The Akora River drains a landscape shaped by both legal and illegal small-scale gold mining, locally known as galamsey, which has repeatedly been linked to sediment plumes, mercury use and disturbed soils across Ghana’s mining belts. Previous studies of mining-affected rivers in the country have documented elevated lead and mercury in water and fish, but they have generally relied on single-point measurements and deterministic risk quotients. That approach has three well-known weaknesses. It ignores samples that fall below the instrument’s detection limit, it treats total metal concentrations as if every chemical form of an element is equally toxic, and it compresses the enormous variability among people—how much fish they eat, how much they weigh, how their bodies handle a dose—into a single average consumer who may not exist.

The new study tackles all three problems at once. The researchers sampled water, sediment and two widely consumed species, Nile tilapia (Oreochromis niloticus) and African catfish (Clarias anguillaris), at eight stations along the river, then analysed eight metal(loid)s: mercury, zinc, arsenic, nickel, chromium, copper, lead and cadmium. Metal quantification used inductively coupled plasma optical emission spectroscopy, with mercury measured separately by cold-vapour atomic fluorescence spectroscopy, a technique sensitive enough to resolve the vanishingly low concentrations at which this element matters for human health. Quality assurance relied on certified reference materials, anchoring the analytical results to internationally traceable standards.

The statistical core of the paper is a censoring-aware hierarchical Bayesian model. In plain terms, the model does not discard or crudely substitute values for samples that registered as non-detects—a practice that can badly bias exposure estimates when many measurements sit near the detection limit. Instead, it treats those values as censored observations, inferring plausible concentrations consistent with the knowledge that they fall somewhere below the limit of detection. The hierarchical structure allows concentrations to vary across sampling stations while borrowing strength from the whole dataset, producing posterior distributions of fish metal concentrations rather than single numbers. Those posteriors were generated with Hamiltonian Monte Carlo using the No-U-Turn Sampler, a modern algorithm that explores high-dimensional probability landscapes efficiently and is now standard in Bayesian computation.

What happens next is where the study earns its title. Total metal concentrations in fish are not directly convertible to dose, because different chemical species of the same element behave very differently in the body. Inorganic arsenic is far more carcinogenic than its organic counterparts; hexavalent chromium is a recognized carcinogen while trivalent chromium is far less hazardous; methylmercury, the organic form that accumulates in fish muscle, crosses the blood-brain barrier far more readily than inorganic mercury. Rather than assuming fixed speciation fractions, the team encoded speciation as prior probability distributions, acknowledging genuine scientific uncertainty about how much of each element’s total burden exists in its most toxic form. Those speciation priors were then propagated through fish ingestion rates, body weights, exposure durations and dose-response models to estimate three endpoints for five receptor groups: children, women of child-bearing age, adults generally, and a distinct category of high-consuming individuals.

The three endpoints capture different dimensions of harm. The non-cancer hazard index sums the ratios of estimated chronic daily intakes to reference doses across elements, flagging potential effects on organs and systems rather than cancer. Total cancer risk estimates the incremental lifetime probability of developing cancer from carcinogenic exposures, conventionally screened against a threshold of one in a million. Blood-lead level predictions, benchmarked against the United States Centers for Disease Control and Prevention reference value of 3.5 micrograms per decilitre, address the neurodevelopmental toxicity of lead, for which no safe threshold is established in children.

The water-column results were stark. Chromium and nickel dominated contamination in the river itself, reaching approximately 55 times and 23 times the World Health Organization guideline values at downstream stations, a pattern consistent with mining and associated industrial activity in the catchment. Sediment quality quotients based on probable effect concentrations added further evidence that the benthic environment is under stress. Yet the fish told a subtler story: metal concentrations showed little separation between the two species, suggesting that neither tilapia nor catfish systematically accumulates more of the metal(loid) burden, and that consumption advice based on species choice alone would offer limited protection.

The human health results hinge on the difference between medians and tails. Median non-cancer hazard indices remained below the screening threshold of one for every receptor group, meaning the typical consumer is not expected to face non-cancer harm. But the probability of exceeding that threshold reached roughly 15 to 17 percent for children and high-consumers—a substantial fraction of the population for whom the assumption of safety is not secure. Total cancer risk exceeded the 10⁻⁶ screening level in 35 to 78 percent of receptor-specific posterior draws, with inorganic arsenic and hexavalent chromium assumptions driving most of the exceedance. Median blood-lead predictions stayed below 3.5 micrograms per decilitre, though the upper tail of the distribution made exceedance plausible for children and heavy consumers. In other words, the average answer is broadly reassuring, but the distribution of possible answers is not.

Perhaps the most actionable finding comes from the sensitivity analysis. Using global sensitivity methods based on total-order Sobol indices, the team decomposed which inputs contributed most to the uncertainty in the risk estimates. Fish ingestion rate, body weight and the speciation and toxicological parameters dominated, while additional precision in total-metal concentration measurements added surprisingly little value. This is a pointed message for monitoring programmes in low-resource settings: buying more sensitive instruments to sharpen total-metal numbers may be less useful than investing in dietary surveys that establish how much fish people actually eat, and in speciation analysis—such as high-performance liquid chromatography coupled to inductively coupled plasma mass spectrometry—that distinguishes toxic chemical forms from benign ones.

The authors argue that Ghana needs nationally codified freshwater fish consumption guidance, and the study’s structure shows what that guidance would need to look like. Blanket advisories built on average consumers and total-metal concentrations would miss the children and high-consumers who carry most of the risk, and would misallocate analytical resources toward measurements that do not reduce uncertainty. A speciation-focused monitoring programme, paired with receptor-specific consumption surveys, would target the parameters that actually drive the posterior tails. The work is also, according to the authors, the first hierarchical Bayesian assessment of fish metal contamination conducted in West Africa, and it offers a template that other mining-affected river basins in the region could adopt. For communities along the Akora, the practical takeaway is measured rather than alarmist: fish remains a vital protein source in a region where food security depends on it, but the safest course lies in knowing who eats how much, and in which chemical form the danger actually arrives.

Subject of Research: Bayesian assessment of metal(loid) exposure and health risk from fish consumption in a mining-influenced Ghanaian river

Article Title: Speciation-informed Bayesian assessment of metal(loid) exposure from Akora River fish, Ghana, for consumption guidance

Article References: Asare, E. A., Abdul-Wahab, D., Kaufmann, E. E., Wahi, R., Ngaini, Z., & Osei, P. (2026). Speciation-informed Bayesian assessment of metal(loid) exposure from Akora River fish, Ghana, for consumption guidance. Environmental Monitoring and Assessment, 198(11), Article 1154. https://doi.org/10.1007/s10661-026-15997-5

Image Credits: AI Generated

DOI: 10.1007/s10661-026-15997-5

Keywords: Ghana, Akora River, heavy metals, Bayesian hierarchical model, fish consumption, arsenic, methylmercury, hexavalent chromium, risk assessment, speciation, galamsey, food safety