Approach
We developed a prioritization methodology that can be used for systematically selecting areas that need to be targeted in an appropriate sequence with ITN distributions while taking into consideration malaria transmission risks, available resources, ongoing emergencies, and other relevant factors. Whereas the exercise was conducted for Nigeria, we present here a general approach that can be adopted by any country or region experiencing malaria transmission that has: (1) access to subnational country-specific malaria intervention data; (2) a reasonably reliable COVID-19 surveillance system; (3) conducted a Demographic and Health Survey (DHS) or Malaria Indicator Survey (MIS) within the past three years; (4) access to open-source spatial layers featuring data pertaining to covariates such as P. falciparum environmental suitability indices, educational attainment, and population density; (5) available conflict and/or disruptive event data (if relevant); (6) insecticide resistance data; and (7) other country-specific information or data relevant to malaria risk. Using a combination of data derived from these elements, we were able to calculate risk scores for each administrative unit within a given country and rank these units to inform prioritization.
The basis of our methodology is derived from a strategy developed by Hanafi-Bojd et al., where a geographic information system (GIS)-based weighted arithmetic and multiplicative approach was used to categorize and rank administrative units based on malaria hazard and risk in the context of targeting interventions in Iran, which experiences P. vivax and P. falciparum transmission 5. Additional conceptual frameworks and methodologies that identified vulnerability and potential hazards from systematic review and expert consultation which led to the creation of spatially explicit malaria risk maps were considered 6–8. Similar indicator weighting methodologies and the creation of a malaria vulnerability/risk index were employed. In lieu of utilizing weighted indicators derived from the coefficients of regression analysis 6, indicator weights were calculated utilizing expert opinion surveys administered among regional and global field experts.
The preceding approaches were adapted for application in settings where P. falciparum is the predominant malaria parasite, while considering practical needs for additional malaria resources due to the COVID-19 pandemic. Therefore, P. falciparum-specific factors were utilised in calculating final risk scores, along with COVID-19 case burden and other factors that have impacts on malaria control. Unlike the approach used by Hanafi-Bojd et al., we focused on producing a single cumulative prioritization map (after quantifying and weighting multiple risk factor) for each of the country, as opposed to individual malaria risk and hazard maps. We did, however, combine similar drivers of falciparum malaria transmission with other malaria risk factors, including intervention coverage, population density, and presence and density of disruptive events. These risk factors were used to characterize administrative units according to their risk of malaria transmission based on their vulnerability for breeding and maintenance of malaria vectors, levels of intervention coverage, and social and biological susceptibility factors.
The steps below summarize the approach used to calculate the final risk score for each administrative unit (e.g., district, local government area, state, etc.). Step-by-step details of the implementation of the approach using the R software are provided under a separate section (Calculation of prioritization scores).
- Most relevant risk factors that need to be considered for prioritization of planned vector control interventions (e.g., ITN campaign) were identified. The types and number of factors used for prioritization could differ between countries and interventions.
- The possible range of values of each factor were classified into a number of classes (typically four). For example, a factor such as "IRS implementation in 2020" has two classes ("Yes" and "No"), whereas "% households with at least 1 ITN for 2 people" could have four classes (e.g., <35%, 35-49%, 50-67%, and >67%). Where appropriate, the cut-off values were determined using "natural breaks" (where classes are based on natural groupings inherent in the data) as obtained using the R software (R Studio version 4.0.3).
- Rank values, defined here as whole numbers typically ranging from 1 to 4 (but can be 0 and 1 for a binary factor, denoting absence of presence of a particular intervention, for example), were assigned to each class of each factor. The higher the rank value the greater the malaria risk associated with that class. For example, the rank value of the class "<35%" for the factor "% households with at least 1 ITN for 2 people" mentioned above would be 4 whereas ">67%" would be assigned a rank value of 1.
- A short survey questionnaire was circulated among malaria professionals and academics with malaria research experience to ask them to rank each factor on a scale of 0 - 10 in terms importance for quantifying malaria risk. An average rank was calculated for each factor, and then standardized on a scale of 0 - 1 by dividing it by 10.
- Each administrative unit was then assigned rank values for each factor, which was multiplied by the respective weight of the factor to obtain the risk scores. Risk scores of each factor were then summated to generate a composite prioritization score for each administrative unit.
Classification intervals featured in Table 1 were calculated using data from Nigeria as an example. We aim to provide a generalized approach that can be adopted in other counties that may have access to different types of data or may adopt different malaria control activities. For example, SMC is typically conducted only in limited regions, therefore this data would not be relevant for countries in other regions of Africa where it is not implemented.
Data inputs
The data related to each of the factors were obtained from various sources, including government documents, such as national malaria control strategic plans, ITN operational plans, vector control coverage data, and entomological and epidemiological reports. Literature covering models and approaches used to quantify and target ITNs in malaria endemic settings were also reviewed in detail 5–8. These documents were used to inform the approach utilised for modelling and targeting ITN distributions.
A combination of environmental covariate data, Demographic and Health Survey (DHS) data, intervention coverage, internally displaced population data and predicted surface layers were used as the primary data inputs to generate prioritisation maps for Nigeria. A temperature suitability index for P. falciparum transmission was obtained from the Malaria Atlas Project (MAP). Mean annual rainfall and rainfall anomalies for each administrative unit were calculated using Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) data. ITN coverage data was provided by Nigeria’s National Malaria Elimination Programme (NMEP), along with projected population estimates for years 2019 and 2020. Built-up area presence data from the Global Human Settlement Layer (GHSL) Project produces global spatial information about the human presence on the planet over time and relies on automatic analysis of satellite imagery to produce fine-scale maps quantifying built-up structures in terms of their location and density and can be utilised. Built-up area presence was used as a proxy for classifying rural and urban areas. Intervention coverage (including ITN availability, or the proportion of households with at least 1 net per 2 people) was interpolated using survey data collected through the most recent DHS or MIS. Data from 2019 featuring distribution of internally displaced persons (IDPs) and high-risk populations (HRPs) per administrative unit was obtained from the Humanitarian Data Exchange.
Spatial interpolation
A major component that influenced our prioritization scheme was previous ITN depth of coverage, as obtained and spatially interpolated using data from DHS or MIS surveys. Using the R-INLA package a Bayesian inference spatial smoothing approach called Integrated Nested Laplace Approximation with Stochastic Partial Differential Equation (INLA-SPDE) was used to fit a geostatistical model predicting the proportion of households with at least 1 ITN per 2 de facto household population at unsampled locations. This process was conducted using cluster-level input data from the most recent DHS or MIS which collected information on household-level ITN ownership. DHS and MIS clusters were georeferenced and cluster-level estimates served as marked data points. DHS and MIS data generally allow for the calculation of representative estimates at the national, regional, and urban/rural levels, but the INLA process allows for pixel-level estimates which can be aggregated to establish subregional estimates based on available administrative boundaries. A logistic regression model was fit in INLA to estimate the probability of ITN availability across the mesh while accounting for spatial autocorrelation. Also included in the INLA function was an estimation stack comprised of the cluster-level ITN estimates as well as corresponding covariate estimates. Raster values were extracted for each DHS or MIS cluster point using a 2km buffer for urban clusters and a 10km buffer for rural clusters. The final output was a surface layer featuring predicted proportion of households with at least 1 ITN per 2 persons. This layer was combined with other factors, as listed in Table 1, to generate a final prioritization score for each administrative unit. The following section details how this was achieved.
Calculation of prioritization scores
Here we present a step-by-step framework illustrating our method for calculating prioritization scores in R Studio version 4.0.3, including only some of the important variables that could be used in most malaria endemic countries (especially in Africa) in Table 2. Other (country-specific) factors could be included in the calculation alongside the input variables described below. Syntax for each step, including a list of required packages for calculation of prioritization scores in R Studio, can be found in the GitHub repository listed in the supplementary files.