Figures
Abstract
Studies have reported that higher circulating levels of total cholesterol (TC), low-density lipoprotein (LDL) cholesterol and lower of high-density lipoprotein (HDL) cholesterol may be associated with increased risk of abdominal aortic aneurysm (AAA). Whether dyslipidemia causes AAA is still unclear and is potentially testable using a Mendelian randomization (MR) approach. We investigated the associations between blood lipids and AAA using two-sample MR analysis with SNP-lipids association estimates from a published genome-wide association study of blood lipids (n = 188,577) and SNP-AAA association estimates from European Americans (EAs) of the Atherosclerosis Risk in Communities (ARIC) study (n = 8,793). We used inverse variance weighted (IVW) MR as the primary method and MR-Egger regression and weighted median MR estimation as sensitivity analyses. Over a median of 22.7 years of follow-up, 338 of 8,793 ARIC participants experienced incident clinical AAA. Using the IVW method, we observed positive associations of plasma LDL cholesterol and TC with the risk of AAA (odds ratio (OR) = 1.55, P = 0.02 for LDL cholesterol and OR = 1.61, P = 0.01 for TC per 1 standard deviation of lipid increment). Using the MR-Egger regression and weighted median methods, we were able to validate the association of AAA risk with TC, although the associations were less consistent for LDL cholesterol due to wider confidence intervals. Triglycerides and HDL cholesterol were not associated with AAA in any of the MR methods. Assuming instrumental variable assumptions are satisfied, our finding suggests that higher plasma TC and LDL cholesterol are causally associated with the increased risk of AAA in EAs.
Citation: Weng L-C, Roetker NS, Lutsey PL, Alonso A, Guan W, Pankow JS, et al. (2018) Evaluation of the relationship between plasma lipids and abdominal aortic aneurysm: A Mendelian randomization study. PLoS ONE 13(4): e0195719. https://doi.org/10.1371/journal.pone.0195719
Editor: Shengxu Li, Tulane University School of Public Health and Tropical Medicine, UNITED STATES
Received: December 22, 2017; Accepted: March 28, 2018; Published: April 12, 2018
Copyright: © 2018 Weng et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All relevant data are provided in the paper and its Supporting Information files.
Funding: The National Heart, Lung, and Blood Institute (NHLBI) supported this research through R01 HL103695. The Atherosclerosis Risk in Communities Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts (HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C), R01HL087641, R01HL59367 and R01HL086694; National Human Genome Research Institute contract U01HG004402; and National Institutes of Health contract HHSN268200625226C. The authors thank the staff and participants of the ARIC study for their important contributions. Infrastructure was partly supported by Grant Number UL1RR025005, a component of the National Institutes of Health and NIH Roadmap for Medical Research. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Abdominal aortic aneurysm (AAA) affects about 1–2% of women and 4%-8% of men over 65 years old based on screening studies.[1] AAA is usually asymptomatic and the mortality rate after its rupture may exceed 90%.[2] Identification of risk factors for AAA are important for primary prevention of AAA as well as decreasing AAA-related mortality.
Several risk factors for AAA have been reported, including greater age[3, 4], male sex[4, 5], white race[5–8], cigarette smoking, and genetic factors.[9] A few prospective studies tested the associations between lipids and AAA risk, and they reported an inverse relationship between high-density lipoprotein (HDL) and AAA and positive relationships between total cholesterol (TC), low-density lipoprotein (LDL), triglycerides and AAA.[10–14]
The Mendelian randomization approach is a method to estimate the causal effect of an exposure on a disease outcome by using genetic variant(s) as instrumental variable(s) (IVs) of the exposure. Based on several assumptions, the causal effect of the exposure on the outcome can be estimated via the IV (genetic variant), regardless of the presence of confounding variables between the exposure and outcome.[15] The three main assumptions for Mendelian randomization include that the IV: (i) is associated with the exposure; (ii) is not associated with any confounder of the exposure—outcome association; and (iii) is conditionally independent of the outcome, given the exposure and confounders. A fourth IV assumption (monotonicity) is also required in order to identify a causal effect. We aimed to conduct a Mendelian randomization study to investigate a potential causal link between plasma lipids and AAA in the European Americans (EAs) of the Atherosclerosis Risk in Communities (ARIC) Study, using lipid-related single nucleotide polymorphisms (SNPs) as IVs.
Materials and methods
Study population
The ARIC study[16] enrolled a total of 15,792 participants aged 45–64 to a baseline exam in 1987–1989 from four U.S. communities, namely Forsyth County, North Carolina; Jackson, Mississippi; suburbs of Minneapolis, Minnesota; and Washington County, Maryland. Trained interviewers collected demographic, medical, and lifestyle information at baseline and four follow-up exams. All medications and supplements taken in the two weeks prior to the baseline exam were also recorded, and medication names were transcribed and coded. Each participant provided informed consent, and the Institutional Review Board at the University of Minnesota, Johns Hopkins University, Wake Forest University, University of North Carolina at Chapel Hill, Baylor College of Medicine, University of Texas Health Sciences Center at Houston, and University of Mississippi Medical Center approved the protocol.
Plasma lipid measurements
Fasting blood samples were collected at the baseline exam and stored at −70°C for a few weeks prior to lipid measurement. The ARIC Central Lipid Laboratory measured plasma cholesterol and triglycerides using a Cobas-Bio centrifugal analyzer (Roche Diagnostics, Montclair, NJ) with enzymatic kits (Boehringer Mannheim Diagnostics, Indianapolis, IN).[17, 18] HDL cholesterol was estimated by the method of Warnick et al,[19] and LDL cholesterol was calculated using the Friedewald formula.[20] If the level of triglycerides was over 400 mg/dL, then LDL cholesterol was not determined. The coefficients of variation within the laboratory for TC, triglycerides, HDL cholesterol, and LDL cholesterol were 2.5%, 2.7%, 3.7%, and 5.2%, respectively.[21]
SNP selection
We selected the SNP IVs for each lipid fraction based on the results of the Global Lipids Genetics Consortium (GLGC) genome-wide association study[22] with MR-Base (www.mrbase.org), [23] using the default thresholds for statistical significance (P < 5 x 10−8) and pruning SNPs for linkage disequilibrium (r2 = 0.001). A total of 78, 85, 53, and 85 SNPs were selected as the IVs for LDL cholesterol, HDL cholesterol, triglycerides, and TC, respectively (see S1 Table for additional information about these SNPs).
Genotyping and imputation
The Affymetrix Genome-Wide Human SNP Array 6.0 (Affymetrix, Santa Clara, CA, USA) was used to genotype SNPs in ARIC; imputation was conducted with the 1000 genome phase I version 3 reference panel. Detailed information on GWAS genotyping, quality control and imputation can be found elsewhere.[24] Imputation quality for the SNPs included in this study was good (quality score = 0.84–1.0 with a mean of 0.99). Additionally, 10 principal components of population stratification based on the GWAS data were generated by EIGENSTRAT[25] to reflect the population structure of the ARIC participants.
Outcome measurement
Incident, clinical AAA was defined using ICD codes from hospitalization records and death records from baseline through 2011, and the Medicare records from Centers for Medicare and Medicaid Services (CMS) for 1991–2011. In ARIC’s annual phone interview, each ARIC participant was asked about all hospitalizations and these records were sought. Additionally, participant identifiers were linked to Medicare claims to identify any missing AAA cases. Clinical AAA cases were defined based on one hospital discharge diagnosis or 2 outpatient diagnoses that occurred at least one week apart (ICD-9-CM: 441.3, or 441.4), procedure codes (38.44 or 39.71), or underlying cause of death codes (ICD-9: 441.3 or 441.4 or ICD-10: I71.3, or I71.4). Other diagnostic codes that indicated a probable diagnosis of AAA were investigated case-by-case to rule out AAA diagnosis. People with known AAA surgery at baseline (n = 11) or uncertain AAA during follow-up (n = 30) were excluded. Details on AAA ascertainment are described elsewhere.[14]
Statistical analysis
We removed individuals from the analysis if they were non-white (n = 4,308), missing genotype data (n = 1,980), taking cholesterol lowering medication (n = 319), not fasting for at least 8 hours (n = 206), or missing any lipid fraction measurements (n = 145).
We estimated odds ratios (OR) and 95% confidence intervals (CI) for AAA per 1 standard deviation increment in each of the lipid fractions using two-sample Mendelian randomization techniques, with the inverse variance weighted (IVW) approach as the primary method and MR-Egger regression and weighted median estimation approaches as sensitivity analyses to validate the results.[26, 27] In the primary analysis, we used the ratio estimator5 to combine the summary estimates and standard errors of the SNP—lipid fraction associations from the GLGC (median sample size: 177,765) with the summary estimates and standard errors of the SNP—AAA associations (adjusted for age, sex, and study center) from ARIC. We calculated the I2 index and Q test6 from the IVW analysis to check for evidence of heterogeneity among the IV estimates.
There might be bias in the estimates obtained in the IVW approach if any of the SNPs do not satisfy the IV assumptions. In particular, the assumption (iii) may be violated due to pleiotropy (i.e., if a SNP is associated with AAA risk through a pathway that does not involve blood lipids). Therefore, we also conducted sensitivity analyses using two other MR methods (MR-Egger regression and weighted median estimation) that allow for some relaxation of the IV assumptions. MR-Egger regression allows all IVs to violate the assumption (iii) as long as they meet a weaker assumption that any pleiotropic effects of the IVs are uncorrelated with the strength of their associations with the exposure.[26] Additionally, the intercept term from MR-Egger regression serves as a check for evidence of directional pleiotropy.[26] Weighted median estimation requires that no more than 50% of the weight contributed by the genetic variants violates the IV assumptions (ii) and (iii).[27] All MR analyses were conducted in Stata (version 12.1) using the mrrobust package.
We also conducted two additional sensitivity analyses. In the first one, the SNP—AAA associations from ARIC were additionally adjusted for the first five principal components of ancestry to account for population stratification. The second sensitivity analysis used ARIC as the data source for both the SNP—lipid fraction associations and SNP—AAA associations, with both sets of summary estimates adjusted for age, sex, and study center. Because ARIC comprised approximately 4.4% of the samples used in the GLGC genome-wide association study, there is a possibility of weak instrument bias in two-sample Mendelian randomization since the summary estimates were based on overlapping samples. Likewise, the second sensitivity analysis is even more prone to this bias given the 100% overlap of samples used to derive the summary estimates. As such, we estimated the strength of the IVs using the F statistic, the expected bias due to sample overlap under the null hypothesis, and the expected type I error rate.[28]
Results
A total of 8,793 EAs had measurements for plasma lipids, SNP dosages, and follow-up for AAA and thus were included in our analysis. Baseline characteristics for those who developed clinical AAA (n = 388) and those who did not (n = 8,405) after a median of 22.7 years of follow-up are shown in Table 1. Mean values for age, height, white blood count, alcohol intake, fibrinogen, LDL cholesterol, triglycerides, and TC were higher in those who developed clinical AAA than the remaining participants. The percentage of men, current smokers, and those with peripheral artery disease and hypertension were higher in incident clinical AAA cases as well. In contrast, mean HDL cholesterol levels and the percentage with diabetes were lower in clinical AAA cases.
Table 2 shows the strength of the SNP IVs in the GLGC and ARIC and their susceptibility to weak instrument bias. The SNP IVs explained more of the variation in LDL cholesterol and TC than in the other lipid fractions. In comparing the data sources, the SNPs generally explained slightly more variation of the lipid fractions in ARIC than in the GLGC, although F-statistics were much larger for the latter study because of its larger sample size (N = 188,577 in GLGC vs N = 8,793 in ARIC). The bias resulting from weak instruments for overlapped two-sample MR analysis would be limited using the GLGC data, and the expected bias under the null hypothesis and type I error rates were generally negligible or low when using ARIC data with 100% sample overlap (Table 2).
Table 3 shows the two-sample Mendelian randomization estimates of ORs (95% CIs) for the association of AAA with lipids, with the GLGC and ARIC as the data sources for the SNP—lipid and SNP—AAA associations, respectively. Higher TC was associated with a greater risk of AAA (OR 1.48, 95% CI: 1.02, 2.16 per 41.8 mg/dL increment) as estimated by the IVW method; ORs estimated using the MR-Egger and weighted median methods were similar in magnitude (OR: 2.04 and 1.66) with wider confidence intervals (p = 0.03 and 0.09, respectively). Similarly, each 38.7 mg/dL increment in LDL cholesterol was associated with 1.55 (95% CI: 1.08, 2.22) times greater odds of AAA based on the IVW methods. The MR-Egger regression and weighted median estimates were similar (1.43 and 1.59) but with wider confidence intervals (p = 0.21 and 0.09, respectively). None of the IV methods showed evidence of associations for HDL cholesterol or triglycerides with the risk of AAA.
SNP—lipid associations are from the GLGC and SNP—AAA associations (adjusted for age, sex, and center) are from ARIC.
There was no evidence of heterogeneity in the effect estimates used in the IVW analysis across the individual SNPs for LDL cholesterol and TC (I2 = 5%, P ≥ 0.35) but possibly modest heterogeneity for HDL cholesterol (I2 = 29%, P = 0.007; Table 3). The intercept estimates in the MR-Egger models for the four lipid fractions indicated no evidence for directional pleiotropy (P ≥ 0.08, Table 3). Sensitivity analyses that additionally adjusted for the first five principal components of ancestry (S2 Table) or used ARIC (instead of the GLGC) as the data source for the SNP—lipid summary estimates (S3 Table) were overall very consistent with the primary results shown in Table 3.
Discussion
Using the IVW method for two-sample MR, we demonstrated positive associations of LDL cholesterol and TC with the incidence of clinical AAA in EAs from the prospective, population-based ARIC study. The association between TC and AAA risk was supported by MR-Egger and median weighted methods with similar effect sizes and p-values, while the association between LDL cholesterol and AAA risk was less consistent due to the wider confidence intervals in both methods. No significant association was observed between HDL cholesterol or triglycerides and AAA.
A pleiotropic effect is a major concern related to the third assumption of Mendelian randomization.[15] An instrumental variable that has an effect on multiple risk factors simultaneously could distort or bias the estimates in our Mendelian randomization study. Using lipid SNPs as the instrument in Mendelian randomization studies of cardiovascular disease has been recognized as a challenge[29] because of the complex nature of the genetic influences for lipids. These complexities include (i) the close correlation among plasma lipid fractions, (ii) many identified SNPs being novel, with unclear function in lipid pathways and (iii) many of the SNPs being associated with multiple lipid fractions. Instead of removing variants related to multiple lipids, which likely reduces statistical power, we performed MR-egger method and demonstrated that the bias resulting from directional pleiotropic effects should be minimal (P > 0.05 for MR-Egger intercepts differing from zero).
In general, weak instruments may bias the estimated association between exposures and health outcomes.[15] In the current study, we included multiple genetic variants as IVs for plasma lipids and observed that our IVs as a whole had high F-statistics and explained a significant percentage of variance in lipid levels. This suggests that bias resulting from weak instruments is unlikely. Furthermore, given the relatively small amount of overlap between samples in our primary analysis (SNP—lipid associations from GLGC and SNP—AAA associations from ARIC, with 4.4% overlap), weak instruments would lead to bias towards the null.[30] In addition, a simulation study conducted by one of our coauthors (Guan et al.) showed that the bias is ignorable when the overlap is less than 20% (unpublished data). Our sensitivity analysis using estimates derived from ARIC for both SNP-lipids and SNP-AAA associations showed concordant results to the primary analysis. Bias due to weak instruments in a one-sample setting leads to bias towards the confounded observational association,[30] although there was also little concern of weak instruments as indicated in our sensitivity analysis. Significant associations between serum TC and AAA risk were previously observed in several perspective cohort studies, with higher serum TC associated with greater risk of AAA.[10–14] Provided the SNPs used in this analysis are valid instruments, our findings suggest a causally positive relation between TC and AAA, and the similar results from MR-egger and median weighted methods support this interpretation. Our results using the IVW method also showed that the LDL cholesterol could be causally related to AAA risk, and the results from the other two methods were very similar in the association estimates, although were less precise as reflected by the wider confidence interval. This is not surprising, because the IVW MR method is generally known to have better precision than MR-Egger regression and weighted median estimation.[27] The similar patterns of associations of TC and LDL cholesterol with AAA risk may imply that the causal effect from TC to AAA risk could be largely attributable to the effect from LDL cholesterol.
Mendelian randomization analysis did not support a causal association of HDL cholesterol or triglycerides with clinical AAA incidence in our sample. One explanation is that the genetic variants for HDL cholesterol and triglycerides only explained a small proportion of the total variance in these lipid levels (r2 = 5% for HDL cholesterol and r2 = 4.4% for triglycerides), which may have reduced our ability to detect a true association between these two lipids and AAA risk. Another interpretation is that both lipids are not causally related to AAA risk.
The biological mechanism underlying the association between TC and AAA is unclear. Animal studies showed that oxidized lipoproteins promote recruitment of inflammatory cells into the elastic media and adventitia of the aortic wall.[31] In vitro studies reported an inverse association between statin use and the activity of matrix metalloproteinase-9, a biomarker for AAA development.[32] Additionally, clinical trials have shown that statin use slows AAA growth.[32] These findings may potentially support a causal pathway between TC or LDL cholesterol and AAA risk; however, statins may prevent AAA by reducing inflammation as well.[32]
Recently, a large meta-analysis MR study between AAA and lipids reported significant associations of AAA with LDL-C, HDL-C, and triglycerides.[33] The strength of association for LDL-C and HDL-C in that study was similar to ours, while that for triglycerides was stronger. Differences in population characteristics might contribute to the heterogeneity in the strength of association for triglycerides. A prior research from Malmö Preventive Study showed increased blood triglycerides and total cholesterol in 126 male subjects later developing large AAA compared to 126 male controls after a median follow-up of 21 years, and the association between triglycerides and AAA was no longer significant after adjusting for total cholesterol and other risk factors.[12] It is noted that plasma levels of triglycerides were not significantly associated with AAA after adjustment for other cardiovascular risk factors in the longitudinal ARIC study as well.[14] The lack of statistical significance for HDL-C in our MR study was likely due to limited power. Compared to the published meta-analysis which pooled data primarily from case-control studies, our study analyzed AAAs ascertained in a prospective cohort. The significant replication of the MR association between AAA and LDL-C in our prospective study as well as the addition of total cholesterol-AAA association further strengthen the data to support a role for these lipid fractions in the etiology of AAA.
Besides the limitations related to the assumptions of Mendelian randomization analysis, there were a few other drawbacks of our study. First, the associations of genetic variants with blood lipid levels were mainly identified in individuals of European descent, and this study included only EAs. Therefore, our findings may not be generalizable to other racial/ethnic groups. Second, we used ICD codes to identify both symptomatic and asymptomatic clinical AAAs from 1991–2011. Certainly, other asymptomatic AAAs could have been present in the sample and missed without additional ultrasound screening. In our analysis, we did not include AAA cases (N = 56) identified by an ultrasound scan in ARIC visit 5 (2011–2013) occurring after our follow-up for clinical AAAs. We excluded these ultrasound AAAs because they were identified in the healthier surviving cohort and pooling them with clinical AAAs might cause bias in the risk factor associations. On the other hand, a separate analysis of the ultrasound-detected AAAs is underpowered.
Conclusions
Using the MR approaches, we identified a significant association between AAA and TC and replicated the previously reported association for LDL-C in EAs. Our data, together with the previously published meta-analysis study, support the hypothesis that TC and LDL-C play a role in the pathogenesis of AAA in EAs. Future studies in populations of different ancestries may be required to evaluate the generalizability of these findings to different ethnic populations.
Supporting information
S1 Table. List of SNPs selected as the instrumental variables for LDL cholesterol, HDL cholesterol, triglycerides, and total cholesterol.
https://doi.org/10.1371/journal.pone.0195719.s001
(XLSX)
S2 Table. (Sensitivity analysis).
Two-sample Mendelian randomization results for the odds of AAA per 1 standard deviation increment in lipid fraction measures. SNP—lipid associations are from the GLGC and SNP—AAA associations (adjusted for age, sex, center, and PC1–PC5) are from ARIC.
https://doi.org/10.1371/journal.pone.0195719.s002
(DOCX)
S3 Table. (Sensitivity analysis).
Two-sample Mendelian randomization results for the odds of AAA per 1 standard deviation increment in lipid fraction measures. SNP—lipid associations and SNP—AAA associations are both from ARIC and are adjusted for age, sex, and center.
https://doi.org/10.1371/journal.pone.0195719.s003
(DOCX)
Acknowledgments
The authors thank the staff and participants of the ARIC study for their important contributions. Additionally, the authors would like to thank the University of Minnesota Supercomputing Institute for use of the Itasca and lab supercomputers.
References
- 1. Nordon IM, Hinchliffe RJ, Loftus IM, Thompson MM. Pathophysiology and epidemiology of abdominal aortic aneurysms. Nat Rev Cardiol. 2011;8(2):92–102. pmid:21079638
- 2. Ernst CB. Abdominal aortic aneurysm. The New England journal of medicine. 1993;328(16):1167–1172. pmid:8455684
- 3. Auerbach O, Garfinkel L. Atherosclerosis and aneurysm of aorta in relation to smoking habits and age. CHEST Journal. 1980;78(6):805–809.
- 4. Cornuz J, Pinto CS, Tevaearai H, Egger M. Risk factors for asymptomatic abdominal aortic aneurysm Systematic review and meta-analysis of population-based screening studies. The European Journal of Public Health. 2004;14(4):343–349. pmid:15542867
- 5. Johnson G Jr, Avery A, McDougal EG, Burnham SJ, Keagy BA. Aneurysms of the abdominal aorta: incidence in blacks and whites in North Carolina. Archives of Surgery. 1985;120(10):1138. pmid:4038055
- 6. Sandiford P, Mosquera D, Bramley D. Ethnic inequalities in incidence, survival and mortality from abdominal aortic aneurysm in New Zealand. Journal of epidemiology and community health. 2012;66(12):1097–1103. pmid:22766777
- 7. Jahangir E, Lipworth L, Edwards TL, Kabagambe EK, Mumma MT, Mensah GA, et al. Smoking, sex, risk factors and abdominal aortic aneurysms: a prospective study of 18 782 persons aged above 65 years in the Southern Community Cohort Study. J Epidemiol Community Health. 2015;69(5):481–488. pmid:25563744
- 8. Baumgartner I, Hirsch AT, Abola MT, Cacoub PP, Poldermans D, Steg PG, et al. Cardiovascular risk profile and outcome of patients with abdominal aortic aneurysm in out-patients with atherothrombosis: data from the Reduction of Atherothrombosis for Continued Health (REACH) Registry. J Vasc Surg. 2008;48(4):808–814. pmid:18639426
- 9. Golledge J, Kuivaniemi H. Genetics of abdominal aortic aneurysm. Curr Opin Cardiol. 2013;28(3):290–296. pmid:23478885
- 10. Törnwall ME, Virtamo J, Haukka JK, Albanes D, Huttunen JK. Life-style factors and risk for abdominal aortic aneurysm in a cohort of Finnish male smokers. Epidemiology. 2001;12(1):94–100. pmid:11138827
- 11. Rodin MB, Daviglus ML, Wong GC, Liu K, Garside DB, Greenland P, et al. Middle age cardiovascular risk factors and abdominal aortic aneurysm in older age. Hypertension. 2003;42(1):61–68. pmid:12796281
- 12. Lindblad B, Borner G, Gottsater A. Factors associated with development of large abdominal aortic aneurysm in middle-aged men. Eur J Vasc Endovasc Surg. 2005;30(4):346–352. pmid:15936229
- 13. Forsdahl SH, Singh K, Solberg S, Jacobsen BK. Risk Factors for Abdominal Aortic Aneurysms A 7-Year Prospective Study: The Tromsø Study, 1994–2001. Circulation. 2009;119(16):2202–2208. pmid:19364978
- 14. Tang W, Yao L, Roetker NS, Alonso A, Lutsey PL, Steenson CC, et al. Lifetime Risk and Risk Factors for Abdominal Aortic Aneurysm in a 24-Year Prospective Study: The ARIC Study (Atherosclerosis Risk in Communities). Arterioscler Thromb Vasc Biol. 2016;36(12):2468–2477. pmid:27834688
- 15. Lawlor DA, Harbord RM, Sterne JAC, Timpson N, Davey Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology. Statistics in medicine. 2008;27(8):1133–1163. pmid:17886233
- 16. The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. American Journal of Epidemiology. 1989;129(4):687–702. pmid:2646917
- 17. Nagele U, Hagele EO, Sauer G, Wiedemann E, Lehmann P, Wahlefeld AW, et al. Reagent for the enzymatic determination of serum total triglycerides with improved lipolytic efficiency. Journal of clinical chemistry and clinical biochemistryZeitschrift fur klinische Chemie und klinische Biochemie. 1984;22(2):165–174.
- 18. Siedel J, Hagele EO, Ziegenhorn J, Wahlefeld AW. Reagent for the enzymatic determination of serum total cholesterol with improved lipolytic efficiency. Clinical chemistry. 1983;29(6):1075–1080. pmid:6851096
- 19. Warnick GR, Mayfield C, Benderson J, Chen JS, Albers JJ. HDL cholesterol quantitation by phosphotungstate-Mg2+ and by dextran sulfate-Mn2+-polyethylene glycol precipitation, both with enzymic cholesterol assay compared with the lipid research method. American Journal of Clinical Pathology. 1982;78(5):718–723. pmid:6182791
- 20. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clinical chemistry. 1972;18(6):499–502. pmid:4337382
- 21. Brown SA, Hutchinson R, Morrisett J, Boerwinkle E, Davis CE, Gotto AM Jr., et al. Plasma lipid, lipoprotein cholesterol, and apoprotein distributions in selected US communities. The Atherosclerosis Risk in Communities (ARIC) Study. Arteriosclerosis and Thrombosis. 1993;13(8):1139–1158. pmid:8343489
- 22. Willer CJ, Schmidt EM, Sengupta S, Peloso GM, Gustafsson S, Kanoni S, et al. Discovery and refinement of loci associated with lipid levels. Nat Genet. 2013;45(11):1274–1283. pmid:24097068
- 23. Telomeres Mendelian Randomization C, Haycock PC, Burgess S, Nounu A, Zheng J, Okoli GN, et al. Association Between Telomere Length and Risk of Cancer and Non-Neoplastic Diseases: A Mendelian Randomization Study. JAMA oncology. 2017;3(5):636–651. pmid:28241208
- 24. Pankow JS, Tang W, Pankratz N, Guan W, Weng LC, Cushman M, et al. Identification of Genetic Variants Linking Protein C and Lipoprotein Metabolism: The ARIC Study (Atherosclerosis Risk in Communities). Arterioscler Thromb Vasc Biol. 2017;37(3):589–597. pmid:28082259
- 25. Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D. Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet. 2006;38(8):904–909. pmid:16862161
- 26. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44(2):512–525. pmid:26050253
- 27. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genet Epidemiol. 2016;40(4):304–314. pmid:27061298
- 28. Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two-sample Mendelian randomization. Genet Epidemiol. 2016;40(7):597–608. pmid:27625185
- 29. Holmes MV, Asselbergs FW, Palmer TM, Drenos F, Lanktree MB, Nelson CP, et al. Mendelian randomization of blood lipids for coronary heart disease. Eur Heart J. 2015;36(9):539–550. pmid:24474739
- 30. Pierce BL, Burgess S. Efficient design for Mendelian randomization studies: subsample and 2-sample instrumental variable estimators. Am J Epidemiol. 2013;178(7):1177–1184. pmid:23863760
- 31. Thompson RW, Curci JA, Ennis TL, Mao D, Pagano MB, Pham CTN. Pathophysiology of abdominal aortic aneurysms. Annals of the New York Academy of Sciences. 2006;1085(1):59–73.
- 32. Van Kuijk JP, Flu WJ, Witteveen OP, Voute M, Bax JJ, Poldermans D. The influence of statins on the expansion rate and rupture risk of abdominal aortic aneurysms. The Journal of cardiovascular surgery. 2009;50(5):599–609. pmid:19741575
- 33. Harrison SC, Holmes MV, Burgess S, Asselbergs FW, Jones GT, Baas AF, et al. Genetic Association of Lipids and Lipid Drug Targets With Abdominal Aortic Aneurysm: A Meta-analysis. JAMA Cardiol. 2018; 3(1):26–33. pmid:29188294