Comparison of Biological Agent Monotherapy and Associations Including Disease-Modifying Antirheumatic Drugs for Rheumatoid Arthritis: Literature Review and Meta-Analysis of Randomized Trials

Objective: Update the available evidence comparing biologic disease-modifying antirheumatic drugs (bDMARDs) in combination with conventional synthetic disease-modifying antirheumatic drugs (CsDMARDs) to bDMARDs in monotherapy in patients with rheumatoid arthritis. Methods: Research was limited to randomized controlled trials. Major outcome: ACR 20 response criteria at 24 weeks. Secondary outcomes: clinical and radiographic criteria at week 24, 52 and 104. Results: 23 trials (6358 patients), including seven bDMARDs and one other molecule: Anbainuo (anti-TNF-R). No study satisfied our search criteria for anakinra, certolizumab and infliximab. Compared to bDMARD monotherapy, combination therapy gives a better ACR 20 at 24 weeks (RR: 0.88 (0.84–0.94)) in fixed and random effect models, and this result is sustained at 52 and 104 weeks. The results were mostly similar for all other outcomes without increasing the risk of adverse effects. Conclusion: This meta-analysis confirms the superiority of combination therapy over monotherapy in rheumatoid arthritis, in accordance to the usual guidelines.


Introduction
Rheumatoid arthritis (RA) is the most common inflammatory rheumatism in adults [1]. The EULAR and French recommendations stipulate that methotrexate (MTX) should be started as soon as possible after the diagnosis of established RA [1,2]. If remission or low disease activity is not achieved after six months of conventional synthetic DMARDs (CsDMARDs) in patients without factors associated with a poor prognosis, treatment with another CsDMARD may be considered. By contrast, in the presence of a poor prognosis factor, biological treatment should be considered in association with the CsDMARD previously used [3]. In total, ten biological agents (bDMARDs) have been approved for RA treatment. Among them, adalimumab, certolizumab and etanercept have also been approved for use in monotherapy, as have abatacept, anakinra, tocilizumab and sarilumab. New strategies for RA treatment based on the inhibition of Janus kinase (JAK) pathways have been developed, but are not reviewed here. Many RA patients find it difficult to adhere to their CsDMARD prescription because of intolerance or contraindications [4][5][6][7][8][9][10]. It is, therefore, important to evaluate the benefits and harm associated with the use of biological agents in monotherapy. We conducted a systematic review of the literature and a meta-analysis, to update the available evidence already established [11] comparing the use of bDMARD and CsDMARD combination therapy with the use of biotherapy in monotherapy in patients with rheumatoid arthritis.

Risk of Bias
We assessed the risk of bias for each trial included, using the Cochrane 'Risk of bias' tool and the following criteria: selection bias, performance bias, detection bias, attrition bias and reporting bias [18]. The risk of bias has been classified as: 'low', 'high' or 'unclear' (due to either a lack of information or uncertainty over the potential for bias). The GRADE score reflects the extent to which we are convinced that the actual effect is close to that estimated in the meta-analysis ( Figure 1).

Statistical Analysis
We performed meta-analyses with fixed and random effects models in R version 3.6.1 (5 July 2019) Copyright © 2022 The R Foundation for Statistical Computing. The relative risk (RR) was the metric of choice for binary outcomes, and the mean difference (MD) or the standardized mean difference (SMD) was used for quantitative variables. Inverse variance weighting was used for the pooling of studies [19]. We used the DerSimonian-Laird method to estimate the variance between studies [20]. Between-study heterogeneity was assessed with the Q-test, considering a p-value < 0.05 to be statistically significant. The I² statistic was calculated to quantify the residual heterogeneity, ranging from 0 to 100% [21]. A leave-one-out method was used to identify outlying studies responsible for heterogeneity. Sensitivity analyses were conducted by meta-regression. The criteria included in the meta-regression analysis included: dose titration allowed in the study protocol, mean duration of disease, history of CsDMARD and bDMARD use before inclusion, presence or absence of a disease stabilization phase before randomization, positivity for RF and/or ACPA, severity on the DAS at inclusion, authorization of corticosteroid therapy use during the study and blinding throughout the study. Publication bias was evaluated by a graphical method, with a rank correlation test for funnel plot asymmetry [22].

Statistical Analysis
We performed meta-analyses with fixed and random effects models in R version 3.6.1 (5 July 2019) Copyright © 2022 The R Foundation for Statistical Computing. The relative risk (RR) was the metric of choice for binary outcomes, and the mean difference (MD) or the standardized mean difference (SMD) was used for quantitative variables. Inverse variance weighting was used for the pooling of studies [19]. We used the DerSimonian-Laird method to estimate the variance between studies [20]. Between-study heterogeneity was assessed with the Q-test, considering a p-value < 0.05 to be statistically significant. The I 2 statistic was calculated to quantify the residual heterogeneity, ranging from 0 to 100% [21]. A leave-one-out method was used to identify outlying studies responsible for heterogeneity. Sensitivity analyses were conducted by meta-regression. The criteria included in the metaregression analysis included: dose titration allowed in the study protocol, mean duration of disease, history of CsDMARD and bDMARD use before inclusion, presence or absence of a disease stabilization phase before randomization, positivity for RF and/or ACPA, severity on the DAS at inclusion, authorization of corticosteroid therapy use during the study and blinding throughout the study. Publication bias was evaluated by a graphical method, with a rank correlation test for funnel plot asymmetry [22].

Study Characteristics
The characteristics of the studies included in the meta-analysis are displayed in Table 1.
Regarding the history of CsDMARD use, three studies included only patients who had never used CsDMARDs (13%), 16 included patients not naïve for CsDMARD use (69.6%) and four were not selective on the basis of these criteria (17.4%). All but two of the studies used MTX as the CsDMARD. For the history of bDMARD use, nine studies included only patients who had never used bDMARDs (39.1%), eight included patients with a possible history of use before inclusion (34.8%) and five included patients not naïve for bDMARD treatment (21.7%). Of the studies, eight included (26.1%) planned treatment adjustment during the trial a priori, with a rescue treatment administered if the main endpoint was not reached within the time allowed.

Principal Characteristics of the Patients
More than 75% of the patients were women, and the mean age was 55 years (range: 45.4 ± 11.9 to 63.3 ± 10.6 years). Disease duration ranged from 26 days to 12 years, and most of the patients for whom the information was available had tested positive for autoantibodies (RF and/or ACPA). The DAS 28 score at baseline ranged from 2.6 to 6.8, and the mTSS score ranged from 0 ± 0 to 39.6 ± 56.1.

Primary Efficacy Endpoint: ACR 20 at 24 Weeks
The ACR 20 results at 24 weeks were reported for 16 studies. Using a random effects model, we found that the therapeutic combination performed significantly better for this endpoint, RR = 0.88 (0.83; 0.93), despite significant heterogeneity between studies (I 2 = 46%, τ 2 = 0.0068, p = 0.02) (Figure 3). We made the results more consistent by performing a sensitivity analysis and a funnel plot asymmetry test. The remaining 14 studies showed that significantly better results for this endpoint were obtained with the combination treatment (RR = 0.91 (0.88; 0.95); I 2 = 0%, τ 2 = 0, p = 0.53). We decided a priori to include open-label studies in the quantitative analysis and to use the qualitative analysis to check whether this decision had any effect on the results. Consistent with the τau 2 test results, we can conclude that the inclusion of SUPRISE, ADORE and JESMR with an open-label design, despite their lower grade of recommendation, had no effect on the results of the analysis.

ACR Reponses
With fourteen studies providing ACR 20 results at 52 weeks, the overall result obtained with the random effects model was significantly in favor of the therapeutic combination (RR= 0.90 (0.84; 0.97)). However, it was not possible to have confidence in the results, due to the degree of heterogeneity (I 2 = 64%, τ 2 = 0.0116, p < 0.01). After controlling for heterogeneity, we selected 11 studies. The results were also in favor of the combination treatment (RR

ACR Reponses
With fourteen studies providing ACR 20 results at 52 weeks, the overall result obtained with the random effects model was significantly in favor of the therapeutic combination (RR= 0.90 (0.84; 0.97)). However, it was not possible to have confidence in the results, due to the degree of heterogeneity (I 2 = 64%, τ 2 = 0.0116, p < 0.01). After controlling for heterogeneity, we selected 11 studies. The results were also in favor of the combination treatment (RR = 0.94 (0.90; 0.98), I 2 = 7%, τ 2 = 0.0004, p = 0.38). The ACR 20 results at 104 weeks were available for eight studies. The results were significantly in favor of the combination treatment both before (RR = 0.89 (0.84; 0.94), I 2 = 66%, τ 2 = 0.0129, p < 0.01) and after the sensitivity analysis (RR = 0.92 (0.87; 0.98)) (I 2 = 0%, τ 2 = 0, p = 0.42). The For the DAS 28-ESR remission scores, the initial data at 24 weeks showed an I 2 value of 0% for heterogeneity, but the funnel plot asymmetry test revealed a publication bias for the GO-FORWARD study, so we excluded this study from the final analysis. The objective was significantly more frequently achieved in the association group at weeks 24

Subgroup Meta-analysis
All our results show an RR close to 1, meaning that the combination therapy was not much more effective than the biological agent used as a monotherapy. We decided to perform a subgroup analysis by successively removing etanercept, tocilizumab and then both molecules from the analysis to try to explain these results. We found that removing tocilizumab resulted in a decrease in RR, whereas removing etanercept did not appear to change the results. This sub-analysis suggests that tocilizumab appears to be more effective as a single agent than other available biotherapies.

Structural Progression
A total of eleven RCTs reported X-ray progression based on mTSS <0 or <0.5 at weeks 24, 52 and/or 104. A significantly lower progression was observed for the combination treatment from 52 weeks ( Figure 5) and remained at 104 weeks before and after sensitivity analysis; the results at 24 weeks were not significant (RR = 0.98 (0.91; 1.04), I 2 = 6%, τ 2 = 0.0003, p = 0. 35

Subgroup Meta-Analysis
All our results show an RR close to 1, meaning that the combination therapy was not much more effective than the biological agent used as a monotherapy. We decided to perform a subgroup analysis by successively removing etanercept, tocilizumab and then both molecules from the analysis to try to explain these results. We found that removing tocilizumab resulted in a decrease in RR, whereas removing etanercept did not appear to change the results. This sub-analysis suggests that tocilizumab appears to be more effective as a single agent than other available biotherapies.

Toxicity
The heterogeneity of the toxicity data made it impossible to perform a reliable statistical analysis. Nevertheless, we were able to compare study outputs for discontinuation due to adverse events. We found no advantage for either group in terms of the number of discontinuations: RR = 0.73 (0.53; 1,01) with I 2 =0% and τ 2 = 0, p = 0.70 at 24 weeks; RR= 0.95 (0.70; 1.31) with I² = 40%, τ2 = 0.1151, p = 0.07 at 52 weeks; and RR = 0.84 (0.67; 1.05) with I 2 = 0%, τ 2 = 0, p = 0.53 at 104 weeks. The I² value for heterogeneity obtained at 52 weeks was > 30%. We decided to retain this result because significance did not differ before and after sensitivity analysis, and a cutoff of 40% has been reported to be acceptable [54].

Discussion
Through this systematic review and meta-analysis, we aimed to compare the use of biological agents in monotherapy and in association with a CsDMARD. Many meta-analyses have compared studies of different therapeutic combinations (for example, biotherapies in monotherapy versus CsDMARDs or placebo), making it difficult to extrapolate results to our arms of interest [55][56][57][58][59]. This is the second systematic review comparing the value of adding MTX to bDMARD treatment with bDMARD monotherapy. Our work confirms the work of Tarp and al. [11], with the difference that our study is more recent, which allowed us to include a larger number of randomized trials and, therefore, patients.
We found that the combination treatment was more effective than monotherapy, as shown by the main endpoint, ACR 20 at 24 weeks. The results are similar for the other endpoints, with, for some, a loss of efficacy at 104 weeks, possibly with a loss of power of the study. The modified Sharp's score was significantly in favor of the combined treatment from 52 weeks onwards. At 24 weeks, the duration of exposure may not have been sufficiently long to distinguish between the two study arms considered. It is interesting to note that the PREMIER study [24] was a source of great heterogeneity for several of the variables studied with no obvious cause found. Our results are comparable with French and international recommendations [1,2].
The purpose of this meta-analysis was not to compare the different bDMARDs between each other. Nevertheless, we could observe that all our results showed an RR close

Toxicity
The heterogeneity of the toxicity data made it impossible to perform a reliable statistical analysis. Nevertheless, we were able to compare study outputs for discontinuation due to adverse events. We found no advantage for either group in terms of the number of discontinuations: RR = 0.73 (0.53; 1.01) with I 2 = 0% and τ 2 = 0, p = 0.70 at 24 weeks; RR= 0.95 (0.70; 1.31) with I 2 = 40%, τ 2 = 0.1151, p = 0.07 at 52 weeks; and RR = 0.84 (0.67; 1.05) with I 2 = 0%, τ 2 = 0, p = 0.53 at 104 weeks. The I 2 value for heterogeneity obtained at 52 weeks was >30%. We decided to retain this result because significance did not differ before and after sensitivity analysis, and a cutoff of 40% has been reported to be acceptable [54].

Discussion
Through this systematic review and meta-analysis, we aimed to compare the use of biological agents in monotherapy and in association with a CsDMARD. Many meta-analyses have compared studies of different therapeutic combinations (for example, biotherapies in monotherapy versus CsDMARDs or placebo), making it difficult to extrapolate results to our arms of interest [55][56][57][58][59]. This is the second systematic review comparing the value of adding MTX to bDMARD treatment with bDMARD monotherapy. Our work confirms the work of Tarp and al. [11], with the difference that our study is more recent, which allowed us to include a larger number of randomized trials and, therefore, patients.
We found that the combination treatment was more effective than monotherapy, as shown by the main endpoint, ACR 20 at 24 weeks. The results are similar for the other endpoints, with, for some, a loss of efficacy at 104 weeks, possibly with a loss of power of the study. The modified Sharp's score was significantly in favor of the combined treatment from 52 weeks onwards. At 24 weeks, the duration of exposure may not have been sufficiently long to distinguish between the two study arms considered. It is interesting to note that the PREMIER study [24] was a source of great heterogeneity for several of the variables studied with no obvious cause found. Our results are comparable with French and international recommendations [1,2].
The purpose of this meta-analysis was not to compare the different bDMARDs between each other. Nevertheless, we could observe that all our results showed an RR close to 1, showing little difference in clinical efficacy between groups. The subgroup analysis showed that excluding tocilizumab from the analysis decreased this RR, suggesting that tocilizumab is probably the most effective single-agent biologic. These results are consistent with the literature. Tarp et al. [60] have shown that most biological agents are effective in monotherapy, with an advantage for etanercept and tocilizumab supported by other network meta-analyses [57,59,61]. Some studies about IL-6 receptor blockers in monotherapy have shown that tocilizumab monotherapy yields response rates close to those obtained in combination with MTX in randomized studies and cohorts [62,63]. The ADACTA and MONARCH studies have shown tocilizumab and sarilumab to be superior to adalimumab in monotherapy [64,65]. In the TOCERRA registry [63], the therapeutic efficacy and maintenance of tocilizumab monotherapy are similar to those of the anti-TNF agents associated with MTX.
Finally, structural damage was not studied for all the biological treatments included in our meta-analysis. Tarp et al. [11] obtained identical results to those reported here and, after a subgroup analysis, no structural differences were found.
We found no difference in terms of safety between the two treatment arms, essentially due to the heterogeneity of the data collection. However, we were able to show that there was no difference between the study arms in terms of the rate of treatment discontinuation due to adverse events. With regard to the risk of infection, Singh et al. [66] showed that, in patients treated with CsDMARDs, the median annual absolute risk of infection was 2%, or 20 per 1000 treated patients per year, whereas there was an increase to 6 per 1000 patients treated with bDMARDs in combination with a CsDMARD, with a significant difference. Ramiro et al. [67] confirmed that patients on bDMARDs (both anti-TNF and no anti-TNF agents) had a higher risk of serious infections than patients on CsDMARDs, and that there was generally no difference between bDMARDs. They also investigated the occurrence of different cancers after exposure to biologics. Relative to both the general population and patients on CsDMARDs, patients on bDMARDs had no higher risk of individual solid cancers or of lymphoma. By contrast, non-melanoma skin cancer may occur more frequently in patients on bDMARDs than in the general population (HR 1.7), but the risk in these patients is no higher than that in patients treated with CsDMARDs. One study with a low risk of bias showed that patients on bDMARDs may have a higher risk of melanoma than patients on CsDMARDs (HR 1.5, 95% CI 1.0 to 2.2) [68].
Nevertheless, our study has several limitations. First, as with all systematic literature reviews, this study is subject to certain publication and selection biases. Second, there was some heterogeneity between these studies. The variables tested with the sensitivity analysis did not significantly influence the results of the meta-analysis. In addition, few studies have used CsDMARDs other than methotrexate, limiting the extrapolation of results for leflunomide or sulfasalazine. We also chose to consult the data collected at weeks 24, 52 and 104; when data for these time points were not available, other time points were used. Comparisons at different time points may limit the interpretation of our results, as may not having been able to contact the authors to recover missing data.

Conclusions
Our meta-analysis confirms the results of a previous one, but with updated research and a larger number of studies included. The results indicate that the combination therapy of a biological agent with CsDMARDs is more effective than monotherapy and should be preferred in uncontrolled RA, in accordance with the usual guidelines. MTX should be switched to another CsDMARD in the case of contraindication or intolerance.