Methodology workshop: Using cluster analysis for disease subtyping

Attendees of the workshop stand for a group picture in front of the Usher Institute logo
12 Feb 2019

Written by Elsie Horne, PhD Student based at the University of Edinburgh

Understanding different types of asthma

Asthma is a condition that can be very different from person to person. It’s difficult to put different types of asthma into exact categories, because we still don’t know exactly what is driving these differences. You can find some information on the current understanding of the different types of asthma on the Asthma UK website.

Understanding asthma types on the Asthma UK website

Using cluster analysis to identify subgroups of patients

Many researchers are working to better understand the different types of asthma and what causes them. One approach to this problem is to apply a method of machine learning called 'cluster analysis' to data collected from asthma patients.

Cluster analysis aims to identify subgroups of patients with similar combinations of symptoms and characteristics. The hope is that these subgroups could help clinicians to better treat the disease, patients to better understand their condition, and researchers find causes and treatments for the specific subtypes. 

Identifying subtypes of asthma using cluster analysis is one of the aims of my PhD project.

The workshop sessions

To address some of the challenges of applying cluster analysis to medical data, I held an AUKCAR methodology workshop on cluster analysis for disease subtyping.

The workshop attracted 12 researchers – four from the University of Edinburgh, six from London Universities, one from Swansea University and one all the way from Lausanne in Switzerland! Not only were these researchers from many different institutions, they were also working on a diverse range of projects from subtyping Alzheimer’s disease based on electronic health records, to subtyping severe malaria in Gambian children.

To kick off the workshop, we spent the morning discussing four research studies which used cluster analysis to identify disease subtypes in asthma, COPD and diabetes. This session was a great way to break the ice and let everyone get to know each other, but we also identified interesting methods which were discussed in more detail later in the afternoon, as well as identifying a few problematic methods!

During the following session we were joined by Dr Susannah McLean, a GP in Edinburgh who has a particular interest in airway disease. Susannah gave a fascinating presentation on precision medicine in airway disease, which got us all thinking about the clinical implications of our projects. We spent much of this session discussing the clinical contexts of our projects in more detail. This discussion benefited greatly from the diverse range of clinical questions being studied.

We spent the afternoon going over specific methods of cluster analysis in detail. To bring the workshop to a close, we discussed the publication of cluster analysis studies in peer-reviewed clinical journals. Several of the researchers shared their experiences of publishing, as well as presenting at conferences. It was an excellent opportunity for PhD students like myself to learn from some of the more experienced researchers at the workshop.

Sharing experiences and making connections

This workshop provided an opportunity to spend a whole day discussing a specific method with like-minded researchers. I’m confident that the knowledge I gained from the workshop will greatly benefit my PhD project. I found it particularly useful to hear about subtyping in clinical areas other than asthma, as this led to new insights on how the methods could be applied in asthma.

Most importantly, the workshop has connected researchers across institutions and clinical areas who are using similar methods. I’m sure we will retain these connections and that discussions will continue as we progress through our projects!

© 2015 AUKCAR