Curbing Six Sources Could Halve the Stretch of Rivers Polluted by Drug Mixtures

Researchers have used a machine-learning model to map how toxic the mix of pharmaceuticals in river water is across the whole world, in cells 2 kilometers across. Jian Chen, of Hong Kong Baptist University and Tsinghua University, and colleagues report that the mixture of 53 drugs passes the study's own exceedance mark in 13% of the river cells on that map, with hotspots in southern China, northern India, central and western Europe, and the southeastern United States.

The work was published Nov. 12, 2026, in Nature Geoscience. The authors write that drugs are now everywhere in surface water and that monitoring every river and testing every mixture is not practical, and they present the map as what they call a policy atlas for monitoring, intervention and regulation.

The exceedance mark is the study's own, and it sits well below the regulatory level it is drawn from: a cell counts as exceeding when the predicted toxicity of the mixture to an invertebrate, an alga and a bacterium reaches 5% of a regulatory threshold.

The paper reports that targeted management of six key sources could cut the river area above that mark by 54–92%, with the largest gains from livestock, aquaculture operations and wastewater treatment plants. That range comes from rerunning the model with those sources cut back, not from a program anyone has carried out. Nine drugs account for most of the predicted mixture toxicity.

The map is a prediction rather than a survey. The model was trained on 1,017 laboratory toxicity assays, and the measured drug concentrations behind it come from a monitoring study of river sites published earlier. Proximity to pollution sources and river chemistry are the direct drivers of exceedance in the analysis, with social, economic and climate factors acting indirectly, by shaping where the sources are, the water's chemistry and the terrain.

The team's Python code is public on GitHub and its machine-learning data on figshare. Funding came from the National Natural Science Foundation of China and the Kwok Chung Bo Fun Charitable Fund.