Al-Araji, Sarmad Adnan Hameed; (2023) Towards Predicting Individual Treatment Response in Patients with Multiple Sclerosis. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Al-Araji_10166926_Thesis_id_removed.pdf Download (2MB) | Preview

Abstract

This thesis demonstrates an observational predominantly retrospective study of nearly 2000 consecutive relapsing remitting multiple sclerosis (RRMS) patients recruited after initiating disease modifying therapies (DMTs); glatiramer acetate, dimethyl fumarate, fingolimod, natalizumab or ocrelizumab, in the National Health Service at the National Hospital for Neurology and Neurosurgery and followed up for a period of two years. Due to the heterogeneity of onset, diagnosis, classification, treatment response and prognosis, it is difficult to predict outcomes in MS patients. Machine learning has been evolving in the medical field over the last two decades and we have therefore reviewed the literature on the use of machine learning (ML) in MS diagnosis, classification, progression and prediction. Despite the enormous application on ML in the field of MS in the last 10 years, we found limited studies that have focused on predicting the response to DMTs that could assist clinicians in choosing the best treatment for MS patients. We therefore explored the association of routinely available clinical and radiological variables in predicting NEDA (no evidence of disease activity) at the group-level in our separate cohorts. We found that baseline variables including older age at MS onset, lower number of previous DMTs, lower number of relapses in the previous 12 months, lower number of new and/or Gd-enhancing MRI brain lesions, and lower EDSS score showed significant association of achieving NEDA in most cohorts. Whilst these factors are associated with clinical response at the group-level, we are unable to make clinically useful predictions of treatment response at the individual-level. We therefore built a Bayesian modelling framework optimal for estimating individual-level uncertainty in predicting NEDA. We combined all cohorts and built five predicting models with increasing complexity. The most comprehensive model with 12 clinical and radiological predictors showed the best accuracy, reaching 73% and 69%, in predicting individualised NEDA at 1 and 2 years, respectively. Finally, since there have been no head-to-head phase III clinical trials comparing multiple DMTs with variable efficacy, we conducted a pilot study to compare the clinical and radiological effectiveness of five different DMTs in our real-world cohort using propensity score adjustment. Although ocrelizumab showed slightly better outcomes, considering the limitations of this study, including the observational and retrospective nature of the study, different therapies used over two decades and the small sample size, further larger studies and using propensity score matching are needed in the absence of randomised controlled trials.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item