Instructor Division of Pharmacoepidemiology, Dept. of Medicine, Brigham & Women’s Hospital, Harvard Medical School St-Laurent, QC, United States
Background: Bariatric surgery is a weight loss treatment that has been shown to be highly effective in clinical trials. Several observational studies have reported significant reductions in mortality when compared to nonsurgical patients, yet it is unclear whether these findings are a result of bias and to what extent.
Objectives: To methodologically review studies of mortality associated with bariatric surgery and to identify and quantify the biases resulting from the use of nonsurgical comparators.
Methods: We searched Pubmed to identify all observational studies investigating the effect of bariatric surgery on all-cause mortality, when compared to nonsurgical patients. We assessed these studies for potential time-related biases and confounding deriving from the lack of exchangeability of exposed patients with selected comparison groups. We conducted bias analyses for each type of bias to quantify the effect of immortal time and residual confounding on effect estimates.
Results: We identified 20 observational studies assessing the association between bariatric surgery and the risk of mortality, compared to no surgery. Among those, 11 were subject to potential time-related biases, including immortal time bias, resulting from improper assignment of cohort entry date. Five studies were subject to potential confounding due to lack of adjustment for body mass index (BMI), whereas improper selection of comparators, due to the lack of eligibility for bariatric surgery in this group, occurred in all studies. Bias analyses to correct for immortal time bias shifted the effect estimates towards the null (example 1: reported crude HR: 0.66 vs. bias-adjusted HR: 0.84). Bias analyses to correct for residual confounding due to lack of adjustment for BMI also shifted the effect estimates towards the null (example 2: reported HR: 0.78 vs. bias-adjusted HR [assuming RRCD=3.5]: 0.90).
Conclusions: When conducting an observational study with a non-active comparator, careful attention must be given to ensuring these comparators are exchangeable with the exposure group with respect to critical patient characteristics and BMI is accounted for in the design or analysis. Additionally, to avoid time-related biases, the cohort entry date for comparators should be anchored to a point in time at which patients are eligible to receive the surgery.
