Predicting failure of methotrexate therapy in patients with rheumatoid arthritis: statistical approaches to multiple outcomes and alternative therapies

Abstract

Rheumatoid arthritis (RA) is a chronic autoimmune condition that causes swelling, stiffness, and pain in the synovial joints of the hands and feet. RA progresses over time, so early diagnosis and pharmacological management is vital to limit disease progression. As recommended by major clinical guidelines, methotrexate (MTX) is the first-line therapy for RA, mainly due to its cost-effectiveness. However, around 40% of patients do not respond to MTX by 6 months and around 20-30% discontinue MTX due to adverse events (AEs) in the first year. Therefore, there is a clinical need to identify which patients are at high-risk of not responding to MTX, so disease activity can be controlled using alternative therapies, such as biologics. Such risk stratification is possible using clinical prediction models (CPMs) but existing models in this area have had little impact on practice. This thesis aimed to identify and address the methodological limitations of CPMs for MTX treatment outcomes in RA. The specific objectives were to (1) conduct a systematic review that summarises and critically appraises previously developed and validated CPMs of MTX outcomes in RA, (2) develop a multinomial prediction model that estimates, at the starting point of treatment, an individual’s risk of not achieving low disease activity (LDA), or discontinuing MTX due to AEs, (3) dynamically update this model at 3 months post baseline, (4) externally validate the models developed in Objectives 2 and 3, (5) present the learnings from Objective 2 to 4 as practical guidance, and (6) explore causal prediction methods that enable predictions under interventions in patients with RA. Previously developed CPMs for MTX outcomes were identified through a systematic search, and a risk of bias assessment was conducted to critically appraise the methodologies used. A total of 20 CPMs were identified, all of which were rated at high risk of bias. This was due to methodological shortcomings, including a lack of internal and/or external validation, being developed using insufficient sample sizes, inadequate handling of missing data, poor reporting, and a lack of accounting for the competing risk of discontinuation due to AEs, which became a particular focus of this thesis. A multinomial prediction model was developed to predict outcomes of disease activity alongside discontinuation due to AEs using data from the UK Rheumatoid Arthritis Medication Study. The multinomial model was dynamically updated at 3 months after commencing treatment using landmarking analysis, and externally validated in the Norwegian Disease-modifying Antirheumatic Drug Register. Results showed that outcomes of disease activity were predicted more accurately than the discontinuation due to AEs. This applied work was used as an illustrative example in new guidance on how to develop, validate, and update multinomial prediction models. The guidance outlined specific methodological considerations for developing and evaluating the performance of multinomial models, which come with added complexities. A reliable CPM could help identify patients that are likely to not respond to, or to discontinue due to AEs related to MTX therapy; however, there is currently no indication that these patients will respond better to alternative therapies. This requires causal inference methods to be incorporated into a CPM to enable predictions under interventions. This thesis considered four causal prediction approaches, exploiting observational and RCT data to demonstrate methods that could be used to develop CPMs that more closely align with how these models are intended to decide optimal treatment in practice. In conclusion, this thesis demonstrated how to address the methodological limitations of CPMs for MTX outcomes in RA, providing a foundation for further research to increase potential for implementation into practice.

Date of Award	1 Aug 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Kimme Hyrich (Supervisor), Suzanne Verstappen (Supervisor), Jamie Sergeant (Supervisor) & Glen Martin (Supervisor)