Predictive modelling

Background

The objective of this course is to provide an overview of predictive modeling technologies. Predictive modelling, which usually refers to the identification of patients at high risk of an event (e.g. emergency readmission, heart failure, risk of death, long term care placement) has been a major concern for many healthcare providers (hospitals, clinicians) or purchasers (primary care trusts, local authorities). A number of predictive tools have been developed, such as the PARR case finding tool however, these are often complex to set up, expensive, and may not be applicable to all domains.

Participants will gain practical experience of developing predictive models based on routinely collected data (e.g. a PCT acute activity data) using a standard statistical computer package such as SPSS.

Aims/Learning objectives

At the end of this module students should be able to:

• Select existing predictive modelling tools in health and social care
• Understand the predictive modelling process
• Prepare data for the purpose of predictive modelling
• Appreciate the use and importance of multiple data sources, e.g. A&E, inpatient care, outpatient care, mental health, etc.
• Compare and contrast the underlying predictive modelling algorithms, e.g. Logistic Regression and Classification Trees
• Select appropriate predictive modelling approaches to identify those cases that are at high risk of an event
• Apply these approaches using SPSS

Syllabus

• What is predictive modelling?
• Predictive modelling in health and social care
• Predictive modelling process
• Data preparation using multiple datasets
• Basic concepts of regression and other algorithms, such as classification trees
• Choosing the right predictive model
• Checking the predictive power of the model (misclassification rates)
• Interpretation of results
• Practical session: Illustration of simple predictive models that can be developed based on routinely collected data

Download Module Data