Delayed treatment effect (DTE), cure fraction (CF) and treatment switching (TS) are known to cause the violation of the proportional hazards assumption required in standard analyses of time-to-event data. We investigated their effect on power under different simulation conditions for various data distributions, durations of the study, accrual periods and analysis methods. Simulations show that power decreases dramatically in the presence of DTE and TS, whereas overpowered situations occur when CF is encountered in the experimental group. Underpowered situations occur when CF is experienced in both experimental and control groups. When DTE, CF and/or TS are considered at the design stage and integrated into the sample size calculation, an adjustment in the number of subjects may be required to maintain the targeted power. In conclusion, omitting DTE, CF and/or TS at the design stage could lead to drastic implications on power and impact the chance to appropriately identify a potential treatment effect. The implementation of Bayesian methods during the design stage is recommended in order to proactively anticipate the effect of DTE, CF and/or TS, allowing for a more efficient design.