Violation of the Constant Genetic Effect Assumption Can Result in Biased Estimates for Non-Linear Mendelian Randomization

Serum vitamin D, blood pressure and hypertension risk in the HUNT study using observational and Mendelian randomization approaches

Limited studies have triangulated the relationship between serum 25-hydroxyvitamin D [25(OH)D] levels and systolic blood pressure (SBP), diastolic blood pressure (DBP) or hypertension risk utilizing both observational and Mendelian randomization (MR) approaches. We employed data from the Norwegian Trøndelag Health Study (HUNT) to conduct cross-sectional (n = 5854) and prospective (n = 3592) analyses, as well as one-sample MR (n = 86,324). We also used largest publicly available data for two-sample MR. Our cross-sectional analyses showed a 25 nmol/L increase in 25(OH)D was associated with a 1.73 mmHg decrease in SBP (95% CI - 2.46 to - 1.01), a 0.91 mmHg decrease in DBP (95% CI - 1.35 to - 0.47) and 19% lower prevalence of hypertension (OR 0.81, 95% CI 0.74 to 0.90) after adjusting for important confounders. However, these associations disappeared in prospective analyses. One-sample and two-sample MR results further suggested no causal relationship between serum vitamin D levels and blood pressure or hypertension risk in the general population.

Non-linear Mendelian randomization: detection of biases using negative controls with a focus on BMI, Vitamin D and LDL cholesterol

Mendelian randomisation (MR) is an established technique in epidemiological investigation, using the principle of random allocation of genetic variants at conception to estimate the causal linear effect of an exposure on an outcome. Extensions to this technique include non-linear approaches that allow for differential effects of the exposure on the outcome depending on the level of the exposure. A widely used non-linear method is the residual approach, which estimates the causal effect within different strata of the non-genetically predicted exposure (i.e. the "residual" exposure). These "local" causal estimates are then used to make inferences about non-linear effects. Recent work has identified that this method can lead to estimates that are seriously biased, and a new method-the doubly-ranked method-has been introduced as a possibly more robust approach. In this paper, we perform negative control outcome analyses in the MR context. These are analyses with outcomes onto which the exposure should have no predicted causal effect. Using both methods we find clearly biased estimates in certain situations. We additionally examined a situation for which there are robust randomised controlled trial estimates of effects-that of low-density lipoprotein cholesterol (LDL-C) reduction onto myocardial infarction, where randomised trials have provided strong evidence of the shape of the relationship. The doubly-ranked method did not identify the same shape as the trial data, and for LDL-C and other lipids they generated some highly implausible findings. Therefore, we suggest there should be extensive simulation and empirical methodological examination of performance of both methods for NLMR under different conditions before further use of these methods. In the interim, use of NLMR methods needs justification, and a number of sanity checks (such as analysis of negative and positive control outcomes, sensitivity analyses excluding removal of strata at the extremes of the distribution, examination of biological plausibility and triangulation of results) should be performed.

Body mass index and all-cause mortality in HUNT and UK biobank studies: revised non-linear Mendelian randomisation analyses

OBJECTIVES

To estimate the shape of the causal relationship between body mass index (BMI) and mortality risk in a Mendelian randomisation framework.

DESIGN

Mendelian randomisation analyses of two prospective population-based cohorts.

SETTING

Individuals of European ancestries living in Norway or the UK.

PARTICIPANTS

56 150 participants from the Trøndelag Health Study (HUNT) in Norway and 366 385 participants from UK Biobank recruited by postal invitation.

OUTCOMES

All-cause mortality and cause-specific mortality (cardiovascular, cancer, non-cardiovascular non-cancer).

RESULTS

A previously published non-linear Mendelian randomisation analysis of these data using the residual stratification method suggested a J-shaped association between genetically predicted BMI and mortality outcomes with the lowest mortality risk at a BMI of around 25 kg/m2. However, the 'constant genetic effect' assumption required by this method is violated. The reanalysis of these data using the more reliable doubly-ranked stratification method provided some indication of a J-shaped relationship, but with much less certainty as there was less precision in estimates at the lower end of the BMI distribution. Evidence for a harmful effect of reducing BMI at low BMI levels was only present in some analyses, and where present, only below 20 kg/m2. A harmful effect of increasing BMI for all-cause mortality was evident above 25 kg/m2, for cardiovascular mortality above 24 kg/m2, for cancer mortality above 30 kg/m2 and for non-cardiovascular non-cancer mortality above 26 kg/m2. In UK Biobank, the association between genetically predicted BMI and mortality at high BMI levels was stronger in women than in men.

CONCLUSION

This research challenges findings from previous conventional observational epidemiology and Mendelian randomisation investigations that the lowest level of mortality risk is at a BMI level of around 25 kg/m2. Our results provide some evidence that reductions in BMI will increase mortality risk for a small proportion of the population, and clear evidence that increases in BMI will increase mortality risk for those with BMI above 25 kg/m2.

A burden of proof study on alcohol consumption and ischemic heart disease

Cohort and case-control data have suggested an association between low to moderate alcohol consumption and decreased risk of ischemic heart disease (IHD), yet results from Mendelian randomization (MR) studies designed to reduce bias have shown either no or a harmful association. Here we conducted an updated systematic review and re-evaluated existing cohort, case-control, and MR data using the burden of proof meta-analytical framework. Cohort and case-control data show low to moderate alcohol consumption is associated with decreased IHD risk - specifically, intake is inversely related to IHD and myocardial infarction morbidity in both sexes and IHD mortality in males - while pooled MR data show no association, confirming that self-reported versus genetically predicted alcohol use data yield conflicting findings about the alcohol-IHD relationship. Our results highlight the need to advance MR methodologies and emulate randomized trials using large observational databases to obtain more definitive answers to this critical public health question.

Plasma adiponectin levels and risk of heart failure, atrial fibrillation, aortic valve stenosis, and myocardial infarction: large-scale observational and Mendelian randomization evidence

AIMS

Adiponectin may play an important protective role in heart failure and associated cardiovascular diseases. We hypothesized that plasma adiponectin is associated observationally and causally, genetically with risk of heart failure, atrial fibrillation, aortic valve stenosis, and myocardial infarction.

METHODS AND RESULTS

In the Copenhagen General Population Study, we examined 30 045 individuals with plasma adiponectin measurements observationally and 96 903 individuals genetically in one-sample Mendelian randomization analyses using five genetic variants explaining 3% of the variation in plasma adiponectin. In the HERMES, UK Biobank, The Nord-Trøndelag Health Study (HUNT), deCODE, the Michigan Genomics Initiative (MGI), DiscovEHR, and the AFGen consortia, we performed two-sample Mendelian randomization analyses in up to 1 030 836 individuals using 12 genetic variants explaining 14% of the variation in plasma adiponectin.In observational analyses modelled linearly, a 1 unit log-transformed higher plasma adiponectin was associated with a hazard ratio of 1.51 (95% confidence interval: 1.37-1.66) for heart failure, 1.63 (1.50-1.78) for atrial fibrillation, 1.21 (1.03-1.41) for aortic valve stenosis, and 1.03 (0.93-1.14) for myocardial infarction; levels above the median were also associated with an increased risk of myocardial infarction, and non-linear U-shaped associations were more apparent for heart failure, aortic valve stenosis, and myocardial infarction in less-adjusted models. Corresponding genetic, causal risk ratios were 0.92 (0.65-1.29), 0.87 (0.68-1.12), 1.55 (0.87-2.76), and 0.93 (0.67-1.30) in one-sample Mendelian randomization analyses, and no significant associations were seen for non-linear one-sample Mendelian randomization analyses; corresponding causal risk ratios were 0.99 (0.89-1.09), 1.00 (0.92-1.08), 1.01 (0.79-1.28), and 0.99 (0.86-1.13) in two-sample Mendelian randomization analyses, respectively.

CONCLUSION

Observationally, elevated plasma adiponectin was associated with an increased risk of heart failure, atrial fibrillation, aortic valve stenosis, and myocardial infarction. However, genetic evidence did not support causality for these associations.

Alcohol consumption and the risk of all-cause and cause-specific mortality-a linear and nonlinear Mendelian randomization study

BACKGROUND

Many observational studies support light-to-moderate alcohol intake as potentially protective against premature death. We used a genetic approach to evaluate the linear and nonlinear relationships between alcohol consumption and mortality from different underlying causes.

METHODS

We used data from 278 093 white-British UK Biobank participants, aged 37-73 years at recruitment and with data on alcohol intake, genetic variants, and mortality. Habitual alcohol consumption was instrumented by 94 variants. Linear Mendelian randomization (MR) analyses were conducted using five complementary approaches, and nonlinear MR analyses by the doubly-ranked method.

RESULTS

There were 20 834 deaths during the follow-up (median 12.6 years). In conventional analysis, the association between alcohol consumption and mortality outcomes was 'J-shaped'. In contrast, MR analyses supported a positive linear association with premature mortality, with no evidence for curvature (Pnonlinearity ≥ 0.21 for all outcomes). The odds ratio [OR] for each standard unit increase in alcohol intake was 1.27 (95% confidence interval [CI] 1.16-1.39) for all-cause mortality, 1.30 (95% CI 1.10-1.53) for cardiovascular disease, 1.20 (95% CI 1.08-1.33) for cancer, and 2.06 (95% CI 1.36-3.12) for digestive disease mortality. These results were consistent across pleiotropy-robust methods. There was no clear evidence for an association between alcohol consumption and mortality from respiratory diseases or COVID-19 (1.32, 95% CI 0.96-1.83 and 1.46, 95% CI 0.99-2.16, respectively; Pnonlinearity ≥ 0.21).

CONCLUSION

Higher levels of genetically predicted alcohol consumption had a strong linear association with an increased risk of premature mortality with no evidence for any protective benefit at modest intake levels.

A data-adaptive method for investigating effect heterogeneity with high-dimensional covariates in Mendelian randomization

BACKGROUND

Mendelian randomization is a popular method for causal inference with observational data that uses genetic variants as instrumental variables. Similarly to a randomized trial, a standard Mendelian randomization analysis estimates the population-averaged effect of an exposure on an outcome. Dividing the population into subgroups can reveal effect heterogeneity to inform who would most benefit from intervention on the exposure. However, as covariates are measured post-"randomization", naive stratification typically induces collider bias in stratum-specific estimates.

METHOD

We extend a previously proposed stratification method (the "doubly-ranked method") to form strata based on a single covariate, and introduce a data-adaptive random forest method to calculate stratum-specific estimates that are robust to collider bias based on a high-dimensional covariate set. We also propose measures based on the Q statistic to assess heterogeneity between stratum-specific estimates (to understand whether estimates are more variable than expected due to chance alone) and variable importance (to identify the key drivers of effect heterogeneity).

RESULT

We show that the effect of body mass index (BMI) on lung function is heterogeneous, depending most strongly on hip circumference and weight. While for most individuals, the predicted effect of increasing BMI on lung function is negative, it is positive for some individuals and strongly negative for others.

CONCLUSION

Our data-adaptive approach allows for the exploration of effect heterogeneity in the relationship between an exposure and an outcome within a Mendelian randomization framework. This can yield valuable insights into disease aetiology and help identify specific groups of individuals who would derive the greatest benefit from targeted interventions on the exposure.

Vitamin D and human health: evidence from Mendelian randomization studies

We summarized the current evidence on vitamin D and major health outcomes from Mendelian randomization (MR) studies. PubMed and Embase were searched for original MR studies on vitamin D in relation to any health outcome from inception to September 1, 2022. Nonlinear MR findings were excluded due to concerns about the validity of the statistical methods used. A meta-analysis was preformed to synthesize study-specific estimates after excluding overlapping samples, where applicable. The methodological quality of the included studies was evaluated according to the STROBE-MR checklist. A total of 133 MR publications were eligible for inclusion in the analyses. The causal association between vitamin D status and 275 individual outcomes was examined. Linear MR analyses showed genetically high 25-hydroxyvitamin D (25(OH)D) concentrations were associated with reduced risk of multiple sclerosis incidence and relapse, non-infectious uveitis and scleritis, psoriasis, femur fracture, leg fracture, amyotrophic lateral sclerosis, anorexia nervosa, delirium, heart failure, ovarian cancer, non-alcoholic fatty liver disease, dyslipidemia, and bacterial pneumonia, but increased risk of Behçet's disease, Graves' disease, kidney stone disease, fracture of radium/ulna, basal cell carcinoma, and overall cataracts. Stratified analyses showed that the inverse association between genetically predisposed 25(OH)D concentrations and multiple sclerosis risk was significant and consistent regardless of the genetic instruments GIs selected. However, the associations with most of the other outcomes were only pronounced when using genetic variants not limited to those in the vitamin D pathway as GIs. The methodological quality of the included MR studies was substantially heterogeneous. Current evidence from linear MR studies strongly supports a causal role of vitamin D in the development of multiple sclerosis. Suggestive support for a number of other health conditions could help prioritize conditions where vitamin D may be beneficial or harmful.

Estimating dose-response relationships for vitamin D with coronary heart disease, stroke, and all-cause mortality: observational and Mendelian randomisation analyses

BACKGROUND

Randomised trials of vitamin D supplementation for cardiovascular disease and all-cause mortality have generally reported null findings. However, generalisability of results to individuals with low vitamin D status is unclear. We aimed to characterise dose-response relationships between 25-hydroxyvitamin D (25[OH]D) concentrations and risk of coronary heart disease, stroke, and all-cause mortality in observational and Mendelian randomisation frameworks.

METHODS

Observational analyses were undertaken using data from 33 prospective studies comprising 500 962 individuals with no known history of coronary heart disease or stroke at baseline. Mendelian randomisation analyses were performed in four population-based cohort studies (UK Biobank, EPIC-CVD, and two Copenhagen population-based studies) comprising 386 406 middle-aged individuals of European ancestries, including 33 546 people who developed coronary heart disease, 18 166 people who had a stroke, and 27 885 people who died. Primary outcomes were coronary heart disease, defined as fatal ischaemic heart disease (International Classification of Diseases 10th revision code I20-I25) or non-fatal myocardial infarction (I21-I23); stroke, defined as any cerebrovascular disease (I60-I69); and all-cause mortality.

FINDINGS

Observational analyses suggested inverse associations between incident coronary heart disease, stroke, and all-cause mortality outcomes with 25(OH)D concentration at low 25(OH)D concentrations. In population-wide genetic analyses, there were no associations of genetically predicted 25(OH)D with coronary heart disease (odds ratio [OR] per 10 nmol/L higher genetically-predicted 25(OH)D concentration 0·98, 95% CI 0·95-1·01), stroke (1·01, [0·97-1·05]), or all-cause mortality (0·99, 0·95-1·02). Null findings were also observed in genetic analyses for cause-specific mortality outcomes, and in stratified genetic analyses for all outcomes at all observed levels of 25(OH)D concentrations.

INTERPRETATION

Stratified Mendelian randomisation analyses suggest a lack of causal relationship for 25(OH)D concentrations with both cardiovascular and mortality outcomes for individuals at all levels of 25(OH)D. Our findings suggest that substantial reductions in mortality and cardiovascular morbidity due to long-term low-dose vitamin D supplementation are unlikely even if targeted at individuals with low vitamin D status.

FUNDING

British Heart Foundation, Medical Research Council, National Institute for Health Research, Health Data Research UK, Cancer Research UK, and International Agency for Research on Cancer.

Mendelian Randomization as a Tool for Cardiovascular Research: A Review

Importance

Mendelian randomization (MR) is a statistical approach that has become increasingly popular in the field of cardiovascular disease research. It offers a way to infer potentially causal relationships between risk factors and outcomes using observational data, which is particularly important in cases where randomized clinical trials are not feasible or ethical. With the growing availability of large genetic data sets, MR has become a powerful and accessible tool for studying the risk factors for cardiovascular disease.

Observations

MR uses genetic variation associated with modifiable exposures or risk factors to mitigate biases that affect traditional observational study designs. The approach uses genetic variants that are randomly assigned at conception as proxies for exposure to a risk factor, mimicking a randomized clinical trial. By comparing the outcomes of individuals with different genetic variants, researchers may draw causal inferences about the effects of specific risk factors on cardiovascular disease, provided assumptions are met that address (1) the association between each genetic variant and risk factor and (2) the association of the genetic variants with confounders and (3) that the association between each genetic variant and the outcome only occurs through the risk factor. Like other observational designs, MR has limitations, which include weak instruments that are not strongly associated with the exposure of interest, linkage disequilibrium where genetic instruments influence the outcome via correlated rather than direct effects, overestimated genetic associations, and selection and survival biases. In addition, many genetic databases and MR studies primarily include populations genetically similar to European reference populations; improved diversity of participants in these databases and studies is critically needed.

Conclusions and Relevance

This review provides an overview of MR methodology, including assumptions, strengths, and limitations. Several important applications of MR in cardiovascular disease research are highlighted, including the identification of drug targets, evaluation of potential cardiovascular risk factors, as well as emerging methodology. Overall, while MR alone can never prove a causal relationship beyond reasonable doubt, MR offers a rigorous approach for investigating possible causal relationships in observational data and has the potential to transform our understanding of the etiology and treatment of cardiovascular disease.

Neutrophil counts and cardiovascular disease

BACKGROUND AND AIMS

Anti-inflammatory trials have shown considerable benefits for cardiovascular disease. High neutrophil counts, an easily accessible inflammation biomarker, are associated with atherosclerosis in experimental studies. This study aimed to investigate the associations between neutrophil counts and risk of nine cardiovascular endpoints using observational and genetic approaches.

METHODS

Observational studies were conducted in the Copenhagen General Population Study (n = 101 730). Genetic studies were firstly performed using one-sample Mendelian randomization (MR) with individual-level data from the UK Biobank (n = 365 913); secondly, two-sample MR analyses were performed using summary-level data from the Blood Cell Consortium (n = 563 085). Outcomes included ischaemic heart disease, myocardial infarction, peripheral arterial disease, ischaemic cerebrovascular disease, ischaemic stroke, vascular-related dementia, vascular dementia, heart failure, and atrial fibrillation.

RESULTS

Observational analyses showed associations between high neutrophil counts with high risks of all outcomes. In the UK Biobank, odds ratios (95% confidence intervals) per 1-SD higher genetically predicted neutrophil counts were 1.15 (1.08, 1.21) for ischaemic heart disease, 1.22 (1.12, 1.34) for myocardial infarction, and 1.19 (1.04, 1.36) for peripheral arterial disease; similar results were observed in men and women separately. In two-sample MR, corresponding estimates were 1.14 (1.05, 1.23) for ischaemic heart disease and 1.11 (1.02, 1.20) for myocardial infarction; multiple sensitivity analyses showed consistent results. No robust associations in two-sample MR analyses were found for other types of leucocytes.

CONCLUSIONS

Observational and genetically determined high neutrophil counts were associated with atherosclerotic cardiovascular disease, supporting that high blood neutrophil counts is a causal risk factor for atherosclerotic cardiovascular disease.

Violation of the Constant Genetic Effect Assumption Can Result in Biased Estimates for Non-Linear Mendelian Randomization

INTRODUCTION

Non-linear Mendelian randomization is an extension of conventional Mendelian randomization that performs separate instrumental variable analyses in strata of the study population with different average levels of the exposure. The approach estimates a localized average causal effect function, representing the average causal effect of the exposure on the outcome at different levels of the exposure. The commonly used residual method for dividing the population into strata works under the assumption that the effect of the genetic instrument on the exposure is linear and constant in the study population. However, this assumption may not hold in practice.

METHODS

We use the recently developed doubly ranked method to re-analyse various datasets previously analysed using the residual method. In particular, we consider a genetic score for 25-hydroxyvitamin D (25[OH]D) used in a recent non-linear Mendelian randomization analysis to assess the potential effect of vitamin D supplementation on all-cause mortality.

RESULTS

The effect of the genetic score on 25(OH)D concentrations varies strongly, with a five-fold difference in the estimated genetic association with the exposure in the lowest and highest decile groups. Evidence for a protective causal effect of vitamin D supplementation on all-cause mortality in low vitamin D individuals is evident for the residual method but not for the doubly ranked method. We show that the constant genetic effect assumption is more reasonable for some exposures and less reasonable for others. If the doubly ranked method indicates that this assumption is violated, then estimates from both the residual and doubly ranked methods can be biased, although bias was smaller on average in the doubly ranked method.

CONCLUSION

Analysts wanting to perform non-linear Mendelian randomization should compare results from both the residual and doubly ranked methods, as well as consider transforming the exposure for the residual method to reduce heterogeneity in the genetic effect on the exposure.