Target trial emulation: A practical framework for credible real-world evidence of healthcare interventions

Randomized controlled trials (RCTs) remain the gold standard for evaluating the causal effect of interventions. However, many clinical and policy questions cannot be fully addressed by RCTs due to ethical, logistical, or resource constraints. Observational studies using real-world data, including electronic health records, disease registries, and administrative claims, are also frequently used to complement trial evidence.[1] Yet, many such studies suffer from suboptimal design or analytical strategies, leading to avoidable confounding and bias.[2] Target trial emulation (TTE), proposed by Hernán and Robins,[3] offers a structured approach to reducing these biases by explicitly designing and analyzing observational data as if conducting a hypothetical randomized trial. Its methodological rigor and practical value are increasingly recognized by regulators, researchers, health technology assessment (HTA) agencies, and major journals.[4–7] This paper introduces the concept of TTE, clarifies where and when it is useful, outlines how to design and implement it, and highlights its strengths and limitations for clinical research [Figure 1].Figure 1: Concept of TTE and its applications. TTE: Target trial emulation.What is TTE? TTE is a methodological framework that uses observational data to emulate a hypothetical “target trial” capable of answering a causal question of interest [Supplementary Figure 1, https://links.lww.com/CM9/C937]. The process begins with specifying the protocol of the ideal trial and then implementing an observational analog of that protocol. The core of TTE lies in its protocol, which covers two dimensions, “target trial specification” and “TTE”, comprising seven key elements: eligibility criteria, treatment strategies, assignment procedures, follow-up period, outcomes, causal contrasts (e.g., intention-to-treat or per-protocol), and statistical analysis plan. Thus, TTE is not a single design or analytical tool but a structured procedure that promotes better practice in study design, conduct, and reporting. The recently published Transparent Reporting of Observational Studies Emulating a Target Trial guideline provides a transparent checklist for reporting studies that emulate target trials and helps improve methodological standards.[8] What makes a data source suitable for TTE? A fit-for-purpose data source is essential for successfully emulating a target trial. Commonly used sources include electronic health records, administrative claims databases, and registries. Determining requires assessing whether the data adequately capture four essential types of variables: eligibility criteria, treatment strategies (initiations, discontinuations, and adherence), outcomes, and key confounders (baseline and time-varying).[1] Clear operational definitions, coding algorithms, and validation strategies must be prespecified to minimize misclassification. Missing or poorly measured variables in any domain may substantially bias causal estimates. Where and when can TTE be used? TTE applies to a broad range of causal questions involving modifiable interventions, including medications, vaccinations, surgical procedures, lifestyle or behavioral interventions, health policies, and system-level changes. Although TTE can be implemented in many contexts, most published studies to date have involved pharmacologic interventions, typically evaluating their effectiveness and safety. TTE has also been recommended for use in policy evaluations and health technology assessments.[9–11] Beyond intervention types, TTE has been applied across a wide variety of clinical and public health areas, and its use has grown rapidly since 2019. Disease areas with particularly active adoption include infectious diseases, cardiology, oncology, and endocrinology.[12–14] Its application scenarios mainly include: (1) Study larger and more representative populations than those typically enrolled in randomized trials; (2) Evaluate long-term or rare outcomes over extended follow-up periods; (3) Compare active treatment strategies or treatment switching as they occur in clinical practice; (4) Assess urgent, emerging, or rapidly evolving interventions, such as those introduced during the coronavirus disease 2019 (COVID-19) pandemic; (5) Replicate or predict RCT findings; (6) Address evidence gaps not covered by trials or conflicting evidence in published observational studies. In these contexts, TTE offers a pragmatic alternative when RCTs are unavailable or infeasible. What TTE should not do? TTE is designed for causal questions about interventions. It cannot be used to study nonmodifiable exposures, such as sex, genetic factors, or gestational age.[15] Similarly, it is inappropriate to directly investigate the causal effect of biomarkers such as body mass index (BMI), blood pressure, or cholesterol on health outcomes, since these biomarkers are not interventions in themselves.[16] They can only be modified through specific treatments (e.g., diet, drugs, or surgery). Therefore, observed associations between biomarker levels and outcomes represent a mixture of effects from many underlying behaviors or treatments and cannot be causally interpreted. When biomarkers are relevant, researchers must reframe questions around interventions that modify those biomarkers—mirroring how trials evaluate targets (e.g., intensive vs. standard blood pressure control). When no intervention exists, the conceptual “set the biomarker to level X” strategy becomes purely hypothetical and inevitably subject to severe confounding. How TTE Strengthens Causal Inference? Evidence supporting the validity of TTE continues to accumulate. The RCT DUPLICATE initiative demonstrated strong concordance between the effect estimates derived from TTE using longitudinal claims data and those obtained from corresponding RCTs.[17] These results have strengthened confidence in the potential of TTE to inform regulatory, clinical, and health policy decision-making. Similar findings have been reported across multiple therapeutic areas, where well-designed TTE studies reproduced randomized trial findings, whereas naïve observational analyses often yielded biased results. TTE improves causal inference by preventing common design and analytical biases that frequently arise in observational studies. For example, naïve designs that require at least two prescriptions before defining treatment may introduce immortal time bias and selection bias due to posttreatment eligibility assessment.[18] Similarly, prevalent-user designs may exaggerate treatment benefits because long-term users represent a selected subgroup who tolerated therapy and survived early risks. A key principle of TTE is the alignment of eligibility assessment, treatment assignment, and time zero, which can be visually clarified using a design diagram. TTE can simulate randomization to control confounding factors using appropriate statistical methods (e.g., propensity score matching, inverse probability weighting). This approach substantially mitigates biases such as selection bias, prevalent-user bias, immortal-time bias, and misclassification. What Should be Done for TTE? To enhance transparency and credibility, TTE studies should preregister a protocol, similar to randomized trials, and explicitly document the analytic plan. Although TTE studies typically adopt cohort designs, nested case-control designs may be used when additional data collection is required to validate key measurements; detailed methodological guidance for applying this approach within a TTE framework has been published.[19] The success of a TTE study largely depends on: (1) the clarity and rigor of the target trial specification and (2) the fidelity with which it is implemented using observational data. Designing a target trial: practical steps. To design a target trial, the following recommended steps include: (1) Define a clear causal question, grounded in the available data. (2) Classify the application scenario as one of the following: replicating or predicting RCT findings, extending evidence to new populations or outcomes, or addressing questions where RCTs are infeasible. (3) Specify all PICO (population, intervention, comparison, outcome) elements that the hypothetical trial would include. (4) Review existing or similar RCTs according to the scenario: emulating an existing RCT with a matched PICO, consulting similar RCTs to inform a hypothetical trial, justifying the design based on clinical rationale in the absence of RCTs. (5) Develop a transparent protocol that details seven core methodological items of TTE framework. When extending evidence (scenario 2), credibility can be strengthened by first replicating RCT findings, then applying the same protocol to answer broader questions. Ultimately, researchers need to balance the ideal target trial with the trial that can be realistically emulated using available data. Implementing a TTE. Implementation requires careful operationalization of the seven core elements of TTE. Eligibility criteria are typically defined using baseline information. Incorporating postbaseline characteristics, for example, requiring participants to remain event-free or to achieve a minimum follow-up duration, may inadvertently introduce selection bias. Treatment strategies should be described as clearly as possible and translated into precise operational definitions with the data. Common types of treatment comparisons include: treatment A vs. no-treatment, treatment A vs. treatment B, combination treatment A+B vs. treatment A, and combination treatment A+B vs. combination treatment A+C. In real-world settings, treatment strategies may vary widely and often fall into three common categories, such as: (1) Point treatment strategies, which compare starting one treatment at baseline (e.g., initiating metformin vs. initiating a sulfonylurea for type 2 diabetes). (2) Sustained static treatment strategies, which require continued treatment over time (e.g., initiating and consistently continuing metformin vs. consistently continuing sulfonylurea). (3) Sustained dynamic treatment strategies, where treatment decisions depend on evolving clinical information such as biomarker levels or symptom changes, including treatment switching strategies. Assignment procedures can emulate randomization through suitable analytical methods, such as propensity score matching, G-methods, or doubly robust estimators. The choice of method often depends on the nature and complexity of the treatment strategy. Follow-up ideally begins at a well-defined time zero that aligns with eligibility assessment and treatment initiation. A visual design diagram is recommended to clearly illustrate this alignment. Outcomes are best prespecified, with explicit definitions and validation procedures to ensure accuracy and relevance. Causal contrast, whether intention-to-treat or per-protocol, should be stated explicitly and selected in accordance with the research question and the intended interpretation. Statistical methods should be chosen to match the specified causal contrast, with sensitivity analyses incorporated to evaluate robustness and to explore the potential influence of unmeasured confounding, especially when studying dynamic treatment strategies. For the effect of sustained treatment strategies, we recommend causal inference methods (e.g., marginal structural model, G-methods, and time-dependent propensity score matching) to adjust for time-dependent confounding. In comparative effectiveness studies, an active comparator new-user design is generally preferred. No-use comparators may be appropriate only when confounding by indication is minimal for the outcomes of interest. In such cases, sequential trial emulation can improve analytic efficiency. The clone-censor-weight approach is particularly useful for handling grace periods (e.g., initiate treatment within one month of eligibility), sustained treatment strategies, or treatment initiation triggered by biomarker levels or repeated prescriptions. Conducting both internal and external validation is recommended to ensure the robustness of the findings.[20] What Works in Practice: Collaboration and Technical Requirements TTE bridges the gap between observational research and randomized trials but involves considerable methodological and analytical complexity.[21] High-quality TTE studies require close collaboration among methodologists, clinicians, data scientists, and artificial intelligence (AI) specialists. Such interdisciplinary work ensures scientific rigor and technical feasibility across all stages, from assessing data sources and specifying protocols to analysis, interpretation, and reporting. The validity of TTE is inherently constrained by the quality of the observational data. TTE cannot eliminate biases such as measurement error, missing data, and selection bias due to loss to follow-up, nor can it fully address unmeasured or residual confounding arising from the absence of randomization. Analytical approaches such as negative control analyses or E-values may help assess the magnitude of possible unmeasured confounding. Furthermore, because TTE emulates pragmatic rather than explanatory trials, blinding is generally infeasible. In summary, TTE represents an important methodological innovation in real-world evidence research, offering a structured and transparent approach to align observational analyses with the principles of randomized trials. By explicitly defining the target trial and operationalizing its key elements, TTE helps minimize common biases and strengthens the credibility of causal inference drawn from routine clinical data [Figure 1]. As data quality improves and analytic techniques continue to advance, TTE is poised to expand its role in addressing questions that randomized trials cannot feasibly answer, providing a robust and timely framework for generating reliable evidence to inform clinical practice and health policy. Funding This study was supported by grants from the National Natural Science Foundation of China (Nos. 82474335 and 82474334), the Key R&D Projects of Sichuan Provincial Department of Science and Technology (No. 2024YFFK0152), special fund for traditional Chinese medicine of Sichuan Provincial Administration of Traditional Chinese Medicine (No. 25ZDAZX008), Traditional Chinese Medicine Innovation Team and Talent Support Program-National Traditional Chinese Medicine Multidisciplinary Cross-Innovation Team Project (No. ZYYCXTD-D-202401), 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University (Nos. ZYYC24010 and ZYGD23004). Conflicts of interest None.