Causal Inference
Age-at-Death Patterns as a Source of Heterogeneity in Mendelian Randomization Studies Across Countries Hong Tang* Hong Tang Tang School of Public Health, The University of Hong Kong Li Ka Shing Faculty of Medicine
Mendelian randomization (MR) is a powerful tool for estimating the causal effect of exposure on outcomes because it is less susceptible to confounding compared to conventional observational study. However, MR studies often recruit middle-aged and older adults, whose survival to recruitment is affected by genetic endowment and occurrence of the outcome or competing risk of the outcome, which can introduce internal bias due to collider selection. In addition, disease-specific age-at-death patterns may vary by country and over time, making MR estimates susceptible to net-external bias related to the transportability of effect estimates, creating apparent effect modification. These two types of selection bias are less considered in MR study.
To address this gap, we compared cause-specific age-at-death patterns over 40 years to identify potential sources of heterogeneity in MR effect estimates across countries. Since this bias is especially concerning in multi-country MR studies with differing age-at-death patterns, we focused on countries with large publicly available genome-wide association studies. Figure 1 compares cause-specific median ages at death and the proportion of deaths across four countries. In the United States and the United Kingdom, individuals tend to die at an older age from stroke compared to chronic obstructive pulmonary disease (COPD), whereas in China and Japan, the pattern is reversed, with higher median ages at death from COPD than from stroke. These country-specific differences in age-at-death patterns may contribute to heterogeneity in MR estimates for the same exposure-outcome relationships.
Age-at-death patterns may vary due to country-specific factors, such as healthcare systems or environmental exposures, which could influence individual survival status and, consequently, who is available for recruitment in MR studies. We explore how differential age-at-death patterns may contribute to heterogeneity in MR studies conducted across countries.
