Mendelian randomisation, lipids, and cardiovascular disease.

In The Lancet, Benjamin Voight and colleagues use mendelian randomisation analysis to invest igate the relation between HDL cholesterol and coronary heart disease. High HDL cholesterol concentration is associated with reduced risk of coronary heart disease in observational studies, but whether the association is causal cannot be unequivocally ascertained from these studies alone, and whether raising of HDL cholesterol would be an eff ective means to reduce risk of coronary heart disease remains uncertain. The ILLUMINATE trial of torcetrapib, a cholesteryl ester transport protein (CETP) inhibitor that raises HDL cholesterol, was stopped early because of an increase in the number of cardiovascular events. This outcome might have been due to an off target action of torcetrapib on blood pressure that appears not to be shared by newer drugs from the same class, and the eff ect of these drugs on risk of coronary heart disease is now being evaluated in phase 3 trials. Mendelian randomisation exploits genetic information to investigate associations between exogenous or endogenous exposures (or both) and disease outcomes. The random allocation of genotype at gametogenesis (like the randomised allocation to a drug in a clinical trial) minimises confounding. If HDL-cholesterol-mediated pathways were causal for coronary heart disease, carriers of genetic variants associated with high concentrations of HDL cholesterol (representing lifelong exposure to high HDL cholesterol) would be expected to have a reduced risk of coronary heart disease. Voight and colleagues used a rare non-synonymous variant in the endothelial lipase gene (LIPG Asn396Ser, minor allele frequency about 2·6%) that showed a consistent association with high HDL cholesterol concentrations, but no association with LDL cholesterol or triglycerides. In view of the observed eff ect of the LIPG variant on HDL cholesterol, this allele was expected to result in roughly a 13% reduced risk of coronary heart disease. However, in pooled data from 20 913 cases and 95 407 controls, the observed odds ratio for coronary heart disease was 0·99 (95% CI 0·88–1·11), suggesting that LIPG-mediated increases in HDL cholesterol do not reduce risk of the disease. In Voight and colleagues’ analysis, a panel of 14 common variants, each with a small eff ect on HDL cholesterol, was combined into a genetic risk score. The risk score was also not associated with coronary heart disease in pooled data from 12 482 cases and 41 331 controls (odds ratio per SD increase in weighted genetic risk 0·93, 95% CI 0·68–1·26). On the basis of these results, genetically raised HDL cholesterol concentrations do not seem to reduce risk of coronary heart disease—an observation that calls into question whether raising of HDL cholesterol therapeutically would translate into the expected clinical benefi t. The validity of a mendelian randomisation analysis is determined by various factors. First, the inter mediate phenotype (HDL cholesterol) must associate with the outcome (coronary heart disease). Second, the genetic instrument must associate with the outcome only through eff ects on the intermediate phenotype. Third, the genetic instrument should be suffi ciently strongly associated with the intermediate phenotype to avoid weak instrument bias. The LIPG variant used here had a fairly large eff ect on HDL cholesterol, but is rare in the population and so might not represent a strong instrument. We should note, however, that the case-control analysis of this variant was adequately powered to detect even a small eff ect, and the negative association of this variant can be regarded as defi nitive. For the genetic risk score, variants with the largest eff ects were excluded because of associations with other lipid fractions, and although each of the Published Online May 17, 2012 DOI:10.1016/S01406736(12)60481-4