Diagnostic and prognostic models are increasingly published in the medical literature each year. But are the results relevant for decision making in practice? How can models be used for risk stratification in populations? What are the critical elements of a well-developed diagnostic or prognostic model? How can we assume that the model makes accurate predictions for our population, and not only for the sample that was used to develop the model (generalizability, or external validity)? Are big data and advanced statistical techniques the solution for the problem of poor generalizability?
In the course we will address these and other questions from an epidemiological, statistical and decision-making perspective, using examples from the clinical literature. The participants will be encouraged to participate in interactive discussions and in practical computer exercises, starting with basic approaches and extending to advanced modelling.
Overall aim of the course (in terms of knowledge, application and attitude of the students):
After the course students:
•
Understand the roles that diagnostic and prognostic models may play in risk stratification, and ultimately medical decision-making.
•
Know the critical factors that determine the validity of predictions from diagnostic and prognostic models.
•
Have insight in the pitfalls of model development with standard statistical techniques.
•
Have both theoretical and practical knowledge on advanced methods in model development and validation, specifically on regression modelling.
•
Understand the possibilities of using Electronic Health Records data for risk stratification.