Malaria transmission has a significant environmental component, resulting in spatial heterogeneity in transmission risk and complex epidemiology (16). Here we attempted to quantify part of that heterogeneity in a useful and reproducible way. In Si Sa Ket and Ubon Ratchathani provinces in northeastern Thailand, the number of annual cases has declined by 96.6% between 2011 and 2021. There were only 36 cases across the two provinces in 2021. During the study period, in 2017, the Thai DVBD launched its malaria elimination strategy to be used at the local administrative level. This strategy recommended using recent history of indigenous transmission and, in those villages with no recent transmission, vector surveys, to determine receptivity to malaria. Vector surveys are, however, resource-intensive, and results can depend on entomological expertise (4, 10). The completeness of the surveillance data collected in Thailand has been improving since 2017 (22). Here we use a combination of this high-quality surveillance data and forest cover as a proxy for receptivity and propose a risk framework which has the potential to facilitate the rationalisation of resources targeting malaria in this part of Thailand. This also has scope to be adapted for use in other countries with similar transmission patterns.
In our dataset, indigenous village API was moderately positively associated with metrics of forest cover. Forested areas are conducive to malaria transmission due to the reproduction of vector species in ideal conditions of vegetation cover; temperature; rainfall; humidity; and a lack of infrastructure (16). The API for cases indigenous to a village was moderately positively correlated with subdistrict percentage forest cover and had lower correlation with the percentage forest cover within a 5km radius of the village. There were many villages with no reported cases in the past 5 years with high forest cover, which may be receptive but likely have reported no indigenous cases due to low overall case numbers and lack of parasite importation. The R number under control (Rc) can be approximated as the ratio of indigenous to imported cases and has been used as a proxy for receptivity in previous studies (6–8, 26). It is generally understood that endemicity will not be established in areas with an Rc of less than 1. In our data, only villages located in subdistricts with greater than 25% average forest cover ever had an annual Rc greater than 1. The Rc has been less than 1 in all villages since 2019, suggesting that progress is being made at a local level towards the goal of elimination in this area.
The greater association of indigenous malaria cases with coarser measures of forest cover such as subdistrict cover, and cover within a 5km radius, is likely to be multifactorial. There is significant heterogeneity in local geography; environmental factors; and behaviour. Known risk factors for malaria foci in Thailand include the presence of tropical forest and plantations; proximity to international borders; and percentage of short-term residents (2). The forest forms a natural border between Thailand and Cambodia/Lao PDR, and border subdistricts have the greatest forest cover in the study area. These areas are vulnerable to malaria importation from both forest-going activities and human mobility around border areas, which can introduce the parasite to receptive areas. It is also possible that 1–2km, typically taken as the flight radius of mosquitos from their breeding sites (30, 31), is too fine a scale to capture the layout of a village. The calculated buffer zones around villages were based on the co-ordinates of a single identifiable site in the village (e.g. a village sign or office) which is not necessarily at the geographic centre, such that the smaller buffers may not fully encompass the village boundaries; nor are villages often laid out in perfect circles. The strength of rank correlation with API was lower when smaller buffer zones were used, and this may be a product of heterogeneity in village layout and location of larval sites.
Risk classification
As Thailand approaches prevention of re-establishment planning, the focus of malaria strategy shifts to ongoing surveillance and response targeted to areas with high malariogenic potential. The highest risk villages under the current risk classification system (those that reported cases consistently each year) had higher average forest cover metrics.
When considering B1 and B2 foci, it is challenging to identify those that are still receptive to malaria using case data as, by definition, they have not had any recent cases. There were no B1/B2 villages (i.e. those which had not reported any cases in 2017–2019) which subsequently reported an indigenous malaria case in 2020 or 2021. While the best indicator of receptivity to malaria is recent case numbers, this lack of recurrence is representative of the current low-burden situation of the local near-elimination setting. The WHO framework for malaria elimination advises that “In practice, in some settings, non-receptive areas are identified as those that have had no vector control and no locally transmitted malaria cases but have had high-quality surveillance for several years...” (3). Due to the high-quality surveillance system in place in Thailand, this would apply to many foci, but doesn’t account for areas that may still be receptive but have not had reported cases due to lack of importation. Here it is useful to consider historical case data and its associations with environmental variables, such as forest cover, in order to assess which areas would have cases if the parasite were to be introduced. Combining this with measures of importation known to be associated with probability of reporting indigenous cases, such as proportion of short-term residents (2), would allow further stratification of areas by risk. There was also greater forest cover surrounding the A1/A2 foci which reported subsequent cases than those foci which did not. On the background of local ongoing reduction of malaria cases, this persistence of foci only in the most forested areas may be due to their higher receptivity and vulnerability. The forest cover is greatest on the border, where there is likely a higher risk of malaria importation from reservoirs both in the forest and across the border.
The proposed risk classification tool (Fig. 11) gave a comparable distribution of high and medium-risk villages to the current classification used by the Thai DVBD. This is to be expected, as both incorporate the number of recent years in which indigenous malaria cases have been reported, although the proposed tool also leverages forest cover metrics. Our tool identified 13 low risk villages which approximate to the B1 classification (no recent cases, but vectors are present). We were unable to compare this to the number of B1 villages under the DVBD system for reasons mentioned previously. There were 1,612 villages which had a score of zero, which is comparable to the total number of B1/B2 villages in 2020 and 2021 (1,612 and 1,618, respectively). The advantage of using this tool over the current approach is that it does not rely on entomological data to determine receptivity. While the absence of malaria vectors in an area can be used to infer that it is not receptive(3), this is based on the assumption that an adequate sample was collected; that sampling covered a sufficient geographical area; and that the vector species can be accurately identified. The Hansen forest dataset, however, has been extensively used and validated in tropical forest settings (20, 32, 33), although it is less accurate for local estimates (27). Both the current and proposed approaches incorporate the high-quality surveillance data currently collected by the Thai DVBD, although it has greater weight in determining the receptivity of low-risk villages in our tool.
However, we could not validate our risk score for the re-introduction of malaria to a village with no recent cases, as there were no villages classed as B1/B2 which went on to report an indigenous case in 2020 or 2021. This is likely due to the low case numbers and success of ongoing local elimination efforts. Instead, we suggest that data from 2022 onwards is used to validate the proposed risk score and adapt it as appropriate.
There is potential for a validated risk score to be adapted for use in other countries in Southeast Asia, particularly those with similar environments, human processes, and forest-based transmission. For it to be reliable, a robust surveillance system would have to be in place. Forest cover data is readily available but would be improved by on the ground validation of satellite data.
Strengths and limitations
This study combined high-quality surveillance data over a 10-year period with publicly available forest data to develop a reproducible scoring system which has the potential for adaptation to different local requirements. We have also used manually geo-located co-ordinates for villages without co-ordinates in the DVBD datasets.
There are some important limitations. There was a high level of incompleteness from 2012–2016 prior to the introduction of 1-3-7, with many cases not classified by likely origin. This means that comparisons with later data should be made with caution, as there is a higher level of completeness from 2017–2021. The malaria dataset did not differentiate introduced cases, where someone has been infected locally by a mosquito which was infected by an imported case, from those with no link to imported cases, where the original case was also infected locally. In very low-burden settings this has been achieved using spatiotemporal modelling (6, 8), but this is a more complex task where case numbers are higher, such as during the outbreak years in this setting. At one point during the 2014 outbreak in Ubon Ratchathani there were more than 1,000 monthly cases.
In Thailand, there are seven Anopheles species known to transmit malaria, which have different environmental optima and geographic distributions (31). If the prevalence of vector species better adapted to urban environments, such as Anopheles stephensii, were to significantly increase, estimates of receptivity based on geographical data would have to change significantly (34). Similarly, the effects of climate change are likely to alter the boundaries of where vectors can breed (35, 36). Including other factors such as temperature, humidity and well-collected entomological data may improve dynamic estimates of receptivity as the environment changes.
We were unable to validate the proposed risk score for the re-introduction of malaria in villages with no recent cases, as there have been no indigenous malaria cases reported in B1 or B2 classified villages since the introduction of the 1-3-7 system in 2017. The data available prior to 2017 is of lower quality and completeness. Instead, we provide an example of how the data available could be used to form a stratification system, which can be validated and refined using future surveillance data.
Lastly, the forest cover variables were calculated assuming only loss in the years since 2001. This is because forest gain is harder to detect due to its gradual nature (33). Therefore an area may be deforested and rapidly reforested for agroforestry but would be marked as deforested. Thailand has seen increases in malaria cases in workers on coffee and rubber plantations (2, 16). Future efforts including validation of satellite forest data with on-the-ground photography of forest cover in at-risk areas may provide better insights into the true forest cover. Other future efforts could also include qualitative studies to explore human factors affecting receptivity and the impact of foci management interventions.