Estimation of malaria parasite reservoir coverage using reactive case detection and active community fever screening from census data with rapid diagnostic tests in southern Zambia: a re-sampling approach

In areas where malaria transmission has been suppressed by vector control interventions many malaria control programmes actively seek new interventions to further reduce malaria prevalence, incidence, and transmission. Additionally, programmes which are considering or have undergone a re-orientation towards malaria elimination may be seeking interventions to actively reduce or eliminate remaining foci of infection. Malaria infection prevalence and incidence have been shown to cluster geographically, especially at lower transmission levels [1‐3], and as such a reactive strategy might be utilized by which index cases identified by a passive surveillance system are used to target small areas for malaria testing and treatment/investigation [reactive case detection (RCD)] or focal drug administration (fDA). The principles behind the deployment of these types of reactive strategies derive from similar epidemiological foundations to ring vaccination [4] and might provide efficient ways to target mass drug administration (MDA) or mass testing and treatment (MTAT) interventions to small geographic areas with unusually high infection or transmission rates.

Little is known, however, about the effectiveness of the use of reactive approaches to guide MDA and MTAT, and few studies have made an attempt to estimate the potential of reactive strategies to cover high proportions of the reservoir of infections [5‐7]. A number of studies have summarized the prevalence of malaria amongst household contacts of passively detected index cases, however, these studies did not include an appropriate comparison group, thus the gain in reservoir coverage of RCD, or related approaches, in those settings could not be ascertained [2, 3, 8‐10]. While malaria is known to cluster geographically, the level and nature of clustering may vary with prevalence, population malaria exposure history, geographic features of the landscape, the built environment, human settlement patterns and various other factors [11].

In order to effectively plan for the deployment and testing of RCD or fDA strategies, when and if they are used, it will be necessary for malaria control programmes to make local assessments of both the expected coverage of the intervention in operational and parasitological terms as well as the resource requirements and the ideal search strategy to use in the response to index case identification.

In Zambia, malaria vector control, intermittent preventive treatment in pregnancy (IPTp) and treatment with artemisinin-based combination therapy (ACT) have been scaled up nationwide. These interventions, along with community case management, have reached high and sustained coverage, resulting in significant reductions in the malaria burden in some areas [12, 13]. In the wake of these successes the Zambian National Malaria Control Centre (NMCC) and partners are in the process of testing new strategies to further reduce the malaria burden with the ultimate goal of malaria elimination and the shorter term goal of creating malaria free areas within the country. These strategies include the expansion of community health worker (CHW) led case management and CHW led RCD. In order to rationally plan and select interventions to achieve these goals, operational, effectiveness and cost-effectiveness information about these relatively un derstudied reactive strategies is urgently needed.

This study utilizes geo-located data from a census of 23 health facility catchment areas with parasitological testing and treatment-seeking data collection, which was conducted in southern Zambia. These data were used to estimate the coverage of RCD or fDA in terms of the total population and the parasite reservoir as well as the operational requirements of such strategies, using a re-sampling algorithm developed exclusively for this purpose.

Based on these re-samples, the likely coverage and resource requirements for RCD in the context of southern Zambia was estimated. Figure 3 shows a small scale representation of the results of a single simulation. These results are for the first dry season round only.

RCD and related strategies are used mainly by malaria elimination programmes, and have been deployed widely in Asia though only to a limited extent in Africa. The literature on operational issues, reservoir coverage and programme effectiveness is limited, but has expanded in recent years [5, 6, 11, 16‐23]. While some studies identify added value in RCD approaches [18, 22, 23] others identify serious limitations and low parasite reservoir coverage despite large operational efforts [17, 19, 20]. This study used data from a census of the population of part of Southern Province Zambia with parasite detection combined with a novel computer algorithm to estimate the coverage of the parasite reservoir and the operational requirements to conduct RCD and related malaria control interventions. These results show that RCD or active fever detection coupled with RCD and fDA have potential to reach significant portions of the malaria parasite reservoir. However, they also point out some serious challenges with these approaches. These include: (1) that meaningful fractions of the parasite reservoir can be found in a short time period only when large numbers of households and individuals are reached, and (2) using the search radii considered, clustering of RDT positive malaria infected persons led to some efficiencies in parasite detection using an RCD approach but in most of the scenarios considered these were surprisingly small compared to random searches of the same areas, and (3) the RCD process is greatly hampered by low levels of treatment-seeking for fever in health facilities which would be used for identification of index cases in a standard passive-active RCD approach.

While a shift from RCD to an active fever search would mitigate some of the problems with low treatment seeking, it would also require wide population sweeps to identify persons with a history of fever or a current febrile illness as candidate index cases. Such an approach would be operationally challenging and costly to carry out on large scales. Other strategies to change treatment-seeking for febrile illness including the expansion of the health system and behaviour change communication may also be options. For example, fever treatment could be improved by extending testing and treatment services into communities and closer to areas where malaria infections occur. Other methods to increase the number number index cases generated such as decreasing the specificty of criteria for selection of index cases at the facility could also improve reservoir coverage.

The data represents a time when no RCD system was in place in the MTAT areas and when the transmission was much higher than currently in Southern Province. In simulations where reported treatment seeking was used these were based on people attending health facilities, not to local CHWs or other community case management (iCCM) implementers. While the treatment-seeking probability was explored in sensitivity analysis, results of improving treatment-seeking for the average person with fever may not reflect the same patterns as occur with the roll out of CHWs as part of iCCM. Currently RCD systems are being piloted in Zambia and expansion of the health system through the roll out of iCCM has also occurred [24, 25]. These results indicate that RCD in these contexts may indeed capture a significant portion of the parasite reservoir, at least in low transmission areas similar to Southern Province Zambia.

One earlier study used population level summaries from southern Zambia survey data to develop an agent-based transmission model to simulate the population data to which an approach to RCD similar to the one described here was applied [6]. This study takes a different approach by applying re-sampling directly to the census data. Additionally, their study was based on data from four catchment areas while the current study is based on data from 23 catchment areas.

Another earlier study using survey data from southern Zambia used a combination of logistic regression methods, and geographic analysis to estimate the proportion of infected individuals living within a specified radius of a household with a positive malaria RDT result for a person who was positive and sought care [7]. The Searle et al. study, however, had to impute most of the malaria diagnostic, symptomatic and treatment seeking results utilized in the analysis of RCD efficiency because only sample survey datasets and household locations were available to them. In addition, they did not consider the sensitivity and specificity of diagnostics used in the index case identification process, nor did they consider search radii smaller than 500 m or imperfect coverage in the search process. These results indicate that search radii of 500 m or more would result in large operational requirements in southern Zambia and significant amounts of overlap between search areas for different index cases. Additionally, the operational coverage [proportion of households actually searched (of those who should have been searched) and individuals searched with these households] achieved during RCD or related activities is an additionally important parameter central to the functioning of these systems.

This study is limited by several threats to external and internal validity. One major factor is that, though it utilizes census data, which mitigates some limitations of previous studies which had to impute outcomes for a majority of households [7], it still relies on cross sectional data, limiting its’ ability to estimate the coverage of RCD and related interventions when repeated over longer time frames. Only well designed prospective studies of the intervention in context could show this. Secondly, the diagnostics used in the census data collection on which this study is based were RDTs. These tests are known to have limited sensitivity for low density infections [26]. As such, this study may have a limited ability to project the achievement of coverage when such low density infections are included, which may be possible with the development of new highly sensitive RDTs. The direction of bias that arises from this limitation will depend on the nature of clustering of these low density infections in relation to index cases. Unfortunately, it is not possible using the data currently available to be certain whether this bias would result in a lower proportion of all infections being identified with RCD or a higher proportion compared to the results we find using only RDT results. Finally, while these datasets derive from large populations with widely varied parasite prevalence, they come from only one part of Zambia. The results of applying RCD or related interventions to other locations in Africa or the world may be very different, even in terms of coverage achieved, due to variation in human behaviour and settlement patterns and epidemiology of malaria transmission.

This study only estimates the reservoir coverage that might be expected in RCD and related systems over a short period of time (~2 weeks). This neither implies that coverage will remain low over extend periods should RCD or related approaches be sustained, nor that there will be no effect on transmission even at low reservoir coverage (though results for previous studies indicate that even high reservoir coverage in a short period are unlikely to have major effects in this setting [14]). Further simulation, modeling and evaluation work around RCD systems can incorporate these results in parameterization and as guidance on measurement. As such these results should be important to development and implementation of RCD and related approaches in malaria control and elimination programmes.

A number of additional factors will be important to consider in future work in this area. These include generating field trial data and mathematical model based estimates of the effectiveness and efficacy of RCD and related strategies, and identifying the determinants which may affect coverage. Some modelling work on limited data sets from this area of Zambia has already been conducted [6].

Additionally, as this study has shown that the resource requirements of conducting RCD and similar strategies may be significant, a finding consistent with qualitative and quantitative work already undertaken in Zambia [21, 27]. An important next step based on these results will be to measure and attach explicit cost functions to these or similar analyses to determine, under budget constraints, what the optimal parameters of an RCD or fDA intervention (i.e. what search radius) are in a given context.

RCD, fDA and active fever screening followed by RCD will only yield limited coverage of the RDT positive parasite reservoir over a short time period. Use of reactive strategies as routine tools for an extended period may increase this proportion. Reactive strategies with a fixed radius around an index case detect a higher proportion of the reservoir of infections than similar searches around randomly selected househods, but this effect appears to be greater in areas of low, but not moderate malaria prevalence in southern Zambia. Changes in the detection limit of RDTs could also affect results. The number of individuals who need to be searched, and thus the resource requirements to do so increase rapidly, but approximately linearly with search radius. RCD, if implemented in southern Zambia, would yield higher fractions of the reservoir detected with similar effort if targeted to areas with prevalence less than 10%. Increasing the probability that febrile individuals seek care or the search radius around index cases can both increase the proportion of the reservoir covered by RCD and related strategies, however, both approaches will increase the coverage most quickly when they start from low levels, and neither appears to greatly increase the fraction detected once moderate levels of sensitivity or treatment seeking are reached. The success of an RCD system appears highly dependent on its ability to actually search the houses and individuals that are within target areas for additional malaria infections—programmes implementing these strategies should not neglect the operational aspects of these systems.