Methods
Study Design and Participants
The Dose REduction Strategy of Subcutaneous TNF inhibitors (DRESS) study was a pragmatic, open label, randomised controlled, non-inferiority trial, stratified by the TNF inhibitor used. The rationale and design have been described extensively elsewhere, and are summarised here. We enrolled consenting patients with rheumatoid arthritis (based on 2010 or 1987 American College of Rheumatology criteria, or clinical diagnosis by the treating rheumatologist) using adalimumab or etanercept at any stable dose and interval for at least six months, with stable low disease activity at two subsequent visits. Low disease activity was determined by the rheumatologist and measured using the DAS28-CRP score. This validated composite score measures disease activity (range 0.9–9) and includes 28 swollen and tender joint counts, patients' judgment of global disease activity, and C reactive protein level. The study was performed at the Sint Maartenskliniek in Nijmegen and Woerden, the Netherlands, from December 2011 to May 2014, and was approved by the local ethics committee (Commissie Mensgebonden Onderzoek region Arnhem-Nijmegen, NL37704.091.11). The study was registered with the Dutch trial register (www.trialregister.nl, NTR 3216) one week after study inclusion started, owing to temporary incapacitation of the research doctor.
Randomisation and Masking
Allocation was stratified by TNF inhibitor using block randomisation in random sized blocks (block size 3–12), and in a ratio of dose reduction versus usual care of 2:1. This ratio was chosen to include more determinants in a prediction model for successful dose reduction or discontinuation. A research physician allocated patients using a randomisation list generated by computer. To conceal the sequence until treatment strategy was assigned, sequentially numbered sealed opaque envelopes that contained the randomly assigned allocations were used.
Procedures
Patients allocated to the usual care group continued a standardised treat-to-target treatment protocol, aimed at maintaining at least low disease activity. Visits were planned every three months and patients were encouraged to contact the outpatient clinic if they experienced more symptoms. DAS28-CRP scores were assessed by nurses and provided to the treating rheumatologists. Nurses had been trained and their joint assessment skills were calibrated repeatedly to optimise inter-rater reliability. Treatment was changed in case of a disease activity flare. We defined a flare using a validated criterion: an increase in DAS28-CRP score of more than 1.2, or a score increase of more than 0.6 compared with baseline score, where the current score was at least 3.2. If a flare occurred, a new visit after four weeks was advised. Intramuscular and intra-articular glucocorticoid injections were allowed.
In the dose reduction group, patients received identical care as the usual care group, with addition of a dose reduction advice given for that particular patient to the treating rheumatologist. The dose reduction strategy consisted of stepwise increases of the time interval between injections every three months. For adalimumab, the steps were (1) 40 mg every 21 days, (2) 40 mg every 28 days, and (3) stop. For etanercept, the steps were (1) 50 mg every 10 days, (2) 50 mg every 14 days, and (3) stop. In the instance of a flare, the last effective interval was reinstated; if the flare persisted, TNF inhibitor use was increased up to the shortest registered interval, and treatment was switched thereafter in case of persisting flare. Only one attempt at dose reduction was done, and in the case of flare and treatment escalation, no further attempts at reduction were made.
Outcomes
Primary outcome was the difference in cumulative incidence of major flare between the dose reduction and usual care groups at 18 month follow-up. Major flare was defined as a DAS28-CRP based flare with a duration of longer than three months, independent of treatment changes. We initially planned the primary outcome to be the cumulative incidence of all flares (now a secondary outcome). The reason for the switch of primary outcome was the developing insight that transient flaring is an inherent trait of all dose tapering strategies based on disease activity. However, when a flare can be treated easily, and does not lead to loss of function, quality of life, persistent increase of disease activity, increase of other treatment, clinically relevant radiographic damage, and prolonged major flare, in our view this can be reasonably viewed as non-inferior care. The outcome change took place after study start but before outcomes were available, as has been published elsewhere. Patients with major flare were reviewed further by two doctors.
Secondary outcomes included cumulative incidence of patients with short lived flare (duration >three months); change in DAS28-CRP score; change in functioning, as measured by the health assessment questionnaire-disability index (HAQ-DI; range 0–3, higher score indicating worse functioning); and quality of life, as measured by EuroQol-5D-5L (EQ5D-5L; range 0–1, higher score indicating better quality of life). These outcomes were assessed at nine and 18 months' follow-up. We recorded the proportions of patients who could successfully taper or stop treatment, as well as any change in the use of glucocorticoids or disease modifying antirheumatic drugs (DMARDs), and the occurrence of severe adverse events.
Radiographs of hands and feet (at baseline and 18 months) were assessed in chronological order by two blinded, trained readers, using the modified Sharp-van der Heijde (SvdH) score (range 0–448; higher scores indicate more joint damage). These values included subscores for erosion (range 0–280) and joint space narrowing (range 0–168). The proportion of patients with a change in SvdH score exceeding the minimal clinical important change of eight points in 18 months was compared between groups. As a sensitivity analysis, the smallest detectable change was calculated and used as cut-off value, as well as a third cut-off value of 0.5 SvdH units. We calculated quality adjusted life years based on EQ5D-5L scores to analyse quality of life differences.
We also assessed costs, cost effectiveness, and the predictive value of serum drug levels and whole body PET/CT scanning (positron emission tomography combined with computed tomography) scans. These results will be published in separate publications.
Statistical Analysis
We assumed that 20% of patients would experience the primary outcome in the dose reduction arm and 15% in the usual care arm. With one sided testing (α=0.05, 1−β=0.8), a non-inferiority margin of 20%, and randomisation ratio of 2:1 dose reduction versus usual care and accounting for 5% dropout, we calculated that 180 patients would be necessary to reject the null hypothesis of inferiority. No interim analyses or stopping rules were defined before study start.
The chosen non-inferiority margin is to some extent arbitrary, because no non-inferiority studies on rheumatoid arthritis have used major flare as outcome. However, in our opinion, a difference in persistent flare of over 20% between the usual care and dose reduction groups would constitute a clinically relevant non-inferiority margin. The underlying reasoning is that half of patients who start another biologic to treat a flare are expected to show a response again within three months. Therefore, half of 20% of the patients would experience a persistent flare—that is, a more prolonged period with uncontrolled disease activity—resulting in a number needed to harm of 10. In our clinical view, this number seems to balance well with an expected chance of being able to reduce the dose or stop the drug of about 60% and 15% respectively (numbers need to treat 1.5 and 6, respectively), because many more patients are expected to benefit than be harmed using this non-inferiority margin.
Primary analyses were done per protocol by including only patients who (1) completed follow-up, (2) actually started dose reduction of TNF inhibitors in the dose reduction arm, and (3) had not stopped or reduced TNF inhibitor use at 18 months' follow-up in the usual care arm. Additional intention to treat analyses were also done. For the primary outcome, we calculated the point estimate and confidence interval of the difference in cumulative incidence of major flare between both groups, and compared the upper limit of the confidence interval with the non-inferiority margin. A t test compared mean values (and mean time averaged) for DAS28-CRP, HAQ-DI, and EQ5D-5L. We used the χ test to analyse the difference in cumulative incidence of flares, and the levels of disease activity at baseline, nine months, and 18 months. The different timepoints were tested separately, with no repeated measure analyses performed.
To identify predictors, we performed two univariate logistic regression analyses. The two main outcomes—successful dose reduction and stopping—were dependent variables, and patient, clinical, and treatment variables at baseline were independent variables. We planned multivariate analyses in case more than one variable was significantly associated with one of the outcomes.