FRI0519 IMPROVING RHEUMATOID ARTHRITIS COMPARATIVE EFFECTIVENESS RESEARCH USING THE TARGET TRIAL EMULATION FRAMEWORK: A SYSTEMATIC REVIEW

Abstract

Background:Methods used in observational comparative effectiveness research (CER) are highly variable. Target trial emulation is an intuitive design approach that encourages researchers to formulate their question as a hypothetical randomised controlled trial (RCT), or the “target trial”. Using observational data to emulate the target trial helps avoid common biases and has been shown to better align results with actual RCTs.Objectives:We systematically reviewed observational CER studies in rheumatoid arthritis to provide examples of design issues that might have been avoided by using target trial emulation.Methods:We searched for head-to-head effectiveness comparisons of biologic DMARDs in RA. Study designs were reviewed for components of target trial emulation: 1) eligibility criteria, 2) treatment strategies, 3) assignment procedures, 4) follow-up period, 5) outcome, 6) causal contrasts of interest (i.e., intention-to-treat or per-protocol effect), and 7) analysis plan. Reported methods were taken as the “emulation” of a corresponding target trial, to assess design issues that might introduce bias.Results:We found 31 CER studies, the majority of which had one design issue belonging to one of the 7 protocol components (Table 1). The most common issues were: 1) 17 out of 31 studies used post-baseline information to define baseline eligibility (e.g. requiring ≥1 follow-up), which can bias results; 2) 26 out of 31 studies did not declare their causal contrast of interest, which is often made difficult by issue 1 and impacts data analysis and interpretation; and 3) 9 out of 31 studies used statistical selection of confounders rather than pre-defining them, which can also introduce bias (e.g. through adjustment of collider or intermediate variables).Table 1.Design issues identified in 31 studies and reasons why they do not correspond to well-defined “target trials”Design issues identified in study methodsHow these issues can be conceptualized in a RCT protocol1. Eligibility criteriaPost-baseline data requirement (17 out of 31 studies).Impossible to use future data at enrolment.Differential eligibility for each arm (5 studies).Breaks the notion of one group of people randomized to 2+ arms.2. Treatment strategiesMixing prevalent users and new users (1 study)Impossible to assign/randomize to “havingused drug A for X months”Not defining treatment strategies beyond “initiate drug A at baseline” (31 studies)Implied protocol leaves everything up to the treating physician and patient3. Assignment proceduresWeak substantive justification for confounder selection (31 studies)Broken randomization (due to insufficient emulation of randomization)4. Follow-upUnspecified follow-up duration in longitudinal analyses (5 studies)Infeasible to conceive an RCT with unspecified duration. Analysis results may lack interpretability.5. OutcomeJoint outcome of remaining on treatmentandhaving a good response, to avoid missingness (3 studies)Unusual outcome for RCT although technically possible.6. Causal contrasts (i.e., ITT or per-protocol effect)Failure to clarify the estimand (26 studies)Problem also common in RCTs7. Analysis planITT-type analysis among those with follow-upDeviates from the ITT principle (all randomized should be analysed)Per-protocol analyses did not account for post-baseline selection biasProblem also common in RCTsConclusion:The majority of observational CER studies in RA have one or more design issues that may introduce bias. Target trial emulation is a structured approach for designing observational CER studies that helps to avoid common biases.Disclosure of Interests:Sizheng Steven Zhao: None declared, Houchen Lyu: None declared, Daniel Solomon Grant/research support from: Funding from Abbvie and Amgen unrelated to this work, Kazuki Yoshida: None declared