Objective To identify patient characteristics associated with responsiveness to tumour necrosis factor inhibitors (TNFi) in rheumatoid arthritis (RA).
Materials and methods Individual patient data from 29 randomised controlled trials (RCTs) evaluating the efficacy of a TNFi versus placebo or conventional therapy were obtained. Response to treatment was assessed in subgroups according to the following baseline characteristics: smoking status, physical activity, sex, age, body mass index, autoantibody profile, disease duration, high initial disease activity defined by Disease Activity Score on 28 joints (DAS28)(C reactive protein (CRP)) >5.1. The primary outcome was the between-treatment group difference in DAS28(CRP) change from baseline to 6 months. The secondary endpoints were the between-treatment group difference in final DAS28(CRP) measured until 6 months and EULAR response criteria until 6 months. Data from each RCT were then pooled by the Mantel-Haenszel method using a random effects model. A linear metaregression was also carried out on two data-sharing platforms separately to support the results.
Results Individual data of 11 617 patients from 29 RCTs were analysed. Until 6 months, a significantly higher EULAR non-response rate was observed in obese patients (OR 0.52 vs 0.36 for non-obese, p=0.01). A multivariable regression model performed on 7457 patients indicated that patients treated by TNFi had a final DAS28(CRP) decreased by 0.02 for each year of disease duration (p<0.001), and a 0.21 decreased for patients with a baseline DAS28(CRP) >5.1 (p<0.001).
Conclusions In RA, patients who are more responsive to TNFi are those who are non-obese, have a long disease duration and have a high initial disease activity.
- biological therapy
- tumor necrosis factor inhibitors
Data availability statement
Data may be obtained from a third party and are not publicly available. The raw data were not extracted but analysed remotely from the Yale Open Data Access and Vivli platforms.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
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.
What is already known about this subject?
Demographics and disease characteristics may modulate the effect of tumour necrosis factor inhibitors (TNFi).
Meta analysis based on individual patient data is a reliable and powerful method to analyse factors related to treatment response.
What does this study add?
High body mass index increased the odds of non-response to TNFi in rheumatoid arthritis (RA).
Patients with RA with high baseline disease activity and those with long disease duration are likely to achieve a better response to TNFi as compared with others.
How might this impact on clinical practice or further developments?
Bodyweight control of RA is essential when starting a TNFi.
Rheumatoid arthritis (RA) is the most common chronic systemic autoimmune disease with a prevalence of 0.3%–1.0%.1 Despite considerable progress in the knowledge of its pathogenesis and therapeutic management,2 3 disease remission or low disease activity is not obtained in all patients.4 Tumour necrosis factor inhibitors (TNFi) are the first biological agents available in RA and are still widely used in patients with inadequate response to conventional disease-modifying antirheumatic drugs (csDMARDs). However, approximately one-third of patients with RA respond insufficiently to TNFi.5–8
The reasons behind the heterogeneity in response remain unclear. Demographic, disease-related and environmental factors could contribute to the variability in clinical response to TNFi. Some factors have been associated with a poor response such as smoking,9–12 being a woman,10 13 14 older age,15 16 obesity,17 presence of rheumatoid factor (RF) and anticitrullinated protein antibodies (ACPAs),18 long disease duration13 and high disease activity,19–22 with conflicting results though.19 21 23 Available data are sparse on the influence of physical activity on the response to TNFi. Physical activity seems to decrease fatigue24 and improve quality of life,25 26 but does not seem to decrease inflammation parameters.27
We therefore aimed to study the influence of these factors on the effect of TNFi by performing a pooled-analysis of randomised clinical trials that evaluated efficacy of TNFi compared with placebo in subgroups of interest.
A systematic review of randomised controlled clinical trials was performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines,28 with the aim of studying the efficacy of TNFi according to different demographic and disease-related factors. The subgroups of interest were selected according to smoking status, physical activity, sex, age, body mass index (BMI), autoantibody status, disease duration and disease activity score at baseline. The protocol was registered in the PROSPERO database (number CRD42018071079) in January 2018 and was updated in July 2021. We searched for randomised controlled trials (RCTs) and meta-analyses of RCT in Cochrane Central Register of Controlled Trials. This research was performed until January 2017 using the keywords “rheumatoid arthritis”, “infliximab”, “adalimumab”, ‘“etanercept”, “golimumab” and “certolizumab”. Two authors (TB-A and SD) selected eligible RCT on title and abstract, retrieved full text of eligible articles and decided on the final inclusion. We included RCT comparing TNFi to placebo or csDMARDS in patients with RA, with no restriction on the presence or not of previous TNFi. We excluded non-randomised studies and studies comparing two TNFi without a placebo or csDMARD control groups. If included studies did not report efficacy of TNFi according to subgroups of interest, we sought to obtain individual patient data (IPD).
Given the unavailability of published subgroup of interest analyses, we contacted the corresponding authors and/or sponsors of these trials in order to obtain aggregated data and/or IPD and to perform a pooled analysis. Since most of the data were stored securely on data-sharing platforms, we requested each of these platforms an access to the raw dataset. We signed data use agreements for two platforms, Yale Open Data Access (YODA) and Vivli Centre for Global Clinical Research, that allowed us to access IPD for 26 trials. We further obtained IPD for two AMGEN trials and for one academic trial (online supplemental table 1).
Subgroup of interest
After obtaining access to data, we analysed the efficacy of TNFi, regardless of the dose used, compared with control in subgroups of interest. These subgroups were smoking status (never/ever smokers), current physical activity (yes/no), sex (men/women), age (≤50/>50 years), BMI (<30/≥30 kg/m2), RF status (positive/negative), ACPA status (positive/negative), RA disease duration (<2/2 to 10/≥10 years) and baseline DAS28(CRP) (≤5.1/> 5.1).
The predefined primary endpoint was ACR20 score after 6 months of follow-up, and secondary endpoints were ACR50, ACR70, DAS28(CRP) and DAS28(ESR). Due to the impossibility to obtain ACR response from the raw dataset for eight trials, we decided in November 2019 to modify the primary endpoint to between-treatment group differences in DAS28(CRP) change (ΔDAS28(CRP)) from baseline to 6 months (or as close as 6 months, depending on each trial available data). Secondary endpoints were between-treatment group differences in final DAS28(CRP) measured at 6 months (or as close as 6 months) and EULAR response criteria at 6 months (or as close as 6 months).29 EULAR response criteria were used to stratify the groups of response, that is, good response if final DAS28 was ≤3.2 with a DAS28 improvement from baseline of at least 1.2 points; non-response if final DAS28 was >5.1 with a DAS28 improvement from baseline ≤1.2 points or DAS28 improvement from baseline ≤0.6 points, and moderate response if DAS28 did not meet these criteria. For trials with missing erythrocyte sedimentation rate (ESR), DAS28(CRP) was used to categorise EULAR response using the same cut-offs as for DAS28(ESR).
Statistical analyses were performed using R Studio software. Descriptive results are presented as median (min–max) or mean (IQR) unless stated otherwise. Pooled ORs or mean difference with 95% CIs for EULAR response and for DAS28(CRP) differences, respectively, between TNFi and placebo were calculated using two-step meta-analyses. First, aggregate data regarding treatment response in each subgroup of interest were estimated from IPD. Second, a random-effect Mantel-Haenszel model was applied to calculate pooled effect. We considered a significant difference between subgroups if the p value is <0.05. Between-study heterogeneity was quantified using Cochrane Q and I² statistics. Heterogeneity was considered low if I2 was <30%, moderate if I2 was 30%–50%, substantial if I2 was 50%–70% and considerable if I2 was >70%. IPD from YODA and Vivli platforms were used to perform two separate linear metaregressions with final DAS28(CRP) as independent variable adjusted on baseline DAS28(CRP), trial and treatment by subgroup variable interaction (bivariate analyses). Multivariable metaregressions of final DAS28(CRP) adjusted on baseline DAS28(CRP), trial, age and subgroup variables with a p value for interaction of <0.20 in bivariate analyses were also performed.
The analysis was performed following the intention-to-treat principle, with no replacement of missing data. Since this was an exploratory post hoc analysis, no p value adjustment for multiplicity was performed. Therefore, results should be considered with caution.
We found 496 articles published between 1994 and 2017, and 220 of them were eligible after selection on title and abstract, and 79 fulfilled all inclusion and exclusion criteria (figure 1). At the end of our search, we did not retrieve any RCT that reported aggregated results regarding the efficacy of TNFi in the subgroups of interest. We therefore retrieved IPD for 29 trials. As to June 2021, we obtained access data for 8 trials from the YODA platform, 18 trials from the Vivli platform, 2 trials from AMGEN and 1 trial from an academic author (figure 1).
Characteristics of included clinical trials
This pooled analysis included 29 RCTs evaluating five TNFi: 12 evaluated adalimumab; 3 evaluated etanercept; 6 evaluated certolizumab; 6 evaluated golimumab; and 2 evaluated infliximab (table 1 and online supplemental table 1). All RCTs were double-blinded, placebo-controlled, parallel group trials conducted between 1997 and 2015. The median number of patients randomised was 444 (range: 47–1648). Individual data of 14 838 randomised patients were available. Clinical and biological data were missing in some studies, leading to 11 617 (78%) IPDs available for analyses. Physical activity data were missing for 25 studies. Some data could not be retrieved for confidentiality and anonymisation reasons, such as patients’ age in six certolizumab studies and disease duration in four certolizumab studies that reported these data as intervals. Treatment response could be evaluated at week 24 for 21 trials (72%), at week 26 for 2 trials (7%), at week 30 for 2 trials (7%), and at week 12 for 4 trials (14%).
Difference in ΔDAS28(CRP) between baseline and until 6 months
Smoking status, physical activity, sex, age, BMI, ACPA, RF status, disease duration and baseline DAS28(CRP) did not significantly influence the difference in ΔDAS28(CRP) between TNFi and placebo in subgroup analyses (p>0.05 for subgroup differences, tables 2 and 3). Heterogeneity was considerable (I²>70%) in some subgroups: non-smoking patients, women, non-obese patients, patients with RF and ACPA positive status and patients with high baseline DAS28(CRP).
Difference in final DAS28(CRP)
Smoking status, physical activity, sex, age, BMI, ACPA and RF status, disease duration and baseline DAS28(CRP) did not significantly influence the final DAS28(CRP) between TNFi and placebo in subgroup meta-analyses (p>0.05 for subgroup differences, online supplemental table 2).
Good EULAR response
Smoking status, physical activity, sex, age, BMI, ACPA or RF status, disease duration and baseline DAS28 did not significantly influence the good EULAR response between TNFi and placebo in subgroup meta-analyses (p>0.05, online supplemental table 3).
We observed a qualitative and significant influence of obesity on the odd of being EULAR non-responder between TNFi and placebo (online supplemental table 4). Obese patients had a higher risk of non-response (OR 0.52, 95% CI 0.43 to 0.63) compared with patients with a BMI of <30 (OR 0.36, 95% CI 0.30 to 0.45), with p=0.01 for subgroup difference (table 3 and online supplemental figure 1). There was no influence of other covariates (smoking status, physical activity, sex, age, baseline DAS28(CRP), disease duration, ACPA and RF status) on the odds of non-response between TNFi and placebo (p>0.05 for subgroup differences).
Bivariable and multivariable metaregression results are shown in table 4. Metaregressions were performed on 7457 patients (18 RCTs) from Vivli and on 3767 (8 RCTs) patients from YODA. Bivariate analyses indicated that being treated by TNFi was associated with a significantly lower final DAS28(CRP), while a higher baseline DAS28(CRP) was associated with a significantly higher final DAS28(CRP) in trials from both databases. Men presented a lower final DAS28(CRP) than women, and disease duration was associated with an increased final DAS28(CRP) in trials from the Vivli database. Significant treatment effect modifiers in bivariate analyses were disease duration with a lower 0.02/year (in trials from the Vivli database) and baseline DAS28(CRP) with a higher 0.1/baseline unit (in trials from YODA database) final DAS28(CRP) in treated patients.
Four variables, that is, age, sex, BMI and disease duration, were included in the multivariable models. Using the individual data from Vivli (n=7457), the model indicated that disease duration as a continuous variable and baseline DAS28(CRP) as a categorial variable significantly modified treatment effect on final DAS28(CRP). Patients treated by TNFi had a final DAS28(CRP) decreased by 0.02 for each year of disease duration (p<0.001), and a 0.21 decreased for patients with a baseline DAS28(CRP) >5.1 (p=0.05). These results were not observed in the metagression performed on the 3767 patients from the YODA platform.
From the meta-analysis on pooled data, the sole characteristic associated with a clinical outcome was BMI, which increased the odds of being non-responders. According to the metaregression analysis based on individual data, the multivariable model found that disease duration and baseline DAS28(CRP) categories interacted with the final DAS28(CRP). In the present work, we did not find any influence of smoking status, physical activity, sex, age, RF or ACPA status on response to TNFi, which was in accordance to previous meta-analyses23 and registries21 but not with other retrospective cohorts.19 23
A recent meta-analysis showed a lower minimal disease activity achievement in obese patients with psoriatic arthritis or RA compare with those normal BMI.30 One hypothesis would be a decrease of TNFi concentration in obese patients due to a larger volume of distribution which is increased with body size.31 32 Therefore, an increase in the dosage of TNFi in obese patients could be considered. However, it has also been observed in the literature that obese patients treated with infliximab, whose dosage is based on body weight, have a lesser response,17 suggesting the role of adipose tissue. It has indeed been demonstrated that adipose tissue can produce adipokines, capable of inducing the production of proinflammatory cytokines such as tumour necrosis factor or interleukin-6,33 which may explain the lesser response to TNFi in obese patients. Furthermore, interaction between obesity and other factors such as physical activity may exist in the clinical trials included.34 However, because of lack of information on physical activity in most of trials, such correlations could not be studied.
We observed that patients with long disease duration, that is, ≥10 years, seemed to have a better response than others, which was not reported in the literature so far. In clinical practice, patients with long disease duration who require TNFi are usually those who have already failed to various treatments, which is known as a poor prognostic factor.13 This surprising result was seen in one of the data-sharing platforms and was not attributed to age nor disease activity. Thus, patients with long-standing disease with insufficient control of disease activity are as eligible as those with shorter disease duration, with even an additional effect of TNFi.
High baseline disease activity, that is, DAS28(CRP) of >5.1, was predictive of a favourable, although small, additional reduction in DAS28(CRP) at the time of 3 and 6 months, as compared with patients with moderate disease activity. It seems coherent that patients with the highest inflammatory burden obtain a tangible effect as compared with those with less inflammation, in which the disease improvement is less pronounced. This finding was only observed by the metaregression analysis issued from one datasharing platform. The meta-analysis of pooled data yielded no statistically significant difference but pointed to the same direction.
Our study has some strengths that deserve to be mentioned. This is the first analysis based on a large amount of IPD, studying the effect of demographic and disease factors on response to TNFi in RA, which allowed us to increase the power to show a very small difference between some subgroups of patients. The current knowledge are predominantly based on national registries, retrospective cohorts or aggregate data meta-analyses. The use of IPD from data-sharing platforms enables reusing raw data from a substantial number of studies. Furthermore, in comparison with meta-analyses based on aggregated data published by the investigators, obtaining raw data allowed us to study in a standardised manner various parameters and to pool them together, creating a large database. The results obtained here have important clinical impacts in the context of personalised treatment strategy, for example, to increase awareness on negative predictive factors such as obesity when initiating a TNFi.
The main challenge in our work was the data selection and availibility. In most cases, we chose to exclude from the subgroup analysis patients who could not be categorised, which may have caused bias in the analysis.35 Some adjustments were nevertheless made to limit this loss of data. For instance, we adapted the age intervals provided by the sponsor to our population subgroups, seeking for the best compromise (online supplemental table 5). Similarly, the absence of ESR in some studies led us to use DAS28(CRP) and ΔDAS28(CRP) in the categorisation of responder and non-responder patients instead of DAS28(ESR) and ΔDAS28(ESR), which may have somehow overestimated the EULAR response rate.36
Another limitation of our study is the different time points of response assessment, that is, at week 30 for 2 studies, at week 24 for 21 studies, at week 26 for 2 studies and at week 12 for 4 of them (online supplemental tables 6 and 7). At the time of the protocol writing, the 6-month time point was selected to capture all patients with sustained clinical response, to discriminate from those with early transient response, sometimes related to placebo effect. For studies where evaluation at 6 months was not available, we chose the 3-month time point. We made a compromise in order to include as many patients as possible in the analysis and to stick to the clinical relevance. In the treat-to-target strategy,4 European recommendations allow and require the clinician to evaluate the response treatment as early as 3 months.
For all the studies, we categorised treatment arms into two groups, either placebo or TNFi, whereas there were often several different TNFi groups with different dosages. In case of different treatment groups, all dosing regimens were gathered and compared with placebo. This did not hamper our conclusions, which was to compare the magnitude of response according to demographics and diseases characteristics, not according to the dosing regimen.
We did not plan to study the influence of concomitant treatments such as methotrexate, because the effect of such treatments is already known. Our objective was to study other treatment effect modifiers. Since such concomitant treatments were equally distributed between treatment arms, we considered that their influence on the effect of other covariate was unlikely. We did not analyse the dose of TNFi. The dose–response relationship would deserve a specific attention and further studies but was not the objective of this study. Because of the randomisation, we considered that this would not hamper our conclusions.
Finally, we should acknowledge the limitation of clinical trials and the extrapolation of our conclusions to patients seen in clinical practice. Differences in patients included in clinical trials, duration of treatment and follow-up, and the fact that we did not study the effect of TNFi monotherapy compared with their use in association with methotrexate could lead to differences in extrapolation to clinical practice. However, clinical trials remain the gold standard to assess treatment efficacy and the influence of treatment effect modifiers.
In conclusion, this meta-analysis on individual patients with RA data confirmed that obese patients are less responsive to TNFi as compared with non-obese patients. Those with long disease duration and those with a high baseline disease activity achieve a better response than others.
Data availability statement
Data may be obtained from a third party and are not publicly available. The raw data were not extracted but analysed remotely from the Yale Open Data Access and Vivli platforms.
Patient consent for publication
We thank the University of Tours for assistance with data use agreement. We are indebted to AMGEN and the Yale Open Data Access team for providing access to data. This publication is based on research using data from data contributors AbbVie, Pfizer and UCB that have been made available through Vivli. Vivli has not contributed to or approved, and is not in any way responsible for, the contents of this publication. We thank Professor Laure Gossec and Professor Cécile Gaujoux-Viala for offering their advice.
JL-W and M-AS contributed equally.
Contributors DM coordinated the work, participated in the interpretation of the analyses and helped draft the manuscript. TB-A initiated the project, performed the statistical analyses and helped draft the manuscript. SD wrote the protocol of the meta-analysis, selected the studies and asked for data. M-AS and JL-W ensured obtainment of data, analysed the results and drafted the manuscript. NA handled the database and helped in improving the manuscript. PG contributed to the interpretation of the results and manuscript improvement. JD provided individual patient data from the PMID 22739990 study evaluating adalimumab and helped in improving the manuscript. All authors read and approved the final version of the manuscript. DM is the guarantor for this work.
Funding This work was partly supported by the French Higher Education and Research Ministry under the program Investissements d’avenir (grant agreement: LabEx MAbImprove ANR-10-LABX-53-01).
Competing interests PG participated on behalf of his institution in clinical trials sponsored by Abbvie, Roche, BMS, Boehringer, Lilly, Novartis, Pfizer, UCB, Janssen and MSD; acted as a consultant and given lectures for Abbvie, Biogaran, BMS, Hospira, Janssen, MSD, Pfizer, Sanofi-Genzyme and UCB; had been invited to attend international congresses by MSD, Roche, BMS and Abbvie. DM acted as a consultant and gave lectures on behalf of his institution for Pfizer, Novartis and Grifols; had been invited to attend an international congress by Janssen-Cilag, GSK and Chugai. His institution received grants for research from the non-governmental organisation Lions Club Tours Val de France. M-AS, JL-W, SD, JD and TB-A declared no conflict of interest in relation with the present work.
Provenance and peer review Not commissioned; externally peer reviewed.