Downloaded from http://ard.bmj.com/ on January 29, 2015 - Published by group.bmj.com
ARD Online First, published on January 3, 2014 as 10.1136/annrheumdis-2013-204024 Clinical and epidemiological research
EXTENDED REPORT
Performance of four current risk algorithms in predicting cardiovascular events in patients with early rheumatoid arthritis E E A Arts,1 C Popa,1 A A Den Broeder,2 A G Semb,3 T Toms,4 G D Kitas,4 P L van Riel,1 J Fransen1 Handling editor Iain McInnes 1
Department of Rheumatology, Radboud University Medical Centre, Nijmegen, The Netherlands 2 Department of Rheumatology, Maartenskliniek, Nijmegen, The Netherlands 3 Department of Rheumatology, Diakonhjemmet Hospital, Oslo, Norway 4 Department of Rheumatology, Dudley NHS Hospital Group, Dudley, UK Correspondence to Elke E A Arts, Department of Rheumatology, Radboud University Medical Centre, Geert Grooteplein 8, Nijmegen 6500 HB, The Netherlands; [email protected] Received 27 May 2013 Accepted 26 November 2013
ABSTRACT Objective This study was undertaken to assess the predictive ability of 4 established cardiovascular (CV) risk models for the 10-year risk of fatal and non-fatal CV diseases in European patients with rheumatoid arthritis. Methods Prospectively collected data from the Nijmegen early rheumatoid arthritis (RA) inception cohort was used. Discriminatory ability for CV risk prediction was estimated by the area under the receiver operating characteristic curve. Calibration was assessed by comparing the observed versus expected number of events using Hosmer-Lemeshov tests and calibration plots. Sensitivity and specificity were calculated for the cut-off values of 10% and 20% predicted risk. Results Areas under the receiver operating characteristic curve were 0.78–0.80, indicating moderate to good discrimination between patients with and without a CV event. The CV risk models Systematic Coronary Risk Evaluation (SCORE), Framingham risk score (FRS) and Reynolds risk score (RRS) primarily underestimated CV risk at low and middle observed risk levels, and mostly overestimated CV risk at higher observed risk levels. The QRisk II primarily overestimated observed CV risk. For the 10% and 20% cut-off values used as indicators for CV preventive treatment, sensitivity ranged from 68–87% and 40–65%, respectively and specificity ranged from 55–76% and 77–88%, respectively. Depending on the model, up to 32% of observed CV events occurred in patients with RA who were classified as low risk (10 years, rheumatoid factor or anticyclic citrullinated peptide positivity and presence of extra-articular manifestations. Recently, an updated version of the QRisk algorithm was developed by Hippisley-Cox et al.3 This QRisk II algorithm includes RA as an independent risk factor. However it is not known whether the M-SCORE, RRS and QRisk II risk algorithms predict future CV events in patients with RA more accurately compared with the SCORE or the FRS. The predictive performance—that is, the accuracy of predictions of future CV event(s)—of these risk algorithms has not been evaluated and compared in European patients with RA. Therefore, the objective of this study is to assess the performance of four established CV risk algorithms (SCORE, FRS, RRS and QRiskII) for predicting the 10-year risk of fatal and non-fatal CV disease in European patients with RA.
PATIENTS AND METHODS Study design and patients This retrospective study is based on prospectively collected data from the Nijmegen early RA inception cohort. Patients were included at diagnosis of RA (baseline) in the outpatient clinic of the
Arts EEA, et al. Ann Rheum Dis 2014;0:1–7. 1 Copyright Article author (or theirdoi:10.1136/annrheumdis-2013-204024 employer) 2014. Produced by BMJ Publishing Group Ltd (& EULAR) under licence.
Downloaded from http://ard.bmj.com/ on January 29, 2015 - Published by group.bmj.com
Clinical and epidemiological research departments of rheumatology of the Radboud University Medical Centre (since 1985) or the Maartenskliniek in Nijmegen (since 1990). At inclusion, patients had a disease duration of
ARD Online First, published on January 3, 2014 as 10.1136/annrheumdis-2013-204024 Clinical and epidemiological research
EXTENDED REPORT
Performance of four current risk algorithms in predicting cardiovascular events in patients with early rheumatoid arthritis E E A Arts,1 C Popa,1 A A Den Broeder,2 A G Semb,3 T Toms,4 G D Kitas,4 P L van Riel,1 J Fransen1 Handling editor Iain McInnes 1
Department of Rheumatology, Radboud University Medical Centre, Nijmegen, The Netherlands 2 Department of Rheumatology, Maartenskliniek, Nijmegen, The Netherlands 3 Department of Rheumatology, Diakonhjemmet Hospital, Oslo, Norway 4 Department of Rheumatology, Dudley NHS Hospital Group, Dudley, UK Correspondence to Elke E A Arts, Department of Rheumatology, Radboud University Medical Centre, Geert Grooteplein 8, Nijmegen 6500 HB, The Netherlands; [email protected] Received 27 May 2013 Accepted 26 November 2013
ABSTRACT Objective This study was undertaken to assess the predictive ability of 4 established cardiovascular (CV) risk models for the 10-year risk of fatal and non-fatal CV diseases in European patients with rheumatoid arthritis. Methods Prospectively collected data from the Nijmegen early rheumatoid arthritis (RA) inception cohort was used. Discriminatory ability for CV risk prediction was estimated by the area under the receiver operating characteristic curve. Calibration was assessed by comparing the observed versus expected number of events using Hosmer-Lemeshov tests and calibration plots. Sensitivity and specificity were calculated for the cut-off values of 10% and 20% predicted risk. Results Areas under the receiver operating characteristic curve were 0.78–0.80, indicating moderate to good discrimination between patients with and without a CV event. The CV risk models Systematic Coronary Risk Evaluation (SCORE), Framingham risk score (FRS) and Reynolds risk score (RRS) primarily underestimated CV risk at low and middle observed risk levels, and mostly overestimated CV risk at higher observed risk levels. The QRisk II primarily overestimated observed CV risk. For the 10% and 20% cut-off values used as indicators for CV preventive treatment, sensitivity ranged from 68–87% and 40–65%, respectively and specificity ranged from 55–76% and 77–88%, respectively. Depending on the model, up to 32% of observed CV events occurred in patients with RA who were classified as low risk (10 years, rheumatoid factor or anticyclic citrullinated peptide positivity and presence of extra-articular manifestations. Recently, an updated version of the QRisk algorithm was developed by Hippisley-Cox et al.3 This QRisk II algorithm includes RA as an independent risk factor. However it is not known whether the M-SCORE, RRS and QRisk II risk algorithms predict future CV events in patients with RA more accurately compared with the SCORE or the FRS. The predictive performance—that is, the accuracy of predictions of future CV event(s)—of these risk algorithms has not been evaluated and compared in European patients with RA. Therefore, the objective of this study is to assess the performance of four established CV risk algorithms (SCORE, FRS, RRS and QRiskII) for predicting the 10-year risk of fatal and non-fatal CV disease in European patients with RA.
PATIENTS AND METHODS Study design and patients This retrospective study is based on prospectively collected data from the Nijmegen early RA inception cohort. Patients were included at diagnosis of RA (baseline) in the outpatient clinic of the
Arts EEA, et al. Ann Rheum Dis 2014;0:1–7. 1 Copyright Article author (or theirdoi:10.1136/annrheumdis-2013-204024 employer) 2014. Produced by BMJ Publishing Group Ltd (& EULAR) under licence.
Downloaded from http://ard.bmj.com/ on January 29, 2015 - Published by group.bmj.com
Clinical and epidemiological research departments of rheumatology of the Radboud University Medical Centre (since 1985) or the Maartenskliniek in Nijmegen (since 1990). At inclusion, patients had a disease duration of
