Aging
Quantifying bias from reverse causation in observational studies of dementia risk factors: A simulation study informed by reverse Mendelian Randomization Jingxuan Wang* Jingxuan Wang Wang Wang Wang Wang Wang Wang Wang Wang Wang Wang Wang Wang Harvard T.H. Chan School of Public Health
Background: Observational studies are central to identifying modifiable determinants of dementia, yet its long preclinical phase can induce reverse causation, biasing associations even in longitudinal analyses. Using simulations informed by prior reverse Mendelian randomization (MR) findings, we quantified the magnitude of reverse causation bias in observational analyses of the effects of body mass index (BMI) on dementia.
Methods: For a hypothetical cohort study (n=20,000), we simulated longitudinal trajectories of BMI and dementia risk from ages 45 to 90 years (Figure: causal directed acyclic graph). Data generating rules were calibrated to align with previously published evidence on age-specific dementia incidence, BMI trajectories across age, and age-specific associations of dementia genetic risk with BMI from reverse MR studies. We generated the data such that BMI had no effect on dementia incidence and that any observed association would reflect bias due to reverse causation. For each simulation, we selected a baseline age for study entry (45 to 80), and simulated a study where BMI was measured at baseline, and participants were followed until age 90. We estimated the association between z-standardized BMI at study entry and dementia incidence using Cox models and quantified bias as the deviation of the estimated HR from the null. Each simulation was repeated 300 times.
Results: The magnitude of reverse causation bias increased with older baseline age (HR decreased). For example, when follow-up began at age 65, the estimated HR for BMI and dementia was 0.82 (95% CI: 0.71-0.96). When follow-up began at age 80, the HR decreased to 0.61 (95% CI: 0.53-0.71).
Conclusion: Our flexible, simulation-based tool can help quantify reverse causation bias in dementia research. Our findings, consistent with prior hypotheses, indicate reverse causation bias for dementia risk factors can be large and lead to qualitatively incorrect conclusions.
