Evaluating the Impact of Marijuana Use on Metabolic Syndrome Using Data from the Continuous National Health and Nutrition Examination Survey

Background Marijuana use has increased in recent years in the US, yet much is still unknown about its impacts on health. Empirical analyses of the relationship between marijuana use and metabolic syndrome are inconclusive, with some studies suggesting marijuana is associated with improvements in weight, BMI, and insulin resistance. The linear regression models used in these studies ignore the potential endogeneity of marijuana use and other reasonable alternative model formulations.

Methods This study aims to assess the validity of various methods for examining the relationship between marijuana use and factors of metabolic syndrome using data from Continuous National Health and Nutrition Examination Survey (NHANES) from 2005 to 2010. First, ordinary least squares (OLS) models are evaluated. Next, instrumental variables methods are utilized to test and account for the potential endogeneity of marijuana use in explaining health outcomes.  Finally, alternative model specifications are employed to assess model robustness.

Results Consistent with prior research, naïve OLS models show lower fasting insulin, insulin resistance, glucose, waist circumference, and higher HDL cholesterol in past and/or current marijuana users compared to individuals who reported never having used marijuana. However, these results change when an additional category for most recent marijuana use is added; fasting insulin, insulin resistance, and HDL cholesterol were no different for recent users compared to those who had never used marijuana. In assessing the robustness of multivariate linear models, cross-validation methods suggest that the models were not consistent in comparable samples. In another check on model validity in which alcohol consumption replaces marijuana use as the risk factor of interest, effects of alcohol are more pronounced than the effects of marijuana in suggesting that health outcomes improve with heavy use. When two sexual behavior instrumental variables are added to the model, the direction, magnitude, and significance of the coefficients on marijuana use change for all regressions. The results of two-stage least squares are mostly non-significant. Durbin-Watson-Hausman tests for endogeneity of marijuana use largely fail, but small p-values are observed in regressions for some of the outcomes.

Discussion Redefining the scales for marijuana use categories alters the OLS association of marijuana use with improvements in any of the health benefits found using multivariate linear analysis, with the exception of BMI and waist circumference. When alcohol use replaces marijuana use in the original OLS model specifications, this risk factor appears to show even greater improvement in the health outcomes, in contradiction to the widely acknowledged detrimental health effects of heavy alcohol consumption.  Results of two-stage least squares estimation with instruments for marijuana use are completely inconsistent with the results of the OLS models, challenging the robustness of prior findings.