The tumour necrosis factor alpha (TNF-α) antagonists adalimumab, etanercept and infliximab have proved to be efficacious in several clinical trials1^–3 and have changed the treatment of patients with rheumatoid arthritis (RA). However, treatment with one of the anti-TNF-α agents is expensive, with mean medication costs exceeding €1000 per month. In most European countries4 and in the United States5 there are guidelines and recommendations in order to optimise the effectiveness and to limit the use of anti-TNF-α agents in clinical practice for reasons of cost. These guidelines all state that the prescription of anti-TNF-α agents should be restricted to patients with active RA in whom there is an insufficient response to one or two disease modifying anti-rheumatic drugs (DMARD). Treatment with one of the anti-TNF-α agents should be discontinued if there is no indication of effect within 3 to 6 months. The measure of effect is not explicitly defined in most guidelines and is then based on clinical opinion.

In The Netherlands as well as in the United Kingdom, a decrease in the disease activity score based on 28 joints (DAS28) of more than 1.2 after 3 months had been advised for continuation of the anti-TNF-α agent. In a recent comment on the updated guidelines of the British Society of Rheumatology it was questioned whether failing to decrease the DAS28 by more than 1.2 after 3 months is sufficient grounds for withdrawing therapy.6 It is, however, not completely clear whether continuing treatment is more effective than stopping treatment and switching to another treatment option.7 In the absence of evidence for the latter option, the British National Institute for Health and Clinical Excellence (NICE) actually recommended stopping treatment when there is no response at 6 months after the initiation of anti-TNF-α therapy.8

In general, there is a lack of evidence to support the development of guidelines for the continuation of anti-TNF-α treatment. Therefore, we studied whether rheumatologists complied with the Dutch guidelines for anti-TNF-α treatment, using the Dutch Rheumatoid Arthritis Monitoring (DREAM) register on anti-TNF-α. Furthermore, the consequences on disease activity, functionality, quality of life and costs were compared between patients who were and were not treated in adherence with the guidelines. In addition, possible alternative strategies were modelled with regard to the percentage of successfully (responders) and unsuccessfully treated patients and costs.

METHODS

Patients

Since February 2003 all consecutive RA patients who started for the first time on one of the anti-TNF-α agents in 10 hospitals in The Netherlands have been included in a prospective register, further called the DREAM register. In The Netherlands, RA patients need to fulfill criteria in order to get reimbursement for anti-TNF-α therapy, which are having at least a moderate disease activity (DAS28 ⩾3.2) and having failed on at least two DMARD including methotrexate at an optimal dose up to 25 mg per week. Anti-TNF-α therapy may be continued if there is a decrease of at least 1.2 in the DAS28 at 3 months. A waiver may be acquired if there is a good reason to continue treatment despite non-response at 3 months.

The register contains all start and stop dates, doses, changes in doses and reasons for changes of all medication that is used by the patient. The choice of anti-TNF-α agent, dosing and concomitant treatment was at the discretion of the attending rheumatologist. Data collection in the register is ongoing even when patients have stopped using anti-TNF-α agents.

Outcome measures

The baseline visit took place at the time of initiation of the first anti-TNF-α agent. Patient characteristics such as age, sex, disease duration, rheumatoid factor and previous DMARD use were registered. Trained research nurses assessed the patients at baseline and subsequently every 3 months. If necessary, patients were seen more frequently by the attending rheumatologist.

Disease activity was measured by the disease activity score (DAS28), combining the following variables: the 28 joint count for swelling (SJC28), the 28 joint count for tenderness (TJC28), the erythrocyte sedimentation rate (ESR) and patient’s global health as assessed by a visual analogue scale.9 In the case of missing values for the DAS28 caused by a missing value for the ESR, the DAS28 was calculated using an imputed ESR. The ESR was imputed by means of linear multivariate regression using the values of the SJC28, TJC28 and the global health of the patient.

Because we hypothesised that benefits not captured by the DAS28, such as a decrease in functional disability or wellbeing, were considered in the decision about continuation of anti-TNF-α, we also used questionnaires assessing these aspects. Functional ability was measured by means of the Health Assessment Questionnaire (HAQ).10 11 Quality of life was measured by the EuroQol 5 dimensions (EQ-5D)12 utility score and the physical and the mental component scales13 of the Short-Form 36 (SF-36).14

Because the anti-TNF-α drug and administration costs were considered the cost drivers, only those were included in the cost analysis. Costs were calculated by multiplying the applied doses by the price per milligram of the particular drugs. The prices were based on the tariff lists (September 2006) published by the Dutch Health Care Insurance Board. A true cost price was calculated for the inpatient clinic administration of infliximab on the basis of the personnel, materials and devices. Costs for overheads (35%) and housing (10%) were added to the total cost price according to the Dutch guidelines for cost analyses in healthcare.15

Analyses

First, the percentage of patients who were treated in adherence with the guidelines was analysed by calculating the percentage of patients with a response at 3 months and the discontinuation rate after 3 months. In order to evaluate whether the extent of improvement was a predictor for the decision to continue or not, the decrease in DAS28 was tested for a statistically significant difference between patients continuing or discontinuing their current anti-TNF-α agent within 3 to 6 months by means of a Student’s t-test.

Next, the consequence of deviating from the guidelines was evaluated regarding disease activity, functional ability and quality of life. Three groups of patients were analysed: (1) responders at 3 months who continued their current agent; (2) non-responders at 3 months who continued their current agent and (3) non-responders at 3 months who discontinued their current agent. The course over time of DAS28, HAQ, EQ-5D and the physical component scale of the SF-36 was plotted up to 12 months. Differences between the three groups were tested at the 6-month follow-up by means of one-way analysis of variance. The total costs for the first 6 months were tested for significant differences between the three groups by means of the non-parametric Kruskal–Wallis test. If a patient switched within those 6 months from one anti-TNF-α agent to another, the costs of both agents were included.

Finally, a decision tree was used to analyse the expected costs and the percentage of successfully (responders) and unsuccessfully treated patient in four possible strategies (fig 1). The first strategy (A) was daily clinical practice as observed in our DREAM registry. The second strategy (B) was full compliance with the present guidelines. The third strategy (C) was continuation of treatment in the case of at least a partial response (0.6 DAS28 decrease). The last strategy (D) was a treatment of 6 months for everyone with evaluation of response at 6 months.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

A decision tree analysing expected costs and percentages of responders and non-responders in four strategies: clinical practice (A), full adherence to the guidelines (B), a variation of the guidelines (C) and 6 months treatment for every patient (D). The DAS28 score was missing in 6% at 3 months and in 12% at 6 months. This means that there is complete data for 539 patients at 3 months and for 503 patients at 6 months. RA, rheumatoid arthritis; TNF, tumour necrosis factor.

The probabilities used as input for the decision tree were calculated using the data of patients in the DREAM registry who had actually followed one of the strategies. For the full compliance strategy (B), responders who continued and non-responders who stopped treatment after 3 months were selected. For strategy C, patients with a 0.6 or more DAS28 decrease who continued treatment and patients with less than a 0.6 DAS28 increase who stopped treatment were selected. For the last strategy (treatment for 6 months for every patient), patients who were treated for 6 months were selected. In all selected patients the response at 6 months after the initiation of therapy was calculated.

Each branch in the decision tree represented one of the four strategies and led to one of the three possible outcomes: (1) responders at 3 months who continued their current agent; (2) non-responders at 3 months who continued their current agent and (3) non-responders at 3 months who stopped their current agent. Costs were based on the average costs as observed in clinical practice as mentioned above. The expected costs for every strategy (or branch) in the decision tree were then calculated by following the probabilities for the outcomes multiplied by the costs related to these outcomes.

Model uncertainty was explored using a Monte Carlo probabilistic sensitivity analysis. During that procedure, the model was run for 1000 times and each time values from distributions surrounding the model parameters were drawn. This provided a non-parametric 95% CI for the expected costs and the percentage of successfully and unsuccessfully treated patients.

RESULTS

Patients

A total of 625 patients started anti-TNF-α treatment between February 2003 and January 2005; 234 on adalimumab, 254 on etanercept and 137 on infliximab. All patients completed 6 months of follow-up, which was the time frame of this analysis. For 49 patients, the occurrence of adverse events was the reason to stop their first anti-TNF agent within 6 months after initiation. These patients were left out of the analyses because they followed a different strategy. The DAS28 score was missing in 6.3% at 3 months and in 12.5% at 6 months mainly because of planning problems equally divided over responders and non-responders; thus the missing nature of these data was considered to be completely at random. As a result, complete data of 539 patients at 3 months and of 503 patients at 6 months were available.

Compliance

At 3 months 56% (306/539) of the patients were classified as responders and 44% (233/539) as non-responders (fig 1). The majority (81%; 189/233) of the non-responders continued treatment. There was no difference in the percentage of non-responders who continued treatment within 3 to 6 months between the three agents (adalimumab 80%, etanercept 82% and infliximab 82%; p = 0.616). Of all responders at 3 months, 96% (295/324) of patients continued treatment.

Table 1 shows the baseline characteristics of the patients in three different groups as defined by the response at 3 months and the decision to continue or not. The responders had a statistically significantly (p<0.001) higher baseline DAS28 than the non-responders, possibly indicating regression to the mean. The mean decrease in DAS28 at 3 months was −0.32 (95% CI −0.21 to −0.43) in the non-response group who continued the current anti-TNF-α agent, compared with 0.09 (95% CI −0.11 to 0.31) in the non-response group who discontinued, which was significantly different (p value 0.002).

Effects on effectiveness and costs

At 6 months, 37% (67/182) of the non-responders at 3 months who continued the current anti-TNF-α agent were eventually classified as responders, indicating a time lag in the effect. This late effect is visualised in fig 2 on the disease activity as well as on functional ability and quality of life. However, non-responders did not perform as well as responders on all outcomes over time. There were no significant differences in most of the outcome measures between non-responders who continued or discontinued the current anti-TNF-α agent. Concerning the SF-36, patients did have mental component scores comparable to the US population norm, which did not change over the 12-month follow-up, and therefore only the physical component scale is presented.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

The DAS28 (disease activity score based on 28 joints), the Health Assessment Questionnaire (HAQ–DI), EuroQol 5 (EQ-5D) dimensions and the physical component scale of the Short-Form 36 (PCS) over time for the responders who continued, the non-responders who continued and the non-responders who discontinued.

During the use of anti-TNF, the dose of anti-TNF-α was increased in 5% (14/295) of the responders who continued, in 11% (21/189) of the non-responders who continued treatment and in 9% (4/44) of the non-responders who stopped. Patients who discontinued their first anti-TNF-α agent because of a non-response switched in 52% (23/44) of the cases to another anti-TNF-α agent. Both the costs of an increased dose and the costs of the second anti-TNF agent in the patients who switched were included in the cost analyses. This resulted in mean 6-month costs of €6.985 (95% CI €6785 to €7185) in the initial responder group, of €7.197 (95% CI €6883 to €7510) in the non-responders who continued the current anti-TNF-α agent and of €5.616 (95% CI €4968 to €6263) in the non-responders who discontinued. The latter group had significantly lower anti-TNF-α costs than the other two groups (p<0.001), but the costs were still rather high.

Decision tree

Table 2 shows the results of the Monte Carlo probabilistic sensitivity analysis regarding the expected costs and percentage of successfully and unsuccessfully treated patients in the four different analysed strategies. There were no differences in the total expected costs per patient per 6 months between all alternative strategies. However, there were some differences between the strategies on the expected percentage of successfully and unsuccessfully treated patients. A strategy of treatment for 6 months independent of the response at 3 months (D) revealed the highest total percentage of responders (67%), but also the highest percentage of patients who were unsuccessfully treated with the initial anti-TNF-α agent for the full 6 months (34%). The strategy of full adherence to the guidelines (B) revealed the lowest expected percentage of responders (56%), but on the other hand, also the lowest percentage of unsuccessfully treated patients (9%). There was very little difference in the expected percentage of responders between strategies A, C and D, but the expected percentage of unsuccessfully treated patients was much lower in the “continuation in case of partial response” strategy, (C) namely 17% compared with 31% in the observed clinical practice (A) and 34% in the “6 months treatment for every patient” strategy (D).

DISCUSSION

According to guidelines in The Netherlands, a decrease in DAS28 of more than 1.2 within 3 months is required for continuation of the current anti-TNF-α agent in the treatment of patients with RA. This study used a daily clinical practice registration and showed that response, defined as at least 1.2 points improvement in the DAS28 at 3 months, was not reached in 44%. According to the strict interpretation of the guidelines, those patients should have stopped the current treatment with anti-TNF-α agents. However, for most (81%) of those non-responders their current anti-TNF-α treatment was actually continued. Interestingly, continuation despite non-response at 3 months seemed to have added value because 37% of the non-responders who still continued their anti-TNF-α agent became responders at 6 months of treatment. These patients also showed prolonged improvements on functional ability and quality of life. In the case of full adherence to the guidelines (stopping treatment) this delayed effect would have been missed.

A long drug survival time despite insufficient inflammatory control has been shown before.16 Potential explanations were suggested but not evaluated by the authors, including a physician or patient perception of better efficacy than reflected in the DAS28 or a lack of guidelines. The first explanation was also argued by the British Society for Rheumatology.6 Consequently, the British NICE has updated their guidance on the use TNF-α blocking agents for RA,8 in which they state that treatment with anti-TNF-α agents should be continued only if there is an adequate response at 6 months instead of the 3 months in the previous version.

However, in the case of a lack of response at 3 months a rheumatologist can also decide to switch to another treatment option. One study has recently been published in which patients with a partial response on etanercept were being randomly assigned between continuation of etanercept or switching to infliximab.7 The authors concluded that there was a numerical trend favouring patients to switch to infliximab. Due to the small sample size (in total 28 patients) the results of that study have to be confirmed by larger studies.

When a rheumatologist decides not to stop and switch treatment but to continue treatment despite a lack of response, we can recommend doing this only in patients with at least a partial response (DAS28 reduction >0.6). From decision analytical modelling it could be concluded that the option to make a decision about continuation in all patients at 6 months revealed the most responders at 6 months. However, it would also result in the most unsuccessfully treated patients. The above-mentioned strategy to “continue treatment in the case of at least partial response” would have the best trade-off between successfully and unsuccessfully treated patients in the case that this strategy is fully adhered to.

Our study has some limitations. First, extra costs including the assessment of the DAS28 were not taken into account in the cost analysis because they were not considered as cost drivers and were not different between patients who continued or stopped treatment. Second, for the calculation of the response percentage in the strategy “6 months treatment for all patients” only completers were selected, possibly resulting in an overestimation of the response percentage. As an alternative the response percentage in the complete unselected group of patients can be used. This percentage was 61%, which is equal to the response percentage estimated in strategies A and C, and therefore it would not change our conclusions. Third, for the full adherence strategy it was assumed in the model that the probabilities of having a response at 6 months, in the case of non-response and discontinuation at 3 months, would be the same as for all non-responders who did stop treatment in daily clinical practice. However, this may have led to an underestimation of this probability because patients who did stop treatment after 3 months’ non-response were patients with no improvement at all at 3 months. Finally, an increased number of responders and a decreased number of unsuccessfully treated patients would result in a decrease in morbidity and ultimately quality of life. However, robust data about the long-term effectiveness on quality of life in all analysed strategies would be needed, which is not available.

It can be concluded that adherence to the present Dutch guidelines for treatment with anti-TNF-α for RA patients is suboptimal. Rheumatologists’ behaviour seemed to a certain extent justified by the number of patients who became responders at 6 months. However, it is unclear whether a switch to the next treatment would have better effectiveness than to continue treatment in the case of non-response. If treatment is continued after 3 months, we recommend doing this only in patients with at least a partial response (at least 0.6 DAS28 improvement). This strategy is estimated to have the best ratio between successfully and unsuccessfully treated patients.