Target Trial Emulation: Improving the Quality of Observational Studies in Inflammatory Bowel Disease Using the Principles of Randomized Trials

Abstract

The past decade has seen a substantial increase in the number of randomized controlled trials (RCTs) conducted in inflammatory bowel disease (IBD). Randomized controlled trials are the gold standard method for generating robust evidence of drug safety and efficacy but are expensive, time-consuming, and may have ethical implications. Observational studies in IBD are often used to fill the gaps in evidence but are typically hindered by significant bias. There are several approaches for making statistical inferences from observational data with some that focus on study design and others on statistical techniques. Target trial emulation is an emerging methodological process that aims to bridge this gap and improve the quality of observational studies by applying the principles of an ideal, or "target," randomized trial to routinely collected clinical data. There has been a rapid expansion of observational studies that have emulated trials over the past 5 years in other medical fields, but this has yet to be adopted in gastroenterology and IBD. The wealth of nonrandomized clinical data available through electronic health records, patient registries, and administrative health databases afford innumerable hypothesis-generating opportunities for IBD research. This review outlines the principles of target trial emulation, discusses the merits to IBD observational studies in reducing the most common biases and improving confidence in causality, and details the caveats of using this approach.

Keywords: causal inference; inflammatory bowel disease; randomized controlled trial; target trial.

Figure 1.

Target trial emulation workflow framework.

Figure 2.

Minimizing bias with target trial emulation and the concept of time zero. A, Retrospective observational study (an example of bias introduction). Patients are prescribed treatments based on a number of demographic and clinical characteristics (for example, age, comorbidities, disease type, disease severity), which leads to an unequal distribution across treatment groups and confounding bias. Therefore, confounding factors should be identified and statistically adjusted to reduce bias. Delays in initiating treatment (eg, prescription delays) leads to a period where the study outcome cannot occur and is referred to as “immortal time.” By misaligning time zero, the specification of the eligibility criteria, the treatment assignment and actual treatment initiation can result in immortal time bias. Finally, selecting patients after treatment assignment (and availability of some study outcomes) may lead to selection bias. B, “Target” randomized controlled trial. Participants are first screened based on eligibility criteria, randomly assigned to treatment, and then followed up for outcomes. Randomization ensures equal distribution of measured and unmeasured confounders. Eligibility assessment occurs prior to randomization and patient follow-up starts at the point of randomization. Time zero (baseline) refers to the date when an eligible patient initiates treatment and study follow-up begins—this aligns naturally for RCTs where time zero is the time of randomization and minimizes bias.