Model-Based Geostatistics for Prevalence Mapping in Low-Resource Settings

Application of multivariate binary logistic regression grouped outlier statistics and geospatial logistic model to identify villages having unusual health-seeking habits for childhood malaria in Malawi

BACKGROUND

Early diagnosis and prompt treatment of malaria in young children are crucial for preventing the serious stages of the disease. If delayed treatment-seeking habits are observed in certain areas, targeted campaigns and interventions can be implemented to improve the situation.

METHODS

This study applied multivariate binary logistic regression model diagnostics and geospatial logistic model to identify traditional authorities in Malawi where caregivers have unusual health-seeking behaviour for childhood malaria. The data from the 2021 Malawi Malaria Indicator Survey were analysed using R software version 4.3.0 for regressions and STATA version 17 for data cleaning.

RESULTS

Both models showed significant variability in treatment-seeking habits of caregivers between villages. The mixed-effects logit model residual identified Vuso Jere, Kampingo Sibande, Ngabu, and Dzoole as outliers in the model. Despite characteristics that promote late reporting of malaria at clinics, most mothers in these traditional authorities sought treatment within twenty-four hours of the onset of malaria symptoms in their children. On the other hand, the geospatial logit model showed that late seeking of malaria treatment was prevalent in most areas of the country, except a few traditional authorities such as Mwakaboko, Mwenemisuku, Mwabulambya, Mmbelwa, Mwadzama, Zulu, Amidu, Kasisi, and Mabuka.

CONCLUSIONS

These findings suggest that using a combination of multivariate regression model residuals and geospatial statistics can help in identifying communities with distinct treatment-seeking patterns for childhood malaria within a population. Health policymakers could benefit from consulting traditional authorities who demonstrated early reporting for care in this study. This could help in understanding the best practices followed by mothers in those areas which can be replicated in regions where seeking care is delayed.

Identifying childhood malaria hotspots and risk factors in a Nigerian city using geostatistical modelling approach

Malaria ranks high among prevalent and ravaging infectious diseases in sub-Saharan Africa (SSA). The negative impacts, disease burden, and risk are higher among children and pregnant women as part of the most vulnerable groups to malaria in Nigeria. However, the burden of malaria is not even in space and time. This study explores the spatial variability of malaria prevalence among children under five years (U5) in medium-sized rapidly growing city of Akure, Nigeria using model-based geostatistical modeling (MBG) technique to predict U5 malaria burden at a 100 × 100 m grid, while the parameter estimation was done using Monte Carlo maximum likelihood method. The non-spatial logistic regression model shows that U5 malaria prevalence is significantly influenced by the usage of insecticide-treated nets-ITNs, window protection, and water source. Furthermore, the MBG model shows predicted U5 malaria prevalence in Akure is greater than 35% at certain locations while we were able to ascertain places with U5 prevalence > 10% (i.e. hotspots) using exceedance probability modelling which is a vital tool for policy development. The map provides place-based evidence on the spatial variation of U5 malaria in Akure, and direction on where intensified interventions are crucial for the reduction of U5 malaria burden and improvement of urban health in Akure, Nigeria.

Geographic pair matching in large-scale cluster randomized trials

Cluster randomized trials are often used to study large-scale public health interventions. In large trials, even small improvements in statistical efficiency can have profound impacts on the required sample size and cost. Location integrates many socio-demographic and environmental characteristics into a single, readily available feature. Here we show that pair matching by geographic location leads to substantial gains in statistical efficiency for 14 child health outcomes that span growth, development, and infectious disease through a re-analysis of two large-scale trials of nutritional and environmental interventions in Bangladesh and Kenya. Relative efficiencies from pair matching are ≥1.1 for all outcomes and regularly exceed 2.0, meaning an unmatched trial would need to enroll at least twice as many clusters to achieve the same level of precision as the geographically pair matched design. We also show that geographically pair matched designs enable estimation of fine-scale, spatially varying effect heterogeneity under minimal assumptions. Our results demonstrate broad, substantial benefits of geographic pair matching in large-scale, cluster randomized trials.

Space-time modelling of monthly malaria incidence for seasonal associated drivers and early epidemic detection in Southern Ethiopia

BACKGROUND

Although Ethiopia has made great strides in recent years to reduce the threat of malaria, the disease remains a significant issue in most districts of the country. It constantly disappears in parts of the areas before reappearing in others with erratic transmission rates. Thus, developing a malaria epidemic early warning system is important to support the prevention and control of the incidence.

METHODS

Space-time malaria risk mapping is essential to monitor and evaluate priority zones, refocus intervention, and enable planning for future health targets. From August 2013 to May 2019, the researcher considered an aggregated count of genus Plasmodium falciparum from 149 districts in Southern Ethiopia. Afterwards, a malaria epidemic early warning system was developed using model-based geostatistics, which helped to chart the disease's spread and future management.

RESULTS

Risk factors like precipitation, temperature, humidity, and nighttime light are significantly associated with malaria with different rates across the districts. Districts in the southwest, including Selamago, Bero, and Hamer, had higher rates of malaria risk, whereas in the south and centre like Arbaminch and Hawassa had moderate rates. The distribution is inconsistent and varies across time and space with the seasons.

CONCLUSION

Despite the importance of spatial correlation in disease risk mapping, it may occasionally be a good idea to generate epidemic early warning independently in each district to get a quick picture of disease risk. A system like this is essential for spotting numerous inconsistencies in lower administrative levels early enough to take corrective action before outbreaks arise.

Model-based geostatistical design and analysis of prevalence for soil-transmitted helminths in Kenya: Results from ten-years of the Kenya national school-based deworming programme

Background

Kenya is endemic for soil-transmitted helminths (STH) with over 6 million children in 27 counties currently at-risk. A national school-based deworming programme (NSBDP) was launched in 2012 with a goal to eliminate parasitic worms as a public health problem. This study used model-based geostatistical (MBG) approach to design and analyse the impact of the NSBDP and inform treatment strategy changes.

Methods

A cross-sectional study was used to survey 200 schools across 27 counties in Kenya. The study design, school selection and analysis followed the MBG approach which incorporated historical data on treatment, morbidity and environmental covariates to efficiently predict the helminths prevalence in Kenya.

Results

Overall, the NSBDP geographic area prevalence for any STH was estimated to sit between 2 % and <10 % with a high predictive probability of >0.999. Species-specific thresholds were between 2 % and <10 % for Ascaris lumbricoides, 0 % to <2 % for hookworm, and 0 % to <2 % for Trichuris trichiura, all with high predictive probability of >0.999.

Conclusions

Based on the World Health Organization guidelines, STH treatment requirements can now be confidently refined. Ten counties may consider suspending treatment and implement appropriate surveillance system, while another 10 will require treatment once every two years, and the remaining seven will require treatment once every year.

Risk mapping for Taenia solium: Applying multicriteria decision analysis in Lao PDR

OBJECTIVES

Taenia solium is ranked the most significant global foodborne parasite and the leading cause of epilepsy in low and middle-income countries. Diagnostic challenges have hampered disease control efforts to date and WHO has called for the development of risk mapping tools to assist endemic countries. This study describes the application of multicriteria decision analysis to map the risk of T. solium in Lao PDR and acts as a case study for other endemic countries.

METHODS

Multicriteria decision analysis was completed using census data on relevant risk factors in Lao PDR. Factors were weighted using an analytical hierarchy process. Village risk scores were calculated using a weighted linear combination and categorised using the Fisher-Jenks algorithm into low, medium, and high risk. District risk scores and categories were calculated using the mean village risk score for a district. Sensitivity analysis was completed by doubling and halving risk factor weights, one at a time, and assessing the standard deviation of scores and categories across all scenarios.

RESULTS

A total of 2017 (23.7%) villages were classified as high risk, with 3312 (39.0%) medium and 3170 (37.3%) low risk. This resulted in 21 (14.2%) high-risk districts, 83 (56.1%) medium and 44 (29.7%) low-risk districts. The risk maps highlight two areas of interest which are high risk and low variation. The first is the northern province of Phongsaly, which is consistent with literature and anecdotal reports. The second is the southern Salavan and Xekong provinces, which have yet to be investigated in detail.

CONCLUSIONS

Multicriteria decision analysis has provided a simple, rapid, and flexible approach to mapping the risk of T. solium in Lao PDR. The nature of the method means that it can be completed in any endemic country with available and appropriate risk factor data.

Geographic pair-matching in large-scale cluster randomized trials

Custer randomized trials are often used to study large-scale public health interventions. In large trials, even small improvements in statistical efficiency can have profound impacts on the required sample size and cost. Pair matched randomization is one strategy with potential to increase trial efficiency, but to our knowledge there have been no empirical evaluations of pair-matching in large-scale, epidemiologic field trials. Location integrates many socio-demographic and environmental characteristics into a single feature. Here, we show that geographic pair-matching leads to substantial gains in statistical efficiency for 14 child health outcomes that span growth, development, and infectious disease through a re-analysis of two large-scale trials of nutritional and environmental interventions in Bangladesh and Kenya. We estimate relative efficiencies ≥1.1 for all outcomes assessed and relative efficiencies regularly exceed 2.0, meaning an unmatched trial would have needed to enroll at least twice as many clusters to achieve the same level of precision as the geographically pair-matched design. We also show that geographically pair-matched designs enable estimation of fine-scale, spatially varying effect heterogeneity under minimal assumptions. Our results demonstrate broad, substantial benefits of geographic pair-matching in large-scale, cluster randomized trials.

Estimating the burden of mycetoma in Sudan for the period 1991-2018 using a model-based geostatistical approach

Mycetoma is widespread in tropical and subtropical regions favouring arid areas with low humidity and a short rainy season. Sudan is one of the highly endemic countries for mycetoma. Estimating the population at risk and the number of cases is critical for delivering targeted and equitable prevention and treatment services. In this study, we have combined a large dataset of mycetoma cases recorded by the Mycetoma Research Centre (MRC) in Sudan over 28 years (1991-2018) with a collection of environmental and water and hygiene-related datasets in a geostatistical framework to produce estimates of the disease burden across the country. We developed geostatistical models to predict the number of cases of actinomycetoma and eumycetoma in areas considered environmentally suitable for the two mycetoma forms. Then used the raster dataset (gridded map) with the population estimates for 2020 to compute the potentially affected population since 1991. The geostatistical models confirmed this heterogeneous and distinct distribution of the estimated cases of eumycetoma and actinomycetoma across Sudan. For eumycetoma, these higher-risk areas were smaller and scattered across Al Jazirah, Khartoum, White Nile and Sennar states, while for actinomycetoma a higher risk for infection is shown across the rural districts of North and West Kurdufan. Nationally, we estimated 63,825 people (95%CI: 13,693 to 197,369) to have been suffering from mycetoma since 1991 in Sudan,51,541 people (95%CI: 9,893-166,073) with eumycetoma and 12,284 people (95%CI: 3,800-31,296) with actinomycetoma. In conclusion, the risk of mycetoma in Sudan is particularly high in certain restricted areas, but cases are ubiquitous across all states. Both prevention and treatment services are required to address the burden. Such work provides a guide for future control and prevention programs for mycetoma, highly endemic areas are clearly targeted, and resources are directed to areas with high demand.

Efficient design of geographically-defined clusters with spatial autocorrelation

Clusters form the basis of a number of research study designs including survey and experimental studies. Cluster-based designs can be less costly but also less efficient than individual-based designs due to correlation between individuals within the same cluster. Their design typically relies on ad hoc choices of correlation parameters, and is insensitive to variations in cluster design. This article examines how to efficiently design clusters where they are geographically defined by demarcating areas incorporating individuals and households or other units. Using geostatistical models for spatial autocorrelation, we generate approximations to within cluster average covariance in order to estimate the effective sample size given particular cluster design parameters. We show how the number of enumerated locations, cluster area, proportion sampled, and sampling method affect the efficiency of the design and consider the optimization problem of choosing the most efficient design subject to budgetary constraints. We also consider how the parameters from these approximations can be interpreted simply in terms of 'real-world' quantities and used in design analysis.

Using geospatial models to map zero-dose children: factors associated with zero-dose vaccination status before and after a mass measles and rubella vaccination campaign in Southern province, Zambia

INTRODUCTION

Despite gains in global coverage of childhood vaccines, many children remain undervaccinated. Although mass vaccination campaigns are commonly conducted to reach these children their effectiveness is unclear. We evaluated the effectiveness of a mass vaccination campaign in reaching zero-dose children.

METHODS

We conducted a prospective study in 10 health centre catchment areas in Southern province, Zambia in November 2020. About 2 months before a national mass measles and rubella vaccination campaign conducted by the Ministry of Health, we used aerial satellite maps to identify built structures. These structures were visited and diphtheria-tetanus-pertussis (DTP) and measles zero-dose children were identified (children who had not received any DTP or measles-containing vaccines, respectively). After the campaign, households where measles zero-dose children were previously identified were targeted for mop-up vaccination and to assess if these children were vaccinated during the campaign. A Bayesian geospatial model was used to identify factors associated with zero-dose status and measles zero-dose children being reached during the campaign. We also produced fine-scale zero-dose prevalence maps and identified optimal locations for additional vaccination sites.

RESULTS

Before the vaccination campaign, 17.3% of children under 9 months were DTP zero-dose and 4.3% of children 9-60 months were measles zero-dose. Of the 461 measles zero-dose children identified before the vaccination campaign, 338 (73.3%) were vaccinated during the campaign and 118 (25.6%) were reached by a targeted mop-up activity. The presence of other children in the household, younger age, greater travel time to health facilities and living between health facility catchment areas were associated with zero-dose status. Mapping zero-dose prevalence revealed substantial heterogeneity within and between catchment areas. Several potential locations were identified for additional vaccination sites.

CONCLUSION

Fine-scale variation in zero-dose prevalence and the impact of accessibility to healthcare facilities on vaccination coverage were identified. Geospatial modelling can aid targeted vaccination activities.

Spatiotemporal analysis of the first wave of COVID-19 hospitalisations in Birmingham, UK

OBJECTIVES

To evaluate the spatiotemporal distribution of the incidence of COVID-19 hospitalisations in Birmingham, UK during the first wave of the pandemic to support the design of public health disease control policies.

DESIGN

A geospatial statistical model was estimated as part of a real-time disease surveillance system to predict local daily incidence of COVID-19.

PARTICIPANTS

All hospitalisations for COVID-19 to University Hospitals Birmingham NHS Foundation Trust between 1 February 2020 and 30 September 2020.

OUTCOME MEASURES

Predictions of the incidence and cumulative incidence of COVID-19 hospitalisations in local areas, its weekly change and identification of predictive covariates.

RESULTS

Peak hospitalisations occurred in the first and second weeks of April 2020 with significant variation in incidence and incidence rate ratios across the city. Population age, ethnicity and socioeconomic deprivation were strong predictors of local incidence. Hospitalisations demonstrated strong day of the week effects with fewer hospitalisations (10%-20% less) at the weekend. There was low temporal correlation in unexplained variance. By day 50 at the end of the first lockdown period, the top 2.5% of small areas had experienced five times as many cases per 10 000 population as the bottom 2.5%.

CONCLUSIONS

Local demographic factors were strong predictors of relative levels of incidence and can be used to target local areas for disease control measures. The real-time disease surveillance system provides a useful complement to other surveillance approaches by producing real-time, quantitative and probabilistic summaries of key outcomes at fine spatial resolution to inform disease control programmes.

Model building and assessment of the impact of covariates for disease prevalence mapping in low-resource settings: to explain and to predict

This paper provides statistical guidance on the development and application of model-based geostatistical methods for disease prevalence mapping. We illustrate the different stages of the analysis, from exploratory analysis to spatial prediction of prevalence, through a case study on malaria mapping in Tanzania. Throughout the paper, we distinguish between predictive modelling, whose main focus is on maximizing the predictive accuracy of the model, and explanatory modelling, where greater emphasis is placed on understanding the relationships between the health outcome and risk factors. We demonstrate that these two paradigms can result in different modelling choices. We also propose a simple approach for detecting over-fitting based on inspection of the correlation matrix of the estimators of the regression coefficients. To enhance the interpretability of geostatistical models, we introduce the concept of domain effects in order to assist variable selection and model validation. The statistical ideas and principles illustrated here in the specific context of disease prevalence mapping are more widely applicable to any regression model for the analysis of epidemiological outcomes but are particularly relevant to geostatistical models, for which the separation between fixed and random effects can be ambiguous.

Supporting elimination of lymphatic filariasis in Samoa by predicting locations of residual infection using machine learning and geostatistics

The global elimination of lymphatic filariasis (LF) is a major focus of the World Health Organization. One key challenge is locating residual infections that can perpetuate the transmission cycle. We show how a targeted sampling strategy using predictions from a geospatial model, combining random forests and geostatistics, can improve the sampling efficiency for identifying locations with high infection prevalence. Predictions were made based on the household locations of infected persons identified from previous surveys, and environmental variables relevant to mosquito density. Results show that targeting sampling using model predictions would have allowed 52% of infections to be identified by sampling just 17.7% of households. The odds ratio for identifying an infected individual in a household at a predicted high risk compared to a predicted low risk location was 10.2 (95% CI 4.2-22.8). This study provides evidence that a 'one size fits all' approach is unlikely to yield optimal results when making programmatic decisions based on model predictions. Instead, model assumptions and definitions should be tailored to each situation based on the objective of the surveillance program. When predictions are used in the context of the program objectives, they can result in a dramatic improvement in the efficiency of locating infected individuals.

How useful are malaria risk maps at the country level? Perceptions of decision-makers in Kenya, Malawi and the Democratic Republic of Congo

Background

Declining malaria prevalence and pressure on external funding have increased the need for efficiency in malaria control in sub-Saharan Africa (SSA). Modelled Plasmodium falciparum parasite rate (PfPR) maps are increasingly becoming available and provide information on the epidemiological situation of countries. However, how these maps are understood or used for national malaria planning is rarely explored. In this study, the practices and perceptions of national decision-makers on the utility of malaria risk maps, showing prevalence of parasitaemia or incidence of illness, was investigated.

Methods

A document review of recent National Malaria Strategic Plans was combined with 64 in-depth interviews with stakeholders in Kenya, Malawi and the Democratic Republic of Congo (DRC). The document review focused on the type of epidemiological maps included and their use in prioritising and targeting interventions. Interviews (14 Kenya, 17 Malawi, 27 DRC, 6 global level) explored drivers of stakeholder perceptions of the utility, value and limitations of malaria risk maps.

Results

Three different types of maps were used to show malaria epidemiological strata: malaria prevalence using a PfPR modelled map (Kenya); malaria incidence using routine health system data (Malawi); and malaria prevalence using data from the most recent Demographic and Health Survey (DRC). In Kenya the map was used to target preventative interventions, including long-lasting insecticide-treated nets (LLINs) and intermittent preventive treatment in pregnancy (IPTp), whilst in Malawi and DRC the maps were used to target in-door residual spraying (IRS) and LLINs distributions in schools. Maps were also used for operational planning, supply quantification, financial justification and advocacy. Findings from the interviews suggested that decision-makers lacked trust in the modelled PfPR maps when based on only a few empirical data points (Malawi and DRC).

Conclusions

Maps were generally used to identify areas with high prevalence in order to implement specific interventions. Despite the availability of national level modelled PfPR maps in all three countries, they were only used in one country. Perceived utility of malaria risk maps was associated with the epidemiological structure of the country and use was driven by perceived need, understanding (quality and relevance), ownership and trust in the data used to develop the maps.

Advances in spatiotemporal models for non-communicable disease surveillance

Surveillance systems are commonly used to provide early warning detection or to assess an impact of an intervention/policy. Traditionally, the methodological and conceptual frameworks for surveillance have been designed for infectious diseases, but the rising burden of non-communicable diseases (NCDs) worldwide suggests a pressing need for surveillance strategies to detect unusual patterns in the data and to help unveil important risk factors in this setting. Surveillance methods need to be able to detect meaningful departures from expectation and exploit dependencies within such data to produce unbiased estimates of risk as well as future forecasts. This has led to the increasing development of a range of space-time methods specifically designed for NCD surveillance. We present an overview of recent advances in spatiotemporal disease surveillance for NCDs, using hierarchically specified models. This provides a coherent framework for modelling complex data structures, dealing with data sparsity, exploiting dependencies between data sources and propagating the inherent uncertainties present in both the data and the modelling process. We then focus on three commonly used models within the Bayesian Hierarchical Model (BHM) framework and, through a simulation study, we compare their performance. We also discuss some challenges faced by researchers when dealing with NCD surveillance, including how to account for false detection and the modifiable areal unit problem. Finally, we consider how to use and interpret the complex models, how model selection may vary depending on the intended user group and how best to communicate results to stakeholders and the general public.

Mapping species richness using opportunistic samples: a case study on ground-floor bryophyte species richness in the Belgian province of Limburg

In species richness studies, citizen-science surveys where participants make individual decisions regarding sampling strategies provide a cost-effective approach to collect a large amount of data. However, it is unclear to what extent the bias inherent to opportunistically collected samples may invalidate our inferences. Here, we compare spatial predictions of forest ground-floor bryophyte species richness in Limburg (Belgium), based on crowd- and expert-sourced data, where the latter are collected by adhering to a rigorous geographical randomisation and data collection protocol. We develop a log-Gaussian Cox process model to analyse the opportunistic sampling process of the crowd-sourced data and assess its sampling bias. We then fit two geostatistical Poisson models to both data-sets and compare the parameter estimates and species richness predictions. We find that the citizens had a higher propensity for locations that were close to their homes and environmentally more valuable. The estimated effects of ecological predictors and spatial species richness predictions differ strongly between the two geostatistical models. Unknown inconsistencies in the sampling process, such as unreported observer’s effort, and the lack of a hypothesis-driven study protocol can lead to the occurrence of multiple sources of sampling bias, making it difficult, if not impossible, to provide reliable inferences.

A geostatistical framework for combining spatially referenced disease prevalence data from multiple diagnostics

Multiple diagnostic tests are often used due to limited resources or because they provide complementary information on the epidemiology of a disease under investigation. Existing statistical methods to combine prevalence data from multiple diagnostics ignore the potential overdispersion induced by the spatial correlations in the data. To address this issue, we develop a geostatistical framework that allows for joint modelling of data from multiple diagnostics by considering two main classes of inferential problems: (a) to predict prevalence for a gold-standard diagnostic using low-cost and potentially biased alternative tests; (b) to carry out joint prediction of prevalence from multiple tests. We apply the proposed framework to two case studies: mapping Loa loa prevalence in Central and West Africa, using miscroscopy, and a questionnaire-based test called RAPLOA; mapping Plasmodium falciparum malaria prevalence in the highlands of Western Kenya using polymerase chain reaction and a rapid diagnostic test. We also develop a Monte Carlo procedure based on the variogram in order to identify parsimonious geostatistical models that are compatible with the data. Our study highlights (a) the importance of accounting for diagnostic-specific residual spatial variation and (b) the benefits accrued from joint geostatistical modelling so as to deliver more reliable and precise inferences on disease prevalence.

A protocol for a multi-site, spatially-referenced household survey in slum settings: methods for access, sampling frame construction, sampling, and field data collection

BACKGROUND

Household surveys are a key epidemiological, medical, and social research method. In poor urban environments, such as slums, censuses can often be out-of-date or fail to record transient residents, maps may be incomplete, and access to sites can be limit, all of which prohibits obtaining an accurate sampling frame. This article describes a method to conduct a survey in slum settings in the context of the NIHR Global Health Research Unit on Improving Health in Slums project.

METHODS

We identify four key steps: obtaining site access, generation of a sampling frame, sampling, and field data collection. Stakeholder identification and engagement is required to negotiate access. A spatially-referenced sampling frame can be generated by: remote participatory mapping from satellite imagery; local participatory mapping and ground-truthing; and identification of all residents of each structure. We propose to use a spatially-regulated sampling method to ensure spatial coverage across the site. Finally, data collection using tablet devices and open-source software can be conducted using the generated sample and maps.

DISCUSSION

Slums are home to a growing population who face some of the highest burdens of disease yet who remain relatively understudied. Difficulties conducting surveys in these locations may explain this disparity. We propose a generalisable, scientifically valid method that is sustainable and ensures community engagement.

Geostatistical Methods for Disease Mapping and Visualisation Using Data from Spatio-temporally Referenced Prevalence Surveys

In this paper, we set out general principles and develop geostatistical methods for the analysis of data from spatio-temporally referenced prevalence surveys. Our objective is to provide a tutorial guide that can be used in order to identify parsimonious geostatistical models for prevalence mapping. A general variogram-based Monte Carlo procedure is proposed to check the validity of the modelling assumptions. We describe and contrast likelihood-based and Bayesian methods of inference, showing how to account for parameter uncertainty under each of the two paradigms. We also describe extensions of the standard model for disease prevalence that can be used when stationarity of the spatio-temporal covariance function is not supported by the data. We discuss how to define predictive targets and argue that exceedance probabilities provide one of the most effective ways to convey uncertainty in prevalence estimates. We describe statistical software for the visualisation of spatio-temporal predictive summaries of prevalence through interactive animations. Finally, we illustrate an application to historical malaria prevalence data from 1 334 surveys conducted in Senegal between 1905 and 2014.

Using non-exceedance probabilities of policy-relevant malaria prevalence thresholds to identify areas of low transmission in Somalia

Background

Countries planning malaria elimination must adapt from sustaining universal control to targeted intervention and surveillance. Decisions to make this transition require interpretable information, including malaria parasite survey data. As transmission declines, observed parasite prevalence becomes highly heterogeneous with most communities reporting estimates close to zero. Absolute estimates of prevalence become hard to interpret as a measure of transmission intensity and suitable statistical methods are required to handle uncertainty of area-wide predictions that are programmatically relevant.

Methods

A spatio-temporal geostatistical binomial model for Plasmodium falciparum prevalence (PfPR) was developed using data from cross-sectional surveys conducted in Somalia in 2005, 2007–2011 and 2014. The fitted model was then used to generate maps of non-exceedance probabilities, i.e. the predictive probability that the region-wide population-weighted average PfPR for children between 2 and 10 years (PfPR_2–10) lies below 1 and 5%. A comparison was carried out with the decision-making outcomes from those of standard approaches that ignore uncertainty in prevalence estimates.

Results

By 2010, most regions in Somalia were at least 70% likely to be below 5% PfPR_2–10 and, by 2014, 17 regions were below 5% PfPR_2–10 with a probability greater than 90%. Larger uncertainty is observed using a threshold of 1%. By 2011, only two regions were more than 90% likely of being < 1% PfPR_2–10 and, by 2014, only three regions showed such low level of uncertainty. The use of non-exceedance probabilities indicated that there was weak evidence to classify 10 out of the 18 regions as < 1% in 2014, when a greater than 90% non-exceedance probability was required.

Conclusion

Unlike standard approaches, non-exceedance probabilities of spatially modelled PfPR_2–10 allow to quantify uncertainty of prevalence estimates in relation to policy relevant intervention thresholds, providing programmatically relevant metrics to make decisions on transitioning from sustained malaria control to strategies that encompass methods of malaria elimination.

Electronic supplementary material

The online version of this article (10.1186/s12936-018-2238-0) contains supplementary material, which is available to authorized users.

Geostatistical modelling of malaria indicator survey data to assess the effects of interventions on the geographical distribution of malaria prevalence in children less than 5 years in Uganda

BACKGROUND

Malaria burden in Uganda has declined disproportionately among regions despite overall high intervention coverage across all regions. The Uganda Malaria Indicator Survey (MIS) 2014-15 was the second nationally representative survey conducted to provide estimates of malaria prevalence among children less than 5 years, and to track the progress of control interventions in the country. In this present study, 2014-15 MIS data were analysed to assess intervention effects on malaria prevalence in Uganda among children less than 5 years, assess intervention effects at regional level, and estimate geographical distribution of malaria prevalence in the country.

METHODS

Bayesian geostatistical models with spatially varying coefficients were used to determine the effect of interventions on malaria prevalence at national and regional levels. Spike-and-slab variable selection was used to identify the most important predictors and forms. Bayesian kriging was used to predict malaria prevalence at unsampled locations.

RESULTS

Indoor Residual Spraying (IRS) and Insecticide Treated Nets (ITN) ownership had a significant but varying protective effect on malaria prevalence. However, no effect was observed for Artemisinin Combination-based Therapies (ACTs). Environmental factors, namely, land cover, rainfall, day and night land surface temperature, and area type were significantly associated with malaria prevalence. Malaria prevalence was higher in rural areas, increased with the child's age, and decreased with higher household socioeconomic status and higher level of mother's education. The highest prevalence of malaria in children less than 5 years was predicted for regions of East Central, North East and West Nile, whereas the lowest was predicted in Kampala and South Western regions, and in the mountainous areas in Mid-Western and Mid-Eastern regions.

CONCLUSIONS

IRS and ITN ownership are important interventions against malaria prevalence in children less than 5 years in Uganda. The varying effects of the interventions calls for selective implementation of control tools suitable to regional ecological settings. To further reduce malaria burden and sustain malaria control in Uganda, current tools should be supplemented by health system strengthening, and socio-economic development.