Article Text

OP0247 An improved matrix to predict rapid radiographic progression of early rheumatoid arthritis patients: pooled analyses from several databases
  1. A Vanier1,
  2. J Smolen2,
  3. C Allaart3,
  4. R Van Vollenhoven4,
  5. P Verschueren5,
  6. N Vastesaeger6,
  7. S Saevarsdottir4,
  8. K Visser3,
  9. D Aletaha2,
  10. B Combe7,
  11. B Fautrel1
  1. 1UPMC Univ Paris 6 and AP-HP Pitié-Salpêtrière Hospital, Paris, France
  2. 2University of Vienna, Vienna, Austria
  3. 3Leiden University Medical Centre, Leiden, Netherlands
  4. 4Karolinska Institute, Stockholm, Sweden
  5. 5Leuven University, Leuven, Belgium
  6. 6Merck Sharp and Dohme, Lucerne, Switzerland
  7. 7Université de Montpellier 1, Montpellier, France


Background In early rheumatoid arthritis (RA), some patients exhibit rapid radiographic progression (RRP) after one year, associated with poor functional prognosis. Identifying at the time of diagnosis the characteristics predictive of RRP is of importance. Several matrices predicting this risk have been proposed over the last years. They were limited somewhat in terms of precision or were built using specific populations.

Objectives To develop a matrix to predict RRP with better precision and generalizability by pooling databases from various studies.

Methods The study is based on the pooling of individual data from cohorts (ESPOIR and Leuven) and clinical trials (BeSt, SWEFOT and ASPIRE). Included patients were adult DMARD-naïve patients with recent suspected or confirmed diagnosis of active RA for which the first therapeutic strategy after inclusion was to prescribe MTX or leflunomide in monotherapy for at least 3 months. The main outcome was the presence after one year of RRP defined as an increase in modified Sharp score (vSHS) of at least 5 points between baseline and year one. Baseline characteristics were compared by the presence of RRP to search for predictors. A logistic regression model to predict RRP was built. The best model was selected by 10-fold stratified cross-validation by maximizing the Area Under the Curve (AUC). Calibration and discriminatory power of the model were assessed. Model parameters were extracted to estimate the probability of a RRP for each combination of level of baseline characteristics.

Results The data of 1306 patients were pooled. After one year, 236 exhibited RRP (20.6%, CI95% [18.2–22.9], mean probability of RRP of 0.21). Model of prediction of RRP included as baseline characteristics Rheumatoid Factor (RF) positivity (OR=2.1 CI95% [1.5–3.0]; p<0.001), erosive disease on X-rays (OR=2.3 CI95% [1.7–3.2]; p<0.001), CRP>30mg/l (OR=2.1 CI95% [1.5–3.0]; p<0.001), number of swollen joints>10 (OR=1.5 CI95% [1.01–2.2]; p=0.048). Model calibration was good (Hosmer and Lemeshow test: p=0.79). AUC was 0.68. The matrix proposes estimated RRP probability for 36 combinations of level of baseline characteristics. Its range goes from patients with a 4.1 fold lower risk of RRP compared to average risk (probability of 0.05 CI95% [0.03–0.08], patient with CRP <10mg/l, without RF, without erosive disease on X-ray, with <6 swollen joints), to patients with a 2.3 fold higher risk than average (probability of 0.47 CI95% [0.39–0.55], patient with CRP >30, with RF, with erosive disease on X-ray, with >10 swollen joints).

Conclusions A matrix proposing RRP probability at one year with better precision (i.e. narrower CI95%than those previously published) in early RA for various combinations of levels of a few common baseline characteristics has been built using several databases. However, discriminating power is not ideal. Further investigations will be needed to fully explore the potential complexity of predicting RRP in early RA.

Disclosure of Interest None declared

Statistics from

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.