In randomized clinical trials with time-to-event endpoints, there are commonly one or more planned interim analyses plus a final analysis, spanning over a few years. Predicting the timing of these analyses (i.e predicting the number of events within a certain timeframe) is critical to the resource allocation and the success of the study. The essence of such prediction mainly depends on patient accrual and event rates, which might vary throughout the whole trial. In this paper, we evaluate several prediction methods for short-term and long-term prediction, and discuss their pros and cons. We focus on two practical methods for long-term study milestone prediction. One is exposure-adjusted event rate method, which would utilize the observed study data and update in real-time. The other is using parametric regression estimation. Some simulation results will be carried out to compare the performance.