Therapeutic effectiveness research relies heavily on prediction modeling, as improving therapeutic outcomes for individuals often requires being able to predict which patients are likely to do poorly on a given therapy. In this dissertation, we examine the specific case of patients starting warfarin therapy, many of whom are at higher risk of bleeding and thrombotic events because they take a long time to determine their optimal therapeutic dose. Additionally, we examine the general problem of transportability of clinical prediction models and whether that problem can be improved through sequential model updating. Specifically, we conducted three projects with the following goals: 1) To determine the social, clinical, and genetic factors associated with time to maintenance dose in patients starting warfarin; 2) To develop and externally validate a prediction model of prolonged dose-titration in these patients; and 3) To determine whether sequential model updating can improve model transportability in a simulation study. Being able to predict which patients are likely to experience prolonged dose titration on warfarin could help clinicians and patients decide whether to use warfarin or a less burdensome alternative oral anticoagulant. Furthermore, the overall utility of this and other clinical prediction models could be greatly increased by strategies that improve model transportability, such as sequential model updating.