Background
Malaria remains the most deadly of all vector-borne diseases, infecting an estimated 200 million people and causing over 550,000 deaths globally every year [
1]. Approximately 90 % of all malaria deaths occur in Africa, where an estimated 78 % of deaths are in children under the age of five [
1]. Malaria parasites are transmitted exclusively by
Anopheles mosquitoes [
2], spurring decades of research not only on the parasite, but also on its widespread vector. The spatial variation in vector-borne disease distributions such as malaria can likely be attributed to variations in environmental conditions (e.g. land cover, temperature, precipitation) and human population distribution that disease vectors depend on for survival [
3]. Ecological niche models (ENMs) correlate these environmental factors with georeferenced occurrences of
Anopheles mosquitoes to improve detection of spatial and temporal variation in biophysical determinants of these malaria vectors, often at a countrywide or even continent-wide scale [
4‐
8]. Although these modelling methods are rapidly advancing, they have yet to incorporate anti-malarial intervention data, though such approaches have been suggested and encouraged [
2]. Anti-malarial controls are not only time-intensive but also financially burdensome, with $5.1 billion US planned but $2.7 billion US expended in 2013 [
1]. The synthesis of species distribution modelling (or ecological niche modelling) and large-scale mosquito control methods can offer a new perspective on the potential effects of anti-malarial controls [
4].
Insecticide-treated nets (ITNs) are an excellent example of such vector control methods working on a large scale. The percent reduction in malaria deaths in children under the age of five was estimated in 43 sub-Saharan African countries from 2001 to 2010, compared to a baseline in 2000. Some countries reported no reduction by ITN scale-ups, while others reported up to a 26 % decrease in malaria deaths in children under the age of five [
9]. In 2014, a record number of ITNs were distributed to malaria-endemic African countries, with the total number of ITNs delivered to Sub-Saharan Africa reaching 427 million [
1]. Within the United Republic of Tanzania alone (hereafter Tanzania), an extensive ITN national plan has been implemented over the past 25 years, supported by the Global Fund to Fight AIDS, Tuberculosis and Malaria (GFATM) and the USA President’s Malaria Initiative [
10]. Since 2004, the Tanzania National Voucher Scheme has provided subsidized ITNs to pregnant women during antenatal visits [
11]. Between 2009 and 2010, a mass campaign distributed 8.7 million ITNs across the country, which were free to families with children under the age of five [
12]. Despite increases in coverage, analyses of ITN scale-up impacts [
9] found only a 3 % reduction in malaria deaths in Tanzanian children under the age of five from 2001 to 2010. In 2011, a Universal Coverage Campaign distributed 17.6 million ITNs nationally with the goal to increase global use in the general population to 80 % [
10]. In addition, the GFATM set a national target for Tanzania to increase the proportion of households with at least one ITN to 90 % by 2013 [
13].
Numerous analyses have been conducted to determine the effects of ITN use on malaria endemicity [
14‐
17] and on mosquito populations at local scales [
18‐
20], but none have yet compared ITN coverage with mosquito habitat suitability to determine if coverage has been optimally allocated. Environments that favour
Anopheles mosquito survival and reproduction, as well as more densely populated areas that facilitate access to hosts and accelerate malaria parasite transmission, will have higher malaria risk. Given mass ITN rollouts, is ITN ownership across Tanzania meeting targets in areas with high mosquito habitat suitability and increased malaria risk? Furthermore, if ITN distributions were optimized to target the most at-risk groups, areas with greater mosquito habitat suitability would be expected to have both greater averages and lower variances in ITN ownership rates (because coverage rates should be consistently high in malaria-endemic areas). Yet, ITN ownership relative to vector suitability remains uncertain even in regions, such as Tanzania, where national household surveys exist to permit such assessments.
Here, a species distribution model is created at 1-km resolution of Anopheles mosquitoes across Tanzania in 2001 (before large-scale ITN distributions) and is compared with countrywide ITN ownership by 2012 (number of ITNs owned per house and proportion of houses with at least one ITN) to assess where mosquitoes were most likely to thrive and whether ITN rollouts ensured coverage of such areas. This study uses the largest collection known of Anopheles mosquito occurrence records in Tanzania available for this time period, including 400 published sources, private data collections and online IR Mapper and the Malaria Atlas Project databases. This study relates species distribution models for Anopheles mosquitoes to ITN ownership and may serve as a template for integrating Anopheles mosquito distributions and disease risk with anti-malaria interventions.
Discussion
The World Health Organization (WHO) estimates ~50 % of African households had access to ≥1 ITN by 2010 but only ~3 % in 2004 [
1]. This study’s analysis of the DHS survey data across Tanzania from 2011 to 2012 suggests more optimistic results, with >93 % of Tanzanian households having one or more nets, meeting the 90 % national target set for 2013 [
13]. There is no difference in ITN ownership rates between rural and urban households. However, the effectiveness of ITN distributions varies among communities, many of which have much lower rates of ITN ownership than national averages. Among such communities, the proportion of households with at least one mosquito net is inversely related to mosquito habitat suitability. There is a clear opportunity to improve ITN effectiveness at a national scale by targeting supplemental net distributions in areas where current ownership rates are low but mosquito habitat suitability peaks. The unexpected decrease in ITN presence in some surveyed areas where malaria risk is high likely contributes to persistent, high malaria morbidity and mortality (20,900 deaths in 2013 [
59]). Other factors clearly contribute to malaria persistence, including increasing insecticide resistance among
Anopheles mosquitoes in Tanzania [
60‐
62].
It is imperative that ITNs reach households where mosquito habitat suitability, and thus malaria risk, is highest. Areas with higher mosquito habitat suitability were expected to have the most comprehensive ITN coverage and lowest variance in the number and proportion of households with one or more ITNs, but the opposite trends were observed. In areas with the highest anopheline habitat suitability, proportional ITN presence in households declines below both the national targets for ITN coverage (90 %) and the more modest (≥80 %) international targets (77 % of households have ≥1 ITN; Fig.
7c). In other words, the proportion of households reporting no ITN ownership increases in areas where mosquito habitat suitability is highest. This gap in ITN ownership is consequential from a human health perspective but is not detectable using models describing the response variable’s central tendency. Coarse resolution analyses have revealed similarly uneven mosquito net use relative to malaria endemicity among 23 African countries [
14]. If internal transportation networks through which ITNs and supporting programmes are delivered (e.g. road access) caused local variability in ITN ownership rates, those patterns should have shown strong autocorrelation. Causes of variation in ITN ownership and use need investigation. These include beliefs in malaria risk, trust in health workers distributing nets, perceived benefits of the nets, education, number of children under five in the household, and availability and use of other vector control measures [
63].
This study’s prediction of mosquito habitat suitability relates strongly to georeferenced malaria cases and to independent distribution models of
Anopheles mosquito species in Tanzania. The binomial test of omission was repeated with the fixed threshold of 0.1 to test the model against independently-collected data on unique georeferenced records of
Plasmodium falciparum across Tanzania from the MAP website. This study’s prediction of mosquito habitat suitability corresponded strongly to malaria records from 1985 to 2008 (model omission rate = 0.1246 for 568 geographically-unique
P. falciparum records) as well as those specifically within the 1999–2003 period used in this study (omission rate = 0.1321 for 107
P. falciparum records). Model prediction (based on the fixed threshold of 0.1) was qualitatively consistent with the predicted occurrence maps for
An. gambiae,
An. arabiensis,
An. funestus, and
An. gambiae s.s. developed independently [
64]. However, this study’s model predicted a narrower distribution than Malaria Atlas Project (MAP) predictions, which likely reflects differences in the time period considered and that MAP models are intended for continental applications and have less within-country detail. This study’s predictions for
An. gambiae s.l. distributions were also constructed using high resolution, regional satellite data and more extensive vector observation data, which may reveal more subtle variation in relative suitability in
Anopheles habitats. Although this study’s
Anopheles training records excluded the Lake Victoria region, this study’s model correctly predicted high relative habitat suitability there. This region is a malaria hotspot [
65], with malaria steadily spreading to higher altitudes and aggravated by climate variability and poverty [
66].
The distribution of anopheline mosquitoes and malaria is strongly influenced by land cover [
6,
67] and human population distributions [
43,
68] in addition to climatic and topographical factors [
43,
69‐
71]. Elevation also limits
Anopheles distributions [
4,
43,
72] and its omission from ecological niche models can severely degrade both model fit and prediction accuracy [
73]. In this study, mosquito habitat suitability increased particularly in cropland/natural land cover, where irrigation and water pooling are common, and in dense shrublands. Land features, such as soil type, are not directly measured in these satellite land cover data but may help explain such findings through differences in soil water holding capacity that can alter mosquito breeding success [
67]. Temperature and NDVI explained little unique variation in habitat suitability, though both can relate to mosquito distributions under some circumstances [
4,
8,
72,
74]. Temperature shows relatively little spatial variation across Tanzania except along elevation gradients [
75], which are more directly measured by the elevation variable. NDVI relates to vegetation greenness, productivity, and moisture availability, which are also strongly related to mosquito reproduction [
76]. However, satellite-based land cover measurements provide similar measurements that may more accurately detect the influences of vegetation on mosquito distributions. Furthermore, highly productive vegetation (i.e. areas with high NDVI) can be found in many regions of Tanzania, including at high elevations that are too cold to permit malaria transmission, eroding the capacity for such measures to discriminate between suitable and unsuitable mosquito habitats at this scale of analysis.
While this study’s dataset includes the most comprehensive collection of spatially-unique georeferenced
Anopheles records for this region to date, systematic, randomized, broad-scale sampling that identifies species in the
An. gambiae complex would be extremely valuable. Presence-only species observations assembled from an array of sources that differ in sampling effort and geographical focus could bias models toward areas with easier access (e.g. roadsides, proximity to research centres). In addition, lack of species-by-species evaluation for the historical period forced the combination of
Anopheles species. By combining the occurrence records across the
An. gambiae s.l. complex, the applicability of model predictions to particular species is limited (e.g.
An. arabiensis inhabits more arid regions than
An. gambiae s.s. and
An. funestus, better tolerating decreasing precipitation and interruptions in rainfall patterns in certain regions of Tanzania [
8]), but the relationship with malaria risk remains very strong. Systematic sampling at the species level, including observations of absence, is needed. ITN ownership and subsequent use may differ depending on which
Anopheles species is locally common (e.g.
An. arabiensis is more common outdoors [
72], rendering it less affected by ITNs than
An. gambiae s.s. and
An. funestus [
77]). Systematic, repeated sampling of
Anopheles mosquitoes across Tanzania could significantly improve the capacity to target interventions, and would provide a stronger basis for predicting how distributions of these vectors would change through time [
78].