Numerous biomarkers have been linked to the development of cardiovascular disease. However, given the observational nature of genetic epidemiological research, any given association between a risk factor and disease could be explained by confounding, where the association is explained by a third factor that is associated with the risk factor and the disease, rather than being directly causal. Observed associations may also be due to reverse causation, where the observed relationship is a consequence of the disease rather than the cause. Whether an association is causal or not cannot be resolved definitively from observational epidemiological research.
The use of randomised control trials (RCTs) in cardiovascular disease has confirmed causal factors such as low-density lipoprotein. Despite being thought of as the most robust study design, RCTs are not always feasible. However, ‘Mendelian randomization’, which relies upon the random assortment of genes from parents to offspring at the time of conception, [Davey Smith G, Ebrahim S. (2009) IJE 32(1):1-22] provides a novel epidemiological research design that is far less susceptible to confounding and excludes reverse causation as a possible non-causal explanation for the observed association [Lawlor DA et al. (2008) Stat Med 27(8): 1133-63]. Because genes are randomly assorted from parents to offspring, this results in non-genetic characteristics that might confound any association being equally distributed amongst the relevant genetic variants, and so allows a randomised comparison between groups of individuals defined by a particular genetic variant. The Mendelian randomization strategy thus helps to clarify whether a causal relationship between a biomarker and a disease exists. A triangle of information is required: specifically, the magnitude of the association between, first, the biomarker and disease, second, the genetic variant and the biomarker, and third, the genetic variant and the disease. The first two associations can be used to estimate the third, which when directly measured can provide an unconfounded assessment of the association between the biomarker and the disease.
In the recent issue of JAMA, Kamstrup et al. [Kamstrup et al. (2009) JAMA 301(22): 2331-2339] conducted a Mendelian randomization study to assess the role of lipoprotein(a) in myocardial infarction (MI). Several observational studies and meta-analyses have found an association between lipoprotein(a) and MI, with presumptive evidence of a causal relationship. Using the random assortment of genetic variation in the lipoprotein apo(a) gene (LPA), as their randomization, Kamstrup et al. quantify three associations: that between serum lipoprotein(a) levels and MI events; that between LPA gene KIV-2 variants and serum lipoprotein(a) levels; and that between LPA gene KIV-2 variants and MI events.
Using three independent cohorts from Copenhagen to test the hypothesis that increased levels of lipoprotein(a) cause increased risk of MI, the authors observed an increased risk of MI with elevated levels of lipoprotein(a) in accordance with previous findings. Genotyping of the LPA gene inversely associated the number of KIV-2 repeats with lipoprotein(a) levels. The number of KIV-2 repeats was also found to be inversely associated with risk of MI. Kamstrup et al. were able to demonstrate that a genetically determined doubling of lipoprotein(a) plasma level leads to an increase in risk of MI of 22% (95% confidence intervals: 9% to 37%). These findings were consistently seen in the all three independent cohorts studied, strongly supporting the previous observational studies conducted to date, and are consistent with a causal association of elevated lipoprotein(a) level and increased MI risk.
Comment: An accompanying editorial by Thanassoulis and O’Donnell [Thanassoulis and O’Donnell (2009) JAMA 301(22): 2386-2388] highlight several considerations that can invalidate the Mendelian randomization study design. Whilst some issues such as confounding due to population stratification can be disregarded due to the homogenous population used in the study, others such as the possible pleiotropic nature (i.e. multiple biological effects) of the KIV-2 genetic variant cannot be ignored, as the KIV-2 variant affects both the lipoprotein(a) plasma level and the lipoprotein(a) isoform size. Although this study provides the strongest evidence to date of lipoprotein as a causal factor for MI, clinical implications remain severely limited. This study also does not investigate whether genetic testing of the LPA gene or measurement of lipoprotein(a) levels can lead to improved MI risk estimation (see previous news). Moreover, no currently recommended medication specifically reduces lipoprotein(a) levels, even if such a test recommended lipid-lowering therapy. Despite these limitations, this successful use of the Mendelian randomization study design by Kamstrup et al. moves lipoprotein(a) from risk marker to causal factor, although the authors agree that the “final proof of causality still requires randomized clinical trials demonstrating reduced MI risk in response to lipoprotein(a)-lowering therapy”.