- Record: found
- Abstract: found
- Article: found
Treating Cattle to Protect People? Impact of Footbath Insecticide Treatment on Tsetse Density in Chad
research-article
Author(s):
Noël Ndeledje
1
,
2 ,
Jérémy Bouyer
3
,
4
,
5
,
* ,
Frédéric Stachurski
4
,
5 ,
Patrice Grimaud
6 ,
Adrien Marie Gaston Belem
7 ,
Fidèle Molélé Mbaïndingatoloum
1 ,
Zakaria Bengaly
2 ,
Idriss Oumar Alfaroukh
6 ,
Guiliano Cecchi
8 ,
Renaud Lancelot
4
,
5
Publication date (Electronic):
14 June 2013
Journal:
PLoS ONE
Publisher:
Public Library of Science
Read this article at
There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
Background
In Chad, several species of tsetse flies (Genus: Glossina) transmit African animal trypanosomoses (AAT), which represents a major obstacle to cattle rearing, and sleeping sickness, which impacts public health. After the failure of past interventions to eradicate tsetse, the government of Chad is now looking for other approaches that integrate cost-effective intervention techniques, which can be applied by the stake holders to control tsetse-transmitted trypanosomoses in a sustainable manner. The present study thus attempted to assess the efficacy of restricted application of insecticides to cattle leg extremities using footbaths for controlling Glossina m. submorsitans, G . tachinoides and G. f . fuscipes in southern Chad.
Methodology/Principal Findings
Two sites were included, one close to the historical human African trypanosomiasis (HAT) focus of Moundou and the other to the active foci of Bodo and Moissala. At both sites, a treated and an untreated herd were compared. In the treatment sites, cattle were treated on a regular basis using a formulation of deltamethrin 0.005% (67 to 98 cattle were treated in one of the sites and 88 to 102 in the other one). For each herd, tsetse densities were monthly monitored using 7 biconical traps set along the river and beside the cattle pen from February to December 2009. The impact of footbath treatment on tsetse populations was strong ( p < 10 -3) with a reduction of 80% in total tsetse catches by the end of the 6-month footbath treatment.
Conclusions/Significance
The impact of footbath treatment as a vector control tool within an integrated strategy to manage AAT and HAT is discussed in the framework of the “One Health” concept. Like other techniques based on the treatment of cattle, this technology should be used under controlled conditions, in order to avoid the development of insecticide and acaricide resistance in tsetse and tick populations, respectively.
Related collections
Most cited references23
- Record: found
- Abstract: found
- Article: found
Estimating and Mapping the Population at Risk of Sleeping Sickness
Pere Simarro, Giuliano Cecchi, José Franco … (2012)
Introduction Human African trypanosomiasis (HAT), or sleeping sickness, is a vector-borne disease caused by two sub-species of the parasitic protozoa Trypanosoma brucei (i.e. T. b. gambiense and T. b. rhodesiense). Trypanosomes are transmitted to humans by the infected bite of various species of tsetse fly (genus Glossina). Transmission of the disease only takes place in sub-Saharan Africa, in discrete areas of endemicity, or ‘foci’, within the geographic distribution of the tsetse fly. The Gambian form of sleeping sickness is normally characterized by a long asymptomatic period and it is found in western and central Africa. The Rhodesian form, which is encountered in eastern and southern Africa, displays a much more rapid onset of overt symptoms and a faster progression. In the early 1960s, the reported incidence of the disease was at a trough, with only a few thousand cases being reported annually. However, a decline in surveillance in the post-independence period allowed sleeping sickness to regain ground. By the end of the 20th century, the World Health Organization (WHO) estimated that 300,000 people contracted the infection every year [1]. Since then, a global alliance led by WHO set elimination as the goal of its strategy against HAT [2], [3]. This renewed commitment by international and national institutions, including the private sector, succeeded in reverting the trend. As compared to the peak in 1998, when 37,991 new cases of HAT had been reported at the continental level, 6,743 cases were reported in 2011, corresponding to a reduction of 82.3%. Also, many countries considered as endemic have not reported any cases in recent years [4]. The magnitude of the recent advances in HAT control and surveillance is such that up-to-date estimates of the number and geographic distribution of people at risk are urgently needed. In the past, estimates of sleeping sickness risk at the continental, regional and national levels could only be based on educated guess and rough estimations of experts, rather than on a clearly laid out, objective analysis of the epidemiological evidence. In 1985, a WHO Expert Committee indicated that a population of 78.5 million was at risk of HAT in sub-Saharan Africa [5]. This figure was based on national-level information provided by the Ministries of Health of affected countries. In 1995, a new WHO Expert Committee indicated that 60.8 million people were at risk of contracting sleeping sickness [1], thus providing what was, to date, the latest global estimate of HAT risk. To derive this latest figure, a semi-quantitative method was used, whereby rural populations involved in agricultural activities within known HAT transmission areas were considered at risk. In both estimates, subjectivity remained high and the link to the epidemiological evidence loose. Since the latest estimations were made, HAT control and surveillance were scaled up [6], and data collection and reporting were substantially improved, with WHO coordinating the efforts of the National Sleeping Sickness Control Programmes (NSSCPs), bilateral co-operation, Non-Governmental Organizations (NGOs), Research Institutes and the private sector [7]. Also, over the last 10 to 15 years, the increased availability and utilization of the Global Positioning System (GPS), remote sensing data and Geographical Information Systems (GIS) triggered the development of novel, more objective methodologies to map the risk of many diseases [8], [9], [10], [11]. Till recently, geospatial analysis had never been used to estimate HAT risk at the regional or African scale. In 2008, the Atlas of HAT was launched, aiming at assembling, harmonizing and mapping datasets on the geographic distribution of sleeping sickness in sub-Saharan Africa [12]. Comprehensive and accurate epidemiological maps were generated [4], [13], which laid the foundations for more objective, evidence-based estimations of sleeping sickness risk. Thereafter, a GIS-based methodology for risk estimation was developed and tested in six Central African countries [14]. In this methodology, harmonized epidemiological data and global human population layers were combined, thus enabling different levels of HAT risk to be estimated and mapped. ‘Risk’ was regarded as the likelihood of infection, and the likelihood was estimated as a function of disease intensity and geographical proximity to HAT reported cases. In the present study, the methodology tested in the six Central African countries was applied at the continental level in order to map the risk of sleeping sickness in sub-Saharan Africa and to estimate at-risk population. In an effort to generate comparable estimates for both T. b. gambiense and T. b. rhodesiense infections, the same methodology was applied to all HAT-endemic countries and to both forms of the disease. Materials and Methods Input data Georeferenced layers of sleeping sickness occurrence and human population for the period 2000–2009 constituted the input for the present HAT risk mapping exercise. The number and the geographic distribution of HAT cases were provided by the latest update of the Atlas of HAT (reference date: 31 May 2012), thus including 170,492 cases of T. b. gambiense infection and 5,084 of T. b. rhodesiense, for a total of 175,576 HAT reported cases. Reported cases originated from twenty countries, namely Angola, Cameroon, Central African Republic, Chad, Congo, Côte d'Ivoire, Democratic Republic of the Congo, Equatorial Guinea, Gabon, Ghana, Guinea, Kenya, Malawi, Mozambique, Nigeria, Sudan, Uganda, United Republic of Tanzania, Zambia and Zimbabwe [4]. The Atlas provided village-level mapping for 81.0% of the cases, corresponding to 19,828 different locations mapped. The average spatial accuracy for reported cases mapped was estimated at ≈1,000 m using methods already described [4]. For the remaining 19.0% of the cases, village-level information was unavailable but the area of occurrence was known (e.g. focus, parish, health zone, etc.). For the purpose of risk estimation, these cases were apportioned among the endemic villages of their area of occurrence by means of proportional allocation [14]. Reported cases also included those diagnosed in non-endemic countries – most notably in travellers and migrants – which in the Atlas of HAT are mapped in the probable place of infection and flagged as ‘exported’ [15]. For T. b. rhodesiense exported cases, the place of infection most frequently corresponds to a park or another type of protected area. For the sole purpose of risk estimation, T. b. rhodesiense exported cases were randomly distributed within the boundaries of their respective park/protected area of origin. The geographic distribution of human population was derived from Landscan ™ databases [16]. Landscan provides global grids where census counts are allocated to grid nodes on the basis of probability coefficients. The spatial resolution of Landscan is 30 arcseconds (≈1 km at the equator), and the population layer is updated on a yearly basis. To delineate risk areas, an average of the ten Landscan population datasets from 2000 to 2009 was used. Subsequently, Landscan 2009 was combined with the risk map to provide estimates of people at risk at the end of the study period [14]. Spatial smoothing Both input layers (i.e. sleeping sickness cases and human population) can be regarded as spatial point processes, and thus amenable to spatial smoothing. Spatial smoothing methods are used in epidemiology to facilitate data analysis, and they allow to transform point layers into continuous surfaces of intensity. In this context, the intensity λ( s ) of a point process is the mean number of events per unit area at the point s [17]. The term ‘event’ is used to distinguish the location of an observation ( s i) from any other arbitrary location s within a study region R. Spatial smoothing techniques can be based on localized averages or more complex, three-dimensional mathematical functions (e.g. kernels), but they all rely on a moving window, whose size and shape determines how far the effect of an event will reach [18]. For this study, intensity was estimated through a kernel function k (·), so that the intensity estimate could be expressed as: Here, s was a location anywhere in the study region R, s 1 ,.., s n were the locations of the n observed events, and k(·) represented the kernel weighting function. τ>0 is normally referred to as the bandwidth or search radius, and s i were the events that lay within the area of influence as controlled by τ. There are various shapes of kernel to choose from, all usually represented by symmetric bivariate functions decreasing radially. The choice of shape has relatively little effect on the resulting intensity estimate [19], [20] and we used a quadratic kernel [20]. A more important choice is the selection of the bandwidth τ, the rule being that the higher τ, the smoother the intensity surface. Although different techniques are available for selecting τ [21], [22], no optimal value exists, and characteristics of the biological process under study are often better suited to guide the choice, so that the smoothed surface provide insights into the underlying data [18]. By taking into account the epidemiological features of HAT, the behaviour of the tsetse vector and the mobility of people in the average rural African milieu where HAT occurs, a search radius of 30 km was chosen [14]. In particular, a few studies investigated the daily distance covered by people living in HAT foci [23], [24], [25] and revealed that this tends not to exceed 15 km. The distance of 30 km enabled to take into account, at least in part, also people's movements that do not occur on a daily basis. Figure 1 provides a three-dimensional illustration of the output of spatial smoothing. In the example, the point layer used as input comprised one single ‘event’ (i.e. one HAT case) localized at the centre of the grid. 10.1371/journal.pntd.0001859.g001 Figure 1 Three-dimensional rendering of the disease intensity surface for one case of HAT, as derived from spatial smoothing (Kernel function k(·): quadratic; bandwidth τ: 30 km; output resolution: 1 km). Delineation of risk areas Prior to spatial smoothing, the number of HAT cases reported in 2000–2009 was divided by ten, thus providing the average number of cases per annum (p.a.). Similarly, Landscan human population layers from 2000 to 2009 were averaged [14]. Both averaged layers were subjected to spatial smoothing using the same quadratic kernel function. Importantly, both intensity surfaces were generated using the same 30 km bandwidth [26]. Spatial smoothing resulted in the two surfaces and , which represent the average annual estimates of disease intensity and population intensity respectively. The input to and output of spatial smoothing are exemplified in Figure 2. 10.1371/journal.pntd.0001859.g002 Figure 2 The foci of Bodo-Moissala in Chad and Batangafo-Maitikoulou in Central African Republic. (a) Distribution of HAT cases; (b) Average population distribution (Landscan); (c) Annual intensity of HAT cases as derived from (a) through spatial smoothing; (d) Population intensity as derived from (b) through spatial smoothing. The ratio between the intensity of HAT cases and the population intensity can be defined as the disease risk [18], so that a risk function was estimated as: Thresholds were applied to the risk function in order to distinguish and map different categories of risk, ranging from ‘very low’ to ‘very high’ (Table 1). Outside the areas mapped as at risk of HAT, i.e. in areas where <1 HAT case per 106 inhabitants p.a. was reported, the risk to contract the disease was considered ‘marginal’. These marginal areas were not taken into account further in this study. The term ‘marginal’ was chosen because, in such areas, risk could not be considered as non-existent, since residents of these zones could still expose themselves to infection if visiting transmission areas. 10.1371/journal.pntd.0001859.t001 Table 1 Thresholds for the definition of sleeping sickness risk categories. Category of risk HAT cases per annum Very high ≥10−2 ≥1 per 102 people High 10−3≤R<10−2 ≥1 per 103 people AND<1 per 102 people Moderate 10−4≤R<10−3 ≥1 per 104 people AND<1 per 103 people Low 10−5≤R<10−4 ≥1 per 105 people AND<1 per 104 people Very low 10−6≤R<10−5 ≥1 per 106 people AND<1 per 105 people Estimates of people at risk The map depicting the different categories of HAT risk was combined with Landscan 2009 dataset to estimate the number of people at risk at the end of the study period [14]. Results An area of 1.55 million km2 in Africa is estimated to be at various levels of HAT risk, ranging from ‘very high’ to ‘very low’ (Table 2 and Table 3). Areas at ‘very high’ to ‘moderate’ risk account for 719 thousand km2 (46.3%) and areas at ‘low’ to ‘very low’ risk account for the remaining 833 thousand km2 (53.7%). 10.1371/journal.pntd.0001859.t002 Table 2 Areas at risk of T. b. gambiense infection in western and central Africa (km2 ×102). Country Total country area* (km2 ×102) Area at risk (km2 ×102) Very High High Moderate Low Very Low Total at risk % of total country area Angola 12,538 - 568 597 480 158 1,803 14.4 Cameroon 4,664 - - 22 79 71 173 3.7 Central African Republic 6,244 55 141 204 161 97 659 10.6 Chad 12,725 - 33 34 36 39 142 1.1 Congo 3,385 21 199 388 372 182 1,162 34.3 Côte d'Ivoire 3,214 - - 23 82 182 286 8.9 Democratic Republic of the Congo 23,041 27 996 2,717 2,599 1,563 7,902 34.3 Equatorial Guinea 270 - 4 37 16 8 65 24.1 Gabon 2,660 - 6 57 69 35 167 6.3 Guinea 2,461 - 1 42 53 88 184 7.5 Nigeria 9,089 - - - 16 55 70 0.8 Sierra Leone 728 - - - 7 11 18 2.5 South Sudan 6,334 21 260 379 265 76 1,001 15.8 Uganda 2,055 - 13 91 42 28 175 8.5 Other Endemic Countries** 60,316 - - - - - - - Total 149,722 124 2,222 4,591 4,277 2,594 13,808 9.2 * Land area. The area of surface water bodies as depicted in the Shuttle Radar Topography Mission – River-Surface Water Bodies dataset [43] is not included. ** Countries at marginal risk: Benin, Burkina Faso, Gambia, Ghana, Guinea-Bissau, Liberia, Mali, Niger, Senegal and Togo. 10.1371/journal.pntd.0001859.t003 Table 3 Areas at risk of T. b. rhodesiense infection in eastern and southern Africa (km2 ×102). Country Total country area* (km2 ×102)* Area at risk (km2 ×102) Very High High Moderate Low Very Low Total at risk % of total country area Burundi 251 - - - - 2 2 0.8 Kenya 5,749 - - - 5 26 31 0.5 Malawi 948 - - 33 53 52 138 14.6 Mozambique 7,791 - - - 5 34 39 0.5 United Republic of Tanzania 8,863 - 16 125 229 286 657 7.4 Uganda 2,055 - - 45 146 97 288 14.0 Zambia 7,425 - - 33 221 224 478 6.4 Zimbabwe 3,884 - - - 9 69 78 2.0 Other Endemic Countries** 25,685 - - - - - - - Total 62,650 - 16 236 667 792 1,711 2.7 * Land area. The area of surface water bodies as depicted in the Shuttle Radar Topography Mission – River-Surface Water Bodies dataset [43] is not included. ** Countries at marginal risk: Botswana, Ethiopia, Namibia, Rwanda and Swaziland. The total population at risk of sleeping sickness is estimated at 69.3 million (Table 4 and Table 5). The categories at ‘very high’ to ‘moderate’ risk account for a third of the people at risk (21 million), whilst the remaining two thirds (48.3 million) are at ‘low’ to ‘very low’ risk. 10.1371/journal.pntd.0001859.t004 Table 4 Population at risk of T. b. gambiense infection in western and central Africa (no. persons ×103). Country Total country population* (no. persons ×103) Population at risk (no. persons ×103) Very High High Moderate Low Very Low Total at risk % of total country population Angola 12,799 - 740 749 3,049 229 4,767 37.2 Cameroon 18,879 - - 28 238 365 631 3.3 Central African Republic 4,511 28 41 130 138 99 435 9.6 Chad 10,329 - 109 114 120 123 465 4.5 Congo 4,013 4 109 451 1,825 177 2,566 63.9 Côte d'Ivoire 20,617 - - 230 722 1,720 2,672 13.0 Democratic Republic of the Congo 68,693 23 3,546 10,767 15,674 6,237 36,247 52.8 Equatorial Guinea 633 - 2 27 8 6 43 6.8 Gabon 1,515 - 2 21 19 761 803 53.0 Guinea 10,058 - - 187 488 1,932 2,606 25.9 Nigeria 149,229 - - - 368 1,814 2,183 1.5 Sierra Leone 5,132 - - 1 83 87 170 3.3 South Sudan 6,996 15 401 453 334 67 1,270 18.2 Uganda 32,370 - 142 1,275 456 251 2,124 6.6 Other Endemic Countries** 103,673 - - - - - - - Total 449,447 70 5,092 14,431 23,521 13,869 56,983 12.7 * As per Landscan 2009. ** Countries at marginal risk: Benin, Burkina Faso, Gambia, Ghana, Guinea-Bissau, Liberia, Mali, Niger, Senegal and Togo. 10.1371/journal.pntd.0001859.t005 Table 5 Population at risk of T. b. rhodesiense infection in eastern and southern Africa (no. persons ×103). Country Total country population* (no. persons ×103) Risk (no. persons ×103) Very High High Moderate Low Very Low Total at risk % of total country population Burundi 9,511 - - - 5 33 38 0.4 Kenya 39,003 - - - 254 870 1,124 2.9 Malawi 15,029 - - 194 217 499 910 6.1 Mozambique 21,669 - - - 5 53 58 0.3 United Republic of Tanzania 41,049 - 22 373 621 808 1,824 4.4 Uganda 32,370 - - 847 4,734 2,295 7,877 24.3 Zambia 11,863 - - 14 122 279 416 3.5 Zimbabwe 11,393 - - - 5 88 94 0.8 Other Endemic Countries** 101,420 - - - - - - - Total 283,306 - 22 1,429 5,964 4,927 12,341 4.4 * As per Landscan 2009. ** Countries at marginal risk: Botswana, Ethiopia, Namibia, Rwanda and Swaziland. The geographic distribution of risk areas in central Africa, western Africa and eastern-southern Africa are presented in Figure 3, Figure 4 and Figure 5 respectively. Country-level risk maps are provided in Supporting Information (Maps S1). Focus level risk maps will be provided at HAT/WHO website: http://www.who.int/trypanosomiasis_african/country/en/. 10.1371/journal.pntd.0001859.g003 Figure 3 The risk of T. b. gambiense infection in central Africa (2000–2009). 10.1371/journal.pntd.0001859.g004 Figure 4 The risk of T. b. gambiense infection in western Africa (2000–2009). 10.1371/journal.pntd.0001859.g005 Figure 5 The risk of T. b. rhodesiense infection in eastern and southern Africa (2000–2009). Trypanosoma brucei gambiense A total of 57 million people are estimated to be at risk of contracting Gambian sleeping sickness (Table 4). This population is distributed over a surface of 1.38 million km2 (Table 2). Approximately 19.6 million (34.4%) of the people at risk live in areas classified at moderate risk or higher, which correspond to areas reporting ≥1 HAT case per 104 inhabitants p.a. The remaining 65.6% (≈37.4 million) live in areas classified at low to very low risk. Central Africa accounts for the vast majority of people at risk of T. b. gambiense infection (Figure 3). The risk patterns in Cameroon, Central African Republic, Chad, Congo, Equatorial Guinea, and Gabon have already been described in some detail elsewhere [14]. In essence, areas at very high to high risk are localized in southeastern and northwestern Central African Republic, southern Chad, along lengthy stretches of the Congo river north of Brazzaville, and by the Atlantic coast on both sides of the border between Gabon and Equatorial Guinea. The Democratic Republic of the Congo is, by far, the country with the highest number of people at risk (≈36.2 million) and the largest at-risk area (≈790 thousand km2). Areas at risk can be found in the provinces of Bandundu, Bas Congo, Équateur, Kasai-Occidental, Kasai-Oriental, Katanga, Kinshasa, Maniema, Orientale, and South Kivu. More details on the risk and the geographic distribution of sleeping sickness in the Democratic Republic of the Congo will be provided in a separate paper. In South Sudan, a sizable area (≈100 thousand km2) and over a million people are estimated to be at risk of sleeping sickness, including a number of high to very high risk areas in Central and Western Equatoria provinces. These findings highlight the need for continued surveillance in this country [27]. In neighbouring Uganda, the area at risk of T. b. gambiense infection (≈17 thousand km2) is located in the North-west of the country. It mostly falls in the category ‘moderate’, and it accounts for over two million people at risk. In Angola, sleeping sickness is found in the northwestern part of the country (≈180 thousand km2 – 4.8 million people at risk), and most of the high-risk areas are located in the Provinces of Bengo, Kwanza Norte, Uige and Zaire. In western Africa, the most affected endemic areas are categorized at moderate risk and they are localized in costal Guinea and central Côte d'Ivoire (Figure 4). Areas at lower risk fringe the main foci, but they are also found in other zones such as southern Guinea and southern Nigeria. Trypanosoma brucei rhodesiense Rhodesian sleeping sickness is estimated to threaten a total of 12.3 million people in eastern and southern Africa (Table 5). This population is distributed over a surface of 171 thousand km2 (Table 3 and Figure 5). Of the total population at risk of T. b. rhodesiense, a minor proportion (≈1.4 million – 11.8%) live in areas classified at moderate risk or higher, the rest (≈10.9 million – 88.2%) live in areas classified at low to very low risk. In Uganda, Rhodesian HAT threatens a population of ≈7.9 million, and the risk area (29 thousand km2) stretches from the northern shores of Lake Victoria up to Lira District, north of Lake Kyoga. The areas in Uganda where risk is relatively higher (i.e. ‘moderate’) broadly correspond to the districts of Soroti, Kaberamaido and northwestern Iganga. Because of a comparatively lower human population density, some areas in the United Republic of Tanzania are estimated to be characterized by higher levels of risk than Uganda, despite fewer reported cases of HAT. In particular, risk is estimated to be high in proximity to the Ugalla River Forest Reserve (Tabora Province). Also all of the other risk areas in the United Republic of Tanzania are associated in one way or another to protected areas, most notably the Moyowosi Game Reserve and the natural reservations in the northeast of the country (i.e. Serengeti, Ngorongoro and Tarangire). Overall, ≈1.8 million people (66 thousand km2) are estimated to be at risk in this country. In Kenya, HAT risk ranging from low to very low is localized in the western part of the country, adjacent to risk areas in neighbouring Uganda. Also, although no cases were reported from the Masai Mara National Reserve during the study period, part of its area is estimated to be at risk, as influenced by the risk observed in the neighbouring Serengeti National Park (United Republic of Tanzania). Interestingly, two cases have been reported recently (2012) in travellers visiting the Masai Mara [28]. Nature reserves also shape the patterns of HAT risk at the southernmost limit of T. b. rhodesiense distribution, most notably in Malawi, Zambia and Zimbabwe. In this region, the highest number of people at risk is found in Malawi (≈0.9 million people), where risk is associated to the wildlife reserves of Vwaza Marsh, Nkota-Kota, and the Kasungu National Park. In Zambia (≈0.4 million people at risk), risk areas are scattered across the country, predominantly in the East and most notably around the North and South Luangwa National Parks. In Zimbabwe, an area of 7.8 thousand km2 is estimated to be at risk (94 thousand people). This risk zone in associated to the Mana Pools National Park and the Lake Kariba. Discussion Approximately 70 million people (1.55 million km2) are estimated to be at various levels of HAT risk in Africa. This corresponds to 10% of the total population and 7.4% of the total area of the endemic countries. This figure is not far from estimates made by WHO over the last thirty years, (78.54 million in 1985 [5] and 60 million in 1995 [1]). However, the meaning and interpretation of these various figures substantially differ, and it is unwarranted to make comparisons between the results of the present study and previous figures, especially if the goal is to explore trends. In the early 80 s, the only way to derive country- and continental-level estimates of people at risk of HAT was to collate heterogeneous information from the Ministries of Health of the affected countries [5]. A decade later, an attempt was made to update the estimates [1], but the degree of subjectivity in the methodology and the reliance on expert opinion remained high. By contrast, the present methodology is quantitative, reproducible, based on evidence and provides a categorization of risk. The use of global human population layers [16] and the regular update of the Atlas of HAT [4] will enable regular and comparable updates to be made. The presented maps of different HAT risk categories will help to plan the most appropriate site-specific strategies for control and surveillance, and they will contribute to ongoing efforts aimed at the sustainable elimination of the sleeping sickness. However, the reported incidence levels underpinning the different risk categories differ by orders of magnitude, so that a more accurate representation of HAT risk can be given by focusing on the different risk categories. For example, 21 million of people (0.7 million km2) are estimated to live at ‘moderate’ to ‘very high’ risk of infection. These are the areas where the most intensive control measures need to be deployed. Low to very low risk categories account for ≈48 million people (0.8 million km2). In these areas, cost-effective and adapted measures must be applied for a sustainable control. From the methodological standpoint, assumptions affect all estimates of disease risk, including those presented in this paper. One important assumption in the proposed methodology is that it is possible to use the same approach based on human cases of trypanosomiasis to estimate risk of both forms of sleeping sickness. This assumption met the primary goal of generating continental risk estimates in a consistent fashion. However, especially for T. b. rhodesiense, different approaches could be explored, explicitly addressing the pronounced zoonotic dimension of this form of the disease. Another important choice in the proposed methodology is that of the 30 km bandwidth – the distance from affected locations beyond which disease intensity is considered zero. Sensitivity analysis conducted for six central African countries showed that there is a positive linear relationship between bandwidth on the one hand, and the extent of risk areas and the at-risk population on the other [29], [30]. However, the categories at higher risk were shown to be the least affected by bandwidth. Therefore, as a rule, increasing the bandwidth would inflate the low-risk categories, but it would have a more limited effect on the delineation of areas at higher risk. The estimates presented here also rest on the assumption of isotropy for the risk function. In the future, anisotropy may be explored in an effort to account for the linear nature of some important landscape features such as rivers or roads. When interpreting the presented risk estimates it is important to acknowledge the uncertainty inherent in the human population datasets used as denominator [31]. Also, it has to be borne in mind that no attempt was made to model HAT under-detection and under-reporting, which, despite recent progress in surveillance [12], are still known to occur. HAT under-detection can occur both in areas covered by active or passive surveillance and in areas that, because of remoteness or insecurity, are off the radar of health care services, and therefore sometimes referred to as ‘blind spots’. These two types of under-detection are expected to have different effects on risk estimation and mapping. The former is likely to impinge mainly on the level of risk, with a limited effect on the delineation of risk areas and on the estimates of the total population at risk. By contrast, if under-detection occurs in zones were no surveillance is in place, a few areas at risk will fail to be captured and mapped, which is bound to result in underestimation of the total population at risk. In the proposed risk mapping methodology, the latter areas would have been included in the ‘marginal’ risk category. Efforts should be made to identify and accurately delineate these hypothetical transmission zones, finding adaptive strategies to cope with the constraints of remoteness and insecurity that affect them. Knowing the true epidemiological status of these areas has vast implications not only for risk estimation but most crucially for the prospects of HAT elimination. For the chronic T. b. gambiense infection [32], under-detection can be addressed by continuous passive case detection and regular active screening surveys. The fact that we took into consideration ten-year data on disease occurrence and control activities should contribute to the robustness of the T. b. gambiense risk estimates. However, in the case of T. b. rhodesiense, due to the acuteness and rapid progression of infection, under-detection poses more serious challenges. Although attempt were made to model under-detection for T. b. rhodesiense [33], both data and methodological constraints prevent these methods from being applied at the continental-level. In the future, methodologies should be developed to estimate and map the coverage of active and passive surveillance. These would provide valuable information complementing risk maps, whilst also assisting in optimizing field interventions. The temporal dimension is also crucial when interpreting risk maps. The proposed estimates were based on an average of HAT reported cases for a ten-year period. No weighting for the different reporting years was applied, despite the fact that a reduction in reported cases was observed during the last years of the study period. As a result, all cases contributed equally regardless of when exactly they were reported. Importantly, the estimates of people at risk presented in this paper, being based on reported cases, can not account for the possible future spread of HAT, and the risk thereof, into presently unaffected areas. Other approaches to risk modelling could be more interested in predicting the future risk of sleeping sickness, focusing on the environmental suitability for HAT rather than on its present occupancy. To this end, the relationships are to be explored between HAT occurrence and a range of factors, including human and livestock population movements [34], environmental, climatic and socio-economic variables, as well as disease and vector control. The potential of this type of models has been investigated in a few local contexts, for example in southeastern Uganda for T. b. rhodesiense [35], [36], [37], [38], and coastal Guinea for T. b. gambiense [25]. Recent attempts have also tried to address risk forecasts at the regional level in relation to climate change [39]. The potential of various modelling frameworks could be explored for modelling the future risk of HAT [40], [41]. The growing range of spatially explicit environmental datasets [42] and increased computational power enable these models to be applied even across large geographical areas. Interpretation of model outputs will probably be the most serious challenge. In fact, incompleteness and biases in the real-world epidemiological records often blur the line between concepts such as the theoretical fundamental niche of a pathogen and its realized niche. Where estimates of prevalence are available, most notably in T. b. gambiense areas, model-based geostatistics could also be applied, which utilize Bayesian methods of statistical inference and enable rigorous assessment of uncertainty [11]. Their potential for, and applicability to, a low-prevalence, focal disease such as HAT would be interesting to explore. Supporting Information Map S1 Maps of distribution of population at risk of human African trypanosomiasis in 21 disease endemic countries, where any level of risk has been identified during the period 2000–2009. Countries are organized on geographical order, west to east + north to south, and from T.b.gambiense to T.b.rhodesiense endemic countries: Guinea, Sierra Leone, Côte d'Ivoire, Nigeria, Cameroon, Chad, Central African Republic, South Sudan, Equatorial Guinea, Gabon, Congo, The Democratic Republic of the Congo, Angola, Uganda, Kenya, United Republic of Tanzania, Burundi, Zambia, Malawi, Mozambique and Zimbabwe. (PDF) Click here for additional data file.
- Record: found
- Abstract: found
- Article: not found
A changing environment and the epidemiology of tsetse-transmitted livestock trypanosomiasis.
The distribution, prevalence and impact of vector-borne diseases are often affected by anthropogenic environmental changes that alter the interactions between the host, the parasite and the vector. In the case of tsetse-transmitted livestock trypanosomiasis these changes are a result of the encroachment of people and their livestock into tsetse-infected wild areas. This has created a sequence of new epidemiological settings that is changing the relative importance of the domestic or sylvatic trypanosome transmission cycles and is causing concomitant changes in the impact of the disease on livestock. These changes in the dynamics of the epidemiology have an important impact on the factors that need to be considered when developing area-specific strategies for the future management of tsetse-transmitted livestock trypanosomiasis.
- Record: found
- Abstract: found
- Article: found
Identifying Transmission Cycles at the Human-Animal Interface: The Role of Animal Reservoirs in Maintaining Gambiense Human African Trypanosomiasis
Sebastian Funk, Hiroshi Nishiura, Hans Heesterbeek … (2013)
Introduction Many infections can be transmitted between animals and humans [1]. Human African Trypanosomiasis (HAT, sleeping sickness) is a vector-borne disease caused by parasites of the species Trypanosoma brucei and transmitted by flies of the genus Glossina (tsetse flies) [2]–[7]. While the east African form of HAT, caused by T. brucei rhodesiense, is a zoonosis with a well-described animal cycle in cattle and wild species, the more chronic west African form, caused by T. brucei gambiense, is often considered a human disease and causes more than 95% of reported cases in humans [8]. Gambiense HAT is endemic in 24 countries and deadly if untreated. While T. b. gambiense has been found in numerous domestic and wild species [2]–[5], [9]–[13] and transmission between humans and other species been shown to occur both experimentally [9] and naturally [14], the exact role of animals in gambiense HAT epidemiology remains an unsolved puzzle [15], [16]. Are they sporadic dead-end hosts, or could they be an important factor for maintaining transmission? Generally, the incidence of gambiense HAT can be brought to very low levels just by treating human cases, and indeed the latter strategy alone appeared to be sufficient for eliminating gambiense HAT from the island of Bioko in Equatorial Guinea [17]. Such observations have given rise to the notion that T. b. gambiense does not spread in animal populations without the presence of humans. However, the parasite was recently detected in flies on Bioko [18], suggesting that there is ongoing circulation of the parasite, with the existence of a wild animal reservoir appearing plausible given the lack of detected cases in humans or domestic animals on Bioko. The existence of self-sustained cycles of infection in animals could jeopardise efforts towards gambiense HAT elimination. One of the very few systematic efforts to link the presence of T. b. gambiense in different animal species to human cases was a survey performed over several years in the historical focus of Bipindi, Cameroon, in response to the detection of 44 cases in humans by a newly-installed surveillance network in 1998/99 [19]. Subsequently, data on T. b. gambiense prevalence in domestic [13] and wild animal species [12], as well as in tsetse flies [20], biting preferences [21] and the distribution of species among different types of habitat [22] were collected, providing a rich epidemiological and ecological dataset. Synthesising these data in a common modelling framework presents a mathematical and conceptual challenge. Here, we use the concept of the next-generation matrix (NGM) [23] to understand the transmission dynamics of gambiense HAT in Bipindi. The NGM describes the number of secondary cases caused in each species by an infected host or vector of any (other or the same) species and allows the generalisation of a classical epidemiological quantity, the basic reproduction number , to a situation in which there are different types of hosts or host species. Defining as the spectral radius or largest eigenvalue of the NGM generalises the endemic threshold properties of in single-host systems, in the sense that if there can be sustained transmission and if there cannot. We use a mathematical model of gambiense HAT transmission to understand the prevalence observed in hosts and vectors and estimate the elements of the NGM. Mathematical models of gambiense HAT transmission involving humans only [24] or humans and one animal species [25]–[27] have been derived previously and have yielded valuable insights into HAT epidemiology. For example, it has been shown that there are scenarios in which HAT may require a non-human reservoir host for persistence [25]. From sensitivity analysis of the parameters entering it has been concluded that the proportion of bloodmeals the vector takes from humans is the most important factor, indicating that variation in the exposure to tsetse flies could explain the spatial distribution of T.b. gambiense [26]. Sensitivity of those parameters to expected climate change (albeit for T.b. rhodesiense) suggests a shift in the geographical range of infection risk [27]. All these results and, more generally, estimates of for gambiense HAT have not been based on data collected from animals, vector and human systems within the same focus, and instead have relied on the combination of parameter values estimated or drawn from different literature sources. The method we present here is broadly applicable to vector-borne diseases with a potential animal reservoir, and is designed to be informed by data from field surveys. It is based on the premise that the system is in endemic equilibrium, an assumption we revisit in the Discussion section. We show that, in an equilibrium scenario, both and the contribution of different species or groups of species can be estimated using only data on (a) relative prevalence of infection in different host species and (b) the distribution of bites of the vector on different species. We use this method to assess the potential of each species or combinations of species to maintain gambiense HAT transmission in Bipindi. Further, we extend our method to incorporate ecological data (the distribution of species across different habitats) and use this to perform extensive sensitivity analysis. Methods The analysis is based on the assumption that the system has been observed in an equilibrium state. This allows us to calculate the forces of infection in all species from measured prevalences. Using these, we derive the next-generation matrix (NGM) in all host and vector species participating in the transmission cycle. Assuming that the system is in an endemic equilibrium, implies that (see linear stability analysis in Supporting Text S1). Data sources The human case data come from two active screening campaigns, performed in November 1998 and February 1999, following the discovery of infected blood sera from the Bipindi area, previously practically ignored in medical surveys [19]. The first of these campaigns concentrated on two neighbouring villages and found 26 infected cases. The second one expanded to a total of 15 villages (including the two villages screened in the first survey), detecting 18 further cases, of which 16 were found in the two villages visited during the first campaign. The data from domestic animals come from a survey performed in 5 villages of the Bipindi area in 2003/04 [13], including the two villages containing most of the human cases. The data from wild animals come from surveys performed in Bipindi between 1999 and 2001 [12]. The case data are summarised in Tables 1 and 2. 10.1371/journal.pcbi.1002855.t001 Table 1 Summary of sampling campaigns. Date Survey sampled positive Nov 1998 Humans (2 villages) 1269 26 Feb 1999 Humans (15 villages) 3519 18 1999–2001 Wild animals 832 18 2003/04 Domestic animals 875 27 Number of sampled humans and animals, and number positive for T. b. gambiense. 500 of the 832 wild animals sampled were from the Bipindi area (15 positive). 204 of the 875 domestic animals were from the Bipindi area (8 positive). 10.1371/journal.pcbi.1002855.t002 Table 2 Summary of case data: + indicates positive for T. b. gambiense, and the resulting equilibrium prevalence. Name Scientific name Samples + Source Human Homo sapiens 3641 44 0.012 [19] Sheep exact species unknown 267 18 0.067 [13] Goat exact species unknown 264 8 0.030 [13] Pig exact species unknown 307 1 0.0033 [13] White-eyelid mangabey Cercocebus torquatus 5 1 0.20 [12] Greater white-nosed monkey Cercopithecus nictitans 80 4 0.050 [12] Blackstriped duiker Cephalophus dorsalis 16 1 0.062 [12] Blue duiker Cephalophus monticola 200 4 0.020 [12] Brush-tailed porcupine Atherurus africanus 100 2 0.020 [12] Giant rat Cricetomys gambianus 125 3 0.024 [12] Small-spotted genet Genetta servalina 8 1 0.13 [12] Two-spotted palm civet Nandinia binotata 29 2 0.069 [12] For our analysis, since we were interested in the potential for animal reservoirs to maintain gambiense HAT, we attempted to make our estimates conservative in that regard. We included all the villages screened in Bipindi for our basic estimate of prevalence in humans, as the area comprising these villages region compares well to where the tested animals came from (see the Results section for sensitivity analysis on the human prevalence estimate). Moreover, we combined the two surveys in human populations into a single prevalence estimate, which is equivalent to assuming that the two surveys took place at the same time and ensures we do not underestimate prevalence due to medical interventions in response to the first survey (i.e., to estimate prevalence we took all infected cases found in both screening surveys as enumerator and the combined population of the villages screened as denominator). The data from both domestic and wild animals were collected later, and are very likely to be affected by vector control installed after the human cases were detected, which could be expected lower the prevalence in all species. Since we did not have access to animal case data separated by location and species, we used all the survey data. As a consequence, in both the data from domestic and wild animals, the prevalence we are using is lower than the one reported from Bipindi alone (all species combined). In summary, we are likely to underestimate equilibrium prevalence in animals, in line with our attempt to be conservative in that regard. In the analyses presented below we assumed infection among a given species to be binomially distributed with fixed infection probability corresponding to an average equilibrium prevalence. The likelihood for equilibrium prevalence in species (equivalent to the probability of being infected), given cases detected among sampled animals, is then proportional to a beta distribution, (1) This quantifies the uncertainty resulting from small sampling sizes (the smallest being White-eyelid mangabeys with only 5 sampled animals), with correspondingly wide confidence intervals. All other parameters are drawn from flat distributions using Latin Hypercube Sampling [28], with ranges given in Supporting Text S2. Model assumptions In setting up the model, we made the following biological assumptions: Population sizes are constant with no demographic stochasticity. The duration of the first stage of the disease (equivalent to the duration of infectiousness in our model) is exponentially distributed, as the evidence suggests [29]. Moreover, we assume that there is no long-term chronic carriage, although there is some evidence of that they sometimes occur [29]. We do not have to distinguish between teneral and non-teneral flies. Generally, the susceptibility of a tsetse fly to midgut infection with trypanosomes decreases if they are not infected after the first bloodmeal. We found no qualitative difference when considering a model in which only teneral flies (i.e., the ones that have not had their first blood meal) can be infected (see Supporting Text S1). Moreover, the probability of infection we estimate for flies ( ) is consistent with what one would expect as average probability of infection of tsetse flies [30]. The transmission rate of an infected host or vector does not change over time. This is consistent with findings that transmissibility of trypanosomes is independent of parasitemia [31]. Biting preference is as measured by Polymerase Chain Reaction (PCR) on blood in flies that have fed. This implies that blood specimens were randomly sampled and that the test is equally sensible to all bitten species. Basic model Assuming random mixing and uncorrelated bites, a simple transmission model for gambiense HAT transmission between host and one vector species is given by the system of ordinary differential equations, based on the Susceptible-Infected-Susceptible (SIS) model (2a) (2b) where is the number of infected of host species , is the number of infected vectors, and are the total population sizes of host species and vectors, respectively, and are the forces of infection acting on host species and the vector, respectively, is the rate at which infected hosts of species lose infectiousness (through recovery or death), and and are the natural death rates (and birth rates, assuming constant population sizes) of host species and the vector, respectively. Forces of infection The forces of infection are (3a) (3b) where is the probability for an infectious bite on a susceptible host of species to lead to infection, rescaled by the ratio of vector to host population sizes, is the force of infection exerted by species on vectors, is the probability that an infectious bite by a susceptible vector leads to transmission of the parasite and establishment in the vector midgut. These transmission probabilities are treated as unknown quantities to be estimated. The other parameters are measured quantities: is the relative population density of species compared to all other hosts, is the biting rate of vectors, is the fraction of bites taken on species , and and are the prevalence of infection in species and vectors, respectively. Assuming that the system is in equilibrium, we get a relation between force of infection and prevalence, (4a) (4b) where the asterisk denotes equilibrium quantities. Next-generation matrix The NGM describes transmission between different vector and host species by mapping the distribution of primary cases to the distribution of secondary cases [23]. Once fully quantified, the matrix allows to identify host species that can maintain transmission of a given infection [32]. That is, we can distinguish between maintenance and non-maintenance hosts by calculating the host-specific reproduction number of (group of or single) host species , which is interpreted as the average number of secondary cases per generation caused (via the vector) by a single primary case belonging to in the absence of hosts other than . If , host(s) can maintain gambiense HAT transmission on its (their) own. This formalises the definition of maintenance hosts given in [33]. Correlated bites To capture the impact of correlated bites on model dynamics, we separate our vector class into classes and denote these , the number of infected vectors that have last fed on host species If is the average time spent feeding on a given species, the dynamical equations for are (5) where is the total number of vectors that have last fed on species . In equilibrium, this can be solved for which is used to parametrise the NGM and can be extended to groups of species (see Supporting Text S1). Habitat separation Extending the scenario of correlated bites to known differences in habitat, we introduce a mixing matrix , the elements of which describes how likely a vector is to switch (and potentially transmit infection) from species (or group of species) to species (or group of species) . The dynamical equations for then become (6) which, again, is used to parametrise the NGM. With the densities (or presence/absence) of the different species in different habitats are given, we estimated mixing rates to (7) Numerical methods Simulations were performed using the Gillespie algorithm [34]. All parameter estimations where there was no analytical solution were performed using Powell's hybrid method [35] as implemented in the GNU Scientific Library [36]. Results We first state the general result relating the basic reproduction number and host- and group-specific reproduction numbers to endemic prevalences and biting preferences, before applying this to the scenario of gambiense HAT transmission in Bipindi. Identifying transmission cycles In a multi-host system, the basic reproduction number is defined as the spectral radius of the NGM. In the Supporting Text S1, we show that when we are dealing with only one vector species the basic reproduction number is (8) where the sum is over all host species and is the average number of infected vectors caused in a completely susceptible vector population by a single host of species , and as the average number of infected hosts of species caused by a single vector in a completely susceptible host population. A special case of this equation for a system composed of humans and one animal species has previously been derived in [26]. The host-specific reproduction number [32] of a group of host species, or their contribution to the basic reproduction number , is (9) This is equivalent to the value would take in a system of only the subset of species in . The summands are related to the forces of infection via (10) In equilibrium, we can use Eqs. (3) and (4) to rewrite this as (11) We can use this to calculate the basic reproduction number given only equilibrium prevalence in the vector ( ) and all host species ( ) and vector biting preference (the fraction of bites taken on species ), (12) This does not require any information on vector biting behaviour, host or vector population sizes, or within-host infection dynamics. Animal reservoirs of gambiense HAT in Bipindi For the focus we investigated, in the baseline scenario of random mixing of vectors with the different host species (proportional to biting preference as measured) we found that the median value of was 1.1 (95% CI 1.0, 1.3) (Fig. 1). The contribution of humans (i.e., the hypothetical value of in a system of only humans and vectors) was 0.5 (0.2, 0.7). When testing for potential cycles of sustained transmission in groups of species, we found that in domestic animals was 0.5 (0.3, 0.8). When adding humans to the system, increased to 0.7 (0.5, 0.9). In wild animals, was 0.8 (0.6, 1.2), with a likelihood of 0.14 of being greater than 1. In all animals (wild and domestic), was 1.0 (0.8, 1.3), with a likelihood of 0.46 of being greater than 1. 10.1371/journal.pcbi.1002855.g001 Figure 1 Contributions of species and species groups to under random mixing. (a) The contributions of different species to under the assumption of random mixing between vectors and hosts. (b) The contribution of different sets of species to under the assumption of random mixing between vectors and hosts. In both plots, the y-axis shows the values of which would be found in a system of only the given (set of) species and vectors, the central line indicating the most likely value, upper and lower edges the interquartile range, the outer lines 1.5 times the interquartile range, and individual dots outlier results. The rightmost data point in (b) shows the estimate for in the whole system (all species combined). Outliers for white-eyelid mangabeys with (0.1% of values) are not shown. These results are in contrast to the notion of gambiense HAT as human disease with only accidental animal hosts [7]. However, we could be underestimating the prevalence in (and, consequently, the importance of) humans for two main reasons: (i) active case detection campaigns might not have detected all cases in the population screened due to problems with diagnostic sensitivity [37], [38] or the presence of asymptomatic carriers with low parasitemia [29] (note that our denominator is the population screened, so screening attendance does not change our estimate as long as individuals screened are chosen randomly), and (ii) the denominator at risk might in fact not be the entire population screened if the risk of infection is unevenly distributed. The effects of these two are equivalent and multiplicative: If a fraction of cases are detected, and a fraction of the population is involved in the transmission cycle, the measured prevalence is and true prevalence is , such that . If we increase the prevalence in humans to account for these potential sources of bias, of the system with only animals and vectors decreases (Fig. 2a). More specifically, if only the 40% of the population of Bipindi living in the two villages with most of the detected cases [19] are at risk of infection, and if we incorporate a low estimate of 90% for screening sensitivity [37], the likelihood for in animals decreases to 0.13, but the likelihood for in humans is still less than 0.01. Only if we further reduce the population at risk to less than 20% of these villages does the likelihood for in animals drop to less than 0.01. In that case, the likelihood for in humans is 0.59. 10.1371/journal.pcbi.1002855.g002 Figure 2 Human and animal contributions to under different model scenarios. (a) The contribution of the human (red, solid) and animal (blue, dashed) populations to as a function of the fraction of the population exposed to bites of the vector, shown here as effective population size . The vertical dashed line indicates the fraction of the population in the main endemic area [19], and the dotted line 90% of that population, a low estimate for screening efficacy [37]. (b) The contribution of the human (red, solid) and animal (blue, dashed) populations to as a function of the rate of host switching between a species, given in units of (number of switches)/year/fly. In both plots, the y-axis shows the values of which would be found in a system of only humans and the vector. The lines show the best estimate, and the light grey areas contain the smoothed (2.5%, 97.5%) quantile range, obtained from the binomial likelihood profiles and Latin hypercube sampling of parameter ranges (see Supporting Text S2). A second source of potential bias could arise if subsequent bites of the same fly were correlated, or if a fly taking a blood meal on a given species or group of species had a higher probability of biting a host of the same species or another species in that group again [39], [40]. Our analysis attributes human infection either to other human infections (via a vector) or to spillover from animal reservoirs (again via a vector). If the two kinds of host population are fully epidemiologically linked (i.e., if we assume random mixing), then the analysis inevitably attributes many of the cases in the population with lower (weighted) prevalence to spillover from the population with higher (weighted) prevalence. The less linkage there is the less likely this is to happen, and eventually in the low-prevalence population is required to explain persistence. When we considered a system of two transmission cycles, one containing humans and domestic animals and the other one wild animals (i.e., a system in which there is a sylvatic cycle separate from the human/domestic animal cycle), the human contribution to the system was not enough to guarantee in the system of humans and domestic animals. When humans were considered to be part of a transmission cycle completely separate from animals, we got in both the human and the (wild and domestic) animal cycle. Introducing only occasional transfer of infection between species, however, means the observed data are not compatible with sustained transmission in the human-vector cycle, with a threshold appearing at a rate of switching of about 1/year (Fig. 2b). in humans was greater than 1 with likelihood greater than 0.01 only when vectors switched between species less than once per year. Comparing these with an average fly life expectancy of about one month, this would mean that most flies never change host species in their lifetime, an unrealistic scenario given that in practice flies cannot afford to restrict themselves to one host type. Independent transmission cycles in animal reservoirs, on the other hand, have a likelihood greater than 0.5 for any rate of switching less than 30/year, corresponding to 2–3 host switches per fly in its lifetime. To inform this analysis with ecological measurements of habitat distributions of the species found to host gambiense HAT in Bipindi [22], we incorporated the overlap of habitat ranges between animals in our derivation of the NGM. This version of the model does not support a human-only transmission cycle, and suggests that a sylvatic cycle is possible. Separating the different species by the habitats they can be found in yielded likelihood 0.48 for in wildlife species only (Fig. 3), and likelihood 0.97 for in all animal species if switches between groups of species happened at a third of the biting rate. 10.1371/journal.pcbi.1002855.g003 Figure 3 Contributions of species groups to under habitat-specific mixing. The contributions of different groups of species to under the assumption of mixing proportional to habitat overlap of hosts. Hosts are grouped according to the habitats they can be found in, with random mixing within these groups and switching occurring at a third of the biting rate between the groups. The y-axis shows the values of which would be found in a system of only the given set of species and vectors, the central line indicating the most likely value, upper and lower edges the interquartile range, the outer lines 1.5 times the interquartile range, and individual dots outlier results. The rightmost data point in shows the estimate for in the whole system (all species combined). We performed simulations of the different model variants, with a particular focus on how long it would take for the disease to become re-established in a human population from which it had previously been eliminated. We tested different rates of vector switching between a human/domestic and a wild animal cycle, as well as other configurations of cycles. As the rate of switching decreased, the time it can take for cases to reappear in the human population increased (Fig. 4). For rates of switching greater than 1/year, reintroduction usually occured within a year or less. When, on the other hand, switches between humans, domestic animals and wild animals were as rare as 0.01/year per fly (i.e., only one in 1000 flies ever switched between these subsystems) it could take 10 years or longer for infection to be transferred between them. 10.1371/journal.pcbi.1002855.g004 Figure 4 Reintroduction periods after elimination from the human and domestic population. The probability distribution of reintroduction periods for different rates of host switching (given in units of (number of switches)/fly/year) between a human/domestic and a wild animal subsystem (with random mixing within each of these two subsystems), given in years. The values were obtained from stochastic simulations, initialised with the prevalence in animal populations as measured in Bipindi, but with no infection present in humans, domestic animals, or human-associated vectors. Simulations were initialised with vectors, based on the number of around 2,000–3,000 flies captured in the area through entomological surveys lasting a few days [45]. We considered reintroduction to have occurred once there were 2 cases in humans at any given time. Discussion We have developed a mathematical model to assess transmission dynamics in a focus of gambiense HAT, and analysed it incorporating a variety of epidemiological and ecological measurements, providing one of the first estimates of in gambiense HAT from field data. If vectors and hosts mix randomly, we only need the prevalence in the different vector and host species, as well as the distribution of bites on host species, to determine the NGM and . In this case, the available data strongly suggest that T. b. gambiense cannot be sustained in a human (and vector) population alone, whereas independent transmission cycles in animal reservoirs are possible in a realistic parameter range. When reducing the human population at risk, we could not rule out the possibility of transmission cycles in humans and vectors. However, these occured only with a very small likelihood corresponding to very specific parameter combinations unless it was only a very small fraction of the human population that was exposed to the potential infection. While there are occupational hazards associated with trypanosomiasis infection (especially hunting [41]), these do not seem enough to explain such strong heterogeneity in risk. When we relaxed the assumption of random mixing to reduce the amount of infection transfer between humans and other species, human transmission cycles were only possible in parameter regimes where there was a parallel transmission cycle in wildlife. When we informed this analysis with measured distributions of species among habitats, independent transmission cycles in animals occured with high probability. Simulating the transmission dynamics of the model with different rates of vector switching between three subsystems of humans, domestic animals and wild animals, we observed that unless switching was rare, reintroduction of infection in humans usually occurred within less than a year. When, on the other hand, such a switch happened only in a minority of vector lifetimes, reintroduction could take many years, and there was the possibility a human-only cycle in parallel with a separate sylvatic cycle. The disease-free periods of 10 or more years subsequent to human case control that have been observed [17] would point to such a scenario. However, the effect of vector control combined with delayed recognition of new outbreaks due to infrequent screening and lack of gambiense HAT testing in routine health services may also explain long delays observed between apparent elimination of T. b. gambiense from a focus and its re-activation. Our analysis hinges on the assumption of equilibrium, which allowed us to estimate the force of infection from observed prevalence. While fluctuations in the density of the different species or the incidence of infection that they experience are likely, the slow dynamics of gambiense HAT combined with the long history of endemic transmission in Bipindi [42] would appear to justify the assumption of stationarity. Still, since the data underlying our study were taken at different points throughout the year, strong seasonality could mean that the measurements were not a good reflection of the average state of the system, as well as raising theoretical issues in linking persistence of an endemic disease to the value of [43]. While we cannot resolve this issue on the basis of the available data, we note that vector density was found not to vary significantly in the study area [44], and that the progression of gambiense HAT is slow relative to the progression of seasons, so that fluctuations in tsetse fly density need not translate into significant changes in prevalence. Further, it is worth noting that more detailed data on incidence would enable relaxation of the model assumptions and direct estimation of the force of infection. Moreover, molecular typing of parasite material could be used to quantify the contribution of non-human hosts to the force of infection in humans. Clarifying the precise role of animal hosts in maintaining transmission has important implications for elimination strategies. If wild animals can maintain T. b. gambiense in a separate transmission cycle, elimination (the permanent interruption of transmission) will be difficult to achieve with a strategy based on human case detection alone. At the same time, all our estimated likely values are very close to 1, suggesting that the disease should be controllable, especially if vector control is introduced and maintained. Beyond maintenance, animals could play a role in transmitting infections between communities within a given focus or indeed (re-)introduction into old, extinct foci or new areas. Gambiense HAT has remained a west and central African disease confined to persistent foci in spite of large-scale population movements around the continent. If transmission could be maintained in a human-vector system alone, one would expect the distribution of the disease to be more diffuse. Instead, one could speculate that restrictions of animal host ranges are at least to some degree responsible for the observed distribution. An intriguing hypothesis that arises from our results is that the apparent decline in gambiense HAT burden in many areas of west Africa (e.g., Gambia, Sierra Leone, Liberia, Nigeria) where it was previously highly endemic might be attributable mainly to the reduction in wildlife habitats and populations in these regions over the past decades. We have concentrated on an gambiense HAT focus in a region with a well-documented history of endemic transmission [42]. Extrapolation of our results to other settings warrants caution. Focus-specific levels of parasite strain virulence, vector competence or human susceptibility could combine to ensure sustained transmission in human-vector systems elsewhere. Similarly, species and distributions of domestic and wild animals vary considerably across foci. Nevertheless, this study offers an attractive explanation for the mysterious disappearance and re-activation of gambiense HAT foci throughout Africa. Our method is easily generalised to other foci, and further studies on the ecology and epidemiology of T. b. gambiense across different areas would firmly establish the role of wild and domestic animals in the maintenance of sleeping sickness, and to systematically assess the prospects of elimination efforts. In this study, we analysed one of the largest systems for which the NGM has been quantified from field data. Combined with efforts to measure infection prevalence in both humans and animals, our model framework could be applied to better characterise the role of animal hosts in the long-term control of many other diseases, such as yellow fever, rift valley fever or Chagas disease. Supporting Information Text S1 Model formulation. Contains details of relating the NGM to the basic and host-specific reproduction numbers in a multi-host, mulit-vector system; a detailed description of the modelling framework for HAT; and a brief overview of possible model extensions. (PDF) Click here for additional data file. Text S2 Data. Contains the gambiense HAT prevalence data in different host and vector species as well as measured biting preferences and other parameters of the gambiense HAT transmission model. (PDF) Click here for additional data file.
Author and article information
Contributors
João Pinto: Role: Editor
Journal
Journal ID (nlm-ta): PLoS One
Journal ID (iso-abbrev): PLoS ONE
Journal ID (publisher-id): plos
Journal ID (pmc): plosone
Title:
PLoS ONE
Publisher:
Public Library of Science
(San Francisco, USA
)
ISSN
(Electronic):
1932-6203
Publication date Collection: 2013
Publication date
(Electronic):
14
June
2013
Volume: 8
Issue: 6
Electronic Location Identifier: e67580
Affiliations
[2
]Centre International de Recherche-Développement sur l’Elevage en Zone subhumide (CIRDES),
Bobo-Dioulasso, Burkina Faso
[3
]Laboratoire National d’Elevage et de Recherches Vétérinaires, Institut Sénégalais
de Recherches Agricoles, Dakar, Sénégal
[4
]Unité Mixte de Recherche Contrôle des Maladies Animales Exotiques et Emergentes, Centre
de Coopération Internationale en Recherche Agronomique pour le Développement (CIRAD),
Montpellier, France
[5
]Unité Mixte de Recherche 1309 Contrôle des Maladies Animales Exotiques et Emergentes,
Institut National de la Recherche Agronomique (INRA), Montpellier, France
Author notes
* E-mail:
bouyer@
123456cirad.fr
Article
Publisher ID: PONE-D-12-37388DOI: 10.1371/journal.pone.0067580
PMC ID: 3682971
PubMed ID: 23799148
SO-VID: 251144bb-1712-40c5-9313-0e688c775445
Copyright statement:
Copyright @
2013
License:
This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
History
Date
received
: 28
November
2012
Date
accepted
: 21
May
2013
Funding
CEVA Santé Animale provided the insecticide (VectocidND) used in this study, and the
field work was funded by the Project “Pôle Regional de Recherche Appliquée au développement
des Systèmes Agricoles d’Afrique Centrale” (PRASAC) and the French Embassy in Chad.
FAO, in the framework of the Programme against African Trypanosomosis (PAAT), provided
technical assistance to this research, in particular through the project "Improving
food security in sub-Saharan Africa by supporting the progressive reduction of tsetse-transmitted
trypanosomosis in the framework of the NEPAD" (GTFS/RAF/474/ITA), funded by the Government
of Italy through the FAO Trust Fund for Food Security and Food Safety. The funders
had no role in study design, data collection and analysis, decision to publish, or
preparation of the manuscript. The boundaries and names shown and the designations
used on the maps presented in this paper do not imply the expression of any opinion
whatsoever on the part of the Food and Agriculture Organization of the United Nations
(FAO) concerning the legal status of any country, territory, city or area or of its
authorities, or concerning the delimitation of its frontiers or boundaries. The views
expressed in this paper are those of the authors and do not necessarily reflect the
views of FAO.
Categories
Subject:
Research Article
ScienceOpen disciplines: Uncategorized
Data availability:
ScienceOpen disciplines: Uncategorized