Improving the efficiency of reactive case detection for malaria elimination in southern Zambia: a cross-sectional study

In countries where a state of low-endemicity for malaria has been established and maintained, strategies and policies geared toward elimination of both vector and parasites have begun to take form. An essential component that has been incorporated in many of these elimination strategies is malaria control using environmental risk factors [1]. These strategies take advantage of the spatial distribution of malaria, which varies depending on the ecology and population, but in regions with low endemicity is often concentrated in small, isolated areas or “hot spots” comprised mostly of asymptomatic individuals still infectious to mosquitoes [2, 3]. Many malaria endemic countries have a surveillance system in place for identifying symptomatic cases in real-time (passive case detection or PCD); however, this system fails to reach asymptomatic individuals [1]. Active case detection (ACD) is a surveillance method recommended by the World Health Organization (WHO) in low transmission settings in which symptomatic and asymptomatic individuals are screened and treated for malaria [4]. Reactive case detection (RCD) is a form of ACD that was designed to take advantage of the spatial and temporal clustering of asymptomatic individuals within “hot spots” by using passively detected cases as triggers to initiate screening and treatment of individuals living in proximity to those cases [5, 6]. RCD is implemented in many countries working towards malaria elimination, including Zambia [6], South Africa [7], Brazil [8], Cambodia [9], and India [10].

The application of RCD in many of these settings differs in important features, such as the optimal screening radius and the number of households investigated [10]. In each instance, however, RCD is labour and time-intensive, requiring significant human resources, many rapid diagnostic tests (RDTs), and often long travel times between households [10, 11]. The utility of RCD in low transmission settings has been debated in part due to operational challenges during implementation and the use of less sensitive diagnostic tools such as RDTs and microscopy, which miss low density infections [1, 2, 10, 12]. Other limitations of RCD are its inability to reach populations with poor access, as well as the reliance on incident symptomatic cases seeking care to find “hot spots” comprised of asymptomatic individuals [10, 12].

Residual transmission in “hot spots” is driven by many local environmental factors such as vegetation and availability of aquatic habitats that determine vector density and heterogeneity [13]. For example, Anopheles larval sites contract and cluster around permanent aquatic habitats during the dry season, and expand during the wet season [14]. Various topographical features can also predict incident cases [13]. To increase efforts towards elimination, RCD may be improved by including environmental risk factors into the screening process, leveraging the heterogeneous nature of malaria transmission as a function of environmental features to guide asymptomatic case detection [13].

The Government of Zambia launched their RCD strategy in 2011 to enhance malaria surveillance and engage health systems at the community level to identify and treat individuals infected with Plasmodium falciparum who did not seek care or had minimal or no symptoms [15‐17]. This RCD strategy is part of the National Malaria Elimination Strategic Plan (NMESP) to eliminate malaria in Zambia by 2021 and is employed in communities where parasite prevalence is approximately 1% and ten or fewer cases are passively detected at health facilities [15, 16, 18]. In Zambia, RCD starts with passive detection of a symptomatic malaria index case using P. falciparum histidine-rich protein-2 (PfHRP2) RDTs at a rural health clinic or by community health workers (CHWs) at rural health posts. CHWs then perform household visits to screen-and-treat residents within the index household as well as neighbouring or secondary households within a 140-m radius [15].

Studies have shown that environmental risk factors can be used to identify households likely to have parasitaemic residents [19‐24]; however, the use of such environmental risk factors have not been explored in southern Zambia in the context of RCD. Building on prior work that assessed the efficiency of RCD in southern Zambia, the predictive ability of environmental risk factors was evaluated at varied spatial scales to identify parasitaemic residents of households located within a larger radius of 250 m from an index household [18, 25, 26].

Environmental risk factors were associated with the probability of finding households with parasitaemic residents using RCD as demonstrated in other studies in Zambia [13]. In the low transmission setting of Choma District, Zambia, identifying streams located near index households to guide and direct screening has the potential to improve RCD and affect transmission by identifying households with asymptomatic infections. These findings are in line with a previous study conducted within the same study area in 2008 where it was shown that households within 1.98 km from a third-order stream were 2.8 (95% CI 1.2–6.9) times more likely to have an RDT positive resident than those within 6 km [26].

Although no associations were found with the other environmental risk factors such as distance to a main road, elevation, season, and number and presence of animal pens, non-parametric comparisons between positive and negative secondary households exhibited a trend of increased malaria risk for these risk factors [19‐21, 26, 36‐38]. The risk associated with animal pens varies in the literature depending on vector behaviour. Anopheles arabiensis has been reported to be anthropophilic in southern Zambia but also displays zoophilic habits by feeding opportunistically on non-human blood sources [36]. Other vectors besides An. arabiensis might also have an important role in transmission as P. falciparum-infected Anopheles squamosus exhibiting outdoor zoophagic feeding behaviour were recently identified in the area. Early studies in Choma District, Zambia found that ownership of cattle reduced the risk of P. falciparum infection by 87% while others have found less conclusive evidence [36, 37]. For elevation, however, it has been clearly shown that increased elevation offers protection against malaria infection [13, 20, 24, 26, 38]. Since index and secondary households in this study were located only < 300 m from each other and variation in elevation was minimal, it is unlikely that the elevation would influence malaria risk at this spatial scale. Distance from the index household marginally increased the probability of finding positive secondary households (OR: 1.24, 95% CI 0.98–1.58), in contrast to other studies. Larsen et al. observed a decreased risk for neighbouring households located further away, and Bulterys et al. found an adjusted OR of 0.26 (95% CI 0.07–0.98) as distance between households increased. Finally, distance to the main road has often been treated as an indicator of increased malaria risk. In Chongwe District, Zambia, the odds of RDT positive households increased by 5% for every 500-m increase in distance from the road [39]. As only proximity to the main road was examined, it is possible that constant use from vehicles, animal carts, and people prevented mosquito breeding sites from developing undisturbed, reducing this as a risk factor. Less frequently used subsidiary roads and rural paths (not included in the analysis) could provide more opportune mosquito breeding sites closer to residences as their composition allows for easier accumulation of aquatic breeding sites compared to the tarmac and concrete main road.

The use of environmental risk factors for malaria risk prediction is a common approach to malaria control and has been employed in various transmission settings around the world. For example, in Chimoio, Mozambique a GIS-based spatial model was designed to estimate areas of risk using temperature, precipitation, altitude, slope, distance to water bodies, distance to roads, normalized difference vegetation index (NDVI), land use, and land cover [40]. The model identified that 4% of Chimoio was at high risk for contracting malaria, with precipitation as a key risk factor for the entire area studied [40]. Another study in south Sumatra, Indonesia used ordinary least square and geographically weighted regression to show that altitude, distance to forest, and rainfall determined overall malaria incidence with considerable heterogeneity at the village level [41]. These findings were consistent with other studies in Cambodia, Addis Ababa, Ethiopia, and Rondôia, Brazil [41]. Despite the extensive literature on environmental risk factors for malaria, their application within the context of RCD has been limited.

Many studies evaluating the efficiency of RCD highlight its inability to halt infections in areas of low transmission due to the use of less sensitive RDTs, travel-related infections, and large budgetary requirements [2, 18]. A major concern for RCD-based strategies is that asymptomatic individuals will be missed if no clinical cases report to CHWs [42]. A survey in coastal Kenya found that asymptomatic and symptomatic infections do not necessarily overlap spatially, and that clusters of symptomatic infections have greater temporal stability over more than 10 years [42]. Another issue often highlighted is the different criteria and screening radii employed by countries to define and recruit neighbouring households [42] For example, RCD data from four villages in the Myanmar-Thailand border determined that RCD would only be successful at a radius of 150 m, and any screening occurring beyond this radius would not perform better than random screening [2]. Another study in Pailin Province in western Cambodia screened the nearest five households for every fifteenth index case and ten nearest households for every 30th index case. Using this approach, they predicted only 40% of infections and concluded that RCD was not recommended in a setting targeting elimination [43]. However, with the shortcomings of a circular radius and the various implementation challenges, for RCD to be an effective method for malaria elimination in these low-endemic countries a tailored approach adapted to the local parasite epidemiology, vector biology, and living/working environment of the community must be considered key for it to succeed.

This study used environmental risk factors for malaria to characterize the low transmission setting to improve RCD efficiency. Previous work on enhancing RCD efficiency in Southern Province, Zambia has also shown that time-invariant measures of the environment, such as the topographic position index (TPI; measure of an area’s relative elevation to find slopes, valleys, and ridges), the convergence index (CI; measure of an area’s propensity to pool water), median enhanced vegetation index (EVI; measure of vegetation density), and the topographical wetness index (TWI; measure of water flow) were stronger predictors for identifying parasitaemic individuals than demographics of incident symptomatic cases [13, 26]. They showed that ridges and upper slopes (at a TPI scale of 270 m) and wetter regions (TWI > 10.2) were associated with finding more parasitaemic individuals during RCD [13]. These findings, along with the current study, support the significance of water bodies in improving the efficiency of RCD. Third through fifth-order streams are mid-level streams that may not always be suitable for larval development; however, larvae have been collected from water at the edges of these streams (unpublished findings). During the dry season, as water accumulates into smaller pools, they become ideal larval development sites. These streams can also serve as important markers for nearby areas with similar high water table harbouring larvae [13, 26, 38, 44]. And as these streams can be challenging to locate depending on size and season, spatial risk maps with topographical measures, such as CI and TWI, can offer guidance to CHWs to possibly reach clusters of asymptomatic carriers otherwise missed during regular RCD screening. Other water sources such as dams, are also important determinants for malaria transmission as was shown in Ethiopia, where reservoir water level management suppressed larval development [45].

In addition to the use of streams as a screening tool, RCD efficiency could benefit from the combined use of RDTs and highly sensitive qPCR. For this study region, the overall parasite prevalence (3.7%) was mostly driven by qPCR as parasite prevalence by RDT was only 1.3%. Although costly, sensitive molecular methods such as qPCR are critical in low endemic settings to detect potential parasite-transmitting asymptomatic carriers. Even ultra-sensitive RDTs (uRDTs), such as the new Alere™ Malaria Ag P.f uRDT which was designed for low transmission settings, may not be sufficiently sensitive alternatives to SD Bioline PfHRP2 RDTs [46]. The Alere™ us-RDT has a ten-fold lower limit of detection for PfHRP2 compared to regular RDTs but missed 56% of PCR-detectable P. falciparum infections in a low-endemic setting in Myanmar, and in Papua New Guinea the test missed 50% of P. falciparum infections otherwise detectable by qPCR [47].

There were several limitations to this study. Restricting environmental variables within set radii raises concerns for edge-effect associations. For example, animal pens located just outside the 100-m radius were not counted as belonging to neighbouring households, thus potentially underestimating the number of animal pens associated with a household. Not all environmental risk factors important for malaria transmission were evaluated. Vegetation cover, an important indicator of available mosquito habitat, could also be a useful screening tool and has yet to be evaluated for RCD strategies [13, 48, 49]. Finally, the risk factors shown to be associated with positive households in this low transmission setting of Choma District, Zambia may not be applicable in other endemic regions.

The effectiveness of RCD ultimately depends on the number of cases found and treated in a timely manner and the resources allocated during implementation [2]. However, it is important to consider the added value of a tailored RCD approach based on demographic and ecological risk factors and more sensitive diagnostic tools to fully reap the benefits of this screening method to achieve malaria elimination [1]. In Cambodia, where infection is linked to occupation and mobility, an expanded RCD approach was implemented in which individuals who were coworkers of a symptomatic index case in settings of high malaria infection, such as forests and plantations, were also screened [1]. The expanded RCD had a detection rate of 3.9% compared to 0–2% using the classic RCD approach [1]. Through this adapted RCD design, Cambodia’s National Malaria Control Programme sought to identify and treat asymptomatic individuals within a discrete population whose members shared a common malaria risk through occupations such as logging, mining, and migrant labour [1]. The environmental risk factors identified in this study demonstrate that, even in low transmission settings, a tailored approach is possible; however, further work is needed to fully understand how these risk factors vary across district and season and how they can be modified to guide RCD strategies nationally in Zambia.

This study identified higher order streams as risk factors for parasitaemia in households neighbouring an index case as part of RCD in rural southern Zambia. These risk factors have the potential to improve the efficiency of RCD in a low transmission setting by not only identifying parasitaemic individuals more efficiently but also potentially reducing the number of households needed to be screened. Combined with other strategies for malaria elimination, tailored RCD approaches can help realize the goal of malaria elimination in Zambia.