Abstract:
|
Decision medicine is considered as a practical version of personalized medicine where patients are categorized based on available information for optimal medical decision making. With the rapid advancement in technology, diverse types and increasing amount of data become available with the potential of improving stratification of patients for personalized care. In cancer prevention setting, developing practical, easy to measure signatures of goof predictive performance is key for improving adoption of chemoprevention strategies. In this talk, we discuss our experience of developing predictive biomarkers for two patient selection endpoints for breast cancer chemoprevention: 1) the level of aromatase expression, a chemoprevention drug target; 2) the presence of a breast white adipose inflammation. We also examine relative importance of factors affecting the predictive performance, such as the measurement accuracy of the endpoints, types of predictors, modeling approaches, and model selection criteria in this setting. We show that predictive performance is largely determined by measurement accuracy of the patient selection endpoint.
|