IMPROVING MEDICAL DECISIONS WITH COGNITIVE DATA SCIENCE

Overview: This course gives participants an overview of how cognitive data science can be used to develop decision aids in health care. Cognitive data science integrates knowledge from statistics, machine learning, and cognitive science to produce tools that are both highly predictive and cognitively simple to use. By the end of the course, participants will be familiar with the most important concepts of cognitive data science and with the steps that are required to develop simple but powerful decision tools.

Background: Participants will get an overview of basic data science concepts such as feature selection, predictability, and overfitting, as well as the theory and application of decision trees, particularly fast and frugal decision trees (FFTs). FFTs are simple decision tools for binary decisions and are well suited for medical contexts due to their predictive power and user friendliness. In this course, participants will aqcuire an overview of the most important data science concepts as well as the theory of fast and frugal trees. Additionally, participants will gain hands-on experience with data science and the development of simple decision trees (in the statistical programming language R and other tools). Participants will also learn to identify (medical) problems to which simple decision trees can be applied. These skills can later be used in medical decision making research to identify predictive variables and to develop predictive models and decision aids.

Format Requirements: The course will be a mix of short input lectures, individual exercises, and group discussions. The individual exercises will be computer-based and require participants to bring their laptop, have the statistical programming language R installed (https://www.r-project.org), and have internet access. Some programming experience is helpful but not required.

Description and Objectives: In some medical decision making situations, for example in emergency rooms, high-stakes decisions have to be made at great speed. Slow and error-prone decision making can threaten patients’ health. Using data from large hospitals, I will show how one can identify useful information to solve a decision problem and how one can estimate the predictability of a data set with state-of-the-art machine learning tools. As a next step, I will show how simple decision aids can improve decision making. As an illustration, I will present a simple fast and frugal decision tree that can identify whether a patient is highly morbid or not.

Fast and frugal decision trees (FFTs) bridge the fields of cognitive and data science. They are a quintessential family of simple heuristics and often perform remarkably well when compared to more complex methods. This workshop will familiarize participants with examples of FFTs. Practical exercises and interactive tools will enable participants to construct and evaluate FFTs for different data sets. Participants will be guided through a series of exercises that will examine the consequences of different cue choices, bias values, and criterion shifts on various measures of classification performance. For example, participants will re-construct an FFT that has been designed to help emergency-room doctors to rapidly decide whether to send a patient with severe chest pain to the coronary care unit. Using these examples, I will introduce participants into the most important concepts of data science and simple decision trees.

Objectives

• Get an overview of basic machine-learning concepts
• Get an overview of decision trees, particularly fast and frugal decision trees
• Gain experience with basic data science
• Gain experience with building fast and frugal decision trees