|
Abstract:
|
When the proportional hazards assumption is violated, using the Cox model can capture the average effect of a covariate within the follow-up period, but the bias due to censoring needs to be adjusted; On the other hand, using time-varying coefficient survival models can capture the underlying changing pattern of the effect over time even with censoring, but the analysis results may be difficult to explain to clinicians and not very useful for decision-making. We combined these two different analysis ideas together through utilizing fractional polynomials, Bayesian dynamic survival model, and multiple imputation techniques, to obtain not only an estimate of the time-varying effect but also a simple overall estimate robust to censoring. The performance of the proposed approach is compared to other alternative approaches through simulation. Application examples are provided.
|
