Predictive Models in Castration Resistant Prostate Cancer

Abstract

The median overall survival from clinical trials evaluating therapies for patients with castration resistant prostate cancer has improved dramatically over the last 30 years from 6 months to over 32 months. As such, the willingness of physicians to administer and of patients to receive these therapies has also increased. However, not all patients will benefit equally from these therapies while still being exposed to, often significant, toxicities. The analysis of risk versus benefit is essential to the medical decision making process and is wholly dependent on the ability of the patient and physician to accurately predict relevant risks and benefits. Traditionally, physicians have made these predictions based on a combination of clinical experience and intuition. However, recent advances in computer based mathematical modeling and the availability of robust datasets from large randomized controlled trials have led to the development of predictive models that can outperform clinicians. In addition, the widespread availability of computers and other electronic devices means that these predictive models can be packaged and delivered to the lay public in an easy to use graphical representation, such as nomograms. While predictive models are powerful tools to aid in clinical decision making, they also have their own unique limitations. A given predictive model’s accuracy and relevance can vary dramatically based on the way the model was constructed, how it was validated, and the dataset from which the model was derived. In addition, predictive models do not consider the patient’s perspective on their disease, the impact of treatment complications or the goals of therapy. When these variables are clearly defined between the clinician and patient, predictive models powerful risk assessments tools for a variety of clinical endpoints.