Study design, patient selection, and outcomes
The Induction therapy with Methotrexate and Prednisone in Rheumatoid Or Very Early arthritic Disease (IMPROVED) study is a multicenter, randomized controlled trial that enrolled 610 patients with early (< 2 years) untreated RA or undifferentiated arthritis. It was steered at disease activity score-remission (DAS44 < 1.6) and for those achieving remission, at drug-free remission (DFR), with treatment adjusted every 4 months according to whether treatment targets had been reached. Initial treatment comprised methotrexate (MTX) and high-dose prednisone [4].
Subjects selected for this study were all 381 patients fulfilling the 2010 ACR/EULAR RA criteria with serum available at least once within the first year and seropositive by routine clinical testing for anti-CCP2 IgG, RF IgM, or our in-house assay for anti-CarP IgG at baseline or at 1 year (details described in [9]). Clinical outcomes investigated were DAS, health assessment questionnaire (HAQ), EULAR response at 4 and 12 months, and long-term sustained DFR. Long-term sustained DFR was defined as disease-modifying anti-rheumatic drug (DMARD)-free remission lasting at least 1 year, starting at any time point and continuing until the last moment of that individual’s follow-up (maximum of 5 years follow-up). Radiographic progression at 1 and 5 years was assessed using the Sharp/van der Heijde Score (SHS), as previously described [4].
Serological measurements
Enzyme-linked immunosorbent assays (ELISAs) were used as described previously [9] to measure at 4-month intervals over the first year of treatment: anti-CCP2 IgG, IgM, and IgA; RF IgM and IgA; anti-CarP IgG, IgM, and IgA; anti-citrullinated-vimentin 59–74 IgG, anti-citrullinated-fibrinogen β 36–52 and α 27–43 IgG, and anti-citrullinated-enolase 5–20 IgG (all in-house assays); and anti-acetylated lysine vimentin IgG and anti-acetylated ornithine vimentin IgG (Orgentec Diagnostika GmbH, Germany) [1]. Samples were considered positive if they fell above a cutoff of the mean arbitrary units (aU) per milliliter plus two standard deviations of 76 sera of healthy controls from the Leiden area.
Composites reflecting the number of autoantibodies present at every time point were constructed: the number of isotypes: anti-CCP2 IgG, IgM, IgA; RF IgM and IgA; anti-CarP IgG, IgM, IgA (range 1–8), and the number of IgG anti-modified peptide antibodies (AMPAs): anti-CCP2, anti-CarP, and the antibodies against citrullinated and acetylated peptides described above (range 1–8).
Statistical analysis
Longitudinal, repeated measures data (autoantibodies, DAS, HAQ) were modeled using generalized estimating equations (GEE), which allow missing data in the outcome. A model with a Toeplitz (m-dependent) correlation structure and a standard Gaussian distribution was chosen (akin to linear regression). For the number of autoantibodies over time (count data), a negative binomial model was specified.
With repeated measurements of the autoantibody levels/number as the dependent variable, we investigated by GEE whether a certain treatment decision (4 months, no change versus escalation of treatment; 8 months: tapering versus escalation of treatment) was associated with a subsequent change in autoantibody levels/number. An interaction term of treatment decision * time was used to assess whether changes in the autoantibody levels over time were different in patients that tapered/did not change versus those that escalated immunosuppressive therapy. For comparison purposes, a normalization of the different measurement units was applied to the final model estimates (which were in aU/mL) by dividing them by the maximum of the autoantibody’s range.
The association between autoantibody levels and DAS over time was investigated using GEE, with DAS as the outcome. The same was conducted for HAQ over time. Ordinal, logistic, and linear regression was used to investigate the association of relative changes in autoantibody levels/absolute changes in number of autoantibodies with EULAR response, long-term sustained DFR, and SHS radiographic progression scores, respectively.
All models were adjusted for gender and age; clinical outcome analyses were adjusted for treatment decisions. Other covariates (i.e., disease duration, smoking, body mass index (BMI), baseline HAQ/DAS) were only included in final models if they were univariably associated with the outcomes of interest (p < 0.1). Holmes-Bonferroni methods were used to correct all analyses for multiple testing, assuming the same number of tests as autoantibodies investigated (14 tests).