Doctors currently choose multiple sclerosis treatments by trial and error, forcing patients to endure relapses and accumulating disability while ineffective drugs fail. Multiple sclerosis is the most common disabling disease of young people in the UK, yet no method exists to predict which of the thirteen approved disease-modifying treatments will work for a given individual. This project translates machine learning techniques from computer science into clinical practice, building a high-dimensional model that forecasts individual treatment response. The researchers will collect deep clinical, genetic, and imaging data from patients starting therapy, then validate the model against independent international registries. If successful, the prototype software could run on NHS computers to guide treatment choice from the outset, sparing patients months or years of unnecessary relapses. The work also aims to clarify why some patients respond well to certain drugs and others do not, improving fundamental understanding of the disease. For the NHS, reducing failed treatment cycles would lower long-term disability costs and position the UK as a leader in personalised neurology.
View original technical description
Background Multiple sclerosis (MS) is the most widespread disabling disease of young people. Thirteen disease-modifying treatments (DMTs) are approved in the UK to reduce the risk of relapses. Patients can switch from one first-line DMT to a more effective medication, if they present with relapses. The consequences of this policy, that requires 'failure' of DMTs before using another DMT are continuous relapses and disability accumulation. We cannot currently predict which DMT will work best for an individual patient. The goal of this project is to predict the individual treatment response in MS by translating machine learning from the computer science field into clinical practice. This will bridge the gap between clinical trials, which focus on the 'average' response to a therapy, to clinical practice, where the focus should be on the individual treatment response. Aims To develop deep phenotyping and genotyping of patients initiating DMTs in the clinical setting To generate and validate a high-dimensional model that predicts treatment response early The study design will offer the possibility to pursue additional aims, which are to: improve our understanding of disease pathobiology by defining responders' profiles collect data for the evaluation of the health-economic impact of stratification Plan of investigation MS patients Retrospective cohort: 1,566 (1,494 adults, 72 children) currently on DMTs at UCLH NHS Trust and in the UK paediatric neuroinflammation centres, who are undergoing clinical and MRI assessments according to standard-of-care protocols. In the NIHR UCLH BRC Imaging initiative, we have already collected all the patients' MRI scans and carried out a pilot study that demonstrates the feasibility of our approach. Prospective cohort: 780 patients (700 adults, 80 children (>12yrs)) who initiate any DMT during the course of two years. Patients will be studied at baseline and after 6 and 18 months. A standardised acquisition of demographic factors, diet quality and life-style, clinical scales, comorbidities, MRI scans and blood tests for safety data and neurofilaments-light levels, will be collected. Genetic factors will be obtained through NIHR UCL BioResource biobank. Independent, international cohorts: MSBase (www.msbase.org,>55,000 patients); Swedish MS registry (http://www.neuroreg.se/en.html/multiple-sclerosis,>19,000 patients); FutureMS (www.stratmed.co.uk/programs-and-projects/future-ms/); EU Paediatric Demyelinating Disease Network. Predictive modelling Treatment response will be defined on clinical and MRI grounds. Machine learning techniques will be used to develop a high-dimensional model that predicts individual treatment response, using all the information collected. The model will be constructed from retrospective and prospective data, and then validated on independent cohorts. Benefit for patients and NHS The outputs of this project are: Access to deep phenotyping and genotyping directly for NHS patients Prototype software running on local machines to guide treatment choice Improved understanding of the mechanisms that lead to a positive treatment response Data registry for the evaluation of health-economic impact of stratification Predicting individual treatment response is crucial to practice personalised medicine, which will have a huge impact at a personal level, societal level and on the NHS. This project will position the UK as a world leader for research in personalised medicine in neurology.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know