top of page

Modelling Duchenne Muscular Dystrophy Progression

Victor Applebaum, 2nd year PhD, University of Exeter

BACKGROUND:

​

Duchenne Muscular Dystrophy is an X-chromosome linked neurodegenerative disease affecting about 0.2% of males born globally. It is characterised by a lack of stability in muscular cells, resulting in muscle weakness and deterioration in ambulatory (movement) abilities.

When treating DMD patients, many of the interventions clinician can make have problematic side-effects. For example, steroids, which helps boost muscle development, can result in behavioural difficulties, difficulty sleeping, and other physical issues.

To aid these decisions, it would be useful for clinicians to be able to predict the trajectory of their patients. This would enable them to, for example, provide stronger interventions if they know a patient’s condition were soon to deteriorate.

I have therefore been working on a model with the purpose of predicting DMD patient trajectories, looking at how their ambulatory abilities change.

​

​

METHODOLOGY:

​

I have access to a data set of NorthStar Ambulatory Assessment (NSAA) scores. This assessment involves the patients performing 17 ambulatory tasks, such as walking, jumping, and getting up off the floor. They receive a score out of 2 for how well they can do each of these, for a total score out of 34, which effectively represents how well they can operate in their lives unaided.

​

There are a range of difficulties with modelling this data set. Each individual doesn’t have that many data points, which means I don’t have much data for each individual to infer parameters in the model from. To deal with this, I use a strategy called hierarchical Bayesian modelling. This allows me to calculate the distribution of parameters in the general population, and this can be used to weight parameter values for individuals.

​

Another issue is how to model treatments. As a result of the side effects of steroids, clinicians do not typically prescribe these unless the patient’s condition is deteriorating. This means that patients who are not taking steroid treatments are not a control for those who are. This creates an issue called ‘non-identifiability’, where it is difficult to calculate the true parameters values for treatments.

​

There are a range of other similar issues, so I have created eight variations of the model, to test, and determine which method of modelling DMD NSAA scores is the most effective.

​

RESULTS:

​

I show two versions of the model below, on two patients. Blue points represent training points, while red ones represent test points. I show the model’s 70% prediction intervals in green (70% of the red test points should ideally be within these), and the median prediction in black. 

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

 

 

 

 

 

 

 

​

VictorApplebaum.png
VictorApplebaum.png

FUTURE WORK:

​

My next steps are to finish fitting the models and start a comprehensive comparison of them. During this I will be analysing how good their fits are (are 70% of the test points in the 70% prediction intervals?), how thin the prediction intervals are, and how good the synthetic data the models can generate is.

​

 

FUNDED BY:

​

​

​

​

                

​

 

 

 

CONTACT: 

​

logo_Victor.png
Logo_victorr.png
  • LinkedIn

Victor Applebaum

  • X Twitter

@victorapplebaum

bottom of page