Introduction Lymphatic filariasis (LF) is a neglected tropical disease of global importance, with an estimated 1.4 billion people in 73 countries at risk of infection. Over 120 million people worldwide are currently affected by lymphatic filariasis and 40 million are disfigured and disabled [1]. Infection is transmitted by mosquito vectors including Anopheles, Aedes, Culex and Mansonia species. The Pacific Programme for Elimination of Lymphatic Filariasis (PacELF) was formed in 1999, and as part of the Global Programme to Eliminate LF (GPELF), aimed to eliminate the disease as a public health problem in 22 Pacific Island countries and territories (PICTs) by 2020 [2]. The Programme in the Pacific covers over 3000 islands and 8.6 million people, and consists of two strategies: firstly, to interrupt transmission through mass drug administration (MDA) using albendazole and diethycarbamazine (DEC) and secondly, to control morbidity and disability of affected persons [2]. Baseline surveys conducted in 1999 and 2000 determined that 11 PICTs were endemic for LF, five partially endemic, and six non-endemic [2]. Since then, variable progress has been made towards reducing prevalence and interrupting transmission on different islands [3], but significant success has been achieved in the Samoan Islands, particularly in American Samoa. Before the 1960s, both Samoa (formerly called Western Samoa) and American Samoa had high prevalence (∼20%) of lymphatic filariasis [4], [5]. Multiple rounds of MDA in the 1960s had considerable impact and reduced the prevalence of microfilaraemia to less than 2%, but neither Samoa nor American Samoa managed to achieve sustained interruption of transmission at that time [6]–[9]. By 1999, antigen prevalence of 16.5% (N = 3018) was recorded in American Samoa and 4.5% (N = 7006) in Samoa. In American Samoa, after seven rounds of MDA from 2000–2006, antigen prevalence dropped to 2.3% (N = 1881) in 2007 in a community cluster survey that involved all age groups [10]. Current WHO guidelines [11] recommend that in areas where W. bancrofti is endemic and Aedes is the principal vector, the target threshold for post-MDA transmission assessment surveys (TAS) is 128 Og4C3>32 Wb123 Bm14 N % N Prevalence OR p N Prevalence OR p N Prevalence OR p N Prevalence OR p Total samples 807 805 805 806 806 Total positive 6 0.7% 26 3.2% 65 8.1% 144 17.9% Gender Females 380 47.1% 0 0.0% - - 8 2.1% 1 11 2.9% 1 41 10.8% 1 Males 423 52.4% 6 1.4% - - 18 4.3% 2.1 0.1 54 12.8% 4.90 0.00 103 24.3% 2.7 0.00 Age (years) 70 36 4.5% 0 0.0% - - 1 2.8% 3.0 0.44 3 8.3% 3.1 0.18 9 25.0% 6.7 0.00 Years lived in AS Whole life 555 68.8% 4 0.7% 13 2.3% 39 7.0% 87 15.7% >10 years 718 89.0% 4 0.6% 1 19 2.6% 1 55 7.7% 1 129 18.0% 1 5 to 10 years 54 6.7% 0 0.0% - - 2 3.7% 1.4 0.65 5 9.3% 1.2 0.67 6 11.1% 0.6 0.21 $30,000 64 7.9% 0 0.0% - - 1 1.6% 0.5 0.45 2 3.1% 0.3 0.12 7 10.9% 0.48 0.08 Unknown 128 15.9% 0 0.0% - - 4 3.1% 0.9 0.89 10 7.8% 0.8 0.63 19 14.8% 0.68 0.18 Chi2/Fisher * Chi2/Fisher * Chi2/Fisher * Chi2/Fisher * Island of residence p p p p Tutuila 721 89.3% 6 0.7% 1.00 26 3.2% 0.10 57 7.1% 0.65 127 15.8% 0.62 Other islands 86 10.7% 0 0.0% 1.00 0 0.0% 0.10 8 1.0% 0.65 17 2.1% 0.62 Statistically significant results (p 128 units (positive according to manufacturer's instructions), and >32 units (positive and equivocal). The mean value recorded for the negative sera over 20 plates was 4.9 units. Wb123 antibody ELISA test The test detects antibody to the Wb123 antigen identified from a library generated from L3 larval stages of W.bancrofti [26]. The assay was performed in ELISA format using plates pre-coated with 10 µg/mL Wb123 antigen. Sera at 1∶50 dilution in PBS/1%BSA/0.05% Tween 20 (PBS/T) were added at 50 µL per well, 50 µL of a known positive control were added at 1∶500 (high positive) and 1∶5000 (low positive) while negative control serum was added at 1∶50. Plates were incubated for 30 minutes at room temperature, washed 6 times in PBS/T and 50 µL per well of HRP conjugated mouse anti-human IgG4 (Invitrogen A10654) at 1∶5000 dilution was added for 45 minutes incubation. After 6 washes TMB substrate was added at 50 µL per well and the reaction was stopped with 1 M HCl after 5 minutes in the dark. Plates were read at 450 nm. Samples were classed as positive if their average optical density (OD) ratio was 9 or more times the negative control (based on the average ratio of of the low positive 1∶5000 serum control to blank). The mean OD ratio of the negative control to blank over 19 plates was 1.0. Bm14 antibody ELISA test The test detects antibody to an antigen identified from a cDNA library screened using sera from microfilaria positive people [27]. Sera were tested at 1∶50 dilution using the method described by [27] and [12]. A known positive control from PNG (S19) and negative serum from lab members were included on each plate. A 7-point standard curve in duplicate using a known high positive serum from PNG (S200) starting at 1∶200 dilution (1000 arbitrary units) and then 2–fold serial dilutions in PBS/T was included on each plate. A 4-parameter calibration curve was used to estimate the units of Bm14 antibody per sample. The cutoff for positivity was 125 units determined empirically as described by [27] using known positive and negative serum panels. Statistical analysis Outcome measures used for statistical analyses were ELISA test results for each LF antigen and antibody. For Og4C3 antigen, statistical analyses were performed using two different cutoff points: >128 units (positive results) and >32 units (equivocal and positive results). Independent variables assessed included age, sex, years lived in American Samoa, occupation, household income, and island of residence. The number of years lived in American Samoa was categorized into 10 years (those who lived in Am Sam during all of the MDA activities). Occupation groups were categorized into those who worked i) predominantly indoors, ii) predominantly outdoors, iii) tuna cannery workers (the largest non-government employer in American Samoa; >90% of employees are migrant workers), and iv) others (including unemployed, unknown occupation, and those who have jobs that include both indoor and outdoor work). Data on household income was available in four categories. Island of residence was categorized into Tutuila and other islands. The serum bank consisted of samples and data on 807 participants. There was sufficient serum in 805 samples to perform ELISA for Og4C3 antigen, and in 806 samples for Wb123 and Bm14 antibodies. Data on gender were available for 803 participants, on age for 798, on years lived in American Samoa for 800, and on household income for 679. Island of residence and geo-locations of households were available for all participants. Chi-squared or Fisher exact tests were used to compare outcomes for categorical independent variables. Variables with p 128 units (positive result) were found in 0.75% (6 persons, 95% CI 0.3–1.6%) of participants, and levels of >32 units (equivocal plus positive results) in 3.2% (26 persons, 95% CI 0.6–4.7%). The seroprevalence of Wb123 and Bm14 antibodies were 8.1% (65 positives, 95% CI6.3–10.2%) and 17.9% (144 positives, 95% CI 15.3–20.7%) respectively. Factors associated with positive LF antigen and antibodies Table 1 provides a summary of the associations between demographic variables and the presence of LF antigen and antibodies. Our results show that both antigen and antibody prevalence were higher in males compared to females (Table 1). Figure 1 shows the age distribution of participants, and the prevalence of antigen (Og4C3>128 and Og4C3>32) and antibodies (Wb123 and Bm14) in each age group. Antigen-positive individuals were identified in all age groups, with no significant difference between ages. Prevalence of both Wb123 and Bm14 antibodies were higher in the older age groups. In participants aged 30 years and older, Bm14 prevalence was two to three times higher than Wb123 prevalence in all age groups. 10.1371/journal.pntd.0003297.g001 Figure 1 Prevalence of filarial antigen and antibodies by age groups, American Samoa 2010. Antibody and antigen prevalence were inversely associated with the number of years lived in American Samoa (Figure 2 and Table 1). Of all study participants, 68.8% (n = 555) had lived in American Samoa for all of their lives. Compared to individuals who had lived in American Samoa for over 10 years, new migrants who had lived there for 128 units, and odds ratio of 6.1 (95% CI: 1.9–19.4) of having Og4C3 antigen of >32 units (Table 1). New migrants also had higher prevalence of Wb123 and Bm14 antibodies compared to those who had lived in American Samoa for >10 years, but differences were not statistically significant. The prevalence of antibodies and antigen were higher in residents on the main island of Tutuila compared to those who lived in smaller islands, but differences were not statistically significant. Tuna cannery workers had significantly higher prevalence of Wb123 antibodies, but there were no other associations between occupational groups and seroprevalence. Our study did not find any association between income and seroprevalence. 10.1371/journal.pntd.0003297.g002 Figure 2 Prevalence of filarial antigen and antibodies by years lived in American Samoa. Geographical clustering of serological indicators For reference, a kernel density map of population distribution in American Samoa is shown in Figure 3 (reproduced from [23]). The household locations of individuals with positive and negative Bm14 and Wb123 antibodies are shown in Figure 4a and 4b, and positive/equivocal Og4C3 levels shown in Figures 5a and 5b. High resolution maps of the villages of Fagalii (Figure 6a) and Ili'ili (Figure 6b) show the locations of participants' households, those with positive/equivocal results for Og4C3, and the location of the elementary school where two ICT-positive children were identified during the 2011 Transmission Assessment Survey. 10.1371/journal.pntd.0003297.g003 Figure 3 Population distribution on the islands of American Samoa 2010 (Reproduced from Lau et al. (23). 10.1371/journal.pntd.0003297.g004 Figure 4 Household locations of individuals with positive and negative antibodies on Tutuila. A. Wb123, B. Bm14. 10.1371/journal.pntd.0003297.g005 Figure 5 Household locations of individuals with positive and negative antigen on Tutuila. A. Og4C3>128 units, B. Og4C3>32 units. 10.1371/journal.pntd.0003297.g006 Figure 6 High resolution village maps of A. Fagali'I and B. Ili'ili, showing household locations of individuals with Og4C3 antigen of >128 units and >32 units, and school where two ICT-positive children identified in 2011 TAS. While the semivariograms for Wb123 and Bm14 antibodies did not reveal any significant small-scale spatial variation, the semivariograms for antigen (both Og4C3>128 units and Og4C3>32 units) showed considerable residual spatial variation (Figure 7 and Table 2). Our results indicate that the average size of a cluster for Og4C3>128 units was 1,242 metres and the proportion of the variation in Og4C3>128 units explained by geographical proximity was 85%. The average size of a cluster for Og4C3>32 units was 1,498 meters and the proportion of the variation in Og4C3>32 units explained by geographical proximity was 62%. Migrants who had lived in American Samoa for 128 units, B. Og4C3>32 units, C. Wb123 positive, D. Bm14 positive. 10.1371/journal.pntd.0003297.t002 Table 2 Spatial parameters of geographical clustering of Og4C3 antigen, and Wb123 and Bm14 antibodies. Spatial parameters Og4C3>128 Og4C3>32 Wb123 Bm14 Range (meters) 1,242 1,498 60 NA Partial sill 0.00965 0.0451 0.015 NA Nugget 0.00173 0.0281 0.075 NA Proportion of variance due to spatial dependence (%) 85 62 17 NA Discussion Our study demonstrates that high-risk populations for LF in American Samoa include adult males and recent migrants. The results also suggest the possible existence of residual foci of antigen-positive individuals in American Samoa. Although our findings do not provide conclusive evidence of recent transmission, further investigation is recommended to confirm (or otherwise) the possible high-risk populations and locations, and determine whether ongoing targeted surveillance of these groups is warranted, particularly in the Samoan Islands where there is a history of resurgence despite achieving very low prevalence [4]. The prevalence of Wb123 and Bm14 antibodies differed significantly, and further research is required to understand the role of each laboratory test in post-MDA surveillance. There was a sharp rise in Bm14 antibody prevalence from age 30–39 years, which was also observed by Mladonicky et al in 2006 in three villages of American Samoa [9]. We found that Wb123 antibody prevalence peaked in participants aged between 30 and 40 years, but at much lower prevalence than Bm14 antibody. Wb123 antibody is a relatively new assay, and the indicative cutoff point used in this study could have contributed to the differences between the prevalence of Wb123 and Bm14 antibodies. Neither Wb123 nor Bm14 antibody prevalence declined with age, but at present we cannot distinguish long-term persistence of antibodies from ongoing transmission. Other studies have noted persistence of Wb123 antibodies in adults for many years after MDA, although significant decline was observed in those who were antigen-negative [28]. Positive Og4C3 antigen was found in all age groups and did not show any age-specific patterns. The presence of Og4C3 is not necessarily associated with microfilaraemia and does not provide evidence of ongoing transmission. Antigen prevalence drops dramatically after MDA, but it is not possible to unequivocally distinguish between recent or past infection based on Og4C3 alone. However, for W. bancrofti areas, the WHO currently supports the use of circulating filarial antigen prevalence (measured by ICT card test) as an indicator of LF infection, and it is one of the options of diagnostic tests used for measuring the prevalence of infection at each stage of the elimination process (pre-MDA mapping, sentinel and spot check sites, and TAS). The Og4C3 antigen has also been used in a similar study in Haiti that investigated clustering of residual antigen-positive persons in low endemic areas [19]. In our study, three aspects of the Og4C3-positive individuals raised suspicion about the possibility of recent transmission. Firstly, one cluster of Og4C3-positive adults was located in very close proximity to the two ICT-positive children found during the 2011 TAS. Secondly, Og4C3 prevalence in our study was higher in migrants (mostly from Samoa) even though baseline antigen prevalence in 1999 was much lower in Samoa (4.5%) than American Samoa (16.5%). If positive Og4C3 in our sample predominantly reflected infections in the remote past, prevalence would be expected to be lower in the migrants. Thirdly, we found significant spatial clustering of Og4C3 antigen, but not of Wb123 or Bm14 antibodies. If the Og4C3-positive adults in our study were predominantly infected in the remote past, clustering would have been much less likely, as demonstrated by the absence of clustering of antibody-positive adults. Data on microfilaraemia would have helped determine the presence of ongoing transmission, but this was not possible with a serum bank. Despite this limitation, we believe that our seroprevalence study of adults provided valuable information about potential residual infections in American Samoa. Similar studies should be considered elsewhere for post-MDA surveillance and for identifying high-risk populations and/or locations that might warrant more intense targeted surveillance. Higher LF seroprevalence in males corroborates findings from some of the previous studies in Samoa [7], [18] and American Samoa [29], and could be explained by more time spent outdoors for work and recreation compared to females. Interestingly, LF prevalence was found to be equivalent in males and females in 1999 prior to MDA in American Samoa, but Liang et al. reported a shift toward higher prevalence in males in sentinel site surveys conducted during and after MDA [29]. Our study (using a much larger and more representative sample of the adult population) confirms the higher prevalence among males post-MDA in American Samoa. Our results also indicate higher antigen prevalence in new migrants, who were mostly from a neighbouring LF-endemic country where transmission is still occurring in some areas. This suggests that human movement could be an important pathway for parasite reintroduction and subsequent resurgence of LF in American Samoa. Visitors and migrants travel for family, work, and economic reasons and usually live and work in close proximity to local American Samoans. Prolonged visits and cross migration are also common, and further increase the chances of parasite reintroduction. In addition, American Samoans also travel frequently to Samoa and other neighbouring Pacific Islands, and could be at significant risk of infection if staying for extended periods in areas of high prevalence. In 2012, there were a total of 67,979 international arrivals to American Samoa (with a local population of ∼56,000). Of these 44,830 were citizens of other Pacific Islands, including 22,600 arrivals of returning citizens of American Samoa. A total of 20,082 arrivals were Samoan citizens, with 158 travelling for business, 4,158 for employment, 7,123 returning residents, and 8,757 visiting relatives [22]. Further research is required to improve understanding of the role of human movement in parasite reintroduction into American Samoa, and the consequent risk of resurgence based on travel patterns between Samoa and American Samoa, and LF prevalence at places of origin of visitors and new migrants. Cross-border strategies to coordinate efforts between Samoa and American Samoa for LF elimination and surveillance should also be considered. American Samoa's population mostly live on ancestral land, and most of the study participants had lived in the same village for most or all of their lives, thus providing an excellent opportunity to examine disease transmission patterns. Our results indicate that most of the spatial distribution of antigenaemia could be accounted for by geographical proximity of place of residence. Geo-spatial analysis provided some evidence of possible micro-spatial clustering of antigen-positive adults at the neighborhood level at two villages. Clustering at the household level suggests that the home environment is important in transmission even though one of the major vectors is day-biting. The close proximity between the elementary school attended by the two ICT-positive young children identified during the 2011 TAS and one of the possible village clusters suggests possible ongoing transmission. Our results indicate an average cluster size of 1,200 meters to 1,500 meters for antigenaemia, and the estimate of cluster size provides important information for the design of further studies to identify local transmission foci. Our study demonstrates the potential value of geospatial databases in post-intervention surveillance, monitoring, and evaluation for identifying possible micro-spatial clusters that might not be captured by routine TAS alone. Early detection of such clusters could be essential for timely intervention to reduce the risk of resurgence. Geospatial analysis could therefore potentially be used as an additional tool for verifying elimination status and for confirming that transmission has been interrupted. Changes in the spatial distribution of serological markers over time would also potentially be useful for identifying focal transmission, but unfortunately results of previous surveys in American Samoa were only located to the village rather than household level, and not of sufficiently high spatial resolution for the types of analyses conducted in this study or for comparing changes over time. Further operational research could also explore the use of geospatial data for informing programme delivery (e.g. by identifying the size of clusters and delineating areas that might warrant targeted surveillance and monitoring); calculating the distance of influence on infection risk that antigen-positive persons have on their near neighbours; and determining transmission threshold targets that include a spatial component rather than just a simple average prevalence for an entire evaluation unit. The accuracy of prevalence estimations in evaluation units will also depend on spatial heterogeneity within the boundaries of the unit. Risk of LF and drivers of transmission are unlikely to be entirely uniform within any evaluation units, and be determined by many factors such as climatic conditions, population density, urban versus rural areas, MDA coverage, and vector species and density. The average prevalence in an evaluation unit could therefore mask focal areas of high prevalence (hotspots) if they are surrounded by large areas of low prevalence. Consequently, estimations of average prevalence in an area could vary greatly depending on how evaluation units were determined. Hotspots are more likely to be missed if they are small, in evaluation areas with greater spatial heterogeneity in risks and drivers, and when prevalence is very low such as in the post-MDA surveillance phase. Careful definition of evaluation units will therefore be crucial for optimising the probability of identifying any residual hotspots of transmission or early resurgence. One of the challenges pertaining to geospatial methods of cluster detection when utilising point location data is that such data are prone to random error and random variation in the presence of rare disease events and/or inadequate representation of the population at risk. We therefore used a robust geostatistical method to identify the presence of geographical clustering in our point location data by partitioning the variation in data that was due to random error and the variation that is due to spatial clustering. Semivariography (as utilized in this study) demonstrated that spatial clustering was present in the study area (Tutuila) but does not identify the location of clusters. The location of clusters could be further investigated by using model-based geostatistics that account for diagnostic uncertainty and variation in factors such as climate, population, and entomological parameters to produce predictive risk maps of LF. Spatial decision support systems are being used for malaria elimination programs, and similar tools could also be useful for LF [30]. A geospatial platform could also be used to integrate environmental and entomological data with human surveillance data, and used to explore possible environmental drivers of disease transmission, the impact of vector control on elimination programs, and the potential for using xenomonitoring to enhance post-MDA surveillance. This study also demonstrates the usefulness of high-quality serum banks for investigating multiple diseases (a dengue seroprevalence study was also conducted using the same serum bank [31]), and provides an example of successful collaboration between researchers of different diseases to improve the cost-effectiveness of field epidemiology investigations, which are often expensive and logistically challenging. We believe that the WHO's recommendations of integrating of LF surveillance activities with other population-based surveys are logistically feasible and practical. Our findings should be interpreted in light of potential limitations. First, the serum bank used for the study was collected for a leptospirosis study, and we could not ascertain whether participants had previously been diagnosed with or treated for LF, or participated in MDA in American Samoa or elsewhere. Only 28 participants (3.9%) were recent migrants (lived 128 units, and 26 participants with Og4C3 of >32 units, and small numbers could have affected the accuracy of spatial analyses. Small numbers generally reduce the likelihood of identifying statistically significant associations, but despite this, we found significant results using robust tests and geospatial analyses. Third, participants in the serum bank included adults of all ages, but did not include children or adolescents. Results of TAS conducted at about the same time provided antigen prevalence data in 6–7 year old children, but data on antigen and antibody levels in children of all ages would help improve understanding of the application of diagnostic tests for post-MDA surveillance. Finally, participants were geo-located to place of residence, but LF infection could occur elsewhere, particularly in the presence of efficient day-biting vectors. If vectors were predominantly night-biting, clustering of infections could potentially be even more readily defined around household locations. This study provides preliminary results to support the importance of further research designed to specifically focus on improving understanding of disease transmission at the last stages of elimination when prevalence is very low; answering operational questions in LF elimination programs, especially the role of migration; developing tools to enhance the effectiveness of post-MDA surveillance and monitoring; and providing an evidence base for elimination strategies and targets. Follow up studies are being conducted in American Samoa to determine whether hotspots truly exist, develop models to quantify the significance of migrants in LF elimination, and explore the use of molecular xenomonitoring in the Pacific Island setting. The study also highlights the importance of assessing locally relevant risks for infection, which could vary significantly between places depending on cultural, societal, and environmental factors, as well as filarial species and mosquito vectors. The approach and results of this study are specifically relevant for the Samoan islands, but could also provide insight into LF transmission in other LF-endemic areas, and be pertinent to other Pacific Islands with similar vectors, lifestyle, culture, climate, environmental conditions, and migration patterns. Supporting Information Checklist S1 STROBE Checklist for observational studies. (DOC) Click here for additional data file.
