AB1435 CLINICAL PREDICTION MODELS FOR METHOTREXATE OUTCOMES IN PATIENTS WITH RHEUMATOID ARTHRITIS: SYSTEMATIC REVIEW AND CRITICAL APPRAISAL

Abstract

BackgroundMethotrexate (MTX) is the preferred first line therapy for rheumatoid arthritis (RA). MTX has several advantages over other treatments including effectiveness and low cost; however, around 40% of patients are classed as non-responders after 6 months (1). Therefore, there is a clinical need to identify patients at high-risk of poor outcomes, such that patients could potentially be fast tracked onto alternative therapies to improve their clinical outcomes and quality of life. Such risk stratification is possible through prognostic prediction models, although models which have previously been developed appear to have had little impact on practice. This may be in part due to methodological features of their development and validation but, to date, no review has collated the evidence in this field.ObjectivesThis systematic review aimed to (i) identify and summarise multivariable prediction models of MTX treatment outcomes in biologic-naïve adult RA patients, and (ii) critically appraise their methodological properties.MethodsThe electronic databases Medline and Embase were searched to identify studies developing or validating prediction models of MTX outcomes in the population of interest, including demographic, disease-specific or treatment-related covariates, published after 2005. Models were stratified by outcome definition, and information on participants, predictors, model performance, handling of missing data and model validation were extracted. A risk of bias (ROB) assessment using PROBAST (prediction model risk of bias assessment tool) was carried out. Two reviewers were independently involved in screening, data extraction, and ROB stages.ResultsThe included studies used three main outcome definitions: a state of disease activity, such as low disease activity or remission; the EULAR response criteria; or discontinuation due to adverse events (AEs). Some studies incorporated AEs into a composite outcome with disease activity and few accounted for potential competing risks, which are events that preclude the occurrence of the primary outcome of interest. Not handling competing risks may result in under-prediction, leading to potentially compromised risk stratification. There was a lack of internal validation using cross sampling techniques, which is critical for reducing overfitting, as well as external validation in new data, a process necessary to ensure reproducibility and generalisability of a prediction model to the larger patient population. Missing data was mostly handled using complete case analysis, leading to potentially biased risk estimates. The ROB assessment showed overall high ROB of the included studies.ConclusionThis systematic review summarises current prediction models of MTX treatment outcomes in RA. It highlights several methodological shortcomings, such as poor handling of missing data and competing risks to the primary outcome, and a lack of internal and external validation. These should be addressed in future model development and validation to improve accuracy of predictions. Without tackling these issues, prediction of MTX treatment outcomes will remain at high risk of bias and should not be recommended for informing risk stratification for RA treatment decisions.