Modelling the spatial distribution of Hepatitis A risk, based on wastewater concentrations in Gauteng

Introduction

Hepatitis A virus (HAV) is a contagious, vaccine-preventable liver infection and the most common cause of viral hepatitis worldwide. In 2024, there were approximately 14 million cases and 28,000 deaths globally (FAO/WHO, 2024). Most infections are asymptomatic or mild, but in severe cases HAV can progress to acute liver failure requiring transplantation. The virus is excreted in stool and spreads primarily through the faecal-oral route via contaminated water, food, or contact with contaminated surfaces and refuse. Its prevalence is therefore tied to water, sanitation, and hygiene conditions (World Health Organization, 2025). HAV disproportionately affects communities with limited access to clean water and adequate sanitation, and children are especially vulnerable in high-endemicity settings: South Africa is classified as highly endemic, meaning at least 90% of children under 5 are exposed to the virus. HAV is being monitored through a national disease surveillance program managed by the National Institute for Communicable Diseases (NICD).

In South Africa, HAV surveillance is traditionally done through clinical surveillance when a patient visits a healthcare facility and is tested for a disease or pathogen. This helps identify at-risk populations and also determines which pathogen variants are present in a community. Clinical surveillance is a key component of disease surveillance, but it is limited to the information provided by people who visit a healthcare facility. Many people, particularly those in poor communities, cannot always access care, meaning that a proportion of diseases in the population go undetected. An additional limitation to understanding HAV epidemiology is that infection in children under the age of five years, and especially under two years, HAV infection is asymptomatic. Therefore, these children will not be taken for clinical assessment, nor will they be tested. In many countries where HAV is endemic, the bulk of infections are in young children, who are initially exposed to HAV through faecal-oral exposure as they learn to crawl, walk and start exploring their environment. Thus, surveillance data derived from laboratory testing results does not give the full picture of HAV epidemiology.

However, because HAV is shed in high concentrations in stool, it is also detectable in wastewater. Wastewater and environmental surveillance (WES) is a novel, complementary surveillance system that fills the gap in clinical surveillance. In a Photo Essay, the GCRO illustrated the wastewater treatment process, as well as details on how a sample is collected and processed. When a person is infected with a disease, they excrete the pathogen through body waste into the wastewater system. By collecting and testing a sample of sewage, it is possible to detect a wide range of pathogens circulating in a community. Because everyone contributes to the sewage system regardless of income, healthcare access, or symptoms, WES can provide a more complete picture of all the diseases circulating in a population compared to clinical surveillance alone (Maree et al., 2024). WES has already been successfully used internationally, and in South Africa, to track diseases such as poliovirus and SARS-CoV-2, helping to detect outbreaks early and monitor emerging variants (Yousif et al., 2023).

South Africa currently runs a national WES system across 48 sites, testing wastewater for SARS-CoV-2, influenza A and B, mpox, tuberculosis, measles, and hepatitis A and E (McCarthy et al., 2025). Because of the faecal-oral route of transmission and the primarily asymptomatic presentation of the disease, WES is an ideal method for monitoring the presence of hepatitis A. However, as it is not feasible to sample everywhere, disease patterns can be modelled to estimate risk in areas where direct wastewater monitoring is not possible.

Spatial modelling is a statistical approach that combines data from multiple sources (such as wastewater surveillance, clinical cases, and socio-economic indicators) to predict disease risk across geographic areas where direct measurement is unavailable (Figure 1). Because it is not feasible to sample wastewater in every ward, this model can be used as a proxy for risk to estimate hepatitis A burden in unsampled or under-sampled wards. By identifying spatial patterns and predictors associated with known high levels of disease transmission inferred from the concentration of HAV in wastewater, this approach allows public health authorities to prioritise surveillance and intervention resources toward areas of higher predicted risk, even in the absence of direct wastewater or case data.

Figure 1: A flow diagram of the modelling process

In this Map of the Month, we set out to determine whether wastewater surveillance data can help identify the communities at greatest risk of HAV exposure. We used machine learning to combine clinical case data, HAV concentrations measured in wastewater, and socio-economic indicators from the GCRO Quality of Life Survey 7 (2023/2024) across Gauteng's 529 wards. The resulting model allows us to predict HAV risk in the rest of Gauteng where wastewater sampling is not currently possible. The objective was to pinpoint communities that are at a higher risk of contracting the HAV. This will ultimately support more targeted and effective public health interventions.

Distribution of modelled HAV concentration in wastewater

Mapping the average modelled concentration of the HAV in wastewater in Gauteng showed that the concentration of HAV is likely to be the highest (i.e. above average concentration) in peripheral areas and among respondents who live in informal dwellings (Figure 2, for example, Soshanguve (A)). Areas where the concentration of HAV is likely very high are located near Soshanguve, Tshepisong, informal settlements near Dainfern, Bekkersdal and Carletonville. Areas where the concentration of HAV is likely very low (i.e. below average concentration) are Roodepoort, Tembisa, Germiston/Wadeville, Katlehong and Sebokeng. In areas such as Tembisa, Katlehong and Sebokeng, where the expected risk would be high based on HAV risk factors (mentioned in the introduction), the lower risk outcome could be attributed to communities with mixed socio-economic status and mixed formal and informal housing structures. Moderate risk (i.e. average concentration) is the most prevalent and is found in both the urban core and peripheral areas.

Figure 2: Modelled concentration (genome copies/µl) of the average hepatitis A virus in wastewater per ward. Source: Modelled using GCRO QoL survey 7 (GCRO, 2024), clinical case reports, and NICD wastewater surveillance data (NICD, 2023-2026).

Discussion

Here we report on Quality of Life (QoL) (2023/2024) indicators in areas that have higher (above average concentration) and lower (below average concentration) modelled concentrations of the HAV in wastewater, signifying increased HAV transmission in these areas. We link these QoL indicators with HAV risk by describing possible transmission pathways and associations. We specifically use QoL indicators that were not used by the machine learning algorithm to generate the model and map (Figure 2). More information on the model input parameters can be found in the method note below.

The HAV is transmitted through contaminated water and food, so exposure is more likely in areas where refuse is not removed regularly. Inadequate refuse removal could add to the possible transmission of HAV if contaminated food or water is present in refuse. In wards with higher risk of contracting HAV, 25% of respondents used alternative means of refuse removal (dumped in a veldt, burned in a pit, etc., implying that refuse is directly handled) and 24% of households have refuse removed less often than once a week.

Figure 3: Modelled HAV concentration in wastewater compared to refuse removal. A range of 1 SD (standard deviation) from the average concentration was used to define the modelled average concentration. Wastewater concentration was measured in genome copies/µl.

In wards where the modelled concentration of HAV is above average, 15% of respondents have piped water compared to 30% who use alternative water sources such as street taps or standpipes, boreholes, rainwater tanks, water tankers or natural water (rivers and dams). HAV can survive in the environment, and so shared water could contribute to higher exposure and therefore higher risk of transmission.

Figure 4: Modelled HAV concentration in wastewater compared to access to water. A range of 1 SD (standard deviation) from the average concentration was used to define the modelled average concentration. Wastewater concentration was measured in genome copies/µl.

In wards with higher risk of contracting HAV, 25% of respondents did not receive adequate (below the Gauteng average) services compared to 15% of respondents who had adequate services. The Services dimension of the Quality of Life Index includes a number of indicators: having a brick or concrete dwelling structure, a flush toilet connected to the sewer system, piped water into the dwelling, formal electricity supply, regular refuse removal from home, and owning a working television. This indicates that respondents who live in informal dwellings and use informal sewage disposal systems are at higher risk of contracting HAV.

Figure 5: Modelled HAV concentration in wastewater compared to the Gauteng average on the Quality of Life Index Services dimension. A range of 1 SD (standard deviation) from the average concentration was used to define the modelled average concentration. Wastewater concentration was measured in genome copies/µl.

Conclusion

Our spatial analysis using case data, quality of life indicators and wastewater concentrations of HAV allowed us to determine areas with the highest modelled HAV wastewater concentrations, and therefore likely the highest HAV disease burden and infection risk. Features of areas with the highest HAV concentrations include higher proportions of residents obliged to use alternative refuse removal strategies because of limited municipal rubbish removal, forced to rely on alternative non-piped water sources, and have overall lower scores on the QoL Index Services dimension. Using multiple data sources such as social surveys with disease indicator data (clinical and wastewater surveillance data) could aid in providing novel ways to measure burden of disease and therefore exposure risk. No single data source fully explains modelled HAV risk, and using a multi-disciplinary approach adds value because it can identify risk in places where conventional service-access indicators alone would not predict it.

Adding wastewater surveillance data to spatial disease modelling offers a valuable complement to household survey data. It may assist in identifying communities at elevated risk for HAV and may fill surveillance gaps that clinical case reporting alone would miss. Areas with high modelled HAV exposure risk, particularly Bekkersdal, Soshanguve, and Tshepisong, should be prioritised for public health interventions, including health promotion, improved sanitation and refuse services. This novel approach and proof of concept to burden of disease estimation for hepatitis A could also be applied to other pathogens of public health importance, such as measles or polio.

Methods notes

This Map of the Month set out to determine whether wastewater surveillance data can help identify the communities at greatest risk of Hepatitis A virus (HAV) exposure, by using machine learning to combine clinical case data, HAV concentrations measured in wastewater, and socio-economic indicators from the Gauteng Quality of Life Survey 7 (2023/2024).

Clinical surveillance data: Clinical case data for March 2024-March 2026 were obtained through the National Institute for Communicable Diseases' Notifiable Medical Conditions surveillance system. Cases were recorded at facility level and geolocated by linking each facility to its corresponding ward, allowing case counts to be aggregated to the ward level for spatial analysis. The NICD surveillance activities are reviewed by the University of the Witwatersrand Human Research Ethics Committee (M215702, reapplication under review).

Wastewater surveillance data: Wastewater samples were collected daily/weekly and analysed at the National Institute for Communicable Diseases' Centre for Vaccines and Immunology, following a standardised laboratory protocol (Mabasa et al., 2025). HAV concentrations were measured for each wastewater testing site (March 2024 - March 2026) and assigned to the wards served by that site's sewershed, providing a ward-level measure of viral concentration in the community. To account for the temporal nature of the wastewater data (data per week), an average of the concentrations for the time period was used.

Socio-economic indicators: Ward-level socio-economic and demographic indicators were drawn from the Gauteng Quality of Life Survey 7. Rather than treating clinical case presence and community risk as independent, the model uses these indicators to identify wards that share a defined set of quality-of-life characteristics associated with elevated HAV risk. The variables included in the model were percentage of respondents who were Black African, under the age of 30, male, completed schooling, unsewered sanitation, use public healthcare, self-reported poor health, low income, and the level of household crowding. Together these variables were combined into an index of ward-level risk, expressed as categorical variables (below average, average and above average), with an above average category reflecting a greater concentration of risk factors within a ward.

Model: A Gaussian Process Regression (GPR) model was used to draw together clinical case data, wastewater HAV concentrations, and Quality of Life data. The dataset was split 70/30 for model training and validation. By linking these three data sources, the model tested whether wards with a higher concentration of risk factors showed a correlated relationship with higher HAV concentrations detected in wastewater i.e. whether socio-economic vulnerability at ward level tracks with viral burden measured through wastewater surveillance. The model achieved a coefficient of determination (R²) of 0.98 and a Root Mean Square Error (RMSE) of 8.17 genome copies/µl on the validation dataset, indicating a strong fit between predicted and observed HAV concentrations across wards. When the model produced a “no data” per ward, we concluded that the representative range of the input parameter in those wards were not covered by the 11 wastewater sampling sites, and the model produced erroneous estimates. To remove this, we could have more sampling data to calibrate and validate the model.

References

Maree, G., Els, F., Naidoo, Y., Naidoo, L., Mahamuza, P., Ndlovu, N., Rachida, S., Iwu-Jaja, C., Taukobong, S., Maposa, S., O’Reilly, K., Yousif, M., & McCarthy, K. (2024). Wastewater surveillance overcomes socio-economic limitations of laboratory-based surveillance when monitoring disease transmission: The South African experience during the COVID-19 pandemic. 2024:2024.09.20.24314039. https://doi.org/10.1101/2024.09.20.24314039.

McCarthy, K., Maposa, S., Mabasa, V., Singh, N., Gwala, S., Sankar, C., Ndlovu, N., Els, F., Phalane, E., Subramony, K., Macheke, M., Madikane, N., Msomi, N., Monametsi, L., Rabotapi, L., Mangena, T., Motloung, M., Kent, C., Quick, J., … Yousif, M. (2025). Wastewater surveillance for infectious agents of measles, rubella, hepatitis, influenza, mpox, and tuberculosis in South Africa, 2024. 3(2), 127–150.

Victor Mabasa, Natasha Singh, Emmanuel Phalane, Mokgaetji Macheke, Sipho Gwala, Thabo Mangena, Lethabo Monamets, Lebohang Rabotapi, Nkosenhle Ndlovu, Fiona Els, Sibonginkosi Maposa, Said Rachida, Kerrigan McCarthy, & Mukhlid Yousif. (2025). Clarification of wastewater samples. protocols.io. https://dx.doi.org/10.17504/protocols.io.5jyl8qo99l2w/v1

World Health Organization. (2025, December 1). Wastewater and environmental surveillance: Summary for Hepatitis A and E viruses. https://cdn.who.int/media/docs/default-source/wash-documents/wash-related-diseases/wes-summary-for-hav-hev---final-1-dec.pdf

Food and Agriculture Organization of the United Nations, World Health Organization. Joint FAO/WHO Expert Meeting on Microbiological Risk Assessment (JEMRA): Microbiological risk assessment of viruses in foods. Rome: FAO/WHO; 2024.

Yousif, M., Rachida, S., Taukobong, S., Ndlovu, N., Iwu-Jaja, C., Howard, W., Moonsamy, S., Mhlambi, N., Gwala, S., Levy, J. I., Andersen, K. G., Scheepers, C., Von Gottberg, A., Wolter, N., Bhiman, J. N., Amoako, D. G., Ismail, A., Suchard, M., & McCarthy, K. (2023). SARS-CoV-2 genomic surveillance in wastewater as a model for monitoring evolution of endemic viruses. Nature Communications, 14(1), 6325. https://doi.org/10.1038/s41467-023-41369-5

Cartography/mapping: Yashena Naidoo

Inputs, edits, and comments: Graeme Götz, Christian Hamann, Dr Samkelisiwe Khanyile

Suggested citation: Els, F., Naidoo, L., Naidoo, Y., Yousif, M., McCarthy, K., (2026). Modelling the spatial distribution of Hepatitis A risk, based on wastewater concentrations in Gauteng. GCRO Map of the Month, July 2027. Gauteng City-Region Observatory, Johannesburg. https://doi.org/10.36634/KQTV7772.