The authors argue that even though the use of hazard ratios (HRs) from the proportional hazards (PHs) models to summarise the results from clinical trials is very popular and the effect estimates may be statistically valid, they are not immediately relevant to individuals as causal effect estimates. At baseline, the assumption of the Cox PH model may be well satisfied, but after baseline there is usually selection bias at work depleting at a higher rate the subjects from the control group when the treatment is indeed effective. In the context of causal inference, the authors argue for the use of survival curve estimates since they may also be used to calculate effect estimates with a causal interpretation, such as restricted mean survival time, or RMST difference reported at several time points.