Mercury levels and trends in fish (2011–2021): A Bayesian approach with multi-group Gaussian processes and hierarchical imputation

Le résumé fourni par la source

Mercury (Hg) is a toxic metal, with fish consumption being the primary source of exposure in humans. This study aimed to describe Hg concentrations in fish species consumed in the Valencian Community (Spain) and their trends during the period 2011-2021. A retrospective study was conducted on Hg levels in fish meat between 2011-2021, using data from the Food Safety Monitoring Program of the Valencian Regional Government. Descriptive analyses and temporal trends were inferred for total Hg (THg) (n=799) and methylmercury (MeHg) (n=271) levels by fish species and fishery origin. Gaussian processes (GPs) with a novel multi-group covariance function were applied, enabling the use of correlations across categories to improve inference on temporal trends in unbalanced groups, with species that have smaller samples borrowing information from correlated species. Swordfish exhibited the highest Hg concentrations (median THg: 0.76 mg/kg; IQR: 0.47-1.17), with 30% of samples exceeding European limit values, followed by fresh tuna (0.46 mg/kg) and canned tuna (0.22 mg/kg). THg and MeHg levels in swordfish tended to decrease by around 0.5 mg/kg from 2011 to 2016, but then increased again to near their initial levels. Fresh and canned tuna showed decreasing trends in the first half of the study period. Data from the second half of the period were limited, except for swordfish; thus, results from this time should be interpreted with caution. Most fish groups showed declining trends between 2011-2016. Our findings on Hg levels in commercially sold fish species could be useful for guiding local fish consumption recommendations. • Swordfish showed the highest Hg levels, and 30% of samples exceeded EU limit values • Hg in swordfish decreased in 2011-2016, then rebounded to initial levels by 2021 • Fresh and canned tuna showed decreasing THg and MeHg trends between 2011-2016 • Gaussian processes are powerful non-parametric probabilistic models for temporal data • Gaussian processes with group correlations improve inference in unbalanced data

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.