A Common KIF6 Polymorphism Increases Vulnerability to Low-Density Lipoprotein Cholesterol: Two Meta-Analyses and a Meta-Regression Analysis

Background We sought to determine if a common polymorphism can influence vulnerability to LDL cholesterol, and thereby influence the clinical benefit derived from therapies that reduce LDL cholesterol. Methods We conducted a meta-analysis of the association between a common Trp719Arg polymorphism in the kinesin-like protein 6 (KIF6) gene and the risk of cardiovascular disease (CVD), and a meta-regression analysis to measure the effect modification of this polymorphism on the association between LDL cholesterol and the risk of CVD. We used this measure of genetic effect modification to predict the expected difference in clinical benefit among KIF6 719Arg allele carriers and non-carriers in response to therapies that reduce LDL cholesterol. We then conducted a meta-analysis of statin trials to compare the expected difference in clinical benefit with the observed difference during treatment with a statin. Results In a meta-analysis involving 144,931 participants, the KIF6 719Arg allele was not associated with the relative risk (RR) of CVD (RR: 1.02, 95%CI: 0.98–1.07, p = 0.288). Meta-regression analysis involving 88,535 participants, however, showed that the 719Arg allele appears to influence the effect of LDL cholesterol on the risk of CVD. KIF6 carriers experienced a 13% greater reduction in the risk of CVD per mmol/L decrease in LDL cholesterol than non-carriers. We interpreted this difference as the expected difference in clinical benefit among KIF6 carriers and non-carriers in response to therapies that lower LDL cholesterol. The difference in clinical benefit predicted by the increased vulnerability to LDL cholesterol among KIF6 carriers (ratio of RR: 0.87, 95%CI: 0.80–0.94, p = 0.001) agreed very closely with the observed difference among 50,060 KIF6 carriers and non-carriers enrolled in 8 randomized trials of statin therapy (ratio of RR: 0.87, 95%CI: 0.77–0.99, p = 0.038). Conclusion The KIF6 719Arg allele increases vulnerability to LDL cholesterol and thereby influences the expected clinical benefit of therapies that reduce LDL cholesterol.


Text S1: Equivalent Measures of Effect Modification
For simplicity and clarity of presentation, the following refers to the effect modification between two dichotomous exposure variables, A and B, and the risk of a dichotomous outcome (DIS). Extension to continuous variables is straightforward.

A) Measuring Effect Modification in a Generalized Linear Model
For the following generalized linear model (logistic regression model) containing an interaction term: Ln(Odds DIS ) = constant + ß 1 (A) + ß 2 (B) + ß 3 (A*B) The interpretation of each term in the model is straightforward: e ß1 = OR (A-DIS) B=0 ; is the Odds Ratio (OR) measuring the association between exposure A and the risk of disease (DIS), among persons not exposed to B (B=0) e ß2 = OR (B-DIS) A=0 ; is the OR measuring the association between exposure B and the risk of disease, among persons not exposed to A (A=0) e ß3 = OR (A-B interaction); is the OR for the interaction, or effect modification, To obtain the OR for the association between exposure A and the risk of disease among persons exposed to B (OR (A-DIS) B=1 ), the OR for the association between exposure A and the risk of disease among persons not exposed to B (OR (A-DIS) B=0 ) is multiplied by the interaction OR: With re-arrangement: The interaction OR is thus the ratio of the OR's for the association between exposure A and the risk of disease among persons exposed to B and not exposed to B, respectively.
Similarly, to obtain the OR for the association between exposure B and the risk of disease among persons exposed to A (OR (B-DIS) A=1 ), the OR for the association between exposure B and the risk of disease among persons not exposed to A (OR (B-DIS) A=0 ) is multiplied by the interaction OR: With re-arrangement: The interaction OR is thus also the ratio of the OR's for the association between exposure B and the risk of disease among persons exposed to A and not exposed to A, respectively.
Therefore, in a generalized linear model, a single interaction term defines the effect modification between any two exposure variables. As a result, the effect modification of exposure B on the association between exposure A and the risk of disease is exactly equal to the effect modification of exposure A on the association between exposure B and the risk of disease, because these two estimates of effect modification are defined by the same interaction term as shown below:

B) Measuring Effect Modification in a Meta-Regression Equation
The regression term in a meta-regression equation is also a measure of effect modification, and it is equivalent to an interaction term in a generalized linear model. Thus the OR for the association between exposure A and the risk of disease among persons exposed to B (OR (A-DIS) B=1 ) is equal to the OR for the association between exposure A and the risk of disease among persons not exposed to B (OR (A-DIS) B=0 ) multiplied by the OR for the meta-regression term estimating the effect modification of exposure B on the association between exposure A and the risk of disease.
With re-arrangement, [4] OR (Effect Modification of B) = Therefore, the inverse natural logarithm of the regression term in this meta-regression equation is the ratio of the OR's for the association between exposure A and the risk of disease among persons exposed to B and not exposed to B, respectively. This is exactly the same interpretation as for the interaction term in the logistic model evaluating

C) Equivalent Measures Effect Modification
We have shown that a single interaction term in a generalized linear model defines the effect modification between any two exposure variables, and as a result the estimate of the effect modification of exposure B on the association between exposure A and the risk of disease is exactly equal to the estimate of the effect modification of exposure A on the association between exposure B and the risk of disease, as shown in equation