Soil-transmitted helminth infections remain the most widespread of the neglected tropical diseases, affecting more than one billion people, primarily in marginalized communities across low- and middle-income countries. The parasites involved include Ascaris lumbricoides, Trichuris trichiura, Necator americanus and Ancylostoma duodenale. Over the past decade, preventive chemotherapy coverage has scaled up substantially, reaching approximately 60 percent of children in need by 2019, and this expansion has contributed to a progressive decline in the prevalence and intensity of these infections worldwide. Yet the surveys that programmes rely on to track that progress are expensive, and their cost has become a serious obstacle for control programmes operating with scarce technical and financial resources.
A new study published in BMC Global and Public Health on 22 October 2026 reports that a model-based geostatistical framework can cut the cost of these impact assessment surveys by 60 to 90 percent while maintaining the accuracy of the treatment decisions they are designed to inform. The research, led by Claudio Fronterre and colleagues including Olatunji Johnson, Emanuele Giorgi, Antonio Montresor and Peter J. Diggle, was applied to real datasets from Kenya, Sierra Leone and Zimbabwe. In the simulation-based evaluation, the average reduction in survey cost was 80 percent, with country-specific reductions of 91 percent for Zimbabwe, 86 percent for Kenya and 62 percent for Sierra Leone.
Impact assessment surveys are central to monitoring and evaluating the effectiveness of mass drug administration programmes targeting soil-transmitted helminths. They are typically conducted after about five years of preventive chemotherapy with albendazole or mebendazole, and their purpose is to estimate current prevalence so that programme managers can decide whether, and at what frequency, treatment should continue. The World Health Organization’s criterion for suspending preventive chemotherapy is that local prevalence of infection with any soil-transmitted helminth species falls below 2 percent. Traditionally, an impact survey is conducted with the same methodology used for the earlier baseline survey but is analysed without reference to the baseline data at all.
The key insight of the new work is that this historical data is being thrown away when it could instead be doing much of the work. Most endemic countries have already conducted a baseline prevalence survey before mass drug administration began, and detailed records exist of the coverage and frequency of treatment rounds delivered since then. The researchers argue that combining these existing resources with environmental risk-factor data and spatial statistical modelling makes it possible to reduce the required sample size drastically, and therefore the cost of an impact survey, without degrading the precision of the survey.
The framework proceeds in three steps. First, a geostatistical model is fitted to the baseline prevalence data, which consist of tests carried out at a sample of geolocated schools or communities before any treatment began. The model treats the number of positive tests at each location as binomial and decomposes variation in prevalence into demographic and environmental covariate effects, such as temperature, precipitation, elevation, vegetation indices and soil moisture, together with residual spatial correlation and unstructured random effects. A separate model is fitted for each of the three soil-transmitted helminth species, because the species differ in how strongly their prevalence is spatially correlated and in how they respond to treatment.
Second, the fitted baseline prevalence surface is projected forward in time to the moment of the proposed impact survey. For each species, a multi-state Markov model estimates transition probabilities between five endemicity classes, ranging from less than 2 percent prevalence to 50 percent or more, as a function of the number of effective treatment rounds, defined as rounds reported to have achieved coverage of at least 75 percent at implementation-unit level. A local spatial thinning algorithm then scales the predicted baseline surface down according to the estimated endemicity class, producing a forecast of the post-treatment prevalence surface. These transition probabilities were estimated from longitudinal data collected in Kenya, where the same set of schools was surveyed at three time points, in 2012, 2015 and 2017, and were then applied to Sierra Leone and Zimbabwe.
Third, the forecast prevalence surface is used to simulate and compare candidate survey designs. The team evaluated 40 unique designs defined by combinations of three parameters: the proportion of schools sampled relative to the baseline survey, ranging from 30 to 100 percent; the number of children tested per school, set at 30, 50, 70 or 100; and the spatial arrangement of sampled schools, comparing simple random selection with a spatially regulated design in which a minimum distance is imposed between any pair of sampled schools. That minimum distance was calibrated so that the expected spatial correlation between the closest pair of points was approximately 0.5, ensuring that spatial correlation could still be exploited while avoiding redundant clustering of nearby schools.
Each candidate design was judged on two criteria: the accuracy with which it classified each implementation unit, typically a district, into the correct endemicity class, and the total cost of the survey, computed using a simple formula that assumes a fixed cost per school visited plus a per-child cost for each child tested. The model-based approach assigns each implementation unit to the most probable class in a predictive distribution over the five endemicity classes, whereas the classical approach relies on observed empirical prevalence from randomly sampled schools and children without accounting for spatial correlation. When pairs of surveys with equal accuracy were compared, the model-based approach consistently achieved the same classification accuracy as the classical approach with a substantially smaller sample size.
The forecast quality varied across the three countries. For Kenya, whose data were used to estimate the Markov transition parameters, the forecasted prevalence surface closely resembled the actual surface, indicating that the treatment history was very effective in projecting prevalence from baseline to impact. For Sierra Leone and Zimbabwe, the researchers found higher discrepancies between forecast and actual surfaces, and they attribute this partly to the fact that the transition parameters were estimated from Kenyan data, whose baseline endemicity differs markedly from the other two countries: baseline prevalence was much lower in Zimbabwe and much higher in Sierra Leone. The authors acknowledge this as a key limitation, noting that the framework assumes the relationship between treatment history and prevalence change is transferable across settings, which may not hold in regions with distinct baseline conditions, intervention quality or transmission dynamics, and that country-specific longitudinal data would be needed to recalibrate the transition probabilities.
The cost function used in the evaluation is also deliberately simple, assuming constant unit costs per school and per child and excluding fixed costs such as planning, training and logistics, which do not affect the relative ranking of designs. Even so, the authors argue that the efficiency gains are substantial enough to matter for programmes operating under real-world constraints, and they note that resources saved through more efficient designs could be reallocated to increase sample sizes and improve decision reliability or to support other monitoring and evaluation activities. The team is preparing a detailed manual to guide national programmes in applying the methodology and is developing a user-friendly web application to support implementation. Both resources are designed to be disease-agnostic, and the researchers suggest the framework could be adapted to other neglected tropical diseases with similar survey design challenges, supporting evidence-based decision-making as countries progress toward control and elimination goals.
Subject of Research: A geostatistical framework for designing lower-cost impact assessment surveys for soil-transmitted helminth control programmes
Article Title: Geostatistical analysis of baseline prevalence and treatment history drastically reduces the cost of soil-transmitted helminth impact assessment surveys
Article References: Fronterre, C., Johnson, O., Giorgi, E., Montresor, A., & Diggle, P. J. (2026). Geostatistical analysis of baseline prevalence and treatment history drastically reduces the cost of soil-transmitted helminth impact assessment surveys. BMC Global and Public Health, 4(1), Article 106. https://doi.org/10.1186/s44263-026-00304-8
Image Credits: AI Generated
DOI: 10.1186/s44263-026-00304-8
Keywords: soil-transmitted helminths, geostatistics, impact assessment surveys, mass drug administration, preventive chemotherapy, neglected tropical diseases, survey design, prevalence mapping, Markov model, Kenya, Sierra Leone, Zimbabwe
News Source: Phoebe Ingram. (October 28, 2026). Research report: Geostatistical analysis of baseline prevalence and treatment history drastically reduces the cost of soil-transmitted helminth impact assessment surveys. Scienmag.
