Global numbers of infection and disease burden of soil transmitted helminth infections in 2010

Summary measures of population health are important to compare the relative importance of different diseases, to track differences in trends across countries and trends over time, and to provide a framework for evaluating the cost-effectiveness of interventions [1]. Estimating the burden of different diseases requires (i) reliable estimates of prevalence of each disease and (ii) an analytic framework within which to estimate morbidity and mortality attributable to each disease. In the case of intestinal nematodes (or soil-transmitted helminths, STH) several efforts have been made to provide worldwide prevalence estimates since those first assembled by Norman Stoll in 1947 [2–6]. The most recent, produced by de Silva et al. in 2003, suggested that despite marked declines in both the Americas and Asia for Ascaris lumbricoides, Trichuris trichiura and the hookworms (Necator americanus and Ancylostoma duodenale), little recent change had occurred in sub-Saharan Africa and STH still infected an estimated one billion people globally [7]. Inevitably, these numbers will have changed during the last decade: on the one hand, prevalence may have declined in some part due to improvements in living conditions and expansion of major deworming efforts; on the other hand, population growth may have increased the numbers infected.

A substantial hurdle when estimating numbers infected is the lack of quality data on infection prevalence [8]. In an effort to collate what data are available into a single resource, the Global Atlas of Helminth Infection (GAHI) was launched with the aim of describing the geographical distribution and prevalence of infection (http://www.thiswormyworld.org) [9, 10]. As a first step, we previously used observed relationships between infection prevalence and environmental factors to delineate the global limits for each STH species, providing an essential basis for identifying the global population at risk of infection [11]. Building on this work, here we quantify the numbers infected within these limits for 1990 and 2010, and estimate potential global morbidity attributable to STH. The primary aim is to update global estimates of infection prevalence, thus providing reliable national and sub-national descriptions of variation in infection risk, highlighting major changes in the global picture of STH between 1990 and 2010, and identifying countries and regions where data are still notably lacking.

As a secondary aim, we use indirect methods to quantify populations at risk of morbidity attributable to infection. For STH, prevalence alone does not provide a useful measure of potential morbidity, as only a fraction of infections will be associated with ill health. Instead morbidity is related to the intensity of infection, with the most intense infections occurring in only a minority of infected individuals [12, 13]. Given the lack of reliable methods for estimating disease directly, we build upon a mathematical modelling approach developed by Chan et al. and used in the original Global Burden of Disease (GBD) study [5] that exploits relationships between infection prevalence, mean intensity and potential morbidity [14, 15]. We further develop these methods, which incorporate heterogeneity between communities and age classes to estimate potential morbidity at regional and global levels, by refining characterisation of geographical heterogeneity at smaller spatial scales. Numbers are generated fully within a Bayesian estimation framework, allowing propagation of uncertainty throughout the modelling process. The work informed estimates of disability-adjusted life years (DALYs) due to STH infection, as part of the GBD 2010 study [16].

It is widely acknowledged that the exact global numbers infected and experiencing morbidity attributable to STH infection will remain an elusive goal, due in part to a paucity of reliable and accurate epidemiological data and in part the non-specificity of clinical signs due to STH [7, 8, 30, 58]. Here, we build upon a modelling framework originally proposed by Chan et al. for use in the first GBD study [5] to provide an update on the global situation in 2010. We estimate that 1.45 billion people were infected worldwide with at least one species of intestinal nematode in 2010, resulting in 4.98 million YLD and 5.18 million DALYs. The vast majority of infections and burden occurred in Asia, where at least one quarter (26.4%) of the population were thought to host at least one STH species. In relative terms however, morbidity attributable to STH infection was more equal across regions, typically falling between 0.6 and 1.4 YLD per 1000 people. Placing these results in context, the GBD 2010 study estimated that the neglected tropical diseases (NTDs) were together attributable for 26.05 million DALYs (1% of all DALYs), and communicable diseases more generally 1.34 billion DALYs (54% of all DALYs) [50].

Analysis of trends between 1990 and 2010 highlight a number of interesting findings. Overall, prevalence of any STH across all endemic regions has dropped from 38.6% in 1990 to 25.7% in 2010, representing a reduction of 140 million infected individuals. Steep declines were seen in countries such as the People’s Republic of China, Indonesia and Republic of Korea, but declines were more modest in other Asian countries and in sub-Saharan Africa and Latin America and the Caribbean. Reductions in DALYs were notably bigger, owing to the non-linear relationship between overall infection prevalence and prevalence of high intensity infection [15]. This highlights the substantial public health gains that have been made over the past 20 years, with sizeable reductions in the number of children suffering the wasting, anaemia and abdominal pain associated with high intensity STH infection.

Our current estimates differ from those produced previously: the first GBD study estimated that in 1990 hookworm prevalence across all endemic regions was 30%, A. lumbricoides was 33.5% and T. trichiura was 24.4%, resulting in an estimated 2.52 billion infections worldwide [58], nearly double the 1.45 billion predicted here for the same year. Prevalence estimates by de Silva et al. in their 2003 update are also substantially higher at 2.15 billion [7]. These discrepancies can be credited to a number of methodological improvements. First, by applying environmental limits we were able to shrink national populations at-risk to include only those living in areas where transmission of infection was plausible [11], thus preventing prevalence assignment to populations living in environmentally inhospitable regions within endemic countries (122–125 million people globally, depending upon species). Second, our estimates have been defined using data from the updated GAHI [20], and thus represent a much larger database than those used in previous estimates. For example, de Silva et al. identified 494 publications with suitable data, in comparison to the 862 publications included in the current analysis. They also relied on much older data (dating back to 1960s or before) when estimating infection prevalence for many Latin American countries, which may partly explain the large differences in 1990 and 1994 estimates from the two analyses. Third, geographical variation of both worms and population within countries were handled more robustly and at higher spatial resolution than previous estimates. For countries outside of sub-Saharan Africa, empirical estimates were applied to the admin2 level (where available) and aggregated to generate population-weighted national estimates, thus potentially preventing unrepresentative point prevalence estimates unduly influencing national estimates; though some countries still lacked suitable data. Within sub-Saharan Africa, Bayesian geostatistical modeling was used to predict the prevalence of infection for 2010, using available data and environmental information. This allowed more accurate predictions to be made for areas with no available survey data.

Reliable estimates of prevalence depend crucially on sampling methods and diagnosis, and thus our estimates inevitably come with some important caveats. First, we emphasize that these results do not all derive from nationally representative, spatially random surveys. Whilst for the majority of countries the total sample size used was at least several thousand individuals, for many epidemiologically important regions (including much of sub-Saharan Africa and south and southeast Asia) data were insufficient. This in part explains higher than expected estimates for Oceania, which in the absence of additional data were driven by evidence of very high prevalence of hookworm infection in Papua New Guinea [59, 60]. The seemingly anomalous high STH prevalence seen in Malaysia (and consequently the large relative burden in terms of YLD/person shown for southeast Asia in Figure 5) can also be ascribed to the few available data, this time from high risk communities in Sarawak [61], Pulau Pinang [62], Selangor [63, 64] and Kelantan [65] and is unlikely to be truly representative of the whole population. This scarcity of data for many Asian countries (and also reliance upon national surveys) can be, at least in part, attributed to necessary restriction of the literature search to English, French and Spanish language sources. However, the national surveys used for the People’s Republic of China, and the Republic of Korea, were powered at the province level and therefore were appropriate for our needs. We thus preferred to give emphasis to the national survey compared to small-scale surveys included in the local literature. By way of comparison, a recent systematic review and associated geostatistical model for China developed by Vounatsou et al. predicted a lower national combined STH prevalence (11% compared with 18% seen here), a consequence of lower predicted STH prevalence in Northern and Eastern Provinces [66]. Lastly, for eight countries (Armenia, Bahrain, Kuwait, Lebanon, Qatar, Syrian Arab republic, United Arab Emirates and western Sahara) no data were available, and it was necessary to rely on data from neighbouring countries with similar environmental and socio-economic conditions. Whilst this risks misestimation of national prevalence for these countries, all these countries are considered very low risk and it is likely that the total populations infected, and hence potential error incurred, will be small.

For sub-Saharan Africa, we were able to overcome sparse availability by modelling prevalence using an MBG framework. This combined empirical and geostatistical modelling approach is now becoming more widely applied in parasitology and infectious disease epidemiology more generally [67], and has been used for example to map the distribution of not only STH, but also schistosomiasis, leishmaniasis, malaria, trachoma, lymphatic filariasis and loaiasis at national, regional or continental scales [24, 68–76]. The generated predictive maps for sub-Saharan Africa only contained information on survey prevalence and environmental and population covariates however, and did not account for other important proximal risk factors including access to clean water and adequate sanitation facilities [77, 78] and implementation of control [79–82]. Results for southern Africa, for example, have been driven by the few available data-points in high prevalence regions, namely low-lying surrounding areas of Cape Town in western Cape Province and the sub-tropical lowlands of KwaZulu-Natal Province, which were collected in informal settlements with poor sanitation and high population crowding [83–92]. This explains in part the very high relative T. trichiura and hookworm burden (in YLD/person) demonstrated for these southern sub-Saharan Africa countries. We were also unable to incorporate measures of uncertainty originating from this model as a single point prediction process was used and thus aggregated certainty estimates would have been invalid. This unfortunately means that the high levels of prediction uncertainty that exist at prediction locations far from any available survey data (that is, for those countries with little or no data) are not reflected by the credible intervals presented in Table 4 We are now working to improve the continental STH map for sub-Saharan Africa by including information on these important covariates and better accounting for geostatistical outliers.

Data may also be systematically biased if survey sites were purposely chosen in areas of known high infection risk. Reassuringly however, 644 (9.7%) of survey sites were negative for all three STH species, and 3185 (47.7%) were negative for at least one species, suggesting that this concern may in general be unfounded. Urban–rural disparities in infection prevalence are also likely to have an impact on national mean prevalence globally, although observed relationships are too inconsistent for us to confidently apply corrections to national prevalence data at global scales [11]. Finally, data were available for only limited time points for many countries. In the absence of a consistent, measurable temporal trend we had to assume that prevalence had not changed between 1990–1999 and 2000–2010 for 70 countries with insufficient data, with prevalence estimates adjusted only for those 30 countries that have implemented large-scale treatment campaigns 2005–2010. However, evidence from Kenya does suggest that STH prevalence may have reduced in the past two decades, even in the absence of large-scale control or major improvements in sanitation [24].

In addition to prevalence, we also present YLD and DALYs for STH produced by the core GBD 2010 team [50, 51]. These estimates are intended to supersede those published previously, and are not directly comparable. Details of the methods used are provided elsewhere [50–52]; however, some issues should be highlighted when reflecting upon the derived estimates. There are a number of major methodological improvements: disability weights were derived from judgments of the general public about the health loss associated with the health state related to each sequela, rather than by health care professionals alone [52]; discounting and age-weighting were removed when estimating DALYs. For hookworm, GBD estimates previously only linked anaemia to high intensity infection. However, based on a recent summary of available evidence which showed that even light infections were associated with lower haemoglobin concentration in adults [44], in this current analysis anaemia outcomes attributable to hookworm infection were applied to all infected individuals regardless of infection intensity.

Controversially however, cognitive impairment has been removed as a sequela for STH, justified by the perceived paucity of evidence of a cognitive impact of intestinal nematodes. This judgment was made primarily on the basis of a 2012 Cochrane review of available randomized controlled trials of deworming, which found contrasting evidence of nutritional benefits and little support for cognitive or education benefits [93]. Critics of this review have argued that not only did this review exclude studies that treated both STH and schistosomiasis, but also that many of the underlying trials of deworming suffer from a number of methodological challenges, non-assessment of treatment externalities, inadequate measurement of cognitive outcomes and school attendance, and sample attrition, and that further, better-designed studies are required [94]. Such studies should ideally be randomized, must be appropriately powered and have sufficient duration of follow-up, care should be taken to collect detailed information on all potential confounders, and outcome measures should include all mediating variables along the hypothesized causal chain from deworming, anaemia, sustained attention and educational development. Further, whilst analysis did take into account the underlying distributions of wasting and haemoglobin density within populations, sparse and low spatial resolution data for many world regions will have lead to inaccurate YLD calculation for STH and other associated conditions.

There are additional limitations specific to the STH burden estimates. Firstly, the indirect methods used to estimate the proportion of the infected population experiencing high intensity infections (and thus at risk of morbidity) are highly sensitive to the choice of model parameters used to define the fitted negative binomial distribution and the chosen intensity cut-offs. In the absence of additional empirical data to better define these relationships, we were limited to using the same parameters and thresholds for A. lumbricoides and T. trichiura as previous efforts [5, 30]. Although additional data for hookworm did allow us to better quantify associations here, these primarily originate from school-aged children living in sub-Saharan Africa and so observed relationships may not be generalisable to other population groups and settings. Future work will concentrate on better quantifying these relationships, and incorporating uncertainty in resulting models. Similarly, mortality estimates for A. lumbricoides are subject to considerable ambiguity since the data sources by which the GBD estimates are derived are unclear.

More generally, the use of DALYs as a measure of disease burden has its advantages and its disadvantages [95]. The main advantage is that DALYs provide a composite, internally consistent measure of population health which can be used to evaluate the relative burden of different diseases and injuries and compare population health by geographic region and over time. Combined with information on the effectiveness and cost of different interventions, such estimates can guide priority setting [96]. The main disadvantage of DALYs is that they focus solely on health and do not capture the broader societal impact of diseases. This is especially true for STH, which have subtle, lasting impacts on child development and education. For example, recent studies in Africa, as well as reanalysis of the extensive Rockefeller Foundation supported efforts to control hookworm in the southern United States at the beginning of the 20th century, have shown remarkable long-run effects on productivity and employment and wages of treating children at school age [97–99]. There is also an important equity issue that goes beyond health: STH and other NTDs affect the poorest communities and are diseases of neglected populations. Thus, tackling STH and other NTDs should be seen part of broader efforts to reduce global poverty.

Improvements in the cartography of helminth infection have enabled us to better estimate the public health burden of STH. The updates presented here, based upon informed approximation using available data, clearly indicate that STH infections remain highly endemic throughout much of the tropics and sub-tropics. Although there have been substantial declines in some regions, in general infection levels over the past twenty years have remained unacceptably high. Whilst it is clear that we still require a better understanding of the full burden of intestinal nematodes, including effects upon child development and cognition, the extraordinary number of STH infections and associated years lost due to disability has helped galvanise the NTD community into reducing the burden of STH infections. Given the current high profile of NTDs on the global public health agenda, and the availability of simple, low-cost interventions that can substantially control the morbidity due to these infections, it’s our collective responsibility to ensure that the next set of STH distribution and burden estimates prove more encouraging.