To maintain the interpretability of the experimental treatment effect obtained from noninferiority trials, current statistical approaches often rely on the constancy assumption. In practice, it has been found difficult to hold this rigorous constancy assumption. We will use a covariate-adjustment approach through propensity score matching to obtained a marginal treatment effect and will compare this approach to our original regression approach, which results a conditional treatment effect. We shall also briefly discuss criticisms (e.g. no patient level data, no pre-specification) of the covariate-adjustment approaches and possible solutions.