Prediction models are the cornerstone of personalized medicine. Yearly over 5,000 publications describe biomarker discovery, or prediction model development and validation. Unfortunately, the development and application of prediction models in medical science is often suboptimal. Moreover, advances in biostatistics and machine learning have led to new insights and opportunities for more accurate and useful prediction models that are able to tap into the wealth of (big) data that is collected in contemporary medical research and clinical practice. However, new technologies for prediction are difficult to apply without a solid understanding of the basic concepts underpinning prediction modeling.
During this course students will learn to address common problems in prediction modeling. First, they will learn the differences between etiological and prediction modelling, then understand the concepts of traditional epidemiological prediction models and learn how to build one themselves and from there learn the ‘machine learning language’ and how the methods compare to traditional epidemiological prediction models. Next, they will learn to create and evaluate a prediction model using advanced statistical techniques and machine learning algorithms (e.g. random forest, support vector machines). They will see and work with interesting biomedical problems such as establishing diagnoses based on large-scale clinical, imaging and molecular data. Guest lectures will show applications in stratification of patients for cancer therapy and artificial intelligence-supported diagnosis of cancers from radiological images.
