Possible selection bias in register-based obesity studies

Some studies of obesity treatments use control groups identified from real-world registers, which may differ from people with obesity in the general population. We evaluated whether such control groups affect the results. The SOS study examines long-term mortality post-bariatric surgery. Among volunteers with obesity, 2,007 individuals underwent surgery, while a control group of 2,040 individuals was matched using 18 variables. Age was 37-60 years and BMI was ≥ 34/≥38 kg/m² for men and women, respectively. We subdivided the control group into those with an obesity diagnosis (n = 177) and those without an obesity diagnosis (n = 1,863) in the Swedish National Patient Register prior to study inclusion. Mortality was determined over a median follow-up period of 26 years. The controls with a prior obesity diagnosis had a higher mortality rate than the controls without a prior obesity diagnosis, with 19.7 (95% CI, 15.5-25.1) and 14.4 (95% CI, 13.3-15.7) deaths per 1000 person-years, respectively. This corresponds to a hazard ratio of 1.45 (95% CI, 1.12-1.89; p = 0.005) and a 3.4-year shorter life expectancy. These results were confirmed in another cohort (n = 2,759, HR = 1.82 [95% CI, 1.47-2.25; p<0.001] and a 6.1-year shorter life expectancy). Controls with obesity identified from real-world datasets may be in poorer health than those who voluntarily participate in clinical studies. Consequently, selection bias could lead to an overestimation of the survival benefits of obesity treatments in research using controls identified by prior obesity diagnosis.