Who is the course for?

This is an introductory/intermediate course for anyone who has exposure to logistic regression. This includes researchers who may wish to use logistic regression and clinicians who read the medical literature and want to gain a better insight into how results are reported. Use of specific statistical packages will not be covered.

Course description

Multivariable analyses are now ubiquitous within the medical literature, with logistic regression being one of the commonest techniques. Logistic regression quantifies the association between several independent variables (e.g. cholesterol, smoking, exercise, and hypertension) and a single, dichotomous outcome (e.g. heart disease).

The growth of statistical packages over the last decade has meant that clinicians can now perform this type of analysis without actually understanding its appropriate usage and limitations, leading potentially to erroneous conclusions.

This course will consider elements of logistic regression at both basic and intermediate levels, and covers:

  • Uses of logistic regression and types of model
  • The concept of odds ratio, how this differs from relative risk
  • Choice of scale for independent variables (e.g. categorical vs. continuous)
  • Variable selection procedures (which variables are in the final model and why?)
  • Interactions
  • Model diagnostics, goodness of fit, discrimination, calibration
  • Potential problems: over fitting, shrinkage, multicollinearity
  • How to report and interpret the final model.

Learning objectives

By the end of the session, students should be able to:

  • Understand potential applications and limitations of logistic regression
  • Interpret results within medical journals
  • Identify analyses that may have significant limitations.

NB: This course will not provide a comprehensive guide to model formulation

NB: Use of specific statistical packages (e.g. SPSS) will NOT be covered

About the trainer

Salma Ayis is a senior lecturer in Medical Statistics at King’s College London, with interest in several statistical methods, and expertise in binary response regression models. Salma has developed theoretical and simulation-based assessments of inference from the logistic model. She has been involved in several studies, where these methods were used to predict possible outcomes, in collaboration with clinicians and other researchers. These were published in peer reviewed journals including the Lancet. Over the last ten years, Salma has advised in many medical research projects and collaborated in a wide range of clinical areas.