- Posted By: Pradeep Virdee
Discussion topic: Statistical Issues in Clinical Trials of Inflammatory Bowel Disease using Endoscopy
I attended the one-day NIHR imaging studies discussion on the 27th April 2015 at the University of Warwick. The discussion was a means of networking for statisticians so as to try and bring forward the statistical challenges in clinical trials with imaging data, in particular for bowel disease, and to develop potential ideas of how to deal with them.
The number of clinical trials using data from images (CT Scans, MRI’s, F-MISO, Endoscopy, etc.) has been increasing over the past couple of years. Endoscopy video imaging for bowel cancer was the main imaging method discussed at this meeting; this has only recently been introduced into clinical trials (guidelines for running imaging trials can be found on the FDA and EMA websites).
At present, there is a lot of uncertainty from a statistician’s perspective as to how data is formulated from images: is it reliable and does it accurately reflect the disease? There may also be disagreement between clinicians – different clinicians have different opinions on what images show – is this disagreement reflected in the data?
At this meeting, upon completion of talks about imaging techniques and clinical challenges, 4 groups of (6-10) statisticians and health researchers were each given a scenario to discuss 4 important aspects to consider in imaging trials which may present statistical challenges:
1. Sample size – Methods such as Bland-Altman for agreement recommends a sample size of at least 100 participants (as a rule of thumb) and sometimes sample size calculations may give large required numbers. However, there are imaging types which can be very expensive so the number of participants who can be treated is very limited (e.g. F-MISO). This limitation could potentially affect the power of detecting a significant change in disease assessed using imaging techniques.
2. Independent central review of endoscopy imaging as an eligibility criterion – As a means to validate imaging outcomes, an independent reviewer can be used to check for agreement between clinicians in terms of what they think images show – any disagreement could potentially be reflected in the data, making it unreliable. Some clinicians may be overoptimistic to enrol participants and may recruit some that may later be found to be ineligible. Multiple reviewers may have different opinions on a participant’s eligibility, in which case, which do you go with? What implications will this have on power and the analysis and how can it be further examined?
3. Independent central review of endoscopy imaging as an outcome criterion - Some issues discussed are the number of central reviewers needed and the associated cost implications as well as what to do if there is disagreement and how to account for the different reviewers in the analyses. Which would you go with if different reviewers had different opinions – this could create different sets of data?
4. Repeatability – Often, images will be taken at multiple time points. Areas in the body move over time so we can never be certain that the images are taken at the same point in the body at each scan. As a result, can we be sure the data accurately represents the true change over time, rather than just natural variability and how would we account for this in the analyses? How can measurement error be identified?
After discussion, potential solutions to some of the above were to ask other clinicians what they did on similar trials they worked on, take images of regions in the body to make sure you capture the same point in each scan at multiple time points, find a standardised measurement of agreement between different types of scans and between different time points for the same type of scan. There are studies on-going that will investigate some of these. It is recommended that the imaging protocols are standardised, particularly when using these in clinical trials and training in the protocols carried out.
Overall, this meeting was very interesting and portrayed very well the statistical issues that come with imaging trials being run by a number of different organisations, reflected by the number of different statisticians attending. It was useful to hear some of the potential solutions that some of the more experienced statisticians proposed and was a good opportunity for networking with other statisticians. NIHR meetings specific to medical imaging are run twice-yearly if you are interested in attending details can be found at the following link:
http://www.statistics-group.nihr.ac.uk/events/upcoming-events-1