Parasitological survey data

This analysis utilizes data on the prevalence of STH species collated in the GAHI (http://www.thiswormyworld.org) [18]. This project aims to provide an open-access, global information resource on the distribution of STH and schistosomiasis, with the specific aims of 1) describing the global distribution and prevalence of infection of each species and 2) highlighting geographical areas for which further survey information is required. The developed maps, along with sources of identified surveys, are presented on an open access website (http://www.thiswormyworld.org). The GAHI database is regularly updated using structured searches of the formal and grey literature, and has strict inclusion criteria: only random or whole community samples are included, excluding data from hospital or clinic surveys or surveys among non-representative sub-populations, such as among refugees, prisoners or nomads. For data from clinical trials or cohort studies, only baseline, pre-intervention estimates of prevalence are included. Efforts are made to geo-position each survey to a single longitude and latitude using a combination of electronic gazetteers, national school and village databases digitised from topographical maps and contact with authors who used a global positioning system (GPS; see Brooker et al. [18]). Where this point geo-position was not possible, efforts were made to geo-position surveys to the second administrative level, using the 2009 version of the Administrative Level Boundaries project (SALB) [19].

Defining the biological limits of transmission

The purpose of this first analysis was to determine biological and climatic suitability for STH transmission, based on post-1980 data that could be geo-located to a single survey location. To reduce the influence of control measures, we excluded data collected after the initiation of national STH control programmes, defined as >25% of the at-risk school-aged population receiving anthelmintic treatment for five years prior to the survey date, as reported to the World Health Organization (WHO) [20]. This resulted in the exclusion of 746 survey points from Burundi, Egypt, Honduras, Lao PDR, Mali, Nepal, Niger, Peru, Sierra Leone, Uganda and Venezuela. Data were also excluded if surveys were conducted in an irrigated area, which was classified as districts with >50% land surface equipped for irrigation [21], excluding a further 25 survey points in Afghanistan, Egypt, India, Madagascar, Mexico, Nepal, Pakistan, South Africa, Thailand, Venezuela and Viet Nam. On the basis of the above, 4,633 independent prevalence surveys were included.

Experimental and field studies indicate that the survival and development of STH free living stages, and hence STH transmission, are crucially dependent on ambient and surface temperature and humidity [22–25]. Indirect estimates of these factors can be identified from high-resolution satellite and meteorological data. Synoptic mean, minimum and maximum monthly values of land surface temperature (LST) for the period 1950–2000 were derived from the WorldClim database of interpolated global weather station temperature data at 1 km spatial resolution [26, 27]. Humidity was indirectly estimated on the basis of (i) annual precipitation rates available in the WorldClim database [26]; (ii) Potential Evapo-Transpiration (PET, a measure of the ability of the atmosphere to remove the water through evapo-transpiration); and (iii) Aridity Index (AI, calculated as mean annual precipitation divided by mean annual PET) [28, 29]. In addition, extremely arid areas, such as deserts and their fringes, were identified using the GlobCover Land Cover product [30, 31] for which the “bare areas” class denotes deserts.

The environmental and climatic data were imported into ArcMap 9.2 (ESRI, Redlands, CA) and linked by geographical location to the parasitological survey data. Analysis was stratified by region (Asia, Latin America and Africa plus the Middle East) and by climatic zone (tropics, sub-tropics/temperate) [32]. Upper and lower thermal and humidity constraints for each species were investigated using scatter plots and box plots. Two sets of biological limits were identified: areas that were biologically unsuitable (where median and mean observed infection prevalence was <0.1%); and areas with low/unstable transmission (where median and mean infection prevalence was <2%). The appropriateness of selected limits were subsequently assessed using box plots of point and district-level district estimates according to the limits and compared using a Kruskal-Wallis non-parametric test. Based on observed relationships, region and species-specific contour maps were developed on a 1 × 1 km grid to mask areas as biologically unsuitable for STH transmission, and areas where transmission is likely to be low/unstable. Districts were masked if >50% of their surface area was covered by at least one of the climatic masks. The identified limits were, however, not applied to irrigated areas.

Investigating the effect of urbanisation and settlement patterns on STH risk

The extent by which population density and urbanisation impacts upon STH transmission is uncertain [33]. The purpose of this analysis therefore was to determine whether STH transmission risk varies according to population density or settlement patterns and whether risk should be modified according to urbanisation, as has been done previously for malaria [34]. Population density was estimated using a global population database (Global Rural Urban Mapping Project (GRUMP)) [35, 36]. To ensure prevalences were contemporaneous with the population data, analysis was restricted to surveys conducted between 2005 and 2011. Survey locations were classified as urban using an updated 2010 urban extents (UE) mask derived from GRUMP. The remaining surveys were classified as peri-urban (<15 km from the UE edge and population density >250/km²), rural (>15 km from the UE edge and/or population density <250/km²), and extreme rural (population density <1/km²) using the Gridded Population of the World version 3 (GPW3) population database, projected to 2010 by applying national, median variant, urban and rural-specific growth rates per country [35, 36].

The effect of urbanisation on infection prevalence was assessed by identifying spatially and temporally matched urban–rural pairs. Here each urban survey prevalence value was matched to surveys in peri-urban and rural areas that were conducted within a 100 km and five year window. When more than one survey originated from the same UE, the mean prevalence was calculated. Averages of the peri-urban and rural sets of surveys were calculated and assigned to their urban counterpart, generating a series of map-defined urban/peri-urban/rural matched pairs. As prevalence distributions were highly skewed, the prevalence values for matched pairs were compared using the Wilcoxon signed-rank test, and overall prevalence distributions by settlement type compared using the Kruskal-Wallis non-parametric test. For subgroup analysis, data were stratified by country-level developmental indicators: the proportion of the urban population with access to improved sanitation (<25%, 25–50%, >50%), and the GINI coefficient (<35%, 35–50%, >50%).

Socioeconomic modification of risk at the country level

The third aim of this work is to determine whether territories can be identified as having no or very low STH transmission, or whether transmission risk should be modified, based on socioeconomic and development factors. Comparable, representative socioeconomic data are usually only available aggregated by country, and as such analysis of these limiting factors was restricted to the country level. National socioeconomic indicators were obtained from the World Bank databank [37], including GDP per capita; percentage of rural and urban populations with access to improved sanitation facilities; literacy rate in females ages 15–24 (proxy for maternal education), and GINI index (an indicator of the distribution of income within society, with 0% representing perfect equality and 100% perfect inequality) for the most recent year available. Parasite prevalence data were only included for the period 2005–2011, and incorporated all surveys conducted in this period that could be assigned to a country level. Scatter and box plots were used to explore ecological relationships between socioeconomic indicators and mean prevalence for each species.

Estimating the population at risk of STH transmission

The identified biological and social limits were overlaid with the 2010 GRUMP population surface to estimate total population at risk (PAR) figures for each species individually, and at risk of infection with one or more species. To standardise to a single, representative age group of relevance to control, the proportion of the population of school-going age (5–14 years) was estimated for each country, based on the World Population Prospects: 2010 Revision Population Database [38].

Defining the biological limits of STH transmission

Prevalence data for at least one helminth species were available for 138 countries: point-level survey data were available for 92 countries (representing 4,840 unique survey locations, 2,063 collected post-2004) and additional prevalence data were available at the second administrative level for a further 46 countries. Only 181 survey points were available for sub-tropical and temperate regions, representing 3.6% of the data. Thus, results presented here are restricted to stratification by world region only. Table 1 presents a more detailed breakdown of this data by region, and Figure 1 shows the geographical distribution of point and district data. The majority of the data comes from sub-Saharan Africa (SSA), but though fewer data are available for Asia and Latin America, the data that are available provide an adequate geographical spread of data points. Since few surveys were available for the North Africa and Middle East region, this area was combined with SSA.

Region	Total number of surveys availablea	Unique survey locationsb	Countries with point-level data	Surveys located to district levelc
Asia and Oceania	1,052	312	32	1174
Sub-Saharan Africa	4,536	4,215	37	207
Latin America and Caribbean	593	294	17	116
North Africa and Middle East	147	19	6	115

^a Includes surveys that can be located to a point, second administrative or country level.

^b Surveys geo-located to longitude and latitude.

^c Surveys geo-located to the second administrative level.

Figure 2 shows the associations between climatic factors and prevalence, stratified by global region. Consistent with previous findings [33, 39], clear relationships exist between prevalence of infection and LST for each species. For Africa and the Middle East, the prevalence of A. lumbricoides and T. trichiura is generally <4% in areas where maximum LST exceeds 35°C, and drops to <1% by 40°C. Hookworm infection remains prevalent throughout the upper end of the thermal range, only dropping to <2% when maximum LST exceeds 40°C. In contrast, relationships in Asia are less clear: although hookworm has similar limits to those seen in Africa and the Middle East, there is no clear upper thermal limit for A. lumbricoides in Asia, and prevalence of T. trichiura only drops below 1% at very high temperatures (>42⁰ C). There is no survey data for areas with maximum LST > 38°C for Latin America and the Caribbean. In contrast, hookworm appears more sensitive to low temperatures in all world regions, with observed prevalence <0.1% in those areas with mean LST below 10°C in the warmest quarter. Conversely, A. lumbricoides and T. trichiura infections remain prevalent throughout this lower thermal range.

All three species demonstrate similar relationships with aridity in Africa and the Middle East (Figure 2). Although there was considerable variation across semi-arid areas (aridity index >0.2–0.5), prevalence typically drops to less than 2% in areas classified as ‘arid’ (aridity index <0.2) and <0.1% in ‘hyper-arid’ areas (aridity index <0.03). This relationship is also seen for hookworm in other world regions. A. lumbricoides and T. trichiura are observed however, throughout much of the lower range of the aridity index in Latin America and the Caribbean, only dropping to <0.1% when the aridity index drops below 0.01. Interestingly, there appears to be no limiting relationship between A. lumbricoides and aridity in Asia. Comparison of the range of infection prevalence by land cover type, as classified by the GlobCover Project using high resolution satellite imagery, shows very large variation in the distribution of prevalence for all major land types (data not shown). Notably, whilst median infection prevalence is 0% for any STH in desert areas, for all three species at least 10% of the surveys conducted in areas classified as desert report infection prevalence of greater than 10%.

Based on the observed relationships, region and species-specific limits were developed and used to classify areas on the basis of climatic suitability as lying outside the plausible limits of transmission, namely unstable (median prevalence <2%) or no risk (median prevalence <0.01%). Districts (second administrative units) were then classified based on the most common transmission category by surface area. Figure 3 shows the spatial extent of transmission as defined by the transmission limits, described in Table 2. The distribution of point and district-level survey data within these limits is provided in Table 3 and Figure 4. For all three infections, there is strong evidence for difference in the distribution of survey data by transmission category: although wide variations in both point and district-level infection prevalence are observed within areas suitable for transmission, median infection prevalence falls below 0.001 in low transmission areas and is zero in areas classified as unsuitable for transmission. As can be seen from Table 3 and Figure 4, there is also strong evidence to support the discriminatory performance of the climatic masks when identifying districts were transmission is implausible.

On the African continent, areas of climatic non-suitability for all three species correspond to the extreme arid Sahara, masking out much of north Africa, whilst unstable transmission is expected for much of the arid Sahelian belt. The application of separate temperature masks for A. lumbricoides and T. trichiura widens the extent of the no risk and unstable transmission bands when compared with hookworm, masking the majority of Mali, Senegal, Burkina Faso, Chad and Republic of Sudan. Risk areas for all three species are however seen along the coast of Morocco, Algeria, Tunisia and Libya, and transmission cannot be excluded from irrigated districts in Egypt close to the Nile (representing 52% of Egypt’s districts). Unstable transmission associated with aridity is also seen in the Namib desert to the south, masking much of Namibia as well as western South Africa; in the horn of Africa (Somalia, Djibouti, Eritrea, northwest Kenya and eastern Ethiopia); and across the Red Sea in Yemen. Transmission is also precluded from the majority of the Arabian Peninsula, due to high temperatures and aridity. Transmission of A. lumbricoides can be ruled out for very few areas in Asia, although transmission is expected to be reduced in arid areas of the Gobi desert and Afghanistan, and in colder parts of the Himalayas. In contrast, hookworm and T. trichiura are masked from some pockets of central India (hookworm: 41 districts, T. trichiura 58 districts) and central Pakistan (hookworm: 34 districts, T. Trichiura: 25 districts), with reduced transmission expected throughout most of Pakistan (hookworm: 71 districts, T. Trichiura: 71 districts), central districts in India (hookworm only: 131 districts), mountainous areas of central Asia (including much of Uzbekistan and Kazakhstan) and in much of the northern part of the People’s Republic of China and Mongolia. Similarly, much of Latin American and the Caribbean can be considered climatically suitable for STH transmission, with transmission mostly limited by low temperatures or aridity. All three species are precluded from northern Chile, and hookworm from southern Chile and coastal regions of Peru, whilst reduced transmission of hookworm and T. trichiura is expected in more arid regions of Argentina.

Investigating the effect of urbanisation and settlement patterns on STH risk

Figure 5 presents box plots of the prevalence of STH species by population settlement type for 2,063 geo-located STH prevalence surveys conducted in 2005–2011. Although wide variation is evident for all three species across population density class, the results of the Kruskal-Wallis non-parametric test do suggest significant differences between groups at a global level (for all three infections p <0.001). For hookworm, urban infection prevalence is substantially lower than all other settings (median infection prevalence 2.4% vs. 7.7% in peri-urban and rural areas), whilst for A. lumbricoides and T. trichiura this situation is reversed, with higher infection prevalence observed in urban and peri-urban than in rural settings (8.2% vs. 0.0% and 3.0 vs 0.0% respectively). Findings were similar when results were stratified by world region, and by country urban development indicators (data not shown).

Socioeconomic modification of risk at the country level

Analysis of the correlation between socioeconomic factors and infection prevalence at a country level was restricted to 109 countries (representing 2,407 unique surveys) with data collected between 2005–2011. Perhaps unsurprisingly, no clear association was seen between mean survey prevalence (2005–2011) and the proportion of households with access to improved sanitation or literacy rates in women aged 15–24 years (a proxy for maternal education) at a country level for any species (data not shown). Although very little contemporary survey data exists for those higher income countries that fall within areas climatically suitable for transmission (which includes much of the Caribbean, and several islands in Oceania), very low infection prevalence for all three countries is seen in those countries with GDP > US$ 20,000 in 2010, suggesting that this may provide a realistic cut-off (Figure 6). This excludes 33 of the 166 potentially endemic countries, representing only 1.4% of the potentially at-risk population.

Total populations at risk

Figure 7 presents the combined extends of STH transmission, highlighting those countries excluded on the basis of socioeconomic status. Also highlighted are irrigated areas (which may still support transmission even when climatic conditions are limiting), and the extents of urban and peri-urban areas in 2010, defined using GRUMP urban extents and population density surfaces.