Article Text

Concise report
Effectiveness of initial treatment allocation based on expert opinion for prevention of rapid radiographic progression in daily practice of an early RA cohort
  1. Anne Durnez1,
  2. Geert Vanderschueren2,
  3. Luc Lateur2,
  4. René Westhovens1,
  5. Patrick Verschueren1
  1. 1Department of Rheumatology, University Hospitals Leuven, Leuven, Belgium
  2. 2Departments of Radiology, University Hospitals Leuven, Leuven, Belgium
  1. Correspondence to Professor Patrick Verschueren, Department of Rheumatology, University Hospital Leuven, Herestraat 49, B-3000 Leuven, Belgium; patrick.verschueren{at}


Objectives To evaluate expert treatment selection for early rheumatoid arthritis and to validate a prediction model for rapid radiographic progression (RRP) in daily practice.

Methods Patients received initial combination therapy with steroids (ICTS) or disease-modifying antirheumatic drug monotherapy (IMT) after informal evaluation of prognostic factors, followed by a tight control strategy. Changes in Sharp/van der Heijde score (total Sharp score (TSS)) of >5 units over 1 year (=RRP) were documented. The mean change in TSS and proportion with RRP were compared between groups. Based on the 28 swollen joint count, rheumatoid factor titre and C reactive protein/erythrocyte sedimentation rate, patients were placed in the ASPIRE prediction matrix, yielding a RRP risk. Numbers needed to treat (NNT) intensively to avoid one RRP after 1 year were calculated.

Results The mean change in TSS after 1 year and the proportion with RRP was lower in the ICTS group (n=37) than in the IMT group (n=43). The mean calculated risk of RRP was higher in patients with radiographic progression. The mean NNT intensively to prevent RRP was lower in the ICTS group than in the IMT group. The positive predictive value of NNT for RRP prevention was 12.6%, but the negative predictive value reached 100%.

Conclusion ICTS seems more effective in preventing RRP than IMT. The predictive matrix model could be helpful in preventing overtreatment in practice.

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.


Combinations of disease-modifying antirheumatic drugs (DMARDs) with glucocorticoids or biological agents have shown superiority to DMARD monotherapy in patients with early rheumatoid arthritis (RA).1,,9 We have shown that step-down treatment, combining DMARDs and glucocorticoids, is feasible in daily practice.10 In this paper we present the 1-year radiographic data of our patient cohort.

The course of RA disease varies considerably between patients. Early treatment response is predictive for outcome.11 ,12 However, from baseline data, the individual disease course is not easily predicted.13 In a recent publication Vastesaeger et al14 described a prediction model for rapid radiographic progression (RRP) in early RA using data from the Active-Controlled Study of Patients Receiving Infliximab for Treatment of Rheumatoid Arthritis of Early Onset (ASPIRE) study.6 The selected risk factors were combined as trichotomous variables in a matrix with 27 cells, each corresponding to a certain probability of RRP after initial monotherapy or combination therapy. This matrix model remains to be validated in a population of patients with early RA in daily practice.

To date, initial treatment decisions in daily practice are essentially based on expert opinion. In the Leuven early RA clinic, treatment allocation is informally inspired by prognostic factors.10 In this study we investigate whether the application of a prediction model for radiographic progression, such as the ASPIRE matrix, would improve therapeutic decision-making.



Patients with DMARD-naive early RA were recruited at the Department of Rheumatology of the University Hospitals Leuven.10 All consecutive patients were included unless they participated in randomised controlled trials (RCTs).

Treatment allocation was free but inspired by the prognostic profile. Patients with rheumatoid factor (RF), anti-citrullinated protein antibodies (ACPA), erosions or active disease preferentially received initial DMARD combination therapy with steroids (ICTS) and the others received initial DMARD monotherapy (IMT). The ICTS group received a modified Combinatie Therapie bij Reumatoïde Artritis (COBRA) treatment.1 The IMT group started DMARD monotherapy. Patients were evaluated at least every 4 months and treatment adjustments were made whenever the target (disease activity score in 28 joints (DAS28) <3.2) was not reached, if feasible and desirable. The disease course of this population was previously published in part.10 ,12 ,15

Radiographic scoring

Radiographs of hands, wrists and feet were obtained at baseline and 1 year. Images were scored according to the modified Sharp/van der Heijde method.16 A total Sharp score (TSS) progression of ≥5 after 1 year was considered RRP. Mean change in TSS, erosion and joint space narrowing score were compared between the ICTS and IMT groups. The proportion of patients with RRP was compared between groups.

Predictive performance of the matrix model in the Leuven population

Patients were systematically placed in one of the 27 cells of the ASPIRE matrix depending on swollen joint count, RF, C reactive protein/erythrocyte sedimentation rate and therapeutic modality, yielding a RRP probability. The mean RRP probability associated with the expert's initial treatment choice was compared between patients with and without RRP. For each risk factor combination, numbers needed to treat (NNT) to avoid one case of RRP were calculated based on the difference between the probabilities associated with initial monotherapy or combination therapy. These NNT values were organised in an NNT matrix. The predictive performance of NNT for RRP was assessed by ROC curve analysis and the optimal NNT cut-off for RRP prevention was determined.


Eighty patients were recruited from 2001 to 2007. The ICTS group (n=37) received a modified COBRA regimen (2 g sulfasalazine (SS) daily, 15 mg methotrexate (MTX) weekly and a step-down dosage of oral prednisolone) except in two cases where MTX was left out (pregnancy, interstitial lung disease). The IMT group (n=43) started monotherapy with MTX (60%), SS (33%), hydrochloroquine or azathioprine (7%). During follow-up, 65% of the IMT group received low-dose steroids at some point and 19% stepped up to a traditional DMARD combination. Baseline demographics and disease characteristics are summarised in table 1. Patients recruited in RCTs (n=20) during the same time period (PREMIER,5 n=1; CIERA,9 n=12; COMET,17 n=6; AGREE,7 n=1) and the ASPIRE population were added as a benchmark. Patients prescribed ICTS had a higher disease activity and more were ACPA-positive than in the IMT group. The RCT and ASPIRE population had a more severe disease profile.

Table 1

Baseline demographics and disease characteristics

DAS28 improved in both treatment groups, but significantly more in the ICTS group than in the IMT group at month 4 (see figure S1 in online supplement).

The proportion of patients with RRP in our observational study (5/80) was lower than in the RCT group (2/20, table 2). The proportion with RRP was lower in the ICTS group (1/37) than in the IMT group (4/43).

Table 2

Radiographic evolution of the Leuven patient population

Despite a more severe disease profile at baseline, the mean (SD) change in TSS score after 1 year was lower in the ICTS group than in the IMT group (p=NS). Similarly, the mean change in erosion score and joint space narrowing score tended to be lower in the ICTS group than the IMT group. Differences between the ICTS and IMT groups and between the observational and RCT population did not reach statistical significance.

As expected, patients with RRP after 1 year tended to occupy cells closer to the upper right corner of the ASPIRE matrix, corresponding to a higher probability of RRP (table 3). The mean calculated probability of RRP was higher in the group of patients who actually developed radiographic progression than in those who did not (see table S1 in online supplement). According to the matrix, the risk taken in patients with RRP was higher than the average risk taken by the experts in our total cohort at treatment initiation.

Table 3

Spatial visualisation of the Leuven observational population in the ASPIRE matrix model depending on initial treatment

Patients with RRP occupied cells closer to the upper right corner of the NNT matrix, corresponding to lower NNT (table 3). The mean NNT intensively to prevent RRP was lower in the ICTS group than in the IMT group, but also in patients with RRP compared with those without RRP (see table S1 in online supplement).

Starting from the NNT to prevent one case of RRP using intensive compared with less intensive therapy, a predictive model for RRP in our population was built. The positive predictive value of the NNT for RRP prevention was 12.6%, but the negative predictive value reached 100% (sensitivity 100%, specificity 53.3%) with 9.17 as a cut-off for NNT. In half of the population (40/80) the NNT was above this level (table 3).


In this study we investigated the effectiveness of initial treatment allocation based on expert opinion for prevention of RRP in a cohort of patients with early RA.

The proportion of patients with RRP was 6.3% (5/80), all therapeutic regimens included. This is similar to the 8.3% for a combination of infliximab and MTX and lower than the 22.8% for MTX monotherapy in the ASPIRE trial. This favourable result can partly be explained by population differences. Patients in the ASPIRE study had more active disease and more were erosive, but other prognostic parameters were similar. More importantly, we can assume that treatment allocation based on the prognostic profile followed by systematic treatment adaptations will lead to better results compared with an RCT setting.

Patients given ICTS had less RRP than those receiving IMT, despite more severe baseline disease characteristics. This is in line with our previous observations and confirms in daily practice the observations of several RCTs.1 ,3 ,8,,10 Most cases of RRP were observed in the IMT group, suggesting a tendency for undertreatment in so-called ‘good prognosis’ patients. The calculated matrix risk in this subgroup was clearly higher than the population average, which illustrates the limitations of informal risk factor evaluation but also the difficulties in complying with theoretical concepts in daily practice. In real life, factors such as comorbidities, conception desires and patient preferences have to be taken into account. However, the relatively low rate of RRP in the IMT group compared with intensive therapy in the ASPIRE study (9.3% vs 8.2%) confirms the effectiveness of tight control, even after a less intensive initial treatment choice.

It appears that in our population the predictive capacity of the ASPIRE model for RRP is low. This might result from selection bias, leading to worse disease profiles in RCTs. Another explanation might be confounding by indication. Preferentially giving intensive therapy in high-risk patients can overrule the effect of prognostic factors.18 ,19 Also, a tight follow-up with therapeutic adjustments can reduce radiographic progression compared with trials with a fixed therapy.

Whereas the matrix does not allow identifying patients at high risk, a low-risk group can be defined. In our population, patients with a risk profile corresponding to a theoretical NNT of >9.17 to prevent RRP by giving intensive therapy can be 100% sure to be spared of radiographic progression even if treated with monotherapy initially (negative predictive value 100%). The importance of this finding cannot be underestimated since this represents half of our population in which the model might allow preventing overtreatment.

Our results should be interpreted with caution given the sample size, but the homogeneity of this monocentric study might also enforce the conclusions.

A new model, better suited for daily practice, would probably incorporate valuable predictors such as ACPA and baseline erosions, like the matrix proposed by Visser.20 However, to be helpful in daily practice, it should be built from a real-life population, avoiding selection bias and taking into account potential treatment adaptations.

Before more accurate prognostic models become available, expert opinion remains reliable for the initial treatment choice, provided it is followed by a tight control approach.


The authors acknowledge the health professional staff of the Rheumatoid Arthritis Care Program of the University Hospitals Leuven and the patients for their participation in the study, and thank Nathan Vastesaeger, Wim Noël and Katrien Van Beneden for their inspiring help.


Supplementary materials


  • Funding This study was supported by a research grant from Centocor BV Medical Affairs.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval This study was conducted with the approval of the ethics committee of the University Hospitals Leuven.

  • Provenance and peer review Not commissioned; externally peer reviewed.