Pre-diagnosis factors may influence a cancer's likelihood to become fatal. Many methods have been used to evaluate associations with lethal cancer among a population that was initially disease-free, but each has limitations. We propose a novel 2-stage method for evaluating risk of lethal disease that separately estimates the association with cancer incidence and survival among cancer cases, then combines the two estimates into a single measure of association with lethal disease among disease-free individuals. We compared 5 methods to evaluate risk factors for lethal cancer: traditional time to event analysis using baseline covariates (TTE);time-to-event analysis with time-dependent covariates (TDC) updated until death;TDC where exposure is only updated until diagnosis (TDX);ordered multiple event analysis (Prentice-Williams-Peterson [PWP]);and our 2-stage model (2S) and applied them to Nurses' Health Study data. In simulations, the PWP model had the most bias and highest mean squared error, while the 2-stage model performed best. Our 2-stage model may be a useful tool for identifying pre-diagnosis factors that lead to more aggressive disease.