It is common in long-term longitudinal clinical studies to collect patients’ characteristics at multiple follow-up times or office visits until the censoring time or the occurrence of a key clinical event (e.g. cardiovascular events). Although the treatment difference after randomization is of high interest, patient safety and efficacy profiles during the entire long-term study may also impact the outcome. With the assumption of no changes in patient profiles, the patient characteristics at a certain single time point (e.g. the baseline or the primary endpoint) could be used as static covariates to predict the outcome; however, patient characteristics usually change over time as the disease progress. If patients characteristics are collected at multiple time points, models with time-varying covariates may yield a more accurate and robust analysis. The discussion will be focused on the current survival models for time-varying covariates and the application on clinical trial data to identify risk factors for event onset. Traditional Cox proportional hazard model can be extended to time-varying covariates, but it assumes covariate independence which is unusual in practice. Tree-based models are flexible and allow interactions/correlations among covariates which are problematic for standard models. The discrete-time survival tree (Bou-Hamad, 2011), one of survival trees with time-varying covariates which can apply to identify important risk factors from data collected at baseline and at endpoint for event onset, will serve as an example of tree-based model in the discussion. The discussion is open to any other more innovative survival models with time-varying covariates.