Association Between Obesity and Cardiovascular Outcomes

1Coronary artery disease (CAD) and its ensuing complications, including myocardial infarction and heart failure, continue to be the leading cause of morbidity and mortality in the developed world and increasingly so in the developing world.,, Dyslipidemia has consistently been shown to be associated with atherosclerosis. The mechanistic link between dyslipidemia and atherogenesis is supported by a large number of studies demonstrating a correlation between levels of low-density lipoprotein cholesterol and major adverse cardiovascular events., Further support is provided by studies with pharmacological agents, including statins, ezetimibe, and proprotein convertase subtilisin/kexin type 9 inhibitors, which reduce low-density lipoprotein cholesterol and improve cardiovascular disease (CVD) outcomes when used for either primary or secondary prevention.,

2While the link between dyslipidemia and CVD is well established, an association between obesity and CVD remains controversial. This is important because the rising prevalence of obesity and metabolic syndrome may eventually offset the public health gains achieved by improved treatment of CAD. To complicate matters further, some investigators have proposed an “obesity paradox,” in which a higher body mass index (BMI) has been paradoxically associated with improved clinical outcomes. Observational studies are limited by bias and confounding variables, and generating randomized data is inherently challenging.

3Mendelian randomization uses genetic variants to estimate the health consequences of phenotypes influenced by these genetic variants. It is a relatively novel epidemiologic study design incorporating genetic information into standard epidemiologic methods. Mendelian randomization offers an opportunity to study associations without many of the typical biases that are inherent in traditional epidemiologic approaches. Thus, mendelian randomization can fill the evidence gaps by minimizing confounding, if variables are randomly and equally distributed in the population of interest.

4To our knowledge, no study to date has pooled data from mendelian randomization studies. Individual mendelian randomization studies have yielded conflicting findings regarding the association between obesity and cardiometabolic outcomes. In an attempt to resolve these inconsistencies, we conducted a systematic review and meta-analysis of mendelian randomized studies to assess the existence and extent of any association between obesity and CVD.

5This systematic review and meta-analysis followed the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline and the American Heart Association guideline., The need for obtaining institutional review board approval or patient informed consent was waived for this study because it is a review of publicly available data.

6MEDLINE and Scopus were searched from the inception of these databases to January 2018 by 2 independent researchers (H.R. and M.S.K.). Detailed search strategies for each database are given in eTable 1 in the Supplement. The reference lists of the retrieved articles and the relevant reviews were then screened to identify any pertinent studies.

7All articles retrieved from the systematic search were exported to EndNote reference library, version X8.1 (Clarivate Analytics), wherein duplicates were sought and removed. Two independent reviewers (H.R. and M.S.K.) assessed the remaining articles, and only those that met the predefined criteria were selected. A third investigator (T.J.S.) was consulted to resolve any discrepancies. Relevant articles were initially selected on the basis of the title and abstract, after which the full text was read to confirm relevance.

8The following 2 eligibility criteria were used to select studies. The study conducted a mendelian analysis to assess the association between any measure of obesity (BMI [calculated as weight in kilograms divided by height in meters squared] or waist to hip ratio [WHR]) and cardiometabolic outcomes, and the reported results included odds ratios (ORs) with 95% CIs, which were estimated using an instrumental variable method.

9We used both BMI and WHR adjusted for BMI (WHRadjBMI) as a measure of obesity for our systematic review. The ORs for the association between 1 SD increase in BMI and cardiometabolic outcomes were abstracted from all studies. For the present study, cardiometabolic traits included CAD, stroke, and type 2 diabetes (T2D). The following baseline and study characteristics were also extracted: sample size, mean age, number of single-nucleotide polymorphisms, measure of obesity used, database used, methods for determining BMI and WHRadjBMI, and outcome ascertainment.

10Mendelian randomization rests on 3 main assumptions. Assumption 1 is that the genotype must be associated with the phenotype (here, obesity). Assumption 2 is that the genotype should not be associated with confounders. Assumption 3 is that the genotype should affect the outcome only through the risk factor. Although the first assumption can be easily evaluated, the second and third assumptions (collectively known as absence of pleiotropy) are hard to prove, and their evaluation rests mainly on the judgment of the investigators. Although several statistical tools have been recently proposed to verify these assumptions and protect against biasing of results through pleiotropic variants, these methods are generally inconclusive and may fail to identify bias in various circumstances. There are no standardized tools to ascertain the risk of bias in mendelian randomization studies while conducting a meta-analysis. Thus, to assess the quality of the included studies, we evaluated whether the 3 assumptions of the mendelian randomization were validated, and, if so, the method used for those validations.

11Review Manager, version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) was used to perform all statistical analyses. The ORs from individual studies were pooled using a random-effects model. Forest plots were created to visually assess the results of pooling. An outcome to be analyzed and reported in our quantitative analysis (meta-analysis) required a minimum of 2 studies reporting nonoverlapping data. When 2 or more studies reported data from the same source or databank, only the study with the most participants was included in the analysis. Studies not included in the meta-analysis were still included in the qualitative analysis (systematic review). The I2 statistic was used to assess heterogeneity across studies, with a value of I2 between 25% and 50% considered mild heterogeneity, between 50% and 75% considered moderate heterogeneity, and greater than 75% considered severe heterogeneity. Visual inspection of the funnel plot was performed to evaluate publication bias. A 2-sided P <.05 was considered statistically significant in all cases.

12An initial search of the 2 databases revealed 4660 potentially relevant articles, of which 2511 remained after excluding duplicates. After applying eligibility criteria, we selected 7 articles for inclusion in the systematic review.,,,,,, Those studies with overlapping data were excluded, leaving 5 articles for inclusion in the present meta-analysis.,,,, An analysis based on WHRadjBMI was not conducted in the present study because an insufficient number of studies used this measure. The Preferred Reporting Items for Systematic Reviews and Meta-analyses flowchart shown in Figure 1 summarizes the literature search.

13The 5 studies selected for quantitative analysis consisted of 881 692 participants. All included studies used BMI as a measure of obesity. The mean (range) age of the individuals in the pooled sample was 60 (50-64) years. These studies adjusted for a mean of 47 single-nucleotide polymorphisms (range, 9-97). Study characteristics of the included studies are given in eTable 2 in the Supplement. The definition of end points varied considerably across the different studies and may have led to higher statistical heterogeneity.

14Visual inspection of the funnel plot suggested a low study bias (eFigure in the Supplement). The Table summarizes the validation of mendelian randomization assumptions by each study. Assumption 1 was validated in 4 studies, and assumptions 2 and 3 (absence of pleiotropy) were verified in 3 studies.

15The results of our meta-analysis are shown in Figure 2. Data on T2D were reported by 4 studies, with 461 871 participants. Our analysis showed a significant association between obesity and T2D (OR, 1.67; 95% CI, 1.30-2.14; P <.001; I2 = 93%). All included studies (n = 570 261 participants) reported CAD as an outcome. Obesity was significantly associated with CAD (OR, 1.20; 95% CI, 1.02-1.41; P =.03; I2 = 87%).

16Three studies with 228 816 participants provided all-cause stroke as a clinical outcome. We found no significant association between obesity and all-cause stroke (OR, 1.02; 95% CI, 0.95-1.09; P =.65; I2 = 0%). An analysis that included only 180 795 patients with ischemic stroke showed no difference in the overall result (OR, 1.02; 95 CI%, 0.92-1.14; P =.65; I2 = 1%).

17The results of the present study of nearly 1 million participants suggested an association between obesity and CAD. To our knowledge, this is the largest meta-analysis of obesity and CVD end points and the first to use mendelian randomization studies to pool findings. Each 1-SD increase in BMI increased the odds of T2D by 67% and of CAD by 20%. In light of the mendelian randomization, those increases are assumed to be independent of other traditional confounders.

18The findings from this meta-analysis fill an important evidence gap and are timely for several reasons. First, obesity and the ensuing metabolic syndrome are becoming an epidemic, and a definitive association is necessary to inform decision making. If recent secular trends continue unabated, up to 20% of the world’s adult population (1.2 billion individuals) is expected to be obese by 2030, with the prevalence of metabolic syndromes such as diabetes and CVD to increase by 54% and 22%, respectively.,, Second, despite a number of efficacious therapies, patients at high risk of CVD or with a history of CVD have substantial residual risk, and it is important to identify factors responsible for this risk. For instance, compelling evidence exists regarding the efficacy of lipid-lowering agents in both primary and secondary prevention of CVD.,, However, despite such treatments and adherence to guideline-directed medical therapy, a substantial proportion of patients continue to experience cardiovascular events, underscoring the importance of the need to identify additional, novel risk factors. Third, although our study shows an association between obesity and CVD, this may not necessarily lead to increased mortality. Obesity may be associated with improved survival in patients with established CVD, a finding that is termed an “obesity paradox.”,, However, this concept is much debated because results from studies are inconsistent.,

19Prior meta-analyses of observational studies on this subject have been limited by bias, such as that caused by smoking or confounding from prediagnostic weight loss associated with disease. Similarly, although some studies have attempted to adjust for confounding variables, such as diabetes, hypertension, and hypercholesterolemia, there has been concern regarding overadjustment. In addition, the substantial heterogeneity of the included studies (approximately 90% for CAD and diabetes) mandates that the results be interpreted with caution. This heterogeneity was anticipated, however, given the variation in study methods, participants, and localities.

20Mendelian randomization assumes that the alleles of interest are randomly and equally distributed in the population of interest. For instance, genetic loci associated with obesity may be randomly (and equally) distributed in some people and not others. Comparing events of interest between people with the alleles of interest (and the ensuing higher BMI) and those without the alleles (and hence lower BMI) should therefore provide unbiased estimates between obesity and outcomes. We also assume that this method addresses several of the limitations conferred by traditional observational studies. For instance, although multivariate regression can be used to adjust for a number of confounding variables, the effect of smoking is so strong that conventional methods cannot sufficiently account for the association of smoking with BMI or with other outcomes of interest. By using mendelian randomization, it is assumed that any potential variables of interest are equally and randomly distributed in the population of interest, similar to a randomized controlled clinical trial.

21Our analysis has several limitations. First, this is a pooled analysis of individual studies, and we did not have access to the individual patient data to conduct a patient-level analysis. Second, we studied obesity as a whole and could not perform a subgroup analysis of specific genetic mutations and their influence on the outcomes of interest. Third, although this analysis provides evidence for the harmful effects of obesity, we are unable to comment on the potential of interventions, such as lifestyle changes or pharmacological management, to attenuate any undesired effects. Fourth, we did not have an adequate number of studies to pool to assess the association of obesity with mortality (cardiovascular or all-cause). Fifth, this study assumed a linear association and therefore could not address the linearity (or lack thereof) of the underlying associations. Sixth, although mendelian randomization methods have several advantages over traditional meta-analysis methods and can provide evidence for an association between obesity and CVD, they remain modeling experiments and assumption dependent.

22Obesity was associated with an increased risk of T2D and CAD and should remain a major focus of public health initiatives. The present analysis of mendelian randomization studies was supportive of a causal association; however, it did not prove causality. Mendelian randomization assumptions were often not verified in individual studies, and this could have contributed bias to the present meta-analysis.