Lipoprotein(a) and progression of aortic valve calcification: a case of collider bias?

This commentary refers to ‘Lipoprotein(a) is associated with the onset but not the progression of aortic valve calcification’, by Y. Kaiser et al., https://doi.org/10.1093/eurheartj/ehac377 and the discussion piece ‘Lipoprotein(a) and progression of aortic valve calcification: a case of collider bias? Reply’, by Y. Kaiser et al., https://doi.org/10.1093/eurheartj/ehac742. Lipoprotein(a) [Lp(a)] has been recognized in observational and human genetic studies as an important risk factor for several forms of cardiovascular disease including coronary artery disease, peripheral artery disease, stroke, and more recently aortic stenosis (AS). The association between Lp(a) and AS has garnered intense interest as a potential therapeutic target, as no pharmacologic therapies for AS are currently available. In their recent article, Kaiser et al.1 investigated associations between Lp(a) and aortic valve calcification (AVC) measured by serial computed tomography examination among participants of the Rotterdam Study. The authors found that Lp(a) was a significant predictor of baseline and follow-up AVC. However, in an analysis of disease progression where ‘every individual had AVC at baseline,’ Lp(a) was not associated with progressive calcification. The authors suggest that ‘Lp(a) drives initiation, but not propagation of aortic valve disease’ and conclude that Lp(a)-lowering interventions may therefore be most effective in pre-calcific stages of the disease. While it is possible that Lp(a) is not a causal risk factor for AVC progression and Lp(a)-lowering treatment should indeed focus on those with pre-calcific disease, the findings of Kaiser et al.1 may potentially be explained by an artefact of their study design. ‘Index-event bias’ is a common but under-recognized form of collider stratification bias, which is responsible for several paradoxical observations within the epidemiologic literature.2 This bias is a form of selection bias that can occur when attempting to study factors associated with disease progression among a group of individuals who are ‘selected’ for inclusion in a study based on having the disease at baseline. Examples of this bias include the ‘aspirin paradox,’ whereby aspirin use is associated with decreased risk of initial myocardial infarction, but increased risk of recurrent myocardial infarction;3 and the ‘thrombophilia paradox,’ whereby thrombophilia increases risk of initial venous thromboembolic events, but not recurrence.4 In situations where multiple risk factors contribute to an outcome, stratifying or conditioning on the outcome of interest can induce statistical dependence among otherwise independent risk factors. These induced correlations may lead to spurious, absent, or biased associations between the risk factor of interest and disease progression (Figure 1).2 In this case, the authors stratified their analysis to consider the subset of individuals with AVC at baseline, and further report ‘additional adjustment for baseline AVC’ representing an additional level of stratification. If individuals have baseline AVC attributable to elevated Lp(a), they may have lower levels of other measured and unmeasured risk factors and spuriously appear ‘protected’ from progressive calcification. Directed acyclic graph of collider bias in the association of lipoprotein(a) with aortic valve calcification progression. Stratifying on baseline aortic valve calcification induces an indirect path (red line) between lipoprotein(a) and other independent risk factors or confounding variables for aortic valve calcification. This ‘collision’ may lead to spurious, absent, or biased associations when attempting to estimate the effect of lipoprotein(a) on aortic valve calcification progression. Statistical approaches including the use of propensity scores in traditional epidemiologic studies and novel correction factors in genetic epidemiologic studies have been proposed to mitigate the effects of this bias in studies of disease progression.5 We caution that establishing causal risk factors for disease progression from observational analyses is challenging and may be impacted by several sources of bias. While we agree that identifying approaches to prevent AVC altogether are key, ultimately additional statistical measures and randomized trials will be needed to prove that Lp(a)-reduction does not meaningfully prevent progressive AVC. M.G.L. is supported by the Institute for Translational Medicine and Therapeutics of the Perelman School of Medicine at the University of Pennsylvania, the National Institutes of Health (NIH)/National Heart, Lung, and Blood Institute (NHLBI) National Research Service Award postdoctoral fellowship (T32HL007843), and the Measey Foundation Physician Scientist Fellowship Award (https://www.measeyfoundation.org/). No new data were generated or analyzed in support of this research.