Review of methods to deal with the misalignment of times of eligibility, start of follow-up, and treatment assignment in studies explicitly aimed at emulating target trials.

OBJECTIVES

Real world evidence based on observational data from cohorts, registries, and health-care databases are increasingly used to assess the effectiveness of therapeutic interventions, often using the target trial emulation framework. One challenge when analyzing observational data are risks of biases due to misalignment of times of eligibility, start of follow-up, and treatment assignment. We aimed to describe the methods used to generate alignment or to account for misalignment in studies explicitly aimed at emulating target trials, and to estimate the proportion of studies for which a low-cost change would limit the risk of biases associated with misalignment of the times.

STUDY DESIGN AND SETTING

We analyzed 199 studies explicitly aiming at emulating a target trial identified in a previous systematic review from Hansford et al. Two reviewers extracted the times of eligibility, start of follow-up, and treatment assignment and the methods used by authors to generate alignment or to account for misalignment of time points.

RESULTS

Out of the 199 studies, 181 (91%) reported the times of eligibility, start of follow-up, and treatment assignment. All time points were aligned for 93/181 (51%), with 73 using no specific method, 18 emulating a sequence of target trials, and 2 using other methods to generate alignment. In contrast, 88/181 (49%) studies had misalignment of time points, of which 29 used a method to correct for misalignment during analysis (24 studies used a cloning, censoring, and weighting approach; 4 randomly allocated patients with early events; and 1 randomly allocated all participants to the study groups with subsequent censoring when they deviated from the allocated intervention). Out of 59/88 (67%) studies that did not use any method to address nonalignment, 46/59 (78%) could have applied low-cost changes to account for misalignment.

CONCLUSION

Approximately, half of the studies explicitly aiming to emulate a target trial had alignment of times of eligibility, start of follow-up, and treatment assignment, either by design or using a method generating alignment. Among studies with misalignment, about 67% did not account for it in the analysis among which 78% could have applied low-cost changes to reduce bias.

PLAIN LANGUAGE SUMMARY

Researchers increasingly use real-world evidence from sources like routinely collected or claims data to assess the efficacy and safety of therapeutic interventions (medications, surgery, physiotherapy, etc). To prevent design errors when analyzing these data, researchers follow a framework called "target trial emulation". A crucial aspect of this framework is to ensure that three key time points of the study are aligned during analysis: the times at which 1) participants in the database are assessed for eligibility in the study (ie, when do we choose to include them or not in the analysis), 2) their follow-up begin, and 3) they are assigned to a treatment group. If these time points are misaligned, the study results can be biased. Here, we reviewed 199 published studies that explicitly aimed to emulate a target trial identified in a previous systematic review by Hansford et al. We found that about half of them had alignment between of times of eligibility, treatment assignment, and follow-up. In the other half, there was misalignment but one-third used specific methods to correct for this problem. Many studies that aim to emulate a trial may have issues in alignment of key time points, which can compromise the validity of their findings. Most of these issues could be corrected with straightforward adjustments.