Background
Malaria causes an estimated 435,000 deaths per year, with the majority of cases occurring in sub-Saharan Africa, affecting children under 5 disproportionately [
1]. Recent advances in reducing case burdens in sub-Saharan Africa through bed net distribution, household level spraying, and rapid clinical diagnostic and treatment responses appeared to slow down in 2017 and 2018, leaving reduction, and eradication goals unmet, and an estimated 219 million cases in 2018 [
1]. The World Health Organization reported that for 10 high burden African countries, there was an increase of 3.5 million cases in 2017 over the prior year. This stall in reduction was largely attributed to a stall in investments in global responses to malaria. The U.S. remained the single largest international donor in 2017, contributing $1.2 billion (39% of the overall investment); it is projected that roughly $6.6 billion annually by 2020 will be needed for the global malaria strategy, underscoring the importance of knowing how much and where to invest.
Geospatial modelling approaches provide a flexible framework in which to explore possible future scenarios of malaria risk as a function of changing climate [
2]. Mordecai et al. introduced a mechanistic nonlinear physiological temperature-driven malaria transmission suitability model in 2013, via incorporating temperature dependent traits of both the mosquito and parasite, based on laboratory data [
3]. This demonstrated that transmissibility of malaria is constrained between 17 and 34 ℃, which will therefore limit the spatial distribution of malaria on the landscape. In addition, this model updated the optimum temperature for malaria transmission from 31 ℃ to 25 ℃, and the model was well validated using 40 years of field observation data matched to specific location month and temperature [
3]. Temperature has also been shown to be an important predictor of incidence in many locations [
4], and the potential effects of climate-induced temperature shifts as an impact on intervention and vector control efforts have been noted [
5]. In previous work, the top quantile of predicted transmission suitability from the Mordecai et al. model, that is, the top 25% of the transmission or
R0 curve, was found to best capture spatial and seasonal risk for Africa, from independent models of malaria risk prediction, based on statistical models of spatial case data from the Mapping Malaria Risk in Africa (MARA) and Malaria Atlas Project (MAP) projects [
2,
6‐
8].
Climate change threatens to the alter the nature of future malaria exposure across Sub-Saharan Africa [
2,
6,
7]. Many countries with a high burden of malaria now have weak surveillance systems and are not well positioned to assess disease distribution and trends, making it difficult to optimize responses and respond to outbreaks [
9]. To date, knowledge on how climate driven changes in malaria risk will manifest at regional and national scales is limited, though such knowledge is critical to designing responses. Changes in both the areas and populations exposed to malaria risk will necessitate adaptive responses to address them. To inform these responses, six scenarios of changing suitability, aligned to potential management strategies to address the changing risks, were explored. This provides an updated view of climate-driven malaria shifts in Africa from the 2015 mapping paper by Ryan et al. [
2], using the newer IPCC AR5 climate change scenario framework, explicitly defining season length to align with policy language, and including a sub-continental approach, aligning changes to regional scale planning.
The goals of this study were to (1) identify new areas that will emerge as suitable for malaria transmission under different scenarios of change; (2) identify areas that may experience reductions in transmission suitability season length; and (3) provide an estimate of the human population at risk under each scenario. These are presented in the language of malaria seasonality risk, to align with surveillance and intervention targeting goals, and summarized as regional scale outcomes, broadly aligned with USAID’s planning scales, as the parent aid organization of much of the US investment in the global malaria strategy.
Discussion
The changes in the geographic range of malaria suitability, broadly consistent across both scenarios of future climate pathways modelled here, suggest that the number of people exposed to conditions of malaria suitability will both increase and decrease in sub-Saharan Africa, depending on the region. Thus, as some populations experience reduced burden of malaria risk in the future, shifting suitability will increasingly place naïve populations at risk for outbreaks, particularly in Southern and Central Africa. Malaria outbreaks that occur where people have little or no immunity to the disease can lead to epidemic conditions, especially among vulnerable groups such as women and children [
1,
20]. This research identifies “hotspots” where current exposure and, therefore, immunity is nonexistent; these areas could see epidemic “flares” as climate conditions affect vector survival and reproduction. This effect may be further exacerbated in novel areas with no previous history of malaria exposure, where both immunity and knowledge regarding malaria prevention are lacking [
21‐
23]. Malaria outbreaks occurring where people have acquired immunity due to prolonged and repeated malaria exposure trigger management actions employing a cadre of tools, including vector control and case management approaches to prevent or reduce transmission [
23,
24].
These results identify regions where interventions need to be revisited to consider how climate will alter risk profiles in the future. The strong seasonal cycle of malaria across Southern Africa is related to climate and weather conditions [
25,
26]. Thus, during some periods of the year, climate conditions are not conducive to spread of the disease. Given the strong empirical relationship between vector survival and temperature, as temperatures rise exposure to malaria transmission is expected to increase in previously unsuitable regions, such as those in the higher elevation regions of Southern and East Africa. A key concern with climate change impacts is whether climate change will lengthen the period of the year during which diseases can establish and be transmitted. For example, areas where spring and autumn are now too cold for the reproduction of malaria vectors may become more suitable in the future. In these areas, increases in temperature may not impact midsummer malaria incidence greatly, but may result in a longer season, extending into both spring and autumn, during which malaria incidences will occur. In some cases, malaria may shift from being a seasonal disease burden to a year-round burden. This will necessitate different types of management and control interventions than those currently in place for short-season malaria [
27,
28]. Where the number of months of suitability for
Anopheles survival decreases, opportunities will emerge to alter and define more targeted seasonal responses—either reducing the cost of interventions or providing a window into potential eradication to malaria exposure. An increase in the number of months where conditions are suitable for mosquito survival will require responses to be extended for longer periods of time, increasing resource needs (e.g. staff time, medicines) as well as costs [
29]. In examining areas where malaria suitability is currently considered seasonally restricted, but will likely become more prevalent throughout the year, public health planners can anticipate which regions may require an extended investment pipeline.
A fundamental underpinning of modelling the response of vector-borne diseases to climate and ecology is the choice of model process. Previous approaches, such as that of the Malaria Atlas Project (MAP) and the Mapping Malaria Risk in Africa (MARA) project, are essentially top-down, wherein empirical data collected on the ground are matched to local climate conditions, and suitability established via geostatistical methods. In contrast, the modelling approach used here is mechanistic and “bottom-up,” wherein the life history of mosquitoes and pathogens, and their responses to temperature, are explicitly quantified based on empirical, laboratory-based data and incorporated into the model to predict where suitability for transmission is likely to occur. A mechanistic model, built independently of case outcome data, allows for validation with empirical, field-collected data, and obviates the bias of modelling data while intervention is ongoing, as is inevitably the case with previous approaches [
30]. In this study, the upper quartile of a curve was the prediction space on temperature, so the model projection contains no description of uncertainty on the mechanistic side. Future explorations of this climate-risk modelling approach could examine the sensitivity of overall findings to either uncertainty bounds of the original model parameterization, or using a range of cutoffs (e.g. 20th and 30th percentile, to describe error range around the 25th percentile).
While substantial progress has been made in recent years in the provision and use of climate projections, considerable uncertainties remain with their use [
31]. Using climate science research results to inform the decision process about which policies or specific measures are needed to tackle climate impacts requires acknowledging the uncertainties inherent in climate projections. These uncertainties may arise from mathematical reductions (parameterizations) of climate phenomena; potential socioeconomic technological pathways and attendant carbon cycle feedbacks that influence atmospheric concentrations of key greenhouse gases; imperfect scientific knowledge and the computational constraints of modelling regional detail while still incorporating relevant large-scale climate patterns; and the relationship between climate models and their relative impacts on key sectors and resources [
31‐
33]. Furthermore, uncertainty can arise over the chance of a single event (for example, crossing a threshold), recurrent events (the return period of a flood, for example), discrete events (hurricane frequency), and complex events (for example, the interplay of different factors that lead to drought) [
34]. Recognizing this, good practice is followed by incorporating a multimodel range of climate projections rather than a single model, as performed in this study [
31,
35,
36]. However, as with the mechanistic transmission model, using a multimodel climate ensemble does not yield an expression of uncertainty bounds. This study is a snapshot of model output at three time horizons: 2030, 2050, 2080, and two future scenarios, in terms of representative concentration pathways (RCPs), one ‘better’ (4.5) and one ‘worse’ (8.5), to give an estimate of potential future values in, rather than picking one scenario. As these RCPs are predicated on potential future cases of mitigation, it is assumed that the future will unfold within this range of possibility, at present there are no grounds to put more credence in one or the other. In future exploration of this modelling approach, unpacking the range of estimated risk (geographic and people) generated under the different models in the ensembled product, and exploring the 4 RCPs currently considered in IPCC projections, would give a richer view of the range of future possibility.
For the population data specifically, it is important to recognize that the projected population for 2020 is used to calculate the numbers of people potentially affected by changing suitability conditions across all future time periods. As with climate models, these projections do not necessarily capture all the factors that drive population movement and growth, and should be taken as best modelled estimates rather than exact values. Future studies can incorporate the additional projected population responses to climate change scenarios themselves, via such projects as the Shared Socioeconomic Pathways (SSP) projections, which offer multiple scenarios of population changes in response to potential social, environmental, and other drivers of societal development and change, as the climate changes into the future [
37]. Given the range of optimal transmission suitability temperatures for malaria, which corresponds with likely optimal human conditions, this study may, unfortunately, be underestimating future vulnerable populations, as people move into areas more suitable for transmission risk.
The study results are based on the temperature response curves of both
Anopheles mosquitoes and malaria pathogens. Nevertheless, many studies point to the critical role that rainfall plays in vector survival across sub-Saharan Africa [
12,
14,
15]. For example, single, intense rainfall events can wash away critical breeding sites, leading to a reduction in transmission potential [
16,
38]. Similarly, too little rainfall can limit mosquito survival as moisture is a prerequisite for breeding habitat [
39]. The approach herein addresses this second issue by masking out areas that are too arid for mosquito survival. While the relationship between rainfall and
Anopheles survival is critical, the available projections of rainfall are uncertain at the geographic scale of this work and, therefore, are not considered in this analysis.
Geographically projected model outputs are a useful component of a planning and intervention framework, providing a means of communicating key areas of risk and affected populations to decision makers. Anticipation of not only the location and time, but the duration of potential outbreak events will facilitate the development of efficient and timely agency responses. Moreover, this framework serves as a foundation for scenario analysis, explicitly modelling risk of exposure for different climate scenarios and time horizons. The range of potential outcomes allows governments and agencies the flexibility needed to reasonably anticipate resource use and funding needs, enabling the development of adaptive intervention strategies for both near and long-term outcomes.
Conclusions
Addressing the changing risk profiles projected in this suitability analysis will require modifying current interventions and programmes and implementing new ones to explicitly consider climate variability and change. While these projections of mechanistic models coupled to ensembled climate predictions for fixed time horizons (2030, 2050, 2080) and fixed population projections are simplified, they present a framework within which those potential futures can be explored, and examine both where and when changing risk is anticipated. Opportunities for improved responses also exist, including detailed geographic targeting, optimizing strategies and seasonal alignment with interventions. Identifying high risks in new areas of suitability present opportunities for informed action. Where malaria suitability is currently nonexistent to newly suitable, whether seasonal or endemic, the risks are critical, especially given that local populations’ immunity will be low. This could lead to the potential emergence of novel strains, rapid resistance, and untimely identification, translating into epidemic outbreaks. To respond, targeted and informed geographic surveillance in these regions could help to prepare timely responses before epidemic outbreaks occur. Knowing where and when more people will potentially be exposed offers an opportunity to increase the investment timeframe (seasonal to year-round), optimize vector control, and improve case management, with the evidence base to support these actions. Moving down the path toward elimination for some regions, where malaria transmission suitability decreases, opportunities will arise to focus resources on making surveillance and response systems increasingly sensitive and focused to identify, track, and respond to malaria cases and any remaining transmission foci.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.