Article Text


Extended report
Optimisation of a treat-to-target approach in rheumatoid arthritis: strategies for the 3-month time point
  1. Daniel Aletaha1,
  2. Farideh Alasti1,
  3. Josef S Smolen1,2
  1. 1Division of Rheumatology, Medical University Vienna, Vienna, Austria
  2. 2Second Department of Medicine, Hietzing Hospital, Vienna, Austria
  1. Correspondence to Professor Daniel Aletaha, Division of Rheumatology, Department of Internal Medicine 3, Medical University Vienna, Vienna 1090, Austria; daniel.aletaha{at}


Background Treat-to-target (T2T) is a widely accepted management strategy for rheumatoid arthritis (RA) with a key decision point at 3 months after treatment initiation. At this time point, it remains unclear which patients will benefit from treatment adaptation or from continuation of existing treatment.

Methods We performed a pooled analysis of patient-level clinical trial data of patients with RA. We used a diagnostic testing methodology and a probabilistic approach employing logistic regression to investigate which levels of response at 3 months can inform treatment decisions in regard to achieving the target at 6 months.

Results To be at least 80% sensitive for achieving the low disease activity (LDA) target at 6 months, a change at 3 months in Simplified Disease Activity Index/Clinical Disease Activity Index (SDAI or CDAI) of 58% needs to be observed at 3 months. Higher changes are needed to sensitively predict remission (REM). Not reaching the (minor) SDAI 50% response level is afflicted with very low negative likelihood ratios (LRs) (0.28 for LDA and 0.07 for REM at 6 months). Experiencing (major) SDAI 85% response has substantial positive LRs of 9.2 for reaching LDA and 6.2 for reaching REM at 6 months. In logistic regression, the change at 3 months is significantly associated with reaching of the target at 6 months.

Conclusions The 3-month time point is a critical decision point. Not achieving minor responses at 3 months makes reaching of the treatment target at 6 months highly unlikely, while reaching major responses is highly predictive of reaching the treatment target.

Statistics from

Rheumatoid arthritis (RA) is characterised by its typical joint inflammation that, if not treated effectively, leads to damage and disability over time. One effective treatment concept that has been successfully used in other diseases, such as diabetes, is termed ‘treat-to-target’ (T2T).1 In RA, T2T suggests that a disease activity target should be set in all patients upon start of therapy, and adaptations should be made if the target is not reached.

While the T2T concept was based on the available literature at the time, it was not sufficiently precise when it came to the type of target (‘remission (REM) or low disease activity (LDA)’) or time prerequisites (‘3 to 6 months’). It was made clear that at the latest by 6 months a good clinical state must be present, but clinical responses should occur within the first 3 months, since there are a number of patients who continue to further improve between 3 and 6 months of the initiation of a new treatment course.2 Patients who have already significantly improved by 3 months will be the ones who should get the chance to show further improvement over the subsequent 3 months to possibly reach their target by 6 months, before their treatment is switched. In fact, responses seen at 3 months have been shown to be a good indicator of REM status at 12 months.3–5 Most importantly, patients in whom disease activity had not improved or improved only to a small extent are considered much less likely to reach the target at a later time point and would so benefit from a change of therapy already at 3 months.

Indeed, the updated T2T recommendations suggest a goal of significant improvement at 3 months and attainment of the treatment target at 6 months.6 The purpose of the present study was to investigate when using the T2T approach what levels of improvement at 3 months are insufficient, and thus require treatment adaptations, and what levels warrant a continuation of the initial regimen for another 3 months even though the target has not yet been reached.

Patients and methods

Data sources

We used patient-level clinical trial data from the past decade. These were pivotal randomised controlled trials in RA, from which data on a random sample of 80–90% of enrolled patients were kindly provided by the respective sponsors. These included the ASPIRE,7 ERA,8 PREMIER9 and IMAGE10 trials of early RA, and the ATTRACT,11 DE01912 and GO-FORWARD13 trials of established RA. Trials comprised of treatment with methotrexate (MTX) or biologicals, or their combination, and data were pooled according to these treatment groups across the trials. More details can be found in the online supplementary methods and table S1.


We obtained data on baseline demographics and core set measures of baseline RA disease activity. From these variables, the Simplified and Clinical Disease Activity Indices (SDAI, CDAI) and the Disease Activity Score 28 (DAS28) were calculated (also see online supplementary methods). In two trials (ATTRACT and DE019), no erythrocyte sedimentation rate (ESR) measures were available, and DAS28-ESR could therefore not be calculated. We calculated absolute and relative changes from baseline to the 3-month time point for all three indices, and applied established cut points to identify patients in American College of Rheumatology (ACR)-European League Against Rheumatism (EULAR) index-based REM (SDAI≤3.3; CDAI≤2.8) or in DAS28 remission; (DAS28<2.6); or patients in LDA or REM (SDAI≤11; CDAI≤10 and DAS28<3.2) at 6 months.14 For calculation of these indices, handling of missing data and exclusion criteria, please see online supplementary methods.


Prior to addressing the main question of our study, we descriptively looked at patients who improved their disease activity state between baseline and 3 months (eg, moving from high to moderate disease activity or from moderate to low). We then calculated which changes in the absolute disease activity scales these switches across disease activity states corresponded to. We also identified the proportion of patients in each group, who were eventually reaching the treatment target at the 6-month time point.

In the main analysis section, we used two different approaches for studying which disease activity responses at 3 months were relevant to predict achievement of treatment targets at 6 months: one was a diagnostic testing approach with receiver operating characteristic (ROC) curve analysis to investigate whether changes during the first 3 months of therapy can be used as tests for achievement of a good state at 6 months. These analyses generated the readout of the area under the ROC curves (as a measure of overall test performance), as well as sensitivities, specificities, positive predictive value (PPV) and negative predictive value (NPV) and likelihood ratios (LRs) for all possible improvement cut points (0–100% improvement) in the disease activity metric.

The second approach was a probabilistic one, using a logistic regression model to predict achievement of 6-month treatment targets (dependent variable) by 3-month changes in the respective disease activity index (independent variable). The readout of these models were ORs (per increment of change in the disease activity scale), and more importantly, splines allowing to address the clinically relevant estimation of the probability of failing to reach the target at 6 months depending on an observed change at 3 months. All these tests and models were performed using SDAI, CDAI, DAS28-ESR or DAS28-C reactive protein (CRP) as the disease activity metrics, and LDA or REM as the target at 6 months. The target of LDA includes remitters.

Prior to these analyses, we used the diagnostic testing approach above to investigate whether relative or absolute changes were more appropriate parameters for use as tests by comparing their area under the ROC curves. Similarly, we investigated whether the type of treatment or the RA population studied had an influence on the results. Regarding the impact of type of treatment, we used one specific analysis setting (disease activity index: SDAI; target at 6 months: LDA) and compared ROC curves analyses across different study arms (within the early RA populations only): MTX monotherapy (MTX naïve), biological monotherapy (MTX-naïve) and the combination of MTX+biological (MTX-naïve). Regarding the impact of study population, we then analogously compared the area under the ROC curves of the MTX+biological groups of the early and the established populations (the latter being MTX-insufficient responders who received add-on biological).

We finally tested the ability of established response criteria at 3 months to predict the presence of the desired target at 6 months. These included the response definitions of the SDAI, the EULAR response definition and the ACR response criteria.15–17


Patient characteristics and overall treatment outcomes

We included a total of 2483 patients (75% women (n=1855), 78% (n=1935) rheumatoid factor positive). Patients had active disease at baseline, with an average SDAI of 43.8 (±14.5), CDAI of 40.7 (±13.1), DAS28-ESR of 6.50 (±1.07) and a health assessment questionnaire of 1.52 (±0.64). The detailed characteristics of the overall study population, and separated by the four treatment groups, can be found in table 1. The details of all trials can be accessed as online supplementary table S1.

Table 1

Patient characteristics in the different treatment groups

At 6 months, 11.4% (n=282) and 41.3% (n=1025) of the patients achieved REM or LDA (including REM), respectively, by the SDAI; 11.8% (n=293) and 40.3% (n=1001), respectively, by the CDAI; 23.1% (n=573) and 39.1% (n=971) by the DAS28-CRP and 19.7% (n=422) and 32.1% (n=689) by the DAS28-ESR.

Determining the value of relative changes versus absolute changes

In an initial step, we found that relative changes showed a significantly better area under the ROC curve than absolute changes for the SDAI (0.80 vs 0.64, respectively) and CDAI (0.80 vs 0.64) and a similar trend for the DAS-based scores ROC curves (DAS28-ESR: 0.80 for relative changes vs 0.74 for absolute changes; DAS28-CRP: 0.79 vs 0.73, respectively). For all subsequent analyses, we therefore focused on the relative changes for all instruments.

Influence of treatment regimen and RA population on the study results

We repeated the analysis in the populations as stratified in table 1: MTX monotherapy (early, MTX naïve), biological monotherapy (early, MTX-naïve), combination of MTX+biological (early, MTX-naïve) and MTX+biological (MTX-insufficient responder population). The first and the last of these groups are key as they correspond to clinical practice (start with MTX, and step up from MTX to combination). The area under the ROC curve using LDA (in parentheses: REM) as the treatment target in the pooled analysis was 0.803 (0.882); for the mentioned subpopulations, it was 0.802 (0.835), 0.801 (0.868), 0.805 (0.897) and 0.833 (0.877), respectively. The data indicate that 3-month changes are more accurate tests for the target of REM, than for the target of LDA, at 6 months, but also that the type of treatment or early versus established RA did not significantly influence this accuracy: the difference of AUCs between treatments was 0.032 for LDA target and 0.062 for REM. For all subsequent analyses, we therefore used the pooled treatment groups.

Improvement in disease activity categories at 3 months and their implications for reaching the target

Table 2 shows the average change scores in patients who improved their disease activity by one or more disease activity categories within the first 3 months after treatment initiation. For the SDAI and CDAI, patients improving by one category experienced a reduction in their disease activity on average by 55–65%, and those improving by two categories by 80–90%. For the DAS28-ESR and DAS28-CRP, improvements by one category were associated with 31–35% change, while two category improvements showed a 51–53% change. The most frequently observed change in disease activity state was a shift from high to moderate disease activity, which was seen in 53–66% of the respective patients for the four indices.

Interestingly, the proportion of patients achieving REM or LDA at 6 months was more strongly related to the state achieved at 3 months than to the number of disease activity categories the patient had improved from baseline to 3 months.

Changes in disease activity at 3 months and their implications for reaching the target

The sensitivities and specificities, as well as PPV and NPV for using different levels of relative change in disease activity at 3 months as a test for achieving LDA or REM at 6 months are shown in figure 1 for the four indices.

Figure 1

Diagnostic test characteristics of relative change scores at 3 months. Sensitivities, specificities and positive and negative predictive values of 3-month relative change scores of the Simplified Disease Activity Index (SDAI) (A and B), Clinical Disease Activity Index (CDAI) (C and D), Disease Activity Score 28 (DAS28)-C reactive protein (CRP) (E and F) and DAS28-erythrocyte sedimentation rate (ESR) (G and H) for predicting low disease activity (A,C,E and G) or remission (B,D,F and H). In each panel, the change score with the highest sensitivity while still maintaining 80% specificity is marked with a red dot, while the change score with the highest specificity while maintaining at least 80% sensitivity is marked with a green dot. The unit-free y-scale of the panels can be multiplied by 100 to obtain performance characteristics in %. For abbreviations, see footnote to table 2.

Identifying the changes with the greatest specificity while still being at least 80% sensitive for reaching LDA at 6 months, these were 58% for the SDAI and CDAI and 30% for the DAS28-based indices (figure 1, left panels, green marks). Vice versa, changes providing the greatest sensitivity while still maintaining a specificity of at least 80% were 65% for SDAI/CDAI and 40% for DAS28 (figure 1, left panels, red marks). To predict REM at 6 months, higher response levels are necessary at 3 months: 70% improvement in SDAI or CDAI (identical for 80% sensitivity and 80% specificity, figure 1) and 34% (80% sensitivity) or 43% (80% specificity) for the DAS-based indices (figure 1, right panels). Given the low prevalence of REM, NPVs remain relatively high throughout, while PPVs increase with increasing size of response.

Table 2

Improvement in disease activity states at 3 months

Predictive capacity of established response definitions

In table 3, we present the test properties of established cut point for RA response. It can be seen that failure to fulfil criteria of minor improvement, such as the ACR20 or the SDAI 50% response, is afflicted with high NPVs and, accordingly, very low negative LRs (0.28 for LDA and 0.07 for REM), making achievement of the target at 6 months highly unlikely. Achieving minor response criteria, on the other hand, is prognostically not helpful given a low PPVs and low positive LRs.

Table 3

Prognostic implications of established response definitions

For the most stringent criteria (ACR70 and SDAI 85%, EULAR good response), it is the opposite: here, failing to fulfil the criteria is not prognostically helpful (based on NPV and LR−), while achieving them at 3 months has quite substantial PPV (up to 87%) and LR+ (of 7.6, 9.2 and 5.5, respectively). For the target of REM, the LR+ of ACR70, SDAI 85% and EULAR good response are 5.3, 6.2 and 4.6, respectively.

Furthermore, predictions were slightly different in the two standard treatment groups of clinical practice, MTX monotherapy in early RA and biological+MTX in late RA. In the latter group, there seems to be a somewhat stronger predictive association with the target at 6 months (see online supplementary tables S2A and S2B).

Probabilistic approach to prediction of achieving treatment goals

Figure 2 depicts the logistic regression models for the different indices and outcomes. It shows that the probability of not reaching the target of LDA decreases clearly and significantly with the extent of relative response at 3 months (figure 2, left panel). For the target of REM, particularly for SDAI/CDAI, large changes in disease activity are necessary at 3 months to decrease the probability of failure (figure 2, right panel). However, in clinical practice, the probability of reaching LDA (including REM; figure 2, left panel) will likely determine the treatment decision at 3 months.

Figure 2

Prediction model of relative change scores at 3 months. Panels depict the probability of not achieving low disease activity (or remission) (A) or remission (B) at 6 months, according to the 3-month relative change scores of the Simplified Disease Activity Index (SDAI), Clinical Disease Activity Index (CDAI), Disease Activity Score 28-erythrocyte sedimentation rate (DAS28-ESR) and DAS28-C reactive protein (CRP).


The first finding of our study is that—regardless of the starting point—6-month success rates are clearly related to the disease activity state reached at 3 months, and not so much to the number of disease activity categories improved. This concept of the importance of a state rather than response has been propagated also by studies in the past looking at functional and structural outcomes, as well as the patient perspective.18 ,19

Based on the results of our study, using response at 3 months as a decision criterion, clear-cut conclusions can be drawn in two clinical situations. First, failure of achieving minor responses (eg, ACR20, SDAI 50) at 3 months has the potential to almost rule out successful attainment of the target at 6 months; second, achievement of major response definitions at 3 months (eg, ACR70, SDAI 85%) can reliably predict achievement of a good state at 6 months. It is also obvious therefore that for many patients in the grey zone between these response levels, the prediction may be less solid, and may—to a greater extent—rest on the physician's decision.

Reliable prediction of REM is more difficult than prediction of LDA, but it is easier to rule out REM than LDA. The choice of the target will depend on a number of patient-related aspects, for example, less chance in patients with long-standing RA1 or isolated (or disease activity-independent) pain limiting achievement of REM20; also the type of treatment is important, as biologicals may inhibit structural progression of RA already in a LDA state.21–24 That said, the meanings of ‘REM’ and ‘LDA’ are different in different instruments25: DAS28 remission apparently more likely reflects LDA,26 ,27 which, in fact, is nicely supported by comparison of the DAS28 probability curves for REM (figure 2B) with the SDAI/CDAI probability curves for LDA (figure 2A).

At the same time, the large amount of change required to specifically predict SDAI/CDAI REM at 6 months (figure 2B) indicates that these stringent targets are generally high bars; this needs to be borne in mind when considering treatment adaptations at 3 months. It is exactly for this stringency (and face validity) that the ACR and EULAR proposed SDAI and CDAI as part of their joint definition of RA REM criteria.28 It is therefore not surprising that the changes required for prediction of 6-month DAS28 REM are only ∼33%, while they are 72% for the SDAI. Also, DAS28 change scores for REM are only 4–5% higher than for LDA, while those required for SDAI REM are 20% higher than those for SDAI LDA. This all supports the frequent notion that DAS28 REM simply does not reflect a clinical state of REM, but rather a state of LDA.29 When we examined cut points for SDAI that would show at least 80% sensitivity for LDA or REM at 6 months, they correspond quite well to the recently established SDAI 50% and SDAI 70%, minor and moderate, improvement criteria,15 respectively.

Several additional factors determine the change of treatment. First, secular trends indicate more aggressive clinical adaptations over the past decade.30 The cost of drug is also an important consideration. However, while this is relevant when deciding when to switch from a conventional synthetic DMARD to a biological DMARD,31 this may be less relevant when switching between—more expensive, but comparably priced—targeted therapies. Also, it needs to be emphasised that T2T approach does not necessarily imply that the current treatment regimen needs to be switched, but speaks about treatment adaptation. Such an adaptation can also be a mere dose increase or route optimisation of existing disease-modifying drugs, application of intra-articular short-term low-dose systemic glucocorticoids or similar.

The limitations of the study include the lack of ESR measurement in some of the trials, which gave lower power for the DAS28-ESR-based analyses. However, we wanted to include these analyses, as the DAS28 is still a widely used disease activity instrument. Further, none of the studies assessed was dedicated to a T2T strategy; however, this fact actually allowed us to assess the ‘natural course’ of disease activity from 0 to 3 to 6 months after treatment initiation without adaptations. In this way, we were able to determine the degree of initial change below which the treatment should have been adapted according to the T2T principles. Finally, clinical trial data do not necessarily reflect all situations in clinical practice, and especially for MTX different escalation regimens, higher doses and switches from oral to subcutaneous therapy may occur. However, all the trials evaluated MTX at doses that are consistent with what is found in many contemporary clinical practices and registries.32 ,33

Taken together, the definition of the treatment target is important, but influenced by a number of considerations, including patient characteristics and comorbidities, type of treatment, method of disease activity assessment and the strategic attitude of the treating physician. Our study provides the data basis to inform physicians (of any attitude) in their clinical decision making in an era when treating to target RA has become a widely accepted paradigm in the community. The data strengthen the 3-month time point as potentially decisive for a large number of patients, especially those who—together with their rheumatologist—are targeting the goal of REM.


We thank AbbVie, Centocor (Janssen), Wyeth (Pfizer) and Roche, for providing the random sample of the original data from their trials. This is a publication of the Joint and Bone Center for Diagnosis, Research and Therapy of Musculoskeletal Disorders of the Medical University of Vienna.


View Abstract
  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Handling editor Tore K Kvien

  • Contributors DA and JSS: study design, data analysis, drafting of manuscript. FA: data analysis.

  • Funding Seventh Framework Programme (HEALTH-F2-2008-223404).

  • Competing interests DA and JSS: received consulting and/or speaking honoraria from Abbvie, MSD, Pfizer and Roche.

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

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.