Violation of the PH assumption

What can you do if the proportional hazard assumption is violated for a predictor?