Limitations of malaria reactive case detection in an area of low and unstable transmission on the Myanmar–Thailand border

In regions with decreasing malaria burden, the distribution of malaria becomes increasingly patchy and heterogeneous across both landscape and population [1, 2]. Given increased efforts towards elimination and evolving epidemiological characteristics, novel approaches at identifying and treating malaria cases may be necessary for disrupting transmission [3].

Reactive case detection (RCD) uses information about passively detected malaria index cases in order to target potential secondary cases [4]. When a symptomatic malaria patient is diagnosed at a health facility, health care workers screen household members, neighbours and other related community members for malaria. This form of case detection is an intuitive approach in that if cases of malaria cluster in space and time, then it may be efficient to target those clusters in order to disrupt transmission [5, 6]. Such an approach is not considered feasible or efficient in high-transmission settings, but may be appropriate for interrupting transmission in a patchy environment [7, 8].

Several studies in a variety of settings (e.g., Zambia, Peru, Thailand) have found strong clustering of malaria within households [9–12]. From this finding it seems logical that reactively screening or presumptively treating individuals who share households with index cases will result in the treatment of cases that could have otherwise been missed. Several projects have explored screening of neighbouring households at varying radii around an index house. For example, studies in Sri Lanka [13] and Swaziland [14, 15] led to a suggested 1-km radius around index households for RCD. Modelling work from Zanzibar suggested that RCD would need to target approximately 100 neighbouring households around each identified case in order to halt transmission [16]. Data from Zambia have suggested clustering of cases within households and up to 50 m around index households [12]. Modelling work from Zambia has suggested that more than three-quarters of cases may be detected through reactively screening in a 500-m radius around index households [17].

In most nations of the Greater Mekong Subregion, RCD is listed as a component of national strategic plans for malaria [18]. However, it is most commonly used as part of research projects with external funding rather than by government officials. If the approach is effective it may be important to consider its inclusion in routine malaria control and elimination activities. Conversely, if the approach fails to halt transmission it could divert important financial and labour resources that could better be directed to other approaches. The relative effectiveness of the RCD approach lies in a balance between the proportion of cases it can detect and treat in a timely manner and the feasibility of the approach, especially with regard to resources (funding, labour, social concerns) needed in order to reach that proportion [12].

Here, empirical clinical malaria case data were used to simulate RCD strategies using varying screening radii, time intervals between activities, and settlement characteristics in order to assess the effectiveness and efficacy of RCD to disrupt transmission along the Myanmar–Thailand border.

Infectious diseases sometimes cluster in space and time and if it is possible to predict the location of clusters it may also be possible to halt transmission through early diagnosis and treatment using an RCD approach [1, 6]. RCD is now a part of the national strategic plans of most Greater Mekong Subregion countries, though the approach (e.g., how RCD is triggered, the screening radius used) varies greatly by location [18].

The results here indicate that in houses with consecutive malaria cases, recurrent falciparum cases occur more quickly in comparison to vivax cases. Using a 1-month time step would have led to detecting a higher proportion of falciparum cases within index houses when compared to vivax cases. This may indicate within-house clustering of falciparum malaria. Also, many vivax infections in these populations are the result of relapse and such cases are less likely to exhibit strong spatial clustering. However, the results of using screening radii from 150 m and above are similar between P. falciparum and P. vivax cases. At this radius the majority of cases would have been detected while screening roughly three-quarters of the houses for both falciparum and vivax malaria. When the time step was decreased to a week it resulted in the same result for vivax (detected most cases while screening about three-quarters of all houses at 150-m radius), but would have only detected about half of all falciparum cases.

In this low-transmission context, 50% of screenings would have occurred in months during which there would have been no falciparum cases to find (median time between index and secondary cases within houses: 39 days; within a 150-m radius: 21 days, Table 3). When there were cases to be found, there was no optimum radius that maximized the number of cases found compared to number of houses treated: house-level screening led to a minimal number of houses screened but only identified 14% of secondary falciparum cases. On the other hand, above 150-m radius, a large proportion of cases was found by screening a large proportion of the village. At such a large proportion of village screened, targeted screening around an index case only had a marginal gain compared to randomly screening the same proportion of village households. These results are consistent with recent findings from Pailin Province, Cambodia where few cases were detected through reactively screening index houses and neighbours [23].

The relative success achieved in detecting larger proportions of cases at larger screening radii must also be weighed against the costs incurred through using larger radii. Screening requires labour and medical supplies (e.g., testing and diagnosis materials). If cases occur within a single village at different times it means there will be overlap in screening. A house that was tested in a previous screening could fall within the screening radius of a subsequent index case, requiring repeated screening. Not only does this require the use of large amounts of medical supplies and labour, it would also mean that participants may need to give blood frequently.

Unless transmission is relatively high, most people who are screened will test negative using standard RDTs. If people are asked to frequently take the time to give their blood and be tested for malaria and they are only rarely malaria positive, it could lead to participation fatigue and even refusal. The same situation can lead to fatigue amongst community-based health workers who may see their efforts as relatively unfruitful (given low numbers of cases that would normally be detected) and even unwanted from the community in a scenario of increasing participant avoidance and refusal. The results could be detrimental not only for malaria efforts but also efforts towards control or elimination of other diseases.

The spatial distribution of malaria cases in groups of houses can arise through complicated transmission patterns. For example, most falciparum cases are likely to be new infections whereas many vivax infections are likely to be relapses. Many vivax cases are therefore the result of exposure that happened in the distant past and clustering is unlikely. Exposure to infectious mosquitoes may occur in different locations. If transmission occurs within the village, clusters of houses with cases can be expected around mosquito breeding sites [6]. In contrast, if transmission occurs outside of the village, e.g., in outlying farms, then patterns within the village are related to shared exposures outside of the village. For example, clustering within households may be the result of shared exposure among household members outside of the village. If most transmission occurs outside of the village there may be very little clustering of cases between houses and spatial targeting within the village may not be effective or efficient. These data indicate spatial clustering within and across houses at very brief points in time, with dispersed spatial patterns in between these brief points in time.

The proportion of symptomatic cases detected increases with the screening radius size. At radii in which most cases are being detected (for example, 150-m radius) most of the inhabitants of the afflicted communities would also be tested. If there are multiple index houses, and if those houses are not clustered in one portion of the village, a screening radius of 150 m is sufficient to cover most of the village. At larger screening radii the location of the index house(s) is increasingly irrelevant as the majority of the village will be included. At 200-m radius, targeted screening is no better than randomly selecting the same proportion of houses.

Furthermore, population densities of these villages and many others in the region change seasonally. During times of the year when farming work is at its peak, usually during the rainy and malarious season, many villagers spend long hours in farm fields and spend nights in their farm huts. The population density of the village is dispersed at these times in an anisotropic fashion. Farms are not evenly distributed around villages but rather follow the contours of the landscape (Fig. 1), along fertile soil patches and in places that are deemed appropriate (e.g., legally and socio-culturally) for farming. Cases can occasionally occur within groups of farm huts. The screening radius necessary to detect high proportions of clinical cases increases heterogeneously and is dependent on the settlement patterns of farm huts and the locations of clinical cases. While these measurements of villager destinations outside of the village are most likely an underestimate, they already show that RCD is limited when accounting for movements outside of the village.

There are several limitations to this work. As with RCD, the clinical cases in this study were detected through the use of RDTs and symptomatic patients that presented at community-based clinics. RDTs miss low-parasitaemia cases, which may be important for ongoing malaria transmission. The simulations also miss those asymptomatic cases with high enough parasitaemia to be RDT positive. Mass blood screenings in the same study villages revealed 71% of falciparum and 82% of vivax RDT-positive villagers to be without fever. These findings are likely to be a conservative estimate of the proportion of symptomatic RDT-positive villagers. Many villagers with symptomatic malaria presented at the clinics and were no longer ill at the time of screening (cases on which the simulations were based). Furthermore, the only symptom considered during screenings was fever. Some of the RDT-positive villagers that are considered asymptomatic here are likely to have exhibited symptoms (e.g., headache, nausea) other than fever but are included in the proportion asymptomatic reported above.

Another limitation is that this approach assumes that screening is constant across the time interval (e.g., month). This is an oversimplification used for the simulation. There were two main reasons for choosing the month time step. Foremost was that preliminary analysis indicated repeat cases of falciparum malaria within a single house tend to occur at monthly intervals (Table 3). This roughly corresponds to the time that would be necessary for a symptomatic primary infection to lead to a symptomatic secondary infection (including incubation periods within both mosquito and human). At the community level, cases occur in uneven time intervals with several cases occurring in a single day followed by days or weeks of no cases, even during months in which there were cases to be detected. If an RCD were triggered by each symptomatic case, it would more frequently result in screenings during days or weeks during which there were no cases to detect. Aggregating the simulation cases into month intervals therefore creates a scenario in which there are fewer times steps with no cases to detect.

Alternatively, RCD including detection of asymptomatic cases may be more effective with a shorter time step as treatment of these cases may prevent them going on to develop symptoms and also further reduce ongoing transmission. However, the efficacy of such a strategy would be limited by the poor sensitivity of RDTs for asymptomatic infections. A strategy that relied on detection of asymptomatic carriers would require a more sensitive diagnostic, ideally PCR to be optimally effective. This would be more expensive, would result in a longer turnaround time between testing and results and be much less practical for use in the field. While there have been major improvements in the sensitivity and specificity of RDTs over the last several years, they still miss many low parasitaemia cases [11, 24, 25]. Novel RDTs are in development, which will allow the detection of much lower parasite densities than currently available RDTs. The use of more sensitive RDTs is likely to increase the impact of RCD.

The locations of farm huts in these simulations are an underestimate of villager time spent outside of the home village, as only villagers with farm huts in near proximity to the village were identified (within several km). Results here should take into account the fact that villagers work in locations much further than these farm huts, meaning that small screening radii would likely be even more inefficient than shown through these simulations.

The simulations also could not control for the impact of treating cases in one time step on the occurrence of cases beyond the next time step. Finding and treating current cases should decrease subsequent cases, but the magnitude of this effect is difficult to quantify. However, all patients with clinical malaria in this study were diagnosed and treated, meaning that this effect should be built into the data that fed the simulations. RCD will only halt transmission if diagnosis and treatment are able to be offered quickly. For now this means that RCD is dependent on RDTs that miss many sub-microscopic malaria infections.

These data suggest that the efficacy for reducing malaria transmission through an RCD approach based on RDTs to detect symptomatic cases is likely to be low in this setting. More intensive approaches, including detection and treatment of asymptomatic cases, may be more efficacious but require far more resources and are challenging to implement, especially in resource-poor settings. Other major public health approaches rely on the distribution of bed nets and community-based health clinics. Bed nets have shown limited effectiveness in the region, most likely because of the diversity of mosquito vectors and the exophilic behaviour of several species [26, 27]. Based on current knowledge, the most appropriate strategy for malaria elimination in most communities in this region is a strong community-based primary health clinic (village health post) in which villagers will receive appropriate early diagnosis and treatment for malaria upon the appearance of symptoms [28].

While attractive in theory, the screen-and-treat approach for RCD of malaria has major limitations. Several previous studies which have evaluated the screen-and-treat approach for malaria control and elimination have invariably failed to detect a significant impact [26–31]. While some studies have indicated that the RCD approach can lead to detecting more cases than normal passive case detection [32, 33], it has not yet shown an ability to halt transmission and in some settings appears to be ineffective [34]. This analysis, based on clinical case data and simulated RCD for clinical cases using RDTs, indicates that the RCD approach in this setting is not effective. In conclusion, spatial targeting approaches require occurrence of spatiotemporal clustering of cases within villages, or in this scenario also in farm huts. Using large screening radii will result in finding most cases, but at potentially much higher financial cost, requiring much effort on behalf of public health workers, and expecting that community members will be willing and available to participate in frequent screenings even in situations where few or no cases will be detected. Ultimately the RCD approach in this setting is limited because of these economic and social costs and because of current limitations in quick malaria diagnosis.