Abstract
Benchmarking an observational analysis against a randomized trial increases our confidence in the use of observational data for causal inference. The Breast International Group (BIG 1-98) randomized trial compared the effect of letrozole and tamoxifen on the risk of death in postmenopausal women with hormone receptor-positive breast cancer. We designed a target trial that aimed to ask the same question as the one asked in BIG 1-98 and emulated it in Swedish registry data. The primary results from our observational analysis showed an increased risk of death in those who initiated aromatase inhibitors compared with tamoxifen [5-year risk difference = 2.5% (95% CI, 0.2-4.6)], which was discordant to the results from BIG 1-98. However, estimates were more closely aligned when our observational analysis was restricted to nonusers of opioids or antidepressants [5-year risk difference = -0.9 (95% CI, -4.2 to 2.0)]. In conclusion, when benchmarking an observational analysis against a trial, alignment of eligibility criteria with the index trial is not always sufficient and further study population restrictions may be required to address unmeasured confounding.