The use of restricted mean survival time to estimate joint treatment effects in the presence of interaction terms under model misspecification - a simulation study