Location: 2240 CIEMAS
This course has two objectives. First, it seeks to develop an understanding of risk prediction and classification in the Omics setting. Second, for researchers who plan to develop risk models, this course seeks to provide concrete steps for study design, analysis, and interpretation. To accomplish these goals, we will discuss how different aspects of a statistical model can provide measures of association or measures of predictive accuracy. This distinction is important in understanding how developing a model for association/etiology/causal inference is conceptually different from using the model to predict. We will then discuss risk models in the conventional setting: larger sample sizes with a smaller number of predictors. We will cover study design, statistical models, and performance metrics. The course seeks to develop an appreciation of challenging considerations in the field, but also seeks to provide clear steps on how to proceed. Finally, we will review areas of active research and in what direction the field is moving. After establishing foundations, we will move into the Omics realm, which is characterized by smaller samples sizes and thousands of predictors. Prediction models in Omics often use machine-learning techniques, so we will cover some common machine-learning techniques and what makes them different from more conventional models. We will review current best practices with an emphasis on estimating performance. This course will not include any hands-on coding because of time limitations, but this will be the topic of a future course. The course focuses on understanding the most important aspects of risk prediction and classification.