Target Trial Emulation vs RCTs

Why this question keeps coming up

Randomised controlled trials are slow, expensive, and sometimes impossible to run, you cannot randomise patients to a surgical complication, and some trials close before enrolling enough patients to answer the question they were designed for. At the same time, health systems now generate enormous volumes of routinely collected data: insurance claims, electronic health records, national registries, cancer databases. The appeal is obvious. If a large observational dataset, analysed with the right causal-inference methods, can reproduce what a randomised trial would have found, it promises answers to questions trials will never be funded to ask, rare outcomes, long-term effects, populations trials routinely exclude. Whether that promise holds, and under what conditions, is a live methodological argument rather than a settled one, and it is worth understanding before citing an observational study as if it carried a trial's weight.

What target trial emulation actually is

Target trial emulation is not a single statistical test but a discipline for designing an observational analysis. The idea, set out by Miguel Hernán and James Robins, is to first write the protocol for the randomised trial you would run if you could, its eligibility criteria, treatment strategies, assignment procedure, follow-up period, outcome, and analysis plan, and then emulate each element as closely as possible using observational data.¹ Where a trial randomises patients to treatment at a defined "time zero," an emulation instead defines time zero as the point a patient in the data met eligibility and started (or did not start) the treatment strategy being compared, and uses methods such as propensity-score matching or inverse-probability weighting to balance the groups on measured confounders. The point of being this explicit is to avoid a set of well-documented, self-inflicted biases that arise when observational analyses skip the step of specifying what trial they are supposed to be emulating, most commonly immortal time bias, where person-time before treatment initiation is miscounted as exposed time, inflating the apparent benefit of the treatment.²

What happens when emulations are tested against real trials

Because the framework makes an explicit claim, that a well-designed emulation should approximate the corresponding trial's result, it can be tested directly, and a growing body of work has done exactly that. The most systematic attempt is the RCT DUPLICATE initiative, which used US insurance claims data to emulate ten completed cardiovascular and diabetes drug trials, prespecifying the protocol and registering it before running the analysis. The regulatory conclusion (whether the drug would have been approved on the basis of that evidence) matched the original trial's conclusion in 6 of the 10 pairs, and the emulation's point estimate fell within the trial's 95% confidence interval in 8 of 10.³ A separate emulation of 32 completed and ongoing trials using claims databases found similar variability: agreement was strong (Pearson correlation 0.93) in the 16 trials where the trial's design and measurements could be closely reproduced in the data, but weaker (Pearson correlation 0.53) in the 16 trials where key design elements, a particular comparator, a specific endpoint definition, simply were not available in claims data.⁶ The most recent and largest synthesis, a 2026 systematic review and meta-analysis of 107 published emulation-RCT pairs across specialties, found an overall Pearson correlation of 0.59 between emulated and trial effect estimates, rising to 0.83 in the 63 pairs that emulated the target trial's design most closely.⁴ The pattern across these studies is consistent: emulation quality, not the general approach, is what determines whether the result can be trusted, and even careful emulations diverge from trials often enough that neither journals nor regulators treat concordance as guaranteed.

Where the disagreement genuinely sits

The honest version of this debate is not "observational data versus randomised data" but a dispute about how much of a trial's advantage survives once causal-inference methods are applied properly. Proponents argue that a poorly specified observational study and a rigorously emulated one are different animals, and that dismissing all real-world evidence because of the failures of the former is a category error; the concordance data above show emulation can work when done well. Sceptics point to the same data and note that "done well" is doing a great deal of work, a review of 100 published studies that claimed to emulate a target trial found that a quarter (24%) did not actually specify what trial they were emulating, and only 40% provided detailed information on all components of the target trial protocol, meaning much of the published literature calling itself target trial emulation does not meet its own standard.⁷ The deeper limitation is structural rather than a matter of diligence: randomisation balances unmeasured as well as measured confounders, and no amount of careful matching or weighting in an emulation can adjust for a confounder that was never recorded in the dataset. An emulation of the REDUCE-AMI trial (beta-blockers after myocardial infarction with preserved ejection fraction), run and published alongside the randomised trial itself, illustrated this directly, the authors concluded that confounding by indication in observational data on this question could only be resolved by the randomisation the emulation was trying to substitute for.⁸ Where the data allow it, target trial emulation clearly outperforms the older, informal habit of comparing treated and untreated patients without defining time zero or eligibility at all. Whether it can stand in for a trial on any given question depends on whether the confounders driving treatment choice are actually measured in the dataset being used, and that has to be argued for each specific clinical question, not assumed for the method as a whole.

Where regulators currently draw the line

The FDA's position has moved, but not as far as "real-world evidence in place of a pivotal trial." Its Advancing Real-World Evidence Program, first established under the 21st Century Cures Act mandate, accepts proposals aimed at supporting new labelling claims, a new indication, an expanded population, a dosing change, for products that already have an approved indication, and at meeting post-approval study commitments; the programme's own scope explicitly excludes proposals intended only as supportive context for interpreting a trial or as methods-development exercises, and the FDA reviews a limited number of requests each submission cycle, selecting them competitively on the basis of the underlying data quality and study design.⁵ In December 2025 the FDA removed a specific procedural barrier, a former requirement that real-world evidence submissions include individual-level identifiable patient data, allowing de-identified sources such as registries and claims databases to be submitted directly for certain medical device reviews, a change described as easing access to real-world evidence rather than expanding what it can be used to prove.⁹ Journals have been similarly measured: reporting guidance for observational studies increasingly asks authors who describe their work as a target trial emulation to actually specify the trial being emulated, but no major clinical journal currently treats a well-conducted emulation as interchangeable with a randomised trial for establishing a novel treatment effect. The practical line, as it stands in 2026, is that real-world evidence and target trial emulation are gaining a defined, audited role in answering questions trials cannot reach, long-term safety, rare subgroups, effectiveness after approval, while the question a pivotal trial is designed to answer still needs a pivotal trial.

What this means for how you frame your own study

If your dataset is retrospective and you are tempted to describe its design as a target trial emulation, the framework is worth using properly rather than as a label, write down the protocol of the trial you are conceptually emulating, define time zero explicitly, and be precise in your Methods about which confounders you adjusted for and which you could not measure. Reviewers increasingly know the difference between a study that did this and one that borrowed the phrase after the fact, and the concordance literature above gives them a specific reason to ask. Where your question genuinely cannot be answered with a trial, because the exposure cannot be ethically randomised, or the outcome is too rare, or the trial that would answer it was never run, a carefully specified emulation is a legitimate and increasingly respected way to generate evidence. It is a different kind of evidence from a randomised trial's, though, and describing it that way in your discussion section will serve the paper better than implying the two are equivalent.

Frequently asked questions

Is target trial emulation the same as a randomized controlled trial?

No. It is a framework for designing an observational analysis to resemble the randomized trial you would run if you could, by writing an explicit protocol and defining time zero, eligibility, and treatment strategies before analyzing the data. Concordance studies show it can approximate a trial's result but does not reliably match it, and it cannot adjust for confounders that were never measured in the dataset.

How often does target trial emulation match the original randomized trial's result?

It varies with how closely the emulation reproduces the trial's design. The RCT DUPLICATE program matched the regulatory conclusion in 6 of 10 emulated cardiovascular drug trials, and a 2026 meta-analysis of 107 emulation-RCT pairs found an overall Pearson correlation of 0.59, rising to 0.83 in the subset that most closely emulated the target trial's design.

Can real-world evidence replace a randomized controlled trial for FDA approval?

Not currently for the pivotal trial that earns initial approval. The FDA's Advancing Real-World Evidence Program reviews real-world evidence to support new labeling claims and post-approval commitments on products that are already approved, not as a substitute for the trial that established the original approval.

References

1. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758-764. https://doi.org/10.1093/aje/kwv254
2. Hernán MA, Wang W, Leaf DE. Target trial emulation: a framework for causal inference from observational data. JAMA. 2022;328(24):2446-2447. https://doi.org/10.1001/jama.2022.21383
3. Franklin JM, Patorno E, Desai RJ, et al. Emulating Randomized Clinical Trials With Nonrandomized Real-World Evidence Studies: First Results From the RCT DUPLICATE Initiative. Circulation. 2021;143(10):1002-1013. https://doi.org/10.1161/CIRCULATIONAHA.120.051718
4. Wang C, et al. Concordance between target trial emulation and randomised controlled trials: systematic review and meta-analysis. BMJ. 2026;393:e086810. https://doi.org/10.1136/bmj-2025-086810
5. US Food and Drug Administration. Advancing Real-World Evidence Program. FDA.gov. Accessed September 2026.
6. Wang SV, Schneeweiss S, Franklin JM, et al. Emulation of Randomized Clinical Trials With Nonrandomized Database Analyses: Results of 32 Clinical Trials. JAMA. 2023;329(16):1376-1385. https://doi.org/10.1001/jama.2023.4221
7. Simon-Tillaux N, et al. Conducting observational analyses with the target trial emulation approach: a methodological systematic review. BMJ Open. 2024;14:e086595. https://doi.org/10.1136/bmjopen-2024-086595
8. Leening MJG, et al. The perpetual need of randomized clinical trials: challenges and uncertainties in emulating the REDUCE-AMI trial. Eur J Epidemiol. 2024;39(4):343-347. https://doi.org/10.1007/s10654-024-01127-3
9. US Food and Drug Administration. FDA eliminates major barrier to using real-world evidence in drug and device application reviews. Press announcement. December 15, 2025.