Figures
Abstract
Background
Fibrinogen is an essential hemostatic factor and cardiovascular disease risk factor. Early attempts at evaluating the causal effect of fibrinogen on coronary heart disease (CHD) and myocardial infraction (MI) using Mendelian randomization (MR) used single variant approaches, and did not take advantage of recent genome-wide association studies (GWAS) or multi-variant, pleiotropy robust MR methodologies.
Methods and findings
We evaluated evidence for a causal effect of fibrinogen on both CHD and MI using MR. We used both an allele score approach and pleiotropy robust MR models. The allele score was composed of 38 fibrinogen-associated variants from recent GWAS. Initial analyses using the allele score used a meta-analysis of 11 European-ancestry prospective cohorts, free of CHD and MI at baseline, to examine incidence CHD and MI. We also applied 2 sample MR methods with data from a prevalent CHD and MI GWAS. Results are given in terms of the hazard ratio (HR) or odds ratio (OR), depending on the study design, and associated 95% confidence interval (CI).
In single variant analyses no causal effect of fibrinogen on CHD or MI was observed. In multi-variant analyses using incidence CHD cases and the allele score approach, the estimated causal effect (HR) of a 1 g/L higher fibrinogen concentration was 1.62 (CI = 1.12, 2.36) when using incident cases and the allele score approach. In 2 sample MR analyses that accounted for pleiotropy, the causal estimate (OR) was reduced to 1.18 (CI = 0.98, 1.42) and 1.09 (CI = 0.89, 1.33) in the 2 most precise (smallest CI) models, out of 4 models evaluated. In the 2 sample MR analyses for MI, there was only very weak evidence of a causal effect in only 1 out of 4 models.
Conclusions
A small causal effect of fibrinogen on CHD is observed using multi-variant MR approaches which account for pleiotropy, but not single variant MR approaches. Taken together, results indicate that even with large sample sizes and multi-variant approaches MR analyses still cannot exclude the null when estimating the causal effect of fibrinogen on CHD, but that any potential causal effect is likely to be much smaller than observed in epidemiological studies.
Citation: Ward-Caviness CK, de Vries PS, Wiggins KL, Huffman JE, Yanek LR, Bielak LF, et al. (2019) Mendelian randomization evaluation of causal effects of fibrinogen on incident coronary heart disease. PLoS ONE 14(5): e0216222. https://doi.org/10.1371/journal.pone.0216222
Editor: Andreas Brodehl, Heart and Diabetes Center NRW, UNiversity Hospital of the Ruhr-University Bochum, GERMANY
Received: November 19, 2018; Accepted: April 16, 2019; Published: May 10, 2019
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: Genotype and phenotype data is available from all cohorts via dbGAp or by request to the respective contact person's or cohort steering committees for researchers who meet the criteria for access to confidential data. Specific data for each cohort can be found below. Participant data for the KORA cohort can be obtained by submitting a request to the online KORA PASST system (https://epi.helmholtz-muenchen.de/) and obtaining proper approvals Data for GENOA participants can be requested via dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000379.v1.p1) Data for ARIC participants can be requested via dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000090.v5.p1). Researhcers may also contact Lisa Reeves (lisa_reeves@unc.edu) and visit https://www2.cscc.unc.edu/aric/contact_the_coord_ctr for more information on obtaining ARIC participant data. Data for CHS participants can be best requested via dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000287.v6.p1) Researcher seeking data on GeneStart participants should visit https://www.hopkinsmedicine.org/gim/research/GeneSTAR/for_researchers for information on how to request data To obtain data on MESA participants please contact Craig Johnson (wcraigj@uw.edu). Some participant data is also available on dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000209.v13.p3) The WGSH study is currently not permitted to release individual level data. Requests for data from WGHS can be made do Dan Chasman (dchasman@research.bwh.harvard.edu) Participant data for LURIC can be obtained by sending a request to Kai Grunwald (Kai.Grunwald@weitnauer.net) and obtaining proper approvals Information on how to obtain participate data for SHIP can be found at the following link - http://www2.medizin.uni-greifswald.de/cm/fv/ship/datennutzung/ Data for FHS can be obtained via dbGaP (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000007.v30.p11) or by submitting a research proposal and obtaining proper approvals (https://www.framinghamheartstudy.org/fhs-for-researchers/research-application-overview/) Participant data for the Rotterdam Study can be obtained by contacting Arfan Ikram (m.a.ikram@erasmusmc.nl).
Funding: No funding sources had a role in the design of the study or the analysis or interpretation of the data. Infrastructure for the CHARGE Consortium is supported in part by the National Heart, Lung and Blood Institute (NHLBI) grant R01HL105756. ACM, NLS and PSdeV were supported by NIH NHLBI 1R01HL139553 and 1R01HL141291. PSdV was additionally supported by American Heart Association grant number 18CDA34110116. Cohort-specific funding sources for each cohort are in the Supplemental Materials. The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the NHLBI; the National Institutes of Health; or the U.S. Department of Health and Human Services. The funder (Synlab Academy, Synlab Holding Deutschland GmbH) provided support in the form of salaries for author WM, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of WM are articulated in the ‘author contributions’ section.
Competing interests: BMP reports serving on the DSMB of a clinical trial funded by the manufacturer (Zoll LifeCor) and on the Steering Committee of the Yale Open Data Access Project funded by Johnson & Johnson. WK reports personal fees from AstraZeneca Novartis, Pfizer, The Medicines Company, GlaxoSmithKline, DalCor, Sanofi, Berlin-Chemie, Kowa, and Amgen. WK also reports grants and non-financial support from Abbott, Roche Diagnostics, Beckmann, and Singulex. All reports from WK are outside the submitted work. WM reports grants and personal fees from Siemens Diagnostics, Aegerion Pharmaceuticals, AMGEN, AstraZeneca, Danone Research, Sanofi/Genzyme, Pfizer, BASF, and Numares. WM reports personal fees from Hoffmann LaRoche, MSD, Sanofi, and Alexion. WM is employed by Synlab Holding Deutschland GmbH and all reports by WM are outside the submitted work. This does not alter our adherence to PLOS ONE policies on sharing data and materials.
Introduction
Fibrinogen is an essential component of the clotting and hemostasis system with a strong genetic basis [1–3]. Although it primarily serves as the precursor to fibrin, it also carries out several other functions, including enhancing platelet aggregation and mediating inflammation [4, 5]. In epidemiologic studies, fibrinogen levels are associated with coronary heart disease (CHD) [6–8], myocardial infarction (MI) [9, 10], ischemic stroke [11, 12], and abdominal aortic aneurysm [13, 14].
Mendelian randomization (MR) is an instrumental variable analysis method which uses genetic variants as instruments to uncover evidence for a causal relationship between a modifiable risk factor and outcome.[15] MR studies utilizing a limited number of genetic variants in the FGB promoter have yielded little evidence of a causal effect of fibrinogen on CHD or MI [16–18]. In a genome-wide association study (GWAS) for fibrinogen, each fibrinogen-associated variant was individually evaluated for association with CHD, but no associations provided substantial evidence of a causal effect [19]. To date MR studies of fibrinogen have been limited to single variant approaches which have not taken into account recent GWAS findings or modern, multi-variant MR methodologies. Here we re-examine the potential for fibrinogen to be a causal biomarker for CHD and MI, taking into account these improved approaches.
Results
For incident CHD there were 3,147 incident events observed in 15,427 participants in the discovery analyses, and 1,482 incident events among the 34,209 participants in the replication analyses. Of the 18,798 participants in the incident MI discovery analyses, 1,711 had an incident MI. For the replication analyses, there were 687 incident MI events out of the 33,288 participants. Table 1 contains the distributions of clinical covariates and fibrinogen. The FGB variant rs1800790 (commonly used in previous fibrinogen MR analyses) had a weaker association (by effect size) than the allele score (Table A in S1 File). In single variant analyses of rs1800790 the estimated causal effect appeared to be centered around the null with little evidence of a causal effect of fibrinogen on CHD or MI (Table B in S1 File), consistent with published literature. In multi-variant MR using the 2SC model, we observed evidence of a causal association of fibrinogen on incident CHD in the discovery and replication analyses which remained in a combined analysis of all cohorts (HR = 1.75; CI = 1.22–2.51; P = 0.002; Fig 1). For incident MI, we observed an elevated HR that included the null, even in the combined analysis (HR = 1.45; CI = 0.85–2.49; P = 0.17; Fig 2).
Forest plot of the CHD MR analysis for the discovery, replication, and combined sets of cohorts. Shown beside each cohort name is the sample size and number of incident CHD events given as (N events; N total). CHD = coronary heart disease; FE = fixed-effects; HR = hazard ratio; CI = confidence interval.
Forest plot of the MI MR analysis for the discovery, replication, and combined sets of cohorts. Shown beside each cohort name is the sample size and number of incident MI events given as (N events; N total). MI = myocardial infarction, FE = fixed-effects, HR = hazard ratio, CI = confidence interval.
Pleiotropy robust models
In sensitivity analyses four MR methods were used each of which is at least partially robust to horizontal pleiotropy under differing assumptions. Sensitivity analyses were performed using summary statistics from the CARDIoGRAMplusC4D consortium [34] which used prevalent cases, but had a larger sample size than the CHARGE cohorts (60,801 prevalent cases and 123,504 controls). For CHD, three of the four models showed a positive effect, albeit smaller than the effect observed in the 2SC model, with the MR PRESSO method having the largest causal OR (OR = 1.18; CI = 0.98, 1.42; Table 2). For MI only the MR PRESSO method showed a causal OR > 1 (OR = 1.16; CI = 0.98, 1.38; Table 2), again substantially reduced from that observed in the 2SC model. All other models for MI showed little evidence of a causal effect of fibrinogen on MI.
We also examined MR associations of fibrinogen on metabolic CHD risk factors using published data available in MR-base as described in the methods. While fibrinogen showed a positive causal effect estimate for some of the CHD risk factors, none provided substantial evidence for excluding the null after accounting for the number of tests performed (Table C S1 File).
Discussion
The attractiveness of fibrinogen as a causal factor in CHD comes from its roles in both thrombosis and inflammation. Fibrinogen is the precursor to fibrin, which interlinks into a mesh that acts as the scaffold of blood clots. Additionally, fibrinogen also has an active role in platelet aggregation,[20] thus contributing to the formation of platelet plugs. By binding the CD11b/CD18 integrin receptor fibrinogen activates the NF-κB pathway [5], an important pathway in inflammation as well as the formation, destabilization, and rupture of atherosclerotic plaques [21, 22]. As a modifiable risk factor [23] even a small causal effect of fibrinogen on CHD could have substantial public health implications.
Using the allele score approach, a 1 g/L higher fibrinogen concentration was causally associated with a HR of 1.75 (CI = 1.22–2.51) in the combined cohort analysis for CHD. However, sensitivity analysis using methods robust to pleiotropy arising from independent effects of SNPs on exposure and outcome (which could invalidate MR analyses) suggested a substantially weaker causal effect on CHD even for the model with the strongest effect estimate (OR = 1.18 per 1 g/L higher fibrinogen; CI = 0.98, 1.42), and the MR Egger model showed virtually no evidence of a causal effect–though the wide 95% confidence interval encompassed effects from all other models. Overall, when accounting for potential horizontal pleiotropy, the accumulated evidence points to a substantially weaker casual effect of fibrinogen on CHD than the observational risk ratio of 1.8 (CI = 1.6, 2.0) previously reported [6]. Using rs1800790 in a single variant MR analysis, there was limited evidence of any causal effect, though the 95% confidence interval could not exclude positive estimated causal effects seen in multi-variant analyses. In combination these analyses suggest that after accounting for horizontal pleiotropy the effect of fibrinogen on CHD is likely to be small, if any at all, and that current MR estimates of the potential causal effect remain unable to exclude the null, despite large sample sizes and the latest methodologies. Discrepancies in the estimates of the causal effect of fibrinogen on CHD obtained by different MR approaches are likely to be due to differences in their approach to accounting for horizontal pleiotropy, given the clear influence of such pleiotropy on estimates from observations studies.
Comparison with previous MR analyses
Previous MR studies assessing the causal effect of fibrinogen on CHD or MI focused exclusively on rs1800790 [24, 25]. In a few studies one additional variant also in the FGB promoter region was examined, however this variant is in nearly complete LD with rs1800790, particularly in Europeans [16, 18]. The allele score was a better predictor of fibrinogen than rs1800790 alone (Table A in S1 File). Though the allele score estimated a causal effect of fibrinogen on CHD similar to observational studies, much of this appeared to be driven by pleiotropy as estimated effects decreased in models more robust to pleiotropy (Table 2). This highlights the need to balance increased power from multi-variant approaches with the potential for increased pleiotropy in these instruments.
For the CHARGE cohorts we used exclusively incident cases whereas previous studies utilized populations composed entirely or primarily of prevalent cases. In some instances, the use of prevalent cases may bias MR studies such as if the disease subsequently what is perceived as a disease risk factor, e.g. if CHD leads to higher fibrinogen as opposed to the reverse, then reverse confounding can still occur even in an MR setting [26]. Additionally, if the risk factor were to affect severity of an event, e.g. the fatality of MI, then use of prevalent cases may dilute the MR-estimated causal effect as the most severe cases may not be observed due to being too ill to participate or suffering a fatal event. This type of prevalence-incidence bias is not exclusive to MR analyses [27–29]. However, care must still be taken when interpreting results from incident case MR studies as the exclusion of prevalent cases is equivalent to conditioning on disease status at baseline. This has the potential to introduce bias in the form of an exclusion restriction violation.[30] Whether bias is introduced and the degree of confounding are dependent on the actual biological processes that account for the relationship between the genetic instrument(s) chosen, the modifiable risk factor, and outcome in the MR analysis. When performing incident case MR it is best to combine the efforts with MR analyses including prevalent cases and interpret results for both with careful consideration towards their underlying assumptions, strengths, and weaknesses.
In general, our results are compatible with previous MR studies, however we use more modern methods, including multi-variant, pleiotropy robust methods, able to produce smaller confidence intervals and which indicate that after accounting for pleiotropy there may be a small positive effect of fibrinogen on CHD. This is particularly true for the methods producing the most precise estimates. However, these results warrant further investigations as confidence intervals for some models were still wide and with results for the single variant and MR Egger analyses possibly more consistent with no causal effect than even a small causal effect.
Strengths and limitations
As with all MR studies the causal effects estimated here are based on regression estimates for genetic variants and are only valid, causal estimates under the assumptions of MR. Additionally, causal estimates generated via MR methodologies are for lifelong, genetically determined increases in the exposure, e.g. fibrinogen, which means that caution should be exercised when applying clinical interpretations or attempting to translate results into estimates of an intervention [31, 32]. In the initial analysis a meta-analysis was performed across studies which were heterogeneous in their distribution of underlying clinical covariates (Table 1). Though there was heterogeneity in the clinical covariate distribution, the allele score variants were evaluated for associations with these clinical covariates. Additionally, in the two sample MR we used approaches robust to confounding from pleiotropy due to associations between SNPs and other clinical covariates, such as body mass index, which might lie outside any causal pathway linking fibrinogen and CHD. This study had some overlap between studies involved in the GWAS used to select fibrinogen variants and those used in the MR analyses. Our approach to mitigate this was to replicate the allele score analysis in an independent set of cohorts. For the pleiotropy robust 2-sample MR approaches this overlap was unavoidable, however there was no overlap for the cases which means that unbiased estimates should be obtained [33]. A strength of the study is the use of incident cases for the allele score model approach which reduces the potential for bias from reverse confounding (which can still affect MR studies) and prevalence-incidence bias. Additionally, even though the allele score approach was sensitive to horizontal pleiotropy we used an array of additional approaches that were each partially robust to horizontal pleiotropy through different assumptions about the nature of the pleiotropy. These models often have lower power than other approaches, which motivated our use of a previously published GWAS which had 60,801 prevalent cases and 123,504 controls [34]. However, to prevent potential bias and more closely align with our initial analyses, a large sample size of incident cases independent of those used to evaluate associations between genetic variants and fibrinogen would have been preferable.
Conclusion
Fibrinogen represents an important role in thrombosis, platelet aggregation, and inflammation making it a promising risk factor for CHD. Despite the epidemiological evidence, MR studies using prevalent cases and single variant approaches have consistently shown no causal effect of fibrinogen on CHD. Out results indicate that epidemiologic studies may substantially over-estimate any causal effect of fibrinogen on CHD. While some MR models which accounted for pleiotropy did show a modest causal effect, the 95% confidence intervals still contained the null indicating that researchers should exercise caution in interpreting these results. Though, it may be tempting to relate these results to the utility of fibrinogen in clinical testing, caution is urged in this endeavor. Causal estimates, null or not, are at best imperfect proxies for the potential utility of a biomarker in a clinical setting, and other estimates, e.g. receiver operating characteristic, positive/negative predictive value, are important to consider. Our results suggest any causal effect of fibrinogen on CHD is likely to be small, and resolving any causal effect will require further analyses using larger sample sizes and more precise methods.
Methods
This study was conducted within the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium [35] using 11 European-ancestry cohorts. For incident CHD, six cohorts participated in the initial (discovery) analyses (N = 15,427), and four cohorts (N = 34,209) contributed data for replication. For incident MI, six cohorts participated in the discovery (N = 18,798), and three cohorts participated in the replication (N = 33,288) analyses. Details on all cohorts are given in the Supplemental Materials and the clinical covariates in Table 1, and Fig 3 outlines all analyses. Data collection analysis for all cohorts was approved by their respective Institutional Review Boards and/or ethical committees, and all cohorts obtained written, informed consent from participants.
Outline of analyses using the allele score, rs1800790 and 2 Sample MR approaches including the analytic method used to estimate the causal effect, subject to valid MR assumptions, for all stages of the analysis. CHD = coronary heart disease; MR = Mendelian Randomization; MI = myocardial infarction.
Assessment of CHD and MI
We defined incident CHD as validated, incident fatal or non-fatal CHD events which included: validated hospitalized MI, CHD-related hospitalizations, definite CHD deaths, likely CHD deaths, and CHD-related revascularization procedures, e.g. percutaneous coronary intervention and coronary artery bypass grafting. Incident MI was defined as a validated fatal or non-fatal MI and included definite MI hospitalizations. For cohorts that used questionnaires as a component of the follow-up procedures, all events were corroborated with medical records and/or review by trained medical personnel. Cohort specific details are given in the Supplemental Online Methods.
Fibrinogen assessment
Fibrinogen was assessed by multiple methods, with seven cohorts using the Clauss method [36]. Of the remaining four cohorts, RS used a clotting time-derived method to assess fibrinogen concentrations, while KORA, MESA, and WGHS used immunological assays to assess total fibrinogen. Comparisons of these three methods have shown that the Class and clotting-time derived methods have very similar means, while the immunological assays can have slightly lower means [37]. However, the correlations between the methods have been reported to be high, with Pearson correlations > 0.95 [38]. Previous studies have also found that effect estimates obtained using the different fibrinogen assessment methods are comparable [1, 39, 40]. We also did not observe heterogeneity in effect estimates by fibrinogen assessment method.
Genotyping and imputation
Genotyping and imputation were performed separately in all cohorts, per published methods [41]. All participating studies used either the HapMap build 36 [42], 1000 Genomes phase I version 3, or 1000 Genomes phase I version 2 reference panel for imputation [43]. Imputation was performed via MACH[44] or IMPUTE [45]. Low quality variants were excluded in line with previously published approaches: MACH imputation quality < 0.3 or IMPUTE imputation quality < 0.4 [41].
Creation of the allele score
We evaluated 69 variants associated with fibrinogen in at least one of three recent genome/exome-wide association studies [2, 19, 41] for inclusion into the allele score [46]. We applied four criteria to each variant to improve the plausibility that each meets the MR assumptions. First, to ensure that the variants were not correlated with known risk factors for cardiovascular disease (CVD), the Spearman correlation between each of the variants and body mass index (BMI), low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, type 2 diabetes mellitus (binary), hypertension (binary), and smoking (ever, never, current) was tested within each cohort and any variants with a Spearman correlation greater than 0.10 in any cohort for any of these outcomes were removed. Second, the variants were tested for linkage disequilibrium (LD) with known CHD loci [34, 47–56] using SNAP from the Broad Institute with LD patterns coming from European ancestry individuals [57]. As no variant had r2 > 0.20 with a CHD locus, they were considered independent of known CHD loci. Next, we reduced pairs of variants in high LD (r2 > 0.70) by preferentially retaining those variants that were found in the largest genome-wide scan [41]. Finally, we eliminated any variants that were missing across any of the discovery cohorts, leaving 38 variants that composed the allele score (Table D in S1 File). We tested the allele score for association with each of the aforementioned CHD risk factors in each cohort as well as in a meta-analysis of all cohorts. The allele score was not associated with any CHD risk factor in the meta-analysis after a Bonferroni correction for the six tests performed (P > 0.008; Table E in S1 File). Six variants from the allele score which were unavailable in one or more replication cohorts were removed from the allele score in the replication phase to ensure a consistent allele score in the replication meta-analysis (Table D in S1 File). In a sensitivity analysis these variants were also removed from the discovery cohorts and the causal effect evaluated in a combined meta-analysis.
Each genotype was aligned prior to summing to create the score so that the designated effect allele corresponded to a positive association with fibrinogen according to the direction of effect in the largest and most recent fibrinogen GWAS [19].
Mendelian randomization
MR is a powerful framework that uses genetic variants as instrumental variables to infer causal relationships between a defined exposure and outcome. The causal effect estimated by MR is the alteration in exposure due to genetic variation and is thus assumed to be over the entire life course. There are three assumptions for a genetic variant to be a valid instrument for MR [58]
- The genetic variant is independent of confounders of exposure and outcome under examination
- The genetic variant is associated with the exposure
- The genetic variant is independent of the outcome conditional on the exposure and any confounders
In addition to these three conditions, valid estimates from MR are dependent on any parametric assumptions of the model being used to estimate relevant coefficients and standard errors.
Our initial MR analyses used a two-stage procedure employing a Cox regression model (2SC). To improve power, we regressed fibrinogen on age and sex and used the resulting residuals as input to the 2SC analyses. In the first stage of the 2SC procedure the fibrinogen residuals were regressed on the allele score. In the second stage the predicted values from the first stage regression were associated with incident MI or CHD via a Cox proportional hazards model. This approach is similar to the two-stage predictor-substitution MR approach [59–61], and results from the 2SC model are given per unit (g/L) increase in the fibrinogen residuals. We used a fixed effects model for all meta-analyses since we observed little heterogeneity according to the Q-statistics [62] (P(Q) > 0.05 for all analyses). We also compared associations with our allele score to those obtained using a single variant, FGB -455G>A (rs1800790), which is a commonly used variant for fibrinogen MR analyses [16, 18]. We tested the allele score in the discovery cohorts and replication cohorts separately as well as in a combined meta-analysis. We separated the discovery and replication cohorts to evaluate any evidence for bias produced by selecting the variants for the allele score and testing the association of the allele score with incident CHD and MI both in the discovery cohort. As the replication cohort was not used to select variants for the allele score it would not suffer from such bias.
We performed sensitivity analyses using four pleiotropy robust methods each of which uses a different approach to partially relax the no horizontal pleiotropy assumption of MR analyses:[63] MR-Egger [58], MR mode based estimate (MBE),[64] MR PRESSO [65], and Weighted median [66]. For these sensitivity analyses, we used the prevalent CHD and MI GWAS results from CARDIoGRAMplusC4D consortium [34] as it had a larger sample size (60,801 prevalent cases and 123,504 controls) and these methods often have lower power to detect effects. For estimates of variant effects on fibrinogen we used fixed-effects meta-analysis estimates from the 11 cohorts in these analyses. Since an individual cannot be both a prevalent and incident CHD or MI case at the same sampling, there was no overlap amongst the cases between our incident analyses and the prevalent cases used in the CARDIoGRAMplusC4D GWAS. There would still be some overlap amongst the non-cases/controls which could bias estimates towards the null.
We also examined whether fibrinogen showed evidence for a causal effect on 7 metabolic CHD risk factors using MR-base (www.mrbase.org), a database of published GWAS available for MR [67]. We focused on metabolic CHD risk factors as initial results indicated that body mass index was the trait with which our allele score showed the strongest evidence for pleiotropy—potentially horizontal (i.e. SNPs affecting fibrinogen and CHD via independent pathways) and vertical (i.e. fibrinogen-associated SNPs also associated with risk factors downstream of fibrinogen) as the associations did not distinguish between the two. The CHD risk factors were body mass index [68], waist circumference [69], waist-to-hip ratio [69], low-density lipoprotein cholesterol [70], triglycerides [70], homeostatic model assessment insulin resistance (HOMA-IR) [71], and Type 2 diabetes [72]. As MR-base only contains published GWAS we used the most recently published GWAS for fibrinogen for our variant-fibrinogen associations [1] but limited to those variants present in our allele score. For the CHD risk factors we compared causal effect estimates obtained from the inverse variance weighted method (which assumes no unbalanced horizontal pleiotropy), to those from the pleiotropy robust MR Egger, and Weighted median methods. All three methodologies were implemented in MR-base.
Statistical analyses were performed in R [73]. Meta-analyses were performed using the R package metaphor [74]. Cox models were estimated via the coxph function in the R package survival [75] with the exception of SHIP where the survreg function was used with an exponential distribution to account for the interval censored data. MR-Egger and weighted median results were performed using the R package MendelianRandomization and Two Sample MR [67]. MR MBE analyses were performed using the methods given by Hartwig et al [64]. The default bandwidth (φ = 1) was used for MR MBE as results did not show sensitivity to the choice of bandwidth. MR PRESSO analyses were performed using code available at the MR PRESSO GitHub repository (https://github.com/rondolab/MR-PRESSO) [65]. We used the robust MR estimates from MR PRESSO which are equivalent to performing an inverse-variance weighted MR analysis after removing outlying variants, which may be influenced by horizontal pleiotropy, as identified by MR PRESSO. Results from the 2SC model are reported in terms of the hazard ratio (HR), while all results that utilize the prevalent disease GWAS are reported in terms of the odds ratio (OR). All HR and OR are given per 1 g/L higher fibrinogen. All confidence intervals (CI) reported are 95% CI.
Supporting information
S1 File. Supplemental Methods and Tables v2.
File containing the Supplemental Online Methods (including cohort specific information) as well as the Supplemental Tables (A-E). Table A. Association between allele score (AS) and fibrinogen as well as rs1800790 (FGB -455G>A) and fibrinogen. Table contains association between AS and fibrinogen as well as the association between FGB variant rs1800790 with fibrinogen. Also given is the association between the AS and fibrinogen after rescaling (RS) the AS in each cohort so that a unit of 1 represented 50% of the range. This transformation allows the allele score associations to be more directly compared with those for rs1800790. Rs1800790 was unavailable in SHIP. The meta-analysis represents a fixed effects meta-analysis Table B. Comparison of allele score and rs1800790 as genetic instruments for CHD. Comparison of the allele score vs rs1800790 as a genetic instrument for CHD. 2SC = two-stage Cox (2SC) model; CI = 95% confidence interval; HR = Hazard Ratio; MR = Mendelian Randomization; OR = Odds Ratio; Q = Cochran’s Q; SE = standard error. Table C. Estimates of causal effect of fibrinogen on metabolic CHD risk factors. As body mass index was the coronary heart disease (CHD) trait with the most evidence for pleiotropic effects from our allele score we used 3 Mendelian Randomization (MR) methods from as implemented in MR-base (www.mrbase.org)[4]. MR-base uses published genome-wide association studies (GWAS) to perform 2-sample MR. For the genetic variant-fibrinogen associations we used the most recently published fibrinogen GWAS.[5] References and sample sizes for each of the CHD risk factor outcomes appear in the table. Only variants that were a part of the allele score were used in the MR analyses, with each GWAS having between 23 and 32 of the 38 variants represented. HOMA-IR = homeostatic model assessment insulin resistance; LDL = low-density lipoprotein, OR = odds ratio; SD = standard deviation; SE = standard error; SNP = single nucleotide polymorphism. Table D. Allele Score Variants. Variants which composed the allele score along with their availability in each cohort as well as selection from either the 1000 Genomes imputation GWAS by de Vries et al,[1] the meta-analysis by Sabater-Lleal et al,[2] or the rare and low-frequency variant meta-analysis by Huffman et al,[3] and the published direction of association. NA indicates that the variant was not available. All variants were available for all discovery cohorts by design, thus only the replication cohorts (MESA, WGHS, LURIC, and SHIP) are listed where one or more variants may have been missing. Closest Gene = gene annotation based on location within a gene or the closest gene for intergenic variants; Locus = genomic location; Published Direction of Association = direction of association for the variant in the given study; Study = published study variant was taken from. Table E. Association of allele score with CHD risk factors. Fixed effects meta-analysis for association between fibrinogen and CHD risk factors. Fixed effects meta-analysis was used for all associations despite some evidence for heterogeneity for HDL P(Q) < 0.05. The association was still not significant in the random effects meta-analysis. Hypertension and Type II Diabetes were binary variables. Smoking was a categorical outcome for current, former, or never smokers. Beta = meta-analysis effect estimate; SE = meta-analysis standard error; P = meta-analysis P; Q = Cochran's Q; P(Q) = P-value associated with Cochran's Q.
https://doi.org/10.1371/journal.pone.0216222.s001
(DOC)
S2 File. PLoS One Fib MR PRISMA 2009 checklist.
https://doi.org/10.1371/journal.pone.0216222.s002
(DOC)
References
- 1. de Vries PS, Chasman DI, Sabater-Lleal M, Chen MH, Huffman JE, Steri M, et al. A meta-analysis of 120 246 individuals identifies 18 new loci for fibrinogen concentration. Hum Mol Genet. 2016;25(2):358–70. Epub 2015/11/13. pmid:26561523; PubMed Central PMCID: PMCPMC4715256.
- 2. Huffman JE, de Vries PS, Morrison AC, Sabater-Lleal M, Kacprowski T, Auer PL, et al. Rare and low-frequency variants and their association with plasma levels of fibrinogen, FVII, FVIII, and vWF. Blood. 2015. pmid:26105150
- 3. Sabater-Lleal M, Huang J, Chasman D, Naitza S, Dehghan A, Johnson AD, et al. Multiethnic meta-analysis of genome-wide association studies in >100 000 subjects identifies 23 fibrinogen-associated Loci but no strong evidence of a causal association between circulating fibrinogen and cardiovascular disease. Circulation. 2013;128(12):1310–24. Epub 2013/08/24. pmid:23969696; PubMed Central PMCID: PMCPMC3842025.
- 4. Mikhailidis DP, Barradas MA, Maris A, Jeremy JY, Dandona P. Fibrinogen mediated activation of platelet aggregation and thromboxane A2 release: pathological implications in vascular disease. Journal of clinical pathology. 1985;38(10):1166–71. Epub 1985/10/01. pmid:3902901; PubMed Central PMCID: PMCPmc499462.
- 5. Davalos D, Akassoglou K. Fibrinogen as a key regulator of inflammation in disease. Seminars in Immunopathology. 2012;34(1):43–62. pmid:22037947
- 6. Danesh J, Collins R, Appleby P, Peto R. Association of fibrinogen, c-reactive protein, albumin, or leukocyte count with coronary heart disease: Meta-analyses of prospective studies. JAMA. 1998;279(18):1477–82. pmid:9600484
- 7. de Maat MP, Pietersma A, Kofflard M, Sluiter W, Kluft C. Association of plasma fibrinogen levels with coronary artery disease, smoking and inflammatory markers. Atherosclerosis. 1996;121(2):185–91. pmid:9125293
- 8. Behague I, Poirier O, Nicaud V, Evans A, Arveiler D, Luc G, et al. β Fibrinogen Gene Polymorphisms Are Associated With Plasma Fibrinogen and Coronary Artery Disease in Patients With Myocardial Infarction. The ECTIM Study. 1996;93(3):440–9.
- 9. Maresca G, Di Blasio A, Marchioli R, Di Minno G. Measuring Plasma Fibrinogen to Predict Stroke and Myocardial Infarction: An Update. Arteriosclerosis, Thrombosis, and Vascular Biology. 1999;19(6):1368–77. pmid:10364066
- 10. Ma J, Hennekens CH, Ridker PM, Stampfer MJ. A prospective study of fibrinogen and risk of myocardial infarction in the physicians’ health study. Journal of the American College of Cardiology. 1999;33(5):1347–52. pmid:10193737
- 11. Rothwell PM, Howard SC, Power DA, Gutnikov SA, Algra A, van Gijn J, et al. Fibrinogen Concentration and Risk of Ischemic Stroke and Acute Coronary Events in 5113 Patients With Transient Ischemic Attack and Minor Ischemic Stroke. Stroke. 2004;35(10):2300–5. pmid:15345800
- 12. Chuang S-Y, Bai C-H, Chen W-H, Lien L-M, Pan W-H. Fibrinogen Independently Predicts the Development of Ischemic Stroke in a Taiwanese Population: CVDFACTS Study. Stroke. 2009;40(5):1578–84. pmid:19286580
- 13. Singh K, Bønaa KH, Jacobsen BK, Bjørk L, Solberg S. Prevalence of and Risk Factors for Abdominal Aortic Aneurysms in a Population-based Study: The Tromsø Study. American Journal of Epidemiology. 2001;154(3):236–44. pmid:11479188
- 14. Al-Barjas HS, Ariëns R, Grant P, Scott JA. Raised Plasma Fibrinogen Concentration in Patients With Abdominal Aortic Aneurysm. Angiology. 2006;57(5):607–14. pmid:17067984
- 15. Davey Smith G, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?*. International Journal of Epidemiology. 2003;32(1):1–22. pmid:12689998
- 16. Keavney B, Danesh J, Parish S, Palmer A, Clark S, Youngman L, et al. Fibrinogen and coronary heart disease: test of causality by ‘Mendelian randomization’. International Journal of Epidemiology. 2006;35(4):935–43. pmid:16870675
- 17. Davey Smith G, Harbord R, Ebrahim S. Fibrinogen, C-reactive protein and coronary heart disease: does Mendelian randomization suggest the associations are non-causal? Qjm. 2004;97(3):163–6. pmid:14976273
- 18. Davey Smith G, Harbord R, Milton J, Ebrahim S, Sterne JA. Does elevated plasma fibrinogen increase the risk of coronary heart disease? Evidence from a meta-analysis of genetic association studies. Arteriosclerosis, thrombosis, and vascular biology. 2005;25(10):2228–33. pmid:16123313
- 19. Sabater-Lleal M, Huang J, Chasman DI, Naitza S, Dehghan A, Johnson AD, et al. A Multi-Ethnic Meta-Analysis of Genome-Wide Association Studies in over 100,000 subjects identifies 23 fibrinogen-associated loci but no strong evidence of a causal association between circulating fibrinogen and cardiovascular disease. Circulation. 2013:CIRCULATIONAHA. 113.002251.
- 20. Bennett JS. Platelet-fibrinogen interactions. Annals of the New York Academy of Sciences. 2001;936:340–54. Epub 2001/07/20. pmid:11460491.
- 21. Ross R. Atherosclerosis—an inflammatory disease. The New England journal of medicine. 1999;340(2):115–26. Epub 1999/01/14. pmid:9887164.
- 22. Pamukcu B, Lip GY, Shantsila E. The nuclear factor—kappa B pathway in atherosclerosis: a potential therapeutic target for atherothrombotic vascular disease. Thrombosis research. 2011;128(2):117–23. Epub 2011/06/04. pmid:21636112.
- 23. Kamath S, Lip GYH. Fibrinogen: biochemistry, epidemiology and determinants. QJM. 2003;96(10):711–29. pmid:14500858
- 24. Tybjaerg-Hansen A, Agerholm-Larsen B, Humphries SE, Abildgaard S, Schnohr P, Nordestgaard BG. A common mutation (G-455—> A) in the beta-fibrinogen promoter is an independent predictor of plasma fibrinogen, but not of ischemic heart disease. A study of 9,127 individuals based on the Copenhagen City Heart Study. Journal of Clinical Investigation. 1997;99(12):3034–9. PMC508156. pmid:9185528
- 25. Leander K, Wiman B, Hallqvist J, Falk G, De Faire U. The G‐455A polymorphism of the fibrinogen BΒ‐gene relates to plasma fibrinogen in male cases, but does not interact with environmental factors in causing myocardial infarction in either men or women. Journal of internal medicine. 2002;252(4):332–41. pmid:12366606
- 26. Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Human Molecular Genetics. 2014;23(R1):R89–R98. pmid:25064373
- 27. Neyman J. Statistics—Servant of All Science. Science. 1955;122(3166):401–6. pmid:13246647
- 28. Hill G, Connelly J, Hebert R, Lindsay J, Millar W. Neyman's bias re-visited. J Clin Epidemiol. 2003;56(4):293–6. Epub 2003/05/28. pmid:12767404.
- 29. Delgado-Rodríguez M, Llorca J. Bias. Journal of Epidemiology and Community Health. 2004;58(8):635–41. pmid:15252064
- 30. Paternoster L, Tilling K, Davey Smith G. Genetic epidemiology and Mendelian randomization for informing disease therapeutics: Conceptual and methodological challenges. PLOS Genetics. 2017;13(10):e1006944. pmid:28981501
- 31. O’Donnell CJ. Mendelian randomization evidence for cardiovascular precision medicine. JAMA Cardiology. 2018;3(7):627–8. pmid:29926078
- 32. Holmes MV, Ala-Korpela M, Smith GD. Mendelian randomization in cardiometabolic disease: challenges in evaluating causality. Nature Reviews Cardiology. 2017;14:577. pmid:28569269
- 33. Burgess S, Davies NM, Thompson SG. Bias due to participant overlap in two‐sample Mendelian randomization. Genetic Epidemiology. 2016;40(7):597–608. pmid:27625185
- 34. The CARDIoGRAMplusC4D Consortium. A comprehensive 1000 Genomes-based genome-wide association meta-analysis of coronary artery disease. Nat Genet. 2015;47(10):1121–30. http://www.nature.com/ng/journal/v47/n10/abs/ng.3396.html#supplementary-information. pmid:26343387
- 35. Psaty BM, O'Donnell CJ, Gudnason V, Lunetta KL, Folsom AR, Rotter JI, et al. Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium: Design of prospective meta-analyses of genome-wide association studies from 5 cohorts. Circulation Cardiovascular genetics. 2009;2(1):73–80. Epub 2009/12/25. pmid:20031568; PubMed Central PMCID: PMCPmc2875693.
- 36. Clauss A. Gerinnungsphysiologische schnellmethode zur bestimmung des fibrinogens. Acta haematologica. 1957;17(4):237–46. pmid:13434757
- 37. Rumley A, Woodward M, Hoffmeister A, Koenig W, Lowe GD. Comparison of plasma fibrinogen by Clauss, prothrombin time-derived, and immunonephelometric assays in a general population: implications for risk stratification by thirds of fibrinogen. Blood coagulation & fibrinolysis: an international journal in haemostasis and thrombosis. 2003;14(2):197–201. Epub 2003/03/13. pmid:12632032.
- 38. Magnani B, Brugnara C, Lapp C, Fenton T. Degree of Agreement in Plasma Fibrinogen Among Two Functional and One Immunonephelometric Assays. American Journal of Clinical Pathology. 1997;107(5):527–33. pmid:9128264
- 39. Ward-Caviness CK, Huffman JE, Everett K, Germain M, van Dongen J, Hill WD, et al. DNA methylation age is associated with an altered hemostatic profile in a multiethnic meta-analysis. Blood. 2018;132(17):1842–50. pmid:30042098
- 40. Peters SAE, Woodward M, Rumley A, Koenig W, Tunstall-Pedoe H, Lowe GDO. Direct comparisons of three alternative plasma fibrinogen assays with the von Clauss assay in prediction of cardiovascular disease and all-causes mortality: the Scottish Heart Health Extended Cohort. British Journal of Haematology. 2013;162(3):392–9. pmid:23701042
- 41. De Vries PS, Chasman DI, Sabater-Lleal M, Chen M-H, Huffman JE, Steri M, et al. A meta-analysis of 120,246 individuals identifies 18 new loci for fibrinogen concentration. Human molecular genetics. 2015:ddv454.
- 42. Frazer KA, Ballinger DG, Cox DR, Hinds DA, Stuve LL, Gibbs RA, et al. A second generation human haplotype map of over 3.1 million SNPs. Nature. 2007;449(7164):851–61. pmid:17943122
- 43. Genomes Project Consortium. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56–65. pmid:23128226
- 44. Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: Using Sequence and Genotype Data to Estimate Haplotypes and Unobserved Genotypes. Genetic epidemiology. 2010;34(8):816–34. PMC3175618. pmid:21058334
- 45. Howie BN, Donnelly P, Marchini J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies. PLoS Genet. 2009;5(6):e1000529. pmid:19543373
- 46. Burgess S, Thompson SG. Use of allele scores as instrumental variables for Mendelian randomization. International Journal of Epidemiology. 2013;42(4):1134–44. pmid:24062299
- 47. Lettre G, Palmer CD, Young T, Ejebe KG, Allayee H, Benjamin EJ, et al. Genome-wide association study of coronary heart disease and its risk factors in 8,090 African Americans: the NHLBI CARe Project. PLoS Genet. 2011;7(2):e1001300. pmid:21347282
- 48. Carty CL, Buzkova P, Fornage M, Franceschini N, Cole S, Heiss G, et al. Associations between incident ischemic stroke events and stroke and cardiovascular disease-related genome-wide association studies single nucleotide polymorphisms in the Population Architecture Using Genomics and Epidemiology study. Circulation Cardiovascular genetics. 2012;5(2):210–6. Epub 2012/03/10. pmid:22403240; PubMed Central PMCID: PMCPmc3402178.
- 49. Chan K, Patel RS, Newcombe P, Nelson CP, Qasim A, Epstein SE, et al. Association between the chromosome 9p21 locus and angiographic coronary artery disease burden: a collaborative meta-analysis. J Am Coll Cardiol. 2013;61(9):957–70. Epub 2013/01/29. pmid:23352782; PubMed Central PMCID: PMCPmc3653306.
- 50. Cheng YC, Anderson CD, Bione S, Keene K, Maguire JM, Nalls M, et al. Are myocardial infarction—associated single-nucleotide polymorphisms associated with ischemic stroke? Stroke. 2012;43(4):980–6. Epub 2012/03/01. pmid:22363065; PubMed Central PMCID: PMCPmc3622211.
- 51. Dichgans M, Malik R, Konig IR, Rosand J, Clarke R, Gretarsdottir S, et al. Shared genetic susceptibility to ischemic stroke and coronary artery disease: a genome-wide analysis of common variants. Stroke. 2014;45(1):24–36. Epub 2013/11/23. pmid:24262325; PubMed Central PMCID: PMCPmc4112102.
- 52. Lieb W, Jansen H, Loley C, Pencina MJ, Nelson CP, Newton-Cheh C, et al. Genetic predisposition to higher blood pressure increases coronary artery disease risk. Hypertension (Dallas, Tex: 1979). 2013;61(5):995–1001. Epub 2013/03/13. pmid:23478099; PubMed Central PMCID: PMCPmc3855241.
- 53. Patel RS, Ye S. Genetic determinants of coronary heart disease: new discoveries and insights from genome-wide association studies. Heart (British Cardiac Society). 2011;97(18):1463–73. Epub 2011/07/28. pmid:21791514.
- 54. Sayols-Baixeras S, Lluís-Ganella C, Lucas G, Elosua R. Pathogenesis of coronary artery disease: focus on genetic risk factors and identification of genetic variants. The Application of Clinical Genetics. 2014;7:15–32. PMC3920464. pmid:24520200
- 55. Zhang X, Johnson AD, Hendricks AE, Hwang SJ, Tanriverdi K, Ganesh SK, et al. Genetic associations with expression for genes implicated in GWAS studies for atherosclerotic cardiovascular disease and blood phenotypes. Hum Mol Genet. 2014;23(3):782–95. Epub 2013/09/24. pmid:24057673; PubMed Central PMCID: PMCPmc3900869.
- 56. Roberts R, Stewart AF. Genes and coronary artery disease: where are we? J Am Coll Cardiol. 2012;60(18):1715–21. Epub 2012/10/09. pmid:23040572.
- 57. Johnson AD, Handsaker RE, Pulit SL, Nizzari MM, O'Donnell CJ, de Bakker PI. SNAP: a web-based tool for identification and annotation of proxy SNPs using HapMap. Bioinformatics (Oxford, England). 2008;24(24):2938–9. Epub 2008/11/01. pmid:18974171; PubMed Central PMCID: PMCPmc2720775.
- 58. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. International Journal of Epidemiology. 2015;44(2):512–25. pmid:26050253
- 59. Burgess S. Identifying the odds ratio estimated by a two-stage instrumental variable analysis with a logistic regression model. Statistics in medicine. 2013;32(27):4726–47. Epub 2013/06/05. pmid:23733419; PubMed Central PMCID: PMCPmc3935453.
- 60. Didelez V, Meng S, Sheehan NA. Assumptions of IV methods for observational epidemiology. Statistical Science. 2010:22–40.
- 61. Dixon SC, Nagle CM, Thrift AP, Pharoah PD, Pearce CL, Zheng W, et al. Adult body mass index and risk of ovarian cancer by subtype: a Mendelian randomization study. International Journal of Epidemiology. 2016;45(3):884–95. pmid:27401727
- 62. Cochran WG. The Comparison of Percentages in Matched Samples. Biometrika. 1950;37(3/4):256–66.
- 63. Hemani G, Bowden J, Davey Smith G. Evaluating the potential role of pleiotropy in Mendelian randomization studies. Human Molecular Genetics. 2018;27(R2):R195–R208. pmid:29771313
- 64. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. International Journal of Epidemiology. 2017;46(6):1985–98. pmid:29040600
- 65. Verbanck M, Chen C-Y, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nature Genetics. 2018;50(5):693–8. pmid:29686387
- 66. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent Estimation in Mendelian Randomization with Some Invalid Instruments Using a Weighted Median Estimator. Genetic Epidemiology. 2016;40(4):304–14. pmid:27061298
- 67. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife. 2018;7:e34408. pmid:29846171
- 68. Locke AE, Kahali B, Berndt SI, Justice AE, Pers TH, Day FR, et al. Genetic studies of body mass index yield new insights for obesity biology. Nature. 2015;518:197. https://www.nature.com/articles/nature14177#supplementary-information. pmid:25673413
- 69. Shungin D, Winkler TW, Croteau-Chonka DC, Ferreira T, Locke AE, Mägi R, et al. New genetic loci link adipose and insulin biology to body fat distribution. Nature. 2015;518:187. https://www.nature.com/articles/nature14132#supplementary-information. pmid:25673412
- 70. Global Lipids Genetics Consortium. Discovery and refinement of loci associated with lipid levels. Nature Genetics. 2013;45:1274. https://www.nature.com/articles/ng.2797#supplementary-information. pmid:24097068
- 71. Dupuis J, Langenberg C, Prokopenko I, Saxena R, Soranzo N, Jackson AU, et al. New genetic loci implicated in fasting glucose homeostasis and their impact on type 2 diabetes risk. Nature Genetics. 2010;42:105. https://www.nature.com/articles/ng.520#supplementary-information. pmid:20081858
- 72. Wood AR, Tyrrell J, Beaumont R, Jones SE, Tuke MA, Ruth KS, et al. Variants in the FTO and CDKAL1 loci have recessive effects on risk of obesity and type 2 diabetes, respectively. Diabetologia. 2016;59(6):1214–21. pmid:26961502
- 73.
R Core Team. R: Language and Environment for Statistical Computing. In: Computing RFfS, editor. Vienna, Austria: R Foundation for Statistical Computing; 2015.
- 74. Viechtbauer W. Conducting meta-analyses in R with the metafor package. J Stat Softw. 2010;36(3):1–48.
- 75. Therneau TM, Lumley T. Package ‘survival’. Verze; 2016.