Article Text

FRI0445 The performance of matrix-based risk models for rapid radiographic progression in an observational cohort of established rheumatoid arthritis patients
  1. S. Lillegraven1,2,
  2. N. Paynter1,
  3. F.H. Prince1,
  4. N.A. Shadick1,
  5. E.A. Haavardsholm2,
  6. M.L. Frits1,
  7. C.K. Iannaccone1,
  8. T.K. Kvien2,
  9. M.E. Weinblatt1,
  10. D.H. Solomon1
  1. 1Brigham and Women’s Hospital, Boston, United States
  2. 2Diakonhjemmet Hospital, Oslo, Norway


Background Matrix-based clinical risk prediction models have been proposed as a tool to predict rapid radiographic progression (RRP, defined as 5 or more units change in Sharp score/year) in rheumatoid arthritis (RA). Three current models are based on data from clinical trials in early RA, and each model includes four of the following variables: treatment, joint erosions, serological status, swollen joint count, CRP and smoking. These models allow calculation of the predicted probabilities for RRP.

Objectives To test the performance of the three risk models in an observational cohort with established RA.

Methods 478 subjects in BRASS, an observational RA cohort with treatment according to clinical practice, were analyzed. Radiographs from baseline and the 2 year follow-up were scored by the van der Heijde Sharp method. The three matrix-based risk models assessed were: “A” - from ASPIRE1, “B” - from BeSt2 and “C” - from the second year of the SWEFOT trial3. Patients were classified as having RRP (cases) or not (non-cases) and placed in the correct matrix cell for each model, with a corresponding predicted risk of RRP. We calculated the mean predicted probability for cases and non-cases, Hosmer-Lemeshow chi-square statistics, integrated discrimination improvement (IDI) and C-statistics.

Results The median (IQR) age for the 478 subjects was 59 (50, 66) years, disease duration 12 (4, 23) years, swollen joint count 6 (2, 13) and tender joint count 7 (1, 4). 84% were female and 86% had presence of erosions at baseline. The percentage of patients with RRP was 12% (32 of 271) in the synthetic DMARD group and 10% (21 of 207) in the biologic DMARD group (either as monotherapy or in combination with synthetic DMARDs). Based on the Hosmer-Lemeshow statistics, neither of the models fit our data very well. The C-statistics and the IDI indicated that Model B performed slightly better than Model A and C in this cohort, but the ability to discriminate cases and non-cases was still limited.

Conclusions Matrix risk models developed in randomized clinical trials of patients with early RA had limited value in this observational cohort of RA patients with established disease. Limitations of this study include lack of feet radiographs, which might have led to fewer patients being classified as RRP, and potential confounding by treatment indication. The value of matrix risk models for RRP might be greater in early RA, or it may be necessary to develop separate risk models for RRP in established disease.

  1. Vastesaeger et al, Rheumatology 2009

  2. Visser et al, ARD 2010

  3. Engström et al, EULAR 2011 (abstract)

Disclosure of Interest S. Lillegraven: None Declared, N. Paynter: None Declared, F. Prince: None Declared, N. Shadick Grant/Research support from: AMGEN, Abbott, Genentech, Cresc. Biosci., MedImm., Biogen Idec, E. Haavardsholm: None Declared, M. Frits: None Declared, C. Iannaccone Grant/Research support from: Cresc. Biosci., MedImmune, T. Kvien: None Declared, M. Weinblatt Grant/Research support from: Biogen Idec, Cresc. Biosci., MedImm., Consultant for: Biogen Idec, Cresc. Biosci., MedImm., D. Solomon Grant/Research support from: Amgen, Abbott, Lilly, Consultant for: CORRONA

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.