developments are implemented in the R package BuyseTest that is available on CRAN.