Real-World Emulation of the SWEFOT Trial

Overview

This study will compare rheumatoid arthritis (RA) patients who have been treated in clinical practice with either infliximab or a combination of sulfasalazine and hydroxychloroquine, after having an active disease despite treatment with methotrexate for at least one month. To establish which patients respond to treatment, DAS28-ESR measurements (disease activity score using 28 joints and erythrocyte sedimentation rate) taken at treatment start and nine months thereafter, and the EULAR (European League Against Rheumatism) definition of a "good response" will be employed. The purpose of the study is to verify if the same conclusion could be reached using data from patients treated in real world clinical practice as in a previous randomized controlled trial comparing the two treatment strategies (SWEFOT -- ClinicalTrials.gov Identifier: NCT00764725). Inclusion criteria similar to the ones used in the emulated trial will be applied. In real clinical practice, patients who receive infliximab may have more severe RA and may also differ in other ways from patients receiving sulfasalazine and hydroxychloroquine. To be able to compare the proportions of responders under each treatment in this "real-world" setting, the data will be re-weighted, so that patient characteristics become balanced between treatment groups.

This observational study is designed to verify if the results of a previously conducted randomized controlled trial (SWEFOT -- ClinicalTrials.gov Identifier: NCT00764725) could be replicated with observational data, by emulating the trial protocol as close as possible. Thus, early rheumatoid arthritis (RA) patients who initiated treatment with infliximab or a combination of sulfasalazine and hydroxychloroquine after at least 30 days of methotrexate will be identified in the Swedish Rheumatology Quality register (SRQ) linked to other Swedish national registers and will be compared. Similar inclusion/exclusion criteria to the ones used in SWEFOT will be applied. Also, in keeping with SWEFOT, the primary outcome is the proportion of patients with a "good EULAR response", measured nine months after study treatment initiation and considering patients who discontinued treatment before this point "non-responders" (i.e. "non-responder imputation" - regardless of the reason for discontinuation). For the combination of sulfasalazine and hydroxychloroquine the start of the second drug in the combination will be considered the study treatment initiation. Since treatment assignment is not random in observational data, inverse probability of treatment weighting will be applied to balance patient characteristics between treatment cohorts, making them comparable (i.e. exchangeable). For each observation (i.e. patient), the probability of receiving the treatment actually received, conditional on a prespecified list of potential confounders will be estimated by logistic regression modelling and will be used to calculate a weight. Potential confounders included in the model are chosen based on knowledge about factors influencing treatment allocation and observed differences in such characteristics between treatment groups. The list of confounders includes: sex, age, country of origin, year of treatment start, duration of RA at treatment start, rheumatoid factor, disease activity at treatment start measured via: DAS28-ESR (disease activity score using 28 joints and erythrocyte sedimentation rate), CDAI (clinical disease activity index), number of swollen and tender joints, pain assessment (measured on a visual analogue scale) and HAQ-DI (health assessment questionnaire disability index), use of NSAIDs and of glucocorticoids, history of: cancer, diabetes, acute coronary syndrome, stroke, venous thromboembolism, peripheral vascular disease, obstructive respiratory disease, anaemia, psoriasis, neuropathies, pain, osteoporosis, depression or anxiety, joint surgery, retinopathy and hospitalized infections, the number of days spent in hospital within 5 years before treatment start, smoking status. All data included in this study is extracted from Swedish national registers. In these registers, data is collected routinely, prospectively and independently from any study. Thus, data on baseline characteristics was collected before treatment initiation, while outcome data was recorded after treatment initiation. The proportion of responders in each group and the ratio between proportions will be estimated using a weighted generalized linear model, specifying a binomial distribution for the dependent variable and a logarithmic identity function. Before analysis, missing data (covariates and outcome) will be imputed using fully conditional specification multiple imputation after non-responder imputation. All data-analysis (from weights estimation to responder-ratio estimation) will be conducted separately in each imputed data-set, the results being subsequently pooled using Rubin's rules.