Detecting univariate, bivariate, and overall effects of drug mixtures using Bayesian kernel machine regression

Area of Science:

• Epidemiology
• Biostatistics
• Toxicology
• Public Health

Background:

• Current drug studies often focus on single substances, limiting understanding of polydrug use effects.
• Innovative analytic methods are crucial for examining complex drug mixtures in illicit drug users.
• Bayesian kernel machine regression (BKMR), established in environmental epidemiology, offers a novel approach for substance use research.

Purpose of the Study:

• To demonstrate the utility of the Bayesian kernel machine regression (BKMR) approach.
• To investigate the health effects of co-using opioids and non-opioids in simulated illicit drug users.
• To analyze the impact of drug substance mixtures on health outcomes.

Main Methods:

• Simulated data for 200 individuals using the Vale and Maurelli method for multivariate non-normal drug exposure.
• Modeled concentrations of xylazine, fentanyl, benzodiazepine, and nitazene.
• Employed 10,000 Markov chain Monte Carlo (MCMC) sampling iterations with diagnostics (trace plots, r-hat, effective sample size) to ensure model stability.

Main Results:

• Higher concentrations of fentanyl and nitazene were significantly associated with increased simulated health outcome levels, after controlling for age.
• BKMR model diagnostics confirmed stability and reliability across multiple Markov chains.
• Visualizations illustrated univariate, bivariate, and cumulative exposure-response relationships within the drug mixture.

Conclusions:

• The Bayesian kernel machine regression (BKMR) approach provides a powerful tool for dissecting the complex effects of drug mixtures.
• BKMR can differentiate the relative health risks posed by individual substances within a mixture.
• This method is applicable for assessing univariate, bivariate, and cumulative drug effects on health outcomes, advancing substance use research.