Defining disease progression in ALS
: A novel analytic approach using existing clinical and imaging datasets

  • Matthew Gabel

Student thesis: Doctoral Thesis


A key aim of medical science is modelling patterns of disease progression; these patterns
increase understanding of the disease, and help construct staging systems that
assist diagnosis and treatment. Within Amyotrophic Lateral Sclerosis (ALS) disease
progression modelling, there is a need to integrate clinical observation-based staging
systems such as Roche et al. (2012), which suffer from low temporal resolution, with
‘unbiased’ staging of biomarkers. To this end, I have adapted and extended an EventBased
Model (EBM) for ALS from previous work in Alzheimer’s disease (Fonteijn et
al., 2012; Young et al., 2014). Unlike traditional models of disease progression, eventbased
models do not rely on a priori staging of patients but extract the event ordering
directly from the data, thus minimising subjective bias. In MR imaging, Fractional
Anisotropy (FA) derived from diffusion tensor imaging is an obvious candidate to test
the hypothesis that imaging events can be staged in the ALS-adapted EBM.
Using contemporary and historical ALS datasets comprised of diffusion MRI, clinical
and neuropsychological data, I have adapted and extended a novel event-based model
to analyse the likely ordering of these biomarkers in the progression of ALS.
Materials and Methods
The contemporary dataset was derived from a cross-sectional sample of 23 ALS patients
and 23 matched controls (Broad et al., 2015). The two historical datasets were
similarly derived from samples of i) 36 ALS patients and 22 matched controls, and
ii) 28 ALS patients and 25 matched controls (Tsermentseli et al., 2015). The ALSspecific
adaptations to the EBM were i) the fitting of Gaussian mixture models by
constrained Expectation Maximisation, ii) the calculation of event probabilities from
the cumulative distribution function to preserve the monotonicity of biomarker reading
progression, and iii) accounting for the clearly delineated patient and control cohorts by
performing Markov Chain Monte Carlo (MCMC) sampling on only the patient cohort.
Finally, a fully Bayesian approach to Event-Based Modelling is demonstrated.
The most likely order of progression of imaging events showed that FA changes in
the lower aspect of the corticospinal tracts (CSTs) occur at an early stage of disease
evolution, with changes in the upper aspect occurring at a later stage. This result was
found individually in all three datasets, as well as when combining them.
This proof-of-principle study shows that data-driven models of ALS progression are
feasible, as well as demonstrating a fully Bayesian approach to Event-Based Modelling.
The diffusion MRI event ordering results suggest very robustly that damage to the
CSTs starts in the lower aspect. Nevertheless, a general important limitation must
be discussed: The small sample size may have biased our results. I have tried to
address this issue by assessing how the results varied across three separate datasets,
both individually and combined. While the CST results were consistent across the
entire process, results for other regions such as the corpus callosum were less constant,
suggesting that the biomarker ordering in the wider population may diverge from this
In order to generalise these results to the wider spectrum of ALS, future studies on
larger datasets are warranted.
These findings provide the first solely data-driven evidence supporting a directional
hypothesis of motor neurone degeneration.
Date of AwardAug 2017
Original languageEnglish
Awarding Institution
  • University of Brighton
SupervisorMara Cercignani (Supervisor)

Cite this