Taking the pain out of subgroup analyses

Interview with Necdet Gunsoy

Why do you think many of these “subgroup” projects are a headache for statisticians?
Have you felt frustrations about such projects yourself?
What are the steps to avoid or at least minimize these frustrations?

Do you have the same questions in mind? These are some of the questions Necdet will be answering and sharing with us.

This episode is based on a presentation at PSI 2019. Necdet won The Effective Statistician Best Presenter award for the amazing delivery of the presentation. But not only the delivery was excellent – the content will help you a lot. Here’s the abstract for the conference:

Abstract: Practical aspects of subgroup detection
Recent years have seen the emergence of new methods for detecting subgroups with enhanced treatment effects and statisticians are now faced with an overwhelming choice of approaches to consider. Confusion around the potential advantages and disadvantages of different methods can often result in the implementation of an approach which is not appropriate for a particular research question. This talk will provide a practical guide to the design and conduct of subgroup detection analyses, providing points to consider for selecting an appropriate method based on the research objective, the context of the analysis, the outcomes, and the covariates under consideration. Two example case studies will be presented to illustrate the discussed considerations.

Necdet and I will also discuss the following

  • Do you have practical examples based on a given data set that leads to very different approaches?
  • As a PSI awardee of best presentation in this year’s conference, How do you approach presentations?
  • What’s your key recommendation for giving a great presentation?

About Necdet Gunsoy (PhD)

Director, GSK.

Necdet is a Director in Analytics and Innovation focusing on health economics and outcomes research at GSK. In this role, Necdet leads the development of tools, solutions, and capabilities to support the generation of evidence using real-world and clinical trial data and advanced analytics. His current research interests include subgroup analysis, preference elicitation, evidence synthesis, prediction modeling, and decision-making.

Necdet graduated with a PhD in Medical Statistics from the Institute of Cancer Research and worked in academic research before joining the GSK clinical statistics team in 2013. He also holds a Masters in Public Health from Imperial College London and a Bachelors in Biomedical Sciences from the University of Mons, Belgium.

Never miss an episode!

Join thousends of your peers and subscribe to get our latest updates by email!

Get the shownotes of our podcast episodes plus tips and tricks to increase your impact at work to boost your career!

We won't send you spam. Unsubscribe at any time. Powered by ConvertKit

2 thoughts on “Taking the pain out of subgroup analyses”

  1. Great interview! Subgroups are so important. In behavioral health, I always emphasize that there are some subgroups we are serving well and there are some subgroups for whom we can improve our offerings. Finding these subgroups is the key to both celebrating our success and CQI. Also, Yes – random forests for reducing collinearity and dimensionality.

  2. Alexander

    Thanks Kate for the comment. Indeed, it’s always an interesting topic and even, if you find, that all the subgroups have no meaningful difference in terms of the outcomes, you have a nice story to tell. This means, that your intervention works very consistent. What tools do you use for the random forest approach and how do you select the variables, which you start with?

Comments are closed.