Students
- learn methods to extract, link and prepare structured and unstructured data from health registers, patient information systems, pharmacy records, health records.
- learn about legal and ethical constraints and how to use privacy enhancing technologies (such as pseudonymisation) to address these constraints
- learn to define which information is required to be able to determine a specific measurement of the effectiveness of an intervention in health care or public health
- learn to retrieve this information from existing observational registries, big data repositories and registry based trials
- learn which provisions to take to deal with legal and ethical issues concerning the use of big data
- learn which methods to use to answer causal questions about the effect of intervention on observational big data
- are familiar with concepts of evaluating probabilistic prediction models, such as discrimination and calibration, and how to asses them using cross-validation
- have profound knowledge of the reasons for over-fitting and complete separation with high-dimensional data
Assessment
There are 3 exams in total during this course.
• 2 assignments in weeks 3 and 5 of the course each counting respectively for 30% of the grade.
• 1 case study, to be handed in at the end of week 10 of the course counting for 40% of the grade.
The average weighted grade will be your final score for this course and the one entered into Osiris.
For a retake the mark of the original test needs to be at least a 4.
Prerequisites
INFOMDWR Data Wrangling
This course is for students in the master Applied Data Science only.
|
 |
|
|
In this course you will learn about the current way in which data from health registries, across Europe, and data from routine care, can be analyzed.
Typically, data from registries are used to answer questions such as the effectiveness or occurrence of side effects of drugs, e.g. COVID vaccines.
These are causal questions, because we want to ascribe effectiveness or side effects to the drug. Because the prescription of drugs for a patient is based on indication by the physician, rather than randomization in an experimental study, it is important to learn how to draw causal conclusions from these observational data.
Further, the way the data from multiple registries are made accessible, may be constrained for legal (e.g. privacy) and ethical reasons.
Data from routine care pose another problem to the analysis of data, depending on the purpose:
- researchers may want to make a classifier or prediction model for the health outcome of patients, that can be used to inform patients, select patients for a certain treatment, or identify patients who are at risk for a bad outcome, to refer them to more advanced care.
- researchers may want use EHRs to compare patients who did receive a certain treatment with patients who received another treatment.
The topics in this course will therefore pay attention to how to extract data from registries, how to draw conclusions on effectiveness, what the legal and ethical considerations are, and what types of analysis should be done on EHRs, depending on the purpose.
We will end with practical methods for causal inference, that are applicable both to registry data and routine care data (EHRs).
Course form
Lectures, exercises, assignments.
Literature
Chapters and articles, listed in the course manual.
|
|
|