Skip to main content

Data-driven models for Progression Of Neurological Disease

Periodic Reporting for period 3 - EuroPOND (Data-driven models for Progression Of Neurological Disease)

Reporting period: 2019-01-01 to 2019-12-31

EuroPOND is developing uniquely powerful data-driven statistical-and-computational models of neurological disease progression. The models are fundamentally new methods for understanding the complexity of the clinical presentation of neurological diseases, and the underlying biology. We demonstrate the models by generating new knowledge in a variety of neurological applications. Further, we are exploring avenues for patient benefit by using the models to underpin prototype support systems for clinical and drug-development applications. Specifically, the models and prototypes enable “Precision Medicine” by providing differential and personalised diagnosis, fine-grained staging, and personalised prognosis. These advances will position Europe as world leaders in one of the biggest challenges facing 21st century healthcare: management of neurological disease.
Our specific objectives are:
A. Develop, and implement as a prototype software tool, a new data-driven computational-and- statistical modelling framework for disease progression.
B. Construct a quantitative validation framework using well-phenotyped and other complementary data sets from across the European arena.
C. Construct models for a range of neurological conditions, specifically dementia, multiple sclerosis, prion diseases, and normal and abnormal neurodevelopment and ageing.
D. Prototype a computer-assisted diagnosis and staging software system for dementia and demonstrate in relevant environments
The key areas of work performed during the first reporting period were:
• Collation of a range of imaging-plus-X data sets in each neurological application (WP2).
• Agreement across the consortium of the set of relevant computational methods and initial best practice guidelines for their usage (WP2).
• Development of a suite of new computational disease progression modelling tools including the Subtype and Stage Inference algorithm, the Discriminative Event-Based Model, various refined Scalar Trajectory Models within a Bayesian framework, and new Bayesian Spatio-temporal models (WP3).
• Design of a simulation framework for enhanced development and validation of models (WP3).
• Prototype of a unified modelling framework for neurological disease progression (WP3).
• Development of an early prototype system for clinical usage exploiting the event-based model in dementia (WP4).
• Exploration of generalizability of event-based models among data sets distinct from their training data (WP5).
• Construction of progression models of normal ageing for comparison with disease progression models (WP6).
• First models of disease progression in multiple sclerosis combining markers from imaging and neuropsychological assessments (WP7).
• First models of disease progression in prion disease, mainly based on imaging (MRI) (WP8).
• First explorations of longitudinal effects in early neurodevelopment primarily using MRI (WP9).

The key areas of work performed during the second reporting period were:
• Delivery of a unified, open-source software toolbox for neurological disease progression modelling (WP3).
• Enhancement of models including improved computational efficiency, handling missing data and correlated variables, shape modelling, and nonparametric mixture modelling (WP3).
• Extensive evaluation of our model enhancements from the first reporting period (WP3).
• New models (DEBM) of MS progression (WP7).
• Application of Subtype and Stage Inference to normal ageing data (WP6, in progress).

The key results achieved during the third reporting period were:
• Performed the first evaluation of submissions to TADPOLE Challenge (WP10) and allocated prizes to the winners. The insights generated from this international challenge to predict Alzheimer’s disease progression have been documented extensively in a scientific article that has been submitted for peer review.
• Methodological (WP3): multiple updates to expand the capabilities of event-based models. These include handling cognitive data (highly non-Gaussian) and high-dimensional imaging data (enabling automatic feature discovery).
• Characterised the effects of fixed variables on the progression of Alzheimer’s disease (WP5) and normal ageing (WP6).
• Developed an ageing-informed model of dementia progression for improved predictive inference in Alzheimer’s disease (WP5, WP6).
• Used our subtyping models to define new subtypes of multiple sclerosis (WP7) and demonstrated their utility for improving clinical trials.
• Developed new insight into prion diseases by revealing and characterising previously unknown subtype-specific differences in the spread of lesions (WP8).
Management and treatment of neurological diseases is one of the biggest challenges facing medicine today, with increasingly devastating socioeconomic outcomes as the global population continues to age. The current paradigm is that of knowledge-driven clinical decisions. However, the snowstorm of data now available defies qualitative clinical evaluation; the heterogeneity of data types complicates use of traditional statistical methods; and the large datasets becoming available remain far from the big-data sizes necessary for fully data-driven machine-learning approaches.

EuroPOND aims to replace the current paradigm with data-driven disease progression modelling. We develop new models to be predictive of disease progression in individual patients and informative in terms of revealing underlying biological patterns. We translate models into prototype clinical tools.
Early progress advanced the models, produced a prototype tool, and released an open source software toolbox. We added model capabilities for precision diagnosis and prognosis that revealed new insight into understanding of neurological diseases, with our subtyping results having considerable impact and media coverage: the discovered data-driven subgroups of dementia are used by researchers to frame their thinking on how to setup new clinical studies (coverage: ALZFORUM, iNews). Additional impact comes via related work in diseases beyond the EuroPOND remit, e.g. respiratory disease.

Progress in the third reporting period added additional model capabilities, demonstrated models in multiple diseases, and advanced our clinical prototype. We are running the ongoing international TADPOLE Challenge to forecast Alzheimer’s disease progression (~90 submissions worldwide); and we discovered data-driven subtypes of Multiple Sclerosis that predict treatment response and progression better than current clinical subtypes. The TADPOLE data set and evaluation metrics have become a standard way to evaluate new algorithms and our challenge winners provide a new benchmark and standard for the field that was not previously available (media coverage: ALZFORUM 1, 2). Our prototype tool was tested extensively by multiple clinical user groups and is undergoing trial deployment as part of icobrain dm, which includes CE/FDA-approved MRI analysis and state of the art data protection measures. This is expected to impact hospitals and clinical research organisations across Europe.

In combination, the expected impact of EuroPOND's work is to facilitate therapy development and deployment through earlier and discriminative diagnosis, as well as clear and consistent staging and stratification. This will ultimately benefit society by improving quality of life and reducing loss of productivity for individuals, families, healthcare workers, and carers