Design the target trial

Before touching registry data, write down the randomized trial you would run. Then emulate it, row by row. The row that most often goes wrong is the one that looks most innocent: when does follow-up start?

Further reading: Hernán and Robins (2016), Using big data to emulate a target trial when a randomized trial is not available, Am J Epidemiol; Hernán, Sauer, Hernández-Díaz, Platt and Shrier (2016), Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses, J Clin Epidemiol; the TARGET reporting guideline for target trial emulations (JAMA, 2025); Suissa (2008), Immortal time bias in pharmacoepidemiology, Am J Epidemiol; FDA, Use of real-world evidence to support regulatory decision-making for medical devices (final guidance, Federal Register notice 18 December 2025; supersedes the 2017 guidance). The registry is simulated with a seeded generator whose true effect is known; the truth curves are exact formulas, the Kaplan–Meier curves are finite-sample estimates, and the step 4 timelines are schematic.