Mind the gap: Characterizing bias due to population mismatch in two-sample Mendelian randomization

Area of Science:

• Genetics
• Epidemiology
• Statistical genetics

Background:

• Mendelian randomization (MR) estimates causal effects using genetic variants as instrumental variables.
• Two-sample MR (2SMR) utilizes separate samples for exposure and outcome associations.
• Valid 2SMR inference requires individuals from the same population to avoid bias.

Purpose of the Study:

• To investigate the impact of using different populations in 2SMR.
• To quantify the bias introduced by population mismatch.
• To explore potential benefits of using mismatched populations for statistical power.

Main Methods:

• Simulated MR analyses to assess bias under population mismatch.
• Empirical analysis of 546 trait pairs across 2-7 ancestries comparing matched and mismatched population estimates.
• Quantified attenuation and assessed false positive rates.

Main Results:

• Violation of the same-population assumption in 2SMR biases causal estimates toward zero on average.
• Bias occurs even between populations of the same continental ancestry.
• Increasing genetic distance between populations correlates with greater attenuation of estimates.
• Larger, mismatched exposure studies can enhance statistical power by increasing precision.

Conclusions:

• Population mismatch in 2SMR introduces bias, typically attenuating estimates.
• Careful consideration of population matching is crucial for accurate MR inference.
• Leveraging larger, mismatched datasets may offer a strategy to boost MR power, especially for understudied populations.