To the Editor:

We thank Didelez1 for their comments on our recent article.2

We wish it were true that we were just criticizing a “straw man.” If the target trial emulation approach was merely that one should have clear research questions, and that it is good to think about how they would be approached by a randomized controlled trial (RCT), then we agree, but these are not new ideas.2,3 However, the advocates of the target trial emulation approach are going much further than this, with studies being judged by how target trial emulation-like they are.4,5 It is argued that one should construct a study that emulates the target trial as closely as possible, and that this is a prerequisite for causal inference.6 The “matched target trial emulation” approach is merely the most extreme example of this.

The problem is that in practice this approach works for some research questions, but not for others. One example is the admonition that all participants should be followed from the initiation of exposure (time zero), as Didelez mention. We have argued elsewhere that for many important questions, for example in many etiologic cancer studies, insisting on follow-up from first exposure is not possible or necessary and may produce right truncation, in which the most important risk periods are missed.7

Another key problem with the target-trial emulation approach is that it cannot deal with intractable confounding. For example, Dickerman et al.8 have produced an excellent target-trial emulation study, comparing the effectiveness of two COVID-19 vaccines. The target-trial emulation approach works (just as a more traditional approach would) in this situation because vaccine allocation is more-or-less random. In our article, we predicted that such an approach would not work when comparing vaccinated and unvaccinated, because the residual confounding would be too great. There is now empirical evidence for this. Hulme et al.9 use a target-trial emulation approach to assessing the effectiveness of COVID-19 vaccination, and produce results that are clearly incorrect—in the first 2 weeks after vaccination, the vaccinated group had only about 1/3 the rate of positive COVID-19 tests, COVID-19 hospitalizations, and death from any cause, even though it is well established that the vaccination has little-or-no effect during this period.10 In contrast, vaccination studies using the test-negative design11 consistently produce findings that are in agreement with those from RCTs.12

There are other examples like this,2 where the target-trial emulation approach can never produce valid effect estimates. In contrast, there are a number of other approaches,13 including the test-negative design, difference-in-differences, regression discontinuity, triangulation, and Mendelian randomization, which can readily deal with these situations, and which have been shown to produce results consistent with RCTs. Thus, for some important research questions, the target-trial emulation adopts a formulaic approach that is transparent, but that can never validly answer the research question under consideration, and that does not naturally lead to exploring the use of other approaches that can answer the question. This is a long way from being best practice.

REFERENCES 1. Didelez V. Re: Are target trial emulations the gold standard for observational studies? Epidemiology. 2024;35:e3. 2. Pearce N, Vandenbroucke J. Are target trial emulations the gold standard for observational studies?. Epidemiology. 2023;34:614–618. 3. Vandenbroucke JP, Pearce N. From ideas to studies: how to get ideas and sharpen them into research questions. Clin Epidemiol. 2018;10:253–264. 4. Guyatt GH, Oxman AD, Vist GE, et al.; GRADE Working Group. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336:924–926. 5. Sterne JA, Hernán MA, Reeves BC, et al. ROBINS-I: a tool for assessing risk of bias in non-randomised studies of interventions. BMJ. 2016;355:i4919. 6. Hernan MA. Invited commentary: hypothetical interventions to define causal effects - afterthought or prerequisite?. Am J Epidemiol. 2005;162:618–20; discussion 621. 7. Vandenbroucke J, Pearce N. Point: incident exposures, prevalent exposures, and causal inference: does limiting studies to persons who are followed from first exposure onward damage epidemiology?. Am J Epidemiol. 2015;182:826–833. 8. Dickerman BA, Gerlovin H, Madenci AL, et al. Comparative effectiveness of BNT162b2 and mRNA-1273 vaccines in U.S. veterans. N Engl J Med. 2022;386:105–115. 9. Hulme WJ, Williamson E, Horne EMF, et al. Challenges in estimating the effectiveness of COVID-19 vaccination using observational data. Ann Intern Med. 2023;176:685–693. 10. Polack FP, Thomas SJ, Kitchin N, et al.; C4591001 Clinical Trial Group. Safety and efficacy of the BNT162b2 mRNA COVID-19 vaccine. N Engl J Med. 2020;383:2603–2615. 11. Vandenbroucke JP, Brickley EB, Vandenbroucke-Grauls C, Pearce N. A test-negative design with additional population controls can be used to rapidly study causes of the SARS-CoV-2 epidemic. Epidemiology. 2020;31:836–843. 12. Cerqueira-Silva T, Katikireddi SV, de Araujo Oliveira V, et al. Vaccine effectiveness of heterologous CoronaVac plus BNT162b2 in Brazil. Nat Med. 2022;28:838–843. 13. Pearce N, Vandenbroucke J, Lawlor D. Causal inference in environmental epidemiology: old and new approaches. Epidemiology. 2019;30:311–316.