Low-Density Lipoprotein Cholesterol and the Risk of Cancer: A Mendelian Randomization Study

Marianne Benn 0 1 Anne Tybjaerg-Hansen 0 1 Stefan Stender 0 1 Ruth Frikke-Schmidt 0 1 Brge G. Nordestgaard 0 1 0 Ringvej 75 , DK-2730 Herlev , Denmark ( 1 Affiliations of authors: Department of Clinical Biochemistry (MB, BGN) and The Copenhagen General Population Study (MB, AT-H, RF-S , BGN), Herlev Hospital, Copenhagen University Hospital , Copenhagen , Denmark; Department of Clinical Biochemistry , Rigshospitalet , Copenhagen University Hospital , Copenhagen, Denmark (AT-H, SS, RF-S); The Copenhagen City Heart Study , Bispebjerg Hospital, Copenhagen University Hospital , Copenhagen, Denmark (AT-H , BGN); Faculty of Health Sciences, University of Copenhagen , Copenhagen, Denmark (MB, AT-H, SS, RF-S, BGN) We studied 10 613 participants in the Copenhagen City Heart Study (CCHS) and 59 566 participants in the Copenhagen General Population Study, 6816 of whom had developed cancer by May 2009. Individuals were genotyped for PCSK9 R46L (rs11591147), ABCG8 D19H (rs11887534), and APOE R112C (rs429358) and R158C (rs7412) polymorphisms, all of which are associated with lifelong reduced plasma LDL cholesterol levels. Plasma LDL cholesterol was calculated using the Friedewald equation in samples in which the triglyceride level was less than 354 mg/dL and measured directly by colorimetry for samples with higher triglyceride levels. Risk of cancer was estimated prospectively using Cox proportional hazards regression analyses and cross-sectionally by logistic regression analyses. Causality was studied using instrumental variable analysis. All statistical tests were two-sided. In the CCHS, compared with plasma LDL cholesterol levels greater than the 66th percentile (>158 mg/dL), those lower than the 10th percentile (<87 mg/dL) were associated with a 43% increase (95% confidence interval [CI] = 15% to 79% increase) in the risk of cancer. The polymorphisms were associated with up to a 38% reduction (95% CI = 36% to 41% reduction) in LDL cholesterol levels but not with increased risk of cancer. The causal odds ratio for cancer for a 50% reduction in plasma LDL cholesterol level due to all the genotypes in both studies combined was 0.96 (95% CI = 0.87 to 1.05), whereas the hazard ratio of cancer for a 50% reduction in plasma LDL cholesterol level in the CCHS was 1.10 (95% CI = 1.01 to 1.21) (P for causal odds ratio vs observed hazard ratio = .03). Low plasma LDL cholesterol levels were robustly associated with an increased risk of cancer, but genetically decreased LDL cholesterol was not. This finding suggests that low LDL cholesterol levels per se do not cause cancer. Methods Results Conclusion Low plasma levels of low-density lipoprotein (LDL) cholesterol are associated with an increased risk of cancer, but whether this association is causal is unclear. In 1974, Rose et al. (1) observed that plasma cholesterol levels were lower than expected in men with colon cancer. Since then, several prospective studies (27) have confirmed an association between low plasma levels of cholesterol and an increased risk of cancer. The reason for this observation is unclear; theoretically, it could reflect a causal relationship (ie, low plasma low-density lipoprotein [LDL] cholesterol causes cancer) or be due to a confounder that causes both low plasma LDL cholesterol and cancer or to reverse causation (ie, a preclinical cancer that reduces plasma LDL cholesterol levels) (3,7). Whether low LDL cholesterol causes cancer is an important question. However, data from randomized intervention trials of LDL cholesterol lowering with statins conducted during the last three decades and several large metaanalyses that include data from more than 90 000 patients [summarized in (8)] have lessened this concern. Nevertheless, combined results from rodent studies (9), an early trial of the lipid-lowering agent clofibrate (10), and a recent lipid lowering trial [Simvastatin and Ezetimibe in Aortic Stenosis (SEAS) Study (11)] have led to renewed concerns that lowering plasma cholesterol via pharmacological interventions might increase the risk of cancer. Mendelian randomization is an epidemiological approach that can be used to study a potential causal relationship because it circumvents confounding and reverse causation (1217). This approach makes use of the random assortment of genetic variants during gamete formation, which is analogous to the random assignment of patients to placebo or active treatment in a clinical intervention trial, to assess the causal relationship between a modifiable risk factor and disease. For example, with this approach, genetic variants that are associated with low plasma levels of LDL cholesterol would be largely unconfounded by other risk factors for low plasma LDL cholesterol, such as sex, smoking history, body mass index (BMI), hypertension, and diabetes mellitus, and can therefore be used to assess the consequences of lifelong low plasma LDL cholesterol levels independent of other risk factors with no risk of reverse causation (14). In this study, we tested the hypothesis that there is a robust and potentially causal association between low LDL cholesterol levels and an increased risk of cancer. We tested the robustness of the association by adjusting the risk estimates for age, sex, BMI, hypertension, diabetes mellitus, current smoking, and statin use; additional adjustments included ischemic heart disease, exclusion of events that occurred within 4 years after LDL cholesterol measurement, and exclusion of participants who were treated with statins. We tested the potential causality of the association by examining whether genotypes that are associated with lifelong reduced plasma LDL cholesterol levels were associated with an increased risk of cancer. For the latter analysis, we compared the association between genotypes and plasma LDL cholesterol levels with the association between genotypes and the risk of cancer, performed instrumental variable analysis to estimate the causal association between genetically reduced levels of LDL cholesterol and the risk of cancer, and compared this estimate with the corresponding estimate from conventional observational epidemiology. Participants and Methods Participants The study population consisted of participants in two studies in Denmarkthe Copenhagen City Heart Study (CCHS) and the Copenhagen General Population Study (CGPS)both of which were approved by the institutional review boards and Danish ethical committees (No. KF-V.100.2039/91, KF-01-144/01, H-KF01-144/01) and conducted according to the Declaration of Helsinki. Written informed consent was obtained from participants. All participants were white and of Danish descent. None of the participants appeared in more than one study, which allowed us to combine the two similar general population studies to obtain maximal statistical power. Follow-up was 100% complete in both studies. The CCHS is a prospective study of the general population of Copenhagen, Denmark, that was initiat (...truncated)