Probabilistic assessment of cyflumetofen dietary exposure risks by a Bayesian framework with Markov Chain Monte Carlo

Traditional probabilistic assessment methods based on @Risk and Crystal Ball focus solely on the randomness of input variables, while neglecting the uncertainty of core model parameters (e.g., α, β, σ in residue regression models). To address this limitation, this study proposes a hybrid approach integrating Bayesian Markov Chain Monte Carlo (MCMC) simulation with probabilistic modeling for quantifying the dietary risk of cyflumetofen in strawberries. A Bayesian linear regression model for pesticide residues was constructed using JASP to obtain the posterior distributions of parameters. MCMC chains of α, β, and σ were extracted in R software, and posterior predictive concentration samples incorporating both parameter and residual uncertainties were generated. These samples were imported into Crystal Ball, where dietary exposure and hazard quotient were calculated by combining strawberry consumption data (Gamma distribution) and human body weight data (normal distribution). After 100,000 Monte Carlo simulations, the 95th percentile (P95) of the hazard quotient distribution was derived and further validated by comparison with traditional methods. Results showed that traditional probabilistic assessment methods based on @Risk and Crystal Ball overestimated the dietary risk of cyflumetofen due to the omission of model parameter uncertainty. The hybrid approach proposed in this study can effectively quantify the impact of parameter uncertainty, and provide a more accurate and scientific basis for risk assessment for the safety regulation of cyflumetofen in strawberry production.