Therapeutic Equivalence of Biosimilar and Reference Biologic Drugs in Rheumatoid Arthritis: A Systematic Review and Meta-analysis

Bruna de Oliveira Ascef; Matheus Oliveira Almeida; Ana Cristina de Medeiros-Ribeiro; Danieli Castro Oliveira de Andrade; Haliton Alves de Oliveira Junior; Patrícia Coelho de Soárez

doi:10.1001/jamanetworkopen.2023.15872

. 2023 May 26;6(5):e2315872. doi: 10.1001/jamanetworkopen.2023.15872

Therapeutic Equivalence of Biosimilar and Reference Biologic Drugs in Rheumatoid Arthritis

A Systematic Review and Meta-analysis

Bruna de Oliveira Ascef ^1,^✉, Matheus Oliveira Almeida ², Ana Cristina de Medeiros-Ribeiro ³, Danieli Castro Oliveira de Andrade ⁴, Haliton Alves de Oliveira Junior ⁵, Patrícia Coelho de Soárez ⁶

¹Departamento de Medicina Preventiva, Faculdade de Medicina, Universidade de Sao Paulo, Sao Paulo, Brazil

²Programa de Pos-Graduaçao em Fisioterapia, Universidade Ibirapuera, Sao Paulo, Brazil

³Disciplina de Reumatologia do Hospital das Clinicas da Faculdade de Medicina da Universidade de Sao Paulo, Sao Paulo, Brazil

⁴Disciplina de Reumatologia do Hospital das Clinicas da Faculdade de Medicina da Universidade de Sao Paulo, Sao Paulo, Brazil

⁵Health Technology Assessment Unit, Hospital Alemao Oswaldo Cruz, Sao Paulo, Brazil

⁶Departamento de Medicina Preventiva, Faculdade de Medicina, Universidade de Sao Paulo, Sao Paulo, Brazil

Accepted for Publication: April 15, 2023.

Published: May 26, 2023. doi:10.1001/jamanetworkopen.2023.15872

^✉

Corresponding Author: Bruna de Oliveira Ascef, PhD, Departamento de Medicina Preventiva, Faculdade de Medicina, Universidade de Sao Paulo, Av. Dr. Arnaldo, 455 – 2° andar – sala 2214, Sao Paulo, SP 01246-903, Brazil (brunaascef16@gmail.com).

Author Contributions: Drs Ascef and de Soárez had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Ascef, Almeida, Medeiros-Ribeiro, Oliveira de Andrade, de Soárez.

Acquisition, analysis, or interpretation of data: Ascef, Almeida, Medeiros-Ribeiro, Oliveira Junior.

Drafting of the manuscript: Ascef, Almeida, Oliveira Junior.

Critical revision of the manuscript for important intellectual content: Almeida, Medeiros-Ribeiro, Oliveira de Andrade, Oliveira Junior, de Soárez.

Statistical analysis: Ascef.

Administrative, technical, or material support: Almeida, Oliveira Junior.

Supervision: Almeida, Medeiros-Ribeiro, Oliveira de Andrade, Oliveira Junior, de Soárez.

Conflict of Interest Disclosures: None reported.

Data Sharing Statement: See Supplement 2.

Additional Contributions: Tiago da Veiga Pereira, PhD (Applied Health Research Centre, Li Ka Shing Knowledge Institute, St Michael’s Hospital, Toronto, Ontario, Canada and Department of Health Sciences, College of Medicine, University of Leicester, Leicester, UK), made substantial contributions to design, analysis, and interpretation of data and writing assistance for this manuscript. No compensation was given. Dr Veiga Pereira has permitted us to include his contribution in this manuscript.

^✉

Corresponding author.

PMCID: PMC10220520 PMID: 37234004

Key Points

Question

Are biosimilars of adalimumab, etanercept, and infliximab associated with equivalent treatment compared with their reference biologic drugs for the management of rheumatoid arthritis?

Findings

In this systematic review and meta-analysis of 25 randomized clinical trials that included data on 10 642 patients with rheumatoid arthritis, patients using biosimilars had equivalent clinical responses and functional capacity compared with patients using reference biologic drugs.

Meaning

These findings suggest that biosimilars may yield therapeutically equivalent outcomes compared with reference biologic drugs for the management of rheumatoid arthritis.

This systematic review and meta-analysis assesses the efficacy, safety, and immunogenicity associated with biosimilars of adalimumab, etanercept, and infliximab compared with their reference biologics for patients with rheumatoid arthritis.

Abstract

Importance

Biosimilar drugs are potentially lower-cost versions of biologics that may improve access to therapy. However, there is a lack of adequate systematic reviews demonstrating equivalence between these drugs for the treatment of rheumatoid arthritis (RA).

Objectives

To assess the efficacy, safety, and immunogenicity associated with biosimilars of adalimumab, etanercept, and infliximab compared with their reference biologics in patients with RA.

Data Sources

MEDLINE via PubMed, Embase, Cochrane Central Register of Controlled Trials, and LILACS databases were searched from inception to September 2021.

Study Selection

Head-to-head randomized clinical trials (RCTs) of biosimilars of adalimumab, etanercept, and infliximab and their biologic reference drugs for RA were assessed.

Data Extraction and Synthesis

Two authors independently abstracted all data. Meta-analysis was conducted with bayesian random effects using relative risks (RRs) for binary outcomes and standardized mean differences (SMDs) for continuous outcomes, with 95% credible intervals (CrIs) and trial sequential analysis. Specific domains were assessed for the risk of bias in equivalence and noninferiority trials. This study was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guideline.

Main Outcomes and Measures

Equivalence was tested using prespecified margins for the American College of Rheumatology criteria, with at least 20% improvement in the core set measures (ACR20) (ie, RR, 0.94 to 1.06), and for the Health Assessment Questionnaire–Disability Index (HAQ-DI) (ie, SMD, −0.22 to 0.22). Secondary outcomes included 14 items measuring safety and immunogenicity.

Results

A total of 25 head-to-head trials provided data on 10 642 randomized patients with moderate to severe RA. Biosimilars met equivalence with reference biologics in terms of ACR20 response (24 RCTs with 10 259 patients; RR, 1.01; 95% CrI, 0.98 to 1.04; τ² = 0.000) and change of HAQ-DI scores (14 RCTs with 5579 patients; SMD, −0.04; 95% CrI, −0.11 to 0.02; τ² = 0.002) considering prespecified margins of equivalence. Trial sequential analysis found evidence for equivalence for ACR20 since 2017 and HAQ-DI since 2016. Overall, biosimilars were associated with similar safety and immunogenicity profiles compared with reference biologics.

Conclusion and Relevance

In this systematic review and meta-analysis, biosimilars of adalimumab, infliximab, and etanercept were associated with clinically equivalent treatment effects compared with their reference biologics for the treatment of RA.

Introduction

Rheumatoid arthritis (RA) is a debilitating inflammatory condition that primarily affects the joints, reducing physical function and quality of life.¹ Globally, approximately 20 million people have RA.² Tumor necrosis factor-α inhibitors (TNFIs), such as infliximab, adalimumab, and etanercept, are biologic disease-modifying antirheumatic drugs used worldwide to treat RA.^3,4,5,6 While robust evidence has demonstrated the efficacy and safety of TNFIs for managing RA, the high costs of TNFIs can limit treatment access to these drugs worldwide.^6,7

Biosimilar drugs are potentially lower-cost versions of TNFIs that have been used in RA treatment and have the potential to improve access to therapy.^{6,8,9,10,11,12} International guidelines and consensus recommend biosimilars for RA management.^{5,6,8,13,14,15,16,17,18} Nonetheless, these recommendations have relied mainly on single trials or expert consensus, with only 2 guidelines using meta-analysis.^17,18

Previous systematic reviews^{18,19,20,21,22,23,24,25,26,27,28,29} attempted to compile evidence on similar efficacy and safety between biosimilars and reference products for RA. Nonetheless, most of these reviews^{18,19,21,23,24,27} provided only qualitative summaries, which are insufficient for decision-making. The few reviews^{20,22,25,26,28,29} that attempted quantitative analyses did not use appropriate equivalence testing methods, resulting in imprecise conclusions.³⁰

In this study, we conducted a systematic review and meta-analysis of randomized clinical trials (RCTs) comparing biosimilar drugs and reference products of the most prescribed TNFIs (infliximab, adalimumab, and etanercept) for RA.^{31,32,33,34,35,36} We addressed previous reviews’ limitations by using equivalence testing via prespecified margins and bayesian meta-analyses.

Methods

For this systematic review and meta-analysis, a multidisciplinary panel developed the protocol (PROSPERO identification: CRD42019137152) and published it elsewhere.³⁷ Changes to the protocol are presented in eAppendix 1 in Supplement 1. We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline³⁸ and a specific guidance for equivalence and noninferiority systematic reviews.³⁹

Eligibility Criteria

We included RCTs investigating patients of both sexes with RA. No limitations regarding sociodemographic characteristics or disease severity or duration were imposed. Interventions of interest were biosimilars of adalimumab, etanercept, and infliximab. Comparators of interest were the reference biologic drugs (ie, adalimumab, etanercept, and infliximab originals). No restrictions were imposed on trial design, dosages, treatment schedules, or cotreatments. The full eligibility criteria are given in eAppendix 2 in Supplement 1.

Evidence Sources, Search Strategy, and Selection Process

A detailed search strategy description is available in our published protocol³⁷ and reproduced in eAppendix 3 in Supplement 1. We conducted a comprehensive literature search in MEDLINE (via PubMed), Embase, Cochrane Central Register of Controlled Trials, and LILACS from database inception to September 7, 2021. We also searched for unpublished or ongoing trials in 4 trial registry databases and performed citation searches of all included studies. No language limitation was imposed. Two investigators (B.O.A. and M.O.A.) independently assessed titles, abstracts, and full-length articles against the eligibility criteria (eAppendix 4 in Supplement 1).

Data Extraction

Whenever available, we collected the population per-protocol (PP) data (eAppendix 5 in Supplement 1). Two investigators (B.O.A. and M.O.A.) extracted all data independently, and discrepancies were solved via a consensus or consultation with a third reviewer (P.C.S.). The complete list of variables extracted for each trial is shown in eAppendix 5 in Supplement 1.

Outcomes

Primary Outcomes: Efficacy

The prespecified primary efficacy end point was the treatment success at 6 months, according to the American College of Rheumatology 20% response criteria (ACR20), which requires at least a 20% improvement in the core set measures for a patient to reach improvement. ACR20 was summarized as relative risk (RR), with an RR greater 1.0, indicating a higher response probability with biosimilar drugs compared with reference biologics (eAppendix 6 in Supplement 1).

We also prespecified the Health Assessment Questionnaire–Disability Index (HAQ-DI) at 6 months of follow-up as a primary outcome, which measures disability and is patient-reported. HAD-DI was presented as a standardized mean difference (SMD), Cohen effect size and an SMD less than 0 indicate a better outcome for biosimilar drugs than in reference biologics (eAppendix 6 in Supplement 1).

Secondary Outcomes

Efficacy

We included the ACR 50% response criteria (ACR50) and ACR 70% response criteria (ACR70) as secondary efficacy outcomes. Due to the volume of data, additional secondary outcomes of efficacy prespecified in the protocol will be shared in future publications (eAppendix 7 in Supplement 1).

Safety and Immunogenicity

The prespecified safety and immunogenicity outcomes included treatment-emergent adverse events (TEAEs), serious adverse events, special adverse events, mortality, discontinuation rates, positive antidrug antibodies (ADAs) formation, and positive neutralizing antibodies (NAbs) (eAppendix 8 in Supplement 1).

Assessment of Risk of Bias

We used the Cochrane Risk of Bias tool (version 1.0)⁴⁰ to assess the risk of bias in each trial, using the ratings of 2 independent reviewers (B.O.A. and M.O.A.). Additionally, we addressed specific domains of equivalence or noninferiority trials³⁹ (eAppendix 9 in Supplement 1).

Statistical Analysis

Data Synthesis

The approaches to approximate means and SDs from the reported statistics are shown in eAppendix 10 in Supplement 1. Binary outcomes were summarized using the RR as a metric, whereas continuous outcomes were summarized as SMDs. We combined results across trials using random-effects models, which allows for between-trial variability.⁴¹ However, we replaced the prespecified frequentist model with fully bayesian random-effects models (eAppendix 1 in Supplement 1).

For binary outcomes, we used binomial likelihood and modeled the log RR directly. For continuous outcomes, we used the normal likelihood and the identity link. We assumed noninformative but biologically plausible priors for treatment effects. Details on the models, model diagnostics, and estimation methods are presented in eAppendix 11 in Supplement 1.

Summary treatment effect estimates and between-trial variance were derived from the median and 95% credibility intervals (CrIs) from the 2.5th and 97.5th percentile of the posterior distribution.

We conducted prespecified subgroup analyses for primary outcomes based on patient and drug characteristics and sensitive analyses based on the study’s methodological characteristics (eAppendix 12 in Supplement 1). We used prespecified frequentist fixed-effects meta-analysis models as a sensitivity analysis. For continuous outcomes, we used the inverse-variance model. For safety or immunogenicity outcomes, we used the Mantel-Haenszel method.

We investigated the association between trial size and treatment effects in contour-enhanced funnel plots for primary outcomes in the context of equivalence testing (eAppendix 13 and eFigure 1 in Supplement 1) and traditional funnel plots for the remaining outcomes (<10 studies). We also used Egger test for continuous outcomes and Harbord test for binary outcomes. Results with a P < .10 were considered statistically significant for Egger and Harbord tests.

We conducted nonprespecified trial sequential analyses based on large, randomized trials only to assess whether the combined number of analyzed participants was sufficient to draw definitive conclusions about the equivalence associated with biosimilars and reference biologics or whether more trials are still needed (eAppendix 14 in Supplement 1).⁴² We defined a large trial as a study that randomized at least 500 participants.⁴³ We used this cutoff because large trials are less likely to be affected by small-study effects or publication bias.⁴³ All analyses were conducted in MultiBUGS version 2.0 and Stata version 16 (StataCorp). P values were 2-sided for primary outcomes and 1-sided for secondary outcomes. Statistical significance was set at P < .05.

Margins of Equivalence

We considered equivalence as when the 95% CrI of summary estimates fell completely within the lower and upper prespecified equivalence margins.³⁷ The rationale for the choice of the equivalence margins is described elsewhere.³⁷ For the ACR20 outcome, we assumed an equivalence margin for RRs ranging from 0.94 to 1.06. For the HAQ-DI outcome, the equivalence margin in SMD units was −0.22 to 0.22. Thus, the probability of equivalence is the proportion of Markov Chain Monte Carlo simulations in which the random-effects summary estimate was within the equivalence margins. Importantly, the term equivalence used in this study refers to the statistical and clinical comparability of clinical responses between biosimilars and reference biologics and does not have the same value as the regulatory terms biosimilarity, bioequivalence, or interchangeability, provided by regulatory agencies, such as the US Food and Drug Administration.^44,45

Assessment of Certainty of Evidence

We assessed the overall certainty of the evidence using the GRADE system⁴⁶ (eAppendix 15 in Supplement 1). Data were analyzed from January 2022 to April 2023.

Results

Search Results

eFigure 2 in the Supplement summarizes the study selection process. Of 2023 references assessed, 25 RCTs^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94} met the eligible criteria.

RCT Characteristics

The Table summarizes the characteristics of the 25 included trials, which included a total of 10 649 randomized participants, with a median (IQR) sample size of 426 (108 to 596) patients. The median (IQR) baseline age of the participants was 53 (51 to 54) years, and the median (IQR) proportion of females was 81% (80% to 84%).

Table. Main Characteristics of Included Trials and Patients at Baseline.

Source	Biosimilar drug	Reference drug	Study design (efficacy phase)	Follow-up, wk	Randomized patients, No.	Age, mean (y)	Female patients, No. (%)
Jani et al, 2015⁴⁷	ZRC-3197	ADA	Equivalence	12	120	45.0	99 (82.5)
Alten et al, 2017^{48,49,50,51,52,53,54}	FKB327	ADA	Equivalence	24	730	53.3	565 (77.6)
Cohen et al, 2017^55,56	ABP-501	ADA	Equivalence	26	526	55.9	426 (81.0)
Jamshidi et al, 2017⁵⁷	CinnoRA	ADA	Noninferiority	24	136	47.9	118 (86.8)
Fleishmann et al, 2018^58,59,60	PF-06410293	ADA	Equivalence	26	597	52.5	470 (78.7)
Cohen et al, 2018^61,62	BI-695501	ADA	Equivalence	24	645	53.7	536 (83.1)
Weinblatt et al, 2018^63,64	SB5	ADA	Equivalence	24	544	51.2	441 (81.1)
Edwards et al, 2019⁶⁵	MSB11022	ADA	Superiority	52	288	54.0	227 (78.8)
Willand et al, 2019^66,67	GP2017	ADA	Equivalence	24	353	53.3	295 (83.5)
Matsuno et al, 2021⁶⁸^a	LBAL	ADA	Equivalence	24	383	NA	NA
Kay et al, 2021^69,70	CT-P17	ADA	Equivalence	24	648	53.8	514 (79.3)
Emery et al, 2015^71,72,73	SB4	ETN	Equivalence	52	596	51.8	502 (84.2)
Bae et al, 2016⁷⁴	HD203	ETN	Equivalence	48	294	51.2	202 (86.7)
Odell et al, 2016^75,76^a	CHS-0214	ETN	Equivalence	24	647	NA	514 (79.8)
Matsuno et al, 2017^77,78	LBEC0101	ETN	Equivalence	54	374	54.1	316 (84.9)
Matucci-Cerinic et al, 2018^79,80	GP2015	ETN	Equivalence	24	376	54.2	308 (81.9)
Yamanaka et al, 2020⁸¹	YLB113	ETN	Equivalence	56	528	52.3	409 (78.0)
Strusberg et al, 2021⁸²	Enerceptan	ETN	Noninferiority	32	150	48.3	127 (85.2)
Yoo et al, 2013^83,84,85	CT-P13	IFX	Equivalence	54	606	50.0	501 (82.7)
Kay et al, 2014^86,87^a	BOW015	IFX	Equivalence	16	189	NA	NA
Choe et al, 2015^88,89,90	SB2	IFX	Equivalence	54	584	52.1	468 (80.1)
Takeuchi et al, 2015⁹¹	CT-P13	IFX	Equivalence	54	108	54.2	81 (80.2)
Matsuno et al, 2018⁹²	NI071	IFX	Equivalence	30	242	53.9	204 (84.3)
Lila et al, 2019⁹³	BCD-055	IFX	Equivalence	54	426	53.0	337 (80.6)
Genovese et al, 2020⁹⁴	ABP710	IFX	Equivalence	50	558	54.9	437 (78.3)
Total	25	3	22 Equivalence, 2 noninferiority, 1 superiority	26 (24-52)^b	426 (288-596)^b	53.1 (51.3-54.0)^b	81.1 (79.5-83.8)^b

Open in a new tab

Abbreviations: ADA, adalimumab; ETN, etanercept; IFX, infliximab; NA, not available.

^{^a}

Unpublished trials in peer-reviewed journals (abstract congress).

^{^b}

Expressed as median (IQR).

All trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94} included patients with moderate to severe RA and experience with methotrexate. Sixteen (64%) trials^{48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,69,70,71,72,73,74,75,76,77,78,79,80,81,83,84,85,88,89,90,91,93} reported the use of concomitant methotrexate in both treatment groups, while in 9 trials^{47,63,64,65,66,67,68,82,86,87,92,94} (36%), it was unclear. eTable 1 and eTable 2 in Supplement 1 provide additional information on the included trials. Overall, the most investigated biosimilars were those of adalimumab (11 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70} [44%]), followed by biosimilars of etanercept^{71,72,73,74,75,76,77,78,79,80,81,82} and infliximab^{83,84,85,86,87,88,89,90,91,92,93,94} biosimilars, with 7 trials each (28%) (Table).

All 25 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94} were industry-sponsored studies, and 3 trials^68,75,86 (12%) were unpublished investigations. Most trials (22 trials^{47,48,49,50,51,52,53,54,55,56,58,59,60,61,62,63,64,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,83,84,85,86,87,88,89,90,91,92,93,94} [88%]) were equivalence trials, 2 trials^57,82(8%) were noninferiority trials, and 1 trial⁶⁵ (4%) was a superiority trial. The median (IQR) follow-up was 26 (24-52) weeks (Table). Additional methodological characteristics of the included trials are provided in eTable 3 and eTable 4 in Supplement 1.

Risk of Bias

eFigure 3 in Supplement 1 shows the risk of bias in 25 trials.^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94} Fourteen trials^{47,48,49,50,51,52,53,54,55,56,61,62,63,64,65,69,70,71,72,73,74,79,80,81,82,88,89,90,93} (56%) had a low risk of bias for random sequence generation, 16 trials^{48,49,50,51,52,53,54,55,56,57,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,77,78,79,80,81,82,88,89,90,92,93} (64%) had a low risk of bias for allocation concealment, 19 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,69,70,71,72,73,74,77,78,81,82,83,84,85,88,89,90,91,94} (76%) had a low risk of bias for inconsistent application criteria, 12 trials^{47,48,49,50,51,52,53,54,57,63,64,69,70,71,72,73,74,77,78,79,80,88,89,90,91,93} (48%) had a low risk of bias for blinding of patients and investigators, 19 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,69,70,74,77,78,79,80,81,82,83,84,85,88,89,90,91,92,94} (76%) had a low risk of bias for patient’s behavior changes, 10 trials^{63,64,69,70,71,72,73,74,77,78,79,80,82,88,89,90,91,93} (40%) had a low risk of bias for blinding of outcome assessors, 22 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,69,70,71,72,73,74,77,78,79,80,81,82,83,84,85,88,89,90,91,92,93,94} (88%) had a low risk of bias for outcomes measures, and 8 trials^{47,57,58,59,60,61,62,63,64,65,69,70,81} (32%) had a low risk of bias for incomplete outcome data. Details on the risk of bias are provided in eTables 5-8 in Supplement 1.

Primary Efficacy Outcomes

ACR20 After 6 Months of Treatment

A total of 24 trials^{47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94} involving 10 259 randomized patients contributed data for the primary outcome of ACR20 response at 6 months. As shown in Figure 1A, when data were combined, the bayesian random-effects summary was tiny (RR, 1.01; 95% Crl, 0.98 to 1.04), with no evidence of heterogeneity (τ² = 0.000). The 95% bayesian CrI was entirely contained within the prespecified equivalence margin of 0.94 to 1.06 and met our prespecified definition of equivalence. The posterior probability of equivalence was 100% (eTable 9 in Supplement 1). Overall, 79% of patients receiving biosimilars and 78% of patients receiving reference biologics experienced an ACR20 response after 6 months of treatment. The prespecified summary estimate using a frequentist fixed-effect model indicated similar conclusions (eTable 10 in Supplement 1). Nonprespecified exploratory analyses that focused only on studies reporting intention-to-treat (ITT) analyses (eFigure 4 in Supplement 1) or PP analyses (eFigure 5 in Supplement 1) found identical conclusions compared with the main analysis.

Figure 1. — A, Meta-analysis of ACR20, including 24 trials comparing biosimilars vs reference biologic drugs (10 259 randomized patients). The shaded area denotes the margins of equivalence (relative risk [RR], 0.94 to 1.06). The estimate of the between-trial variance, τ², was 0.000 (low heterogeneity). B, Meta-analysis of HAQ-DI, including 14 trials of biosimilars and reference biologic drugs (5579 randomized patients). The shaded area denotes the margins of equivalence (standardized mean difference [SMD], −0.22 to 0.22). The estimate between-trial variance, τ², was 0.002 (low heterogeneity). Summary results are based on a bayesian random-effects model. Point estimates for primary studies are displayed with a 95% CI. Meta-analysis estimates are shown with a 95% credible interval (CrI). The number of participants analyzed may be smaller than the number of randomized participants.

HAQ-DI After 6 Months of Treatment

A total of 14 trials^{47,48,49,50,51,52,53,54,57,58,59,60,65,66,67,71,72,73,74,75,76,79,80,82,83,84,85,88,89,90,91} with 5579 randomized participants contributed data for HAQ-DI at 6 months. The bayesian random-effects summary SMD was −0.04 (95% CrI, −0.11 to 0.02), with low statistical heterogeneity (τ² = 0.002) (Figure 1B), which falls entirely within the prespecified margins of equivalence of −0.22 to 0.22 (eTable 11 in Supplement 1). The summary SMD corresponds to a difference of −0.03 units (95% CrI, −0.08 to 0.01) on the HAQ-DI scale (range, 0-3). The prespecified sensitivity analysis using a frequentist fixed-effect model indicated analogous results (eTable 12 in Supplement 1). Exploratory analyses based on ITT studies only (eFigure 6 in Supplement 1) and PP analyses only (eFigure 7 in Supplement 1) found identical conclusions as the main analysis.

Subgroup Analysis for ACR20 and HAQ-DI

The ACR20 response by subgroups yielded similar conclusions compared with the main analysis. In 8 of the 14 subgroups in Figure 2A, the 95% CrIs of the Bayesian summary RRs fell entirely within the predefined equivalence range, with low heterogeneity (τ² ranging from 0 to 0.05). The subgroup analyses for HAQ-DI followed the same patterns, with the 95% CrIs of most subgroups within the equivalence margins, with posterior probabilities of equivalence between 73% to 100% (Figure 2B).

Trial Sequential Analysis Based on Large Trials: ACR20 and HAQ-DI

For the primary outcome of ACR20, trial sequential analysis based on 12 large trials^{48,49,50,51,52,53,54,55,56,58,59,60,61,62,63,64,69,70,71,72,73,75,76,81,83,84,85,88,89,90,94} with 7207 randomized participants demonstrated that the random-effects cumulative Z score crossed the boundary for equivalence in 2017, before the required information size was reached (Figure 3A). Additional trials after that year did not change the results, suggesting that the results are definitive. For the HAQ-DI, trial sequential analyses based on 6 large trials^{48,49,50,51,52,53,54,58,59,60,71,72,73,75,76,83,84,85,88,89,90} with 3758 randomized participants found that the accumulated number of participants reached the required information (eg, additional trials will not change the results). The random-effects Z score crossed the boundary of equivalence in 2016, indicating conclusive results (Figure 3B).

Figure 3. — A, Results are based on 12 large, head-to-head trials of biosimilars and reference biologic drugs (7207 randomized participants). RIS indicates required information size (vertical line) to show equivalence considering the prespecified margins (relative risk [RR], 0.94 to 1.06]) with 90% of power at an α = .005. B, Results based on 6 large, head-to-head trials of biosimilars and reference biologic drugs for HAQ-DI (3758 randomized participants). The RIS was calculated as the sample size that gives a trial with 90% power to show equivalence (standardized mean difference [SMD] −0.22 to 0.22) with 2-sided α = .05. For A and B, cumulative z scores were estimated using a random-effects model with the restricted maximum likelihood estimator. The blue lines represent the O’Brien-Fleming monitoring boundaries. The orange characterizes the inner wedges (ie, futility boundaries), indicating the limits to the equivalence region. Circles indicate the cumulative z score for each additional trial added to the analysis. We assumed a between-trial variation using diversity (D2) index-adjusted sample sizes of 50%. The number of participants analyzed was shown by year and may be smaller than the number of randomized participants.