Assessing violations of the proportional hazards assumption in Cox regression: does the chosen method matter?