Mendelian Randomization

The idea

Genetic variants are randomly assigned at conception, much like treatment allocation in an RCT. If a variant influences a specific exposure (e.g. higher LDL), then comparing health outcomes between carriers and non-carriers of the variant mimics randomising people to "high-LDL throughout life" vs "low-LDL throughout life."

What it answers well

• Lifetime exposure to a modifiable factor (LDL, BP, BMI, alcohol).
• Causal direction for ambiguous observational associations.
• Avoids most confounding because variants are inherited randomly.

Famous examples

• LDL is causally atherogenic: PCSK9 LOF carriers have lifetime low LDL and dramatically reduced CV events — supporting trials of PCSK9 inhibitors.
• HDL is not causally protective (Voight 2012, Lancet): variants that isolate HDL changes don’t change MI risk — explaining why CETP inhibitors failed.
• Alcohol harms even at moderate intake: ADH/ALDH variants showed cardiovascular risk rises monotonically with alcohol exposure, challenging the J-curve.
• BMI causally drives type-2 diabetes (overwhelming MR evidence).

Limitations / caveats

• Pleiotropy: variants often affect multiple traits, confounding inferences.
• Weak instruments: small effect sizes need very large sample sizes.
• Lifetime exposure ≠ short-term treatment exposure — effects may not transfer linearly to interventions in adulthood.
• Linkage disequilibrium: variants may track with others in unintended ways.

Why it matters for longevity

Many longevity-claim products rely on observational associations. MR provides one of the strongest tools to triage which observed effects are likely causal (and worth pursuing) and which are likely confounded.

Related entries

Bradford-Hill criteria, Confounding, Survivorship bias, APOE genotype.