Probabilistic modelling of pharmaceutical pollution risk from sewage treatment work discharges using a Bayesian Network: application to a Scottish river catchment

Mads Troldborg, Miriam Glendell, Zisis Gagkas, Kerr J. Adams, Camilla Negri , Philip Taylor, Zulin Zhang, Pat Cooper, Alison Brown, Linda May, Ana Corrochano- Fraile, Lindsay Beevers, Andrew Tyler

Pharmaceuticals are increasingly recognised as a class of emerging contaminants of concern in rivers. Their continuous release from human use and variable removal in sewage treatment works (STWs) can produce ecologically relevant concentrations and contribute to antimicrobial resistance. We developed a probabilistic catchment-scale model based on a Bayesian Network (BN) to quantify pharmaceutical concentrations and the probability of exceeding predicted no-effect concentrations at a monthly time step. The BN embeds a stochastic mass-balance linking monthly prescribing rates, excretion fractions, STW removal efficiencies and river discharge to produce posterior distributions of concentrations for 16 pharmaceuticals at 20 monitoring points in a Scottish catchment. Model inputs were derived from Scotland’s National Health Service prescribing records, a literature compilation of excretion and removal data, and a calibrated hydrological model of the catchment. Simulated concentration distributions generally agreed with observations made at the 20 locations over a 16-month period and were typically within one order of magnitude for most compounds, indicating satisfactory performance. Highest exceedance probabilities were predicted for azithromycin, diclofenac, ibuprofen and clarithromycin, particularly at heavily impacted sites and during low-flow summer months. Scenario analyses show that future drier summers (UKCP18 RCP8.5) increase exceedance probabilities, and that substantial reductions in prescribing or markedly improved STW removal efficiencies are needed to reduce risks for high-impact compounds. The BN framework transparently captures uncertainty, supports diagnostic inference (to prioritise interventions) and is readily extensible to include additional sources (e.g., Combined Sewer Overflows or septic tanks) and mixture risk assessment.