This paper introduces multivariable Mendelian randomization (MvMR), an extension of the traditional Mendelian randomization approach that uses multiple genetic variants associated with several risk factors to estimate the causal effects of each risk factor on an outcome. Unlike conventional Mendelian randomization, which uses a single genetic variant associated with a single risk factor, MvMR allows for the simultaneous estimation of causal effects of multiple risk factors. The method is analogous to a factorial randomized trial, where multiple interventions are assessed simultaneously. The paper discusses the assumptions necessary for a valid MvMR analysis, including that genetic variants must not be associated with confounders and must be conditionally independent of the outcome given the risk factors. It also addresses the issue of pleiotropy, where a genetic variant may be associated with multiple risk factors. The paper presents methods for estimating causal effects using both individual-level data and summarized data, and compares the performance of these methods using real and simulated data. The paper applies MvMR to estimate the causal effects of low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and triglycerides on the risk of coronary heart disease (CHD). The results suggest that reductions in LDL-C and triglycerides are causally protective against CHD, while the causal effect of HDL-C is compatible with a null effect. The paper also discusses the limitations of MvMR, including the potential for weak instrument bias and the inability to account for unmeasured pleiotropy. The paper concludes that MvMR provides evidence of a causal effect of triglyceride-related pathways on CHD risk independent of the effects of LDL-C and HDL-C. However, the weight of evidence attributed to the findings depends on the validity of the instrumental variable assumptions. The paper also highlights the importance of careful interpretation of MvMR results, given the uncertainty of the underlying assumptions in any applied analysis.This paper introduces multivariable Mendelian randomization (MvMR), an extension of the traditional Mendelian randomization approach that uses multiple genetic variants associated with several risk factors to estimate the causal effects of each risk factor on an outcome. Unlike conventional Mendelian randomization, which uses a single genetic variant associated with a single risk factor, MvMR allows for the simultaneous estimation of causal effects of multiple risk factors. The method is analogous to a factorial randomized trial, where multiple interventions are assessed simultaneously. The paper discusses the assumptions necessary for a valid MvMR analysis, including that genetic variants must not be associated with confounders and must be conditionally independent of the outcome given the risk factors. It also addresses the issue of pleiotropy, where a genetic variant may be associated with multiple risk factors. The paper presents methods for estimating causal effects using both individual-level data and summarized data, and compares the performance of these methods using real and simulated data. The paper applies MvMR to estimate the causal effects of low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), and triglycerides on the risk of coronary heart disease (CHD). The results suggest that reductions in LDL-C and triglycerides are causally protective against CHD, while the causal effect of HDL-C is compatible with a null effect. The paper also discusses the limitations of MvMR, including the potential for weak instrument bias and the inability to account for unmeasured pleiotropy. The paper concludes that MvMR provides evidence of a causal effect of triglyceride-related pathways on CHD risk independent of the effects of LDL-C and HDL-C. However, the weight of evidence attributed to the findings depends on the validity of the instrumental variable assumptions. The paper also highlights the importance of careful interpretation of MvMR results, given the uncertainty of the underlying assumptions in any applied analysis.