We read with great interest the systematic review by Kizer et al of large-scale trials of statins versus placebo, usual care, or active (lower dose statin) control to calculate updated summary estimates of risk reduction in coronary artery disease (CAD) and all-cause mortality. Their meta–regression analyses showed that different measures of low-density lipoprotein (LDL) cholesterol helped explain the heterogeneity observed in the effect estimates from different studies. In these analyses, the absolute difference between achieved in-trial LDL cholesterol in the control versus treatment group emerged as the strongest determinant of CAD risk reduction. We herein introduce a novel concept, LDL cholesterol–independent effects of statins on CAD, taking our cue from blood pressure–independent effects of blood pressure–lowering drugs, advocated by the Blood Pressure Lowering Treatment Trialists’ Collaboration (BPLTTC).
In the primary analysis of data from the 12 trials (included in the systematic review by Kizer et al ) comparing pravastatin (1 of the low-intensity statins) or atorvastatin (1 of the high-intensity statins) against placebo or another drug class, the association between absolute difference between achieved in-trial LDL cholesterol in the control versus treatment group and the log odds ratio for CAD events was investigated using meta–regression analysis with inverse variance weighting. Standard methods for the comparison of regression models were used. Initially, using a single regression model, separate regression lines were fitted for trials in which pravastatin was the investigational treatment and in which atorvastatin was the investigational treatment. The slopes of these lines were compared to test for a differential effect of LDL cholesterol reduction on risk in trials of pravastatin compared to atorvastatin. If these slopes were not significantly different (p >0.05), a parallel lines regression model was fitted, which assumed equal effects of LDL cholesterol reduction with pravastatin and atorvastatin. Using this model, the null hypothesis of no separation between the 2 parallel lines was tested to explore whether there was evidence of a differential effect of the 2 drug classes independent of LDL cholesterol lowering. Assumptions of linear associations between differences in follow-up LDL cholesterol levels and log odds ratio were tested using standard graphical methods. Analyses were carried out using the metareg routine in Stata version 8.0 (StataCorp LP, College Station, Texas).
The slopes of the separate regression lines using a single regression model were not significantly different (p = 0.562 by Monte Carlo permutation test). The slope of a parallel lines regression was statistically significantly different from zero (p = 0.046), and there was a relatively strong trend toward a difference between the line for pravastatin and the line for atorvastatin (p = 0.060; Figure 1 ).
