Population (ARAD)
ARAD is a voluntary Australian biologic registry established in 2001 to collect patient-reported long-term safety and other outcome data from patients with inflammatory arthritis, including RA, psoriatic arthritis, ankylosing spondylitis and juvenile idiopathic arthritis [12]. It includes participants commenced on bDMARDs as well as control subjects on conventional treatments (enrolment of control subjects since February 2007). Rheumatologists refer patients to the registry with minimal baseline information, including diagnosis, as well as rheumatoid factor and anti-citrullinated protein antibody (ACPA) status (ACPA collected since November 2010). All participants may commence, switch or cease bDMARDs at any point during their follow-up. Following written informed consent, participants complete a baseline ARAD questionnaire, with a follow-up questionnaire completed every 6 months for 2 years and then at 12-monthly intervals. The initial questionnaire is defined as the ‘baseline’ questionnaire, and there are no exclusions for disease duration, prior therapies or associated co-morbidities. The questionnaire was initially available in paper form only, but an electronic version has been available since August 2009. Adult participants with a diagnosis of RA were selected for this analysis, with data censored at October 2015. In order to reflect real-life clinical practice, participants are included in the registry with a diagnosis of RA based on expert clinical opinion (rheumatologist) rather than classification criteria. In Australia, bDMARDs can be prescribed for RA only by rheumatologists, and prescribing is restricted on the basis of the following criteria: (1) The patient has severe, active RA; (2) the patient has failed a 6-month intensive csDMARD trial with a minimum of two agents for a minimum of 3 months each; (3) the patient can demonstrate failure to achieve an adequate response to 6 months of intensive prior treatment by an elevated erythrocyte sedimentation rate > 25 mm/h and/or an elevated C-reactive protein level > 15 mg/L, and the patient has an active joint count of ≥ 20 active (swollen and tender) joints or ≥ 4 major active joints (elbow, wrist, knee, ankle, shoulder and/or hip). In Australia, there is universal access to medications via the Pharmaceutical Benefits Scheme (PBS). bDMARDs have been available on the PBS since August 2004. Prior to this, patients accessed bDMARDs through clinical trials.
Ethics approval for ARAD has been obtained from 18 committees and organisations across Australia (Additional file 1). This study was approved by The University of Adelaide Office of Research Ethics, Compliance and Integrity (approval number H-2015-258).
Outcome measure
For oral GC current use, each questionnaire contains a section ‘medications for arthritis’ where patients indicate their use of oral GCs (prednisolone/prednisone) since their previous questionnaire as ‘never taken’, ‘currently taking’, ‘stopped taking’ or ‘don’t know’. Data regarding dosage are not collected. For this analysis, a time-varying ‘current use’ variable was created for which ‘currently taking’ was coded as ‘yes’ and ‘never taken’, ‘stopped taking’ and ‘don’t know’ responses were coded as ‘no’. The current use variable includes only oral GC use, with injectable GC use described but not included in the analyses.
Predictors
Patient demographics, including age and sex, at baseline/initial questionnaire and a time-varying current age variable were considered as predictors in the analyses. Current use of bDMARDs and csDMARDs were coded as yes/no time-varying variables using the same method described for current oral GC use. Current use of bDMARDs included use of etanercept, adalimumab, anakinra, infliximab, rituximab, abatacept, tocilizumab, golimumab or certolizumab. Current csDMARD use included use of methotrexate, leflunomide, sulphasalazine, hydroxychloroquine, azathioprine, cyclosporin, intramuscular gold or penicillamine. Current use of non-steroidal anti-inflammatory drugs (NSAIDs) included use of celecoxib, diclofenac, ibuprofen, indomethacin, ketoprofen, meloxicam, naproxen, piroxicam or any other NSAID. Current use of opioids included use of aspirin and codeine, paracetamol and codeine, dextropropoxyphene, oxycodone, OxyContin, morphine or tramadol.
The ARAD questionnaire also contains a global evaluation of disease activity section in which patients are asked to indicate their level of pain and overall arthritis activity in the past week on a 0–100 visual analogue scale (0 indicates no pain/arthritis activity, and 100 indicates pain as bad as it could be/extreme arthritis activity). In addition to other measures of health-related quality of life, the questionnaire contains the HAQ [13]. HAQ scores range from 0 to 3, with higher scores reflecting greater disability [14]. These variables were also time-varying.
Statistical analysis
Descriptive statistics were used to determine the patterns of oral GC use at baseline and throughout follow-up. It was hypothesised that GC use might vary according to the date of the baseline questionnaire. Prior to the availability of bDMARDs, there were limited treatment options for patients with RA with ongoing disease activity despite maximal csDMARD therapy. GC use may have been different in these patients who would have joined ARAD in the years closest to its inception, compared with those who joined in more recent years, when bDMARDs were more readily available. Population-averaged logistic regression (generalised estimating equation model) and transition state analysis were used to assess change in GC use over time, according to the date of baseline questionnaire. Date of entry (DOE) categories were created according to the date of the baseline questionnaire: 12 September 2001–15 March 2005, 15 March 2005–15 September 2008, 15 September 2008–15 March 2012, or 15 March 2012–6 October 2015.
A multivariable fixed-effects panel regression model was used to examine whether oral GC current use was associated with current age; disease duration; self-reported pain score; self-reported arthritis activity score; HAQ score; and current medication use, including bDMARDs, csDMARDs, NSAIDs and opioids. Age, self-reported pain score and self-reported arthritis activity score were transformed (divided by 10) for ease of interpreting the results. A fixed-effects model was chosen over a random effects model on the basis of the Hausman test. The fixed-effects model allows within-patient comparisons so that each patient is effectively acting as his or her own control.
Univariate transition state analysis was used to assess how these same factors influenced the HR of either commencing or ceasing oral GCs, with HRs relative to the first time point. In this analysis, two transition states were of interest: the transition from GC non-use at one visit to GC use at the next visit, and the transition from GC use at one visit to GC non-use at the next visit.
The fixed-effects panel regression model and transition state analyses included all patients with at least one follow-up visit after baseline. The panel regression model excluded those who were either on oral GCs at all visits or off oral GCs at all visits. Regression models were carried out using Stata version 12.1 software (StataCorp, College Station, TX, USA). The transition state analysis was carried out using R version 3.2.3 software (library msm version 1.6.4) [15, 16].