{"paper_id":"41923502-b0d9-456d-9b99-8e442efdec0d","body_text":"1\n1 Predictive risk mapping of lymphatic filariasis residual hotspots in American Samoa \n2 using demographic and environmental factors\n3\n4 Short title: Spatial prediction of Lymphatic filariasis in American Samoa\n5 Authors\n6 Angela M. Cadavid Restrepo 1*, Beatris Mario Martin1, Saipale Fuimaono2, Archie C.A. \n7 Clements 3, Patricia M. Graves4, Colleen L. Lau1\n8 Affiliations\n9 1. School of Public Health, Faculty of Medicine, The University of Queensland, Brisbane, \n10 Queensland, Australia\n11 2. American Samoa Department of Health, Pago Pago, American Samoa\n12 3. Curtin School of Population Health, Faculty of Health Sciences, Curtin University, Perth, \n13 Western Australia, Australia\n14 4. College of Public Health, Medical and Veterinary Sciences, James Cook University, \n15 Cairns, Queensland, Australia\n16\n17 *Corresponding author\n18 E-mail: a.cadavidrestrepo@uq.edu.au (AMCR)\n19\n20\n21\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n2\n22 Abstract\n23 Background\n24 American Samoa successfully completed seven rounds of mass drug administration (MDA) \n25 for lymphatic filariasis (LF) from 2000-2006. The territory passed the school-based \n26 transmission assessment surveys in 2011 and 2015 but failed in 2016. One of the key \n27 challenges after the implementation of MDA is the identification of any residual hotspots of \n28 transmission. \n29\n30 Method\n31 Based on data collected in a 2016 community survey in persons aged ≥8 years, Bayesian \n32 geostatistical models were developed for LF antigen (Ag), and Wb123, Bm14, Bm33 \n33 antibodies (Abs) to predict spatial variation in infection markers using demographic and \n34 environmental factors (including land cover, elevation, rainfall, distance to the coastline and \n35 distance to streams). \n36\n37 Results\n38 In the Ag model, females had a 29.6% (95% CrI: 16.0–41.1%) lower risk of being Ag-\n39 positive than males. There was a 1.4% (95% CrI: 0.02–2.7%) increase in the odds of Ag \n40 positivity for every year of age. Also, the odds of Ag-positivity increased by 0.6% (95% CrI: \n41 0.06–0.61%) for each 1% increase in tree cover. The models for Wb123, Bm14 and Bm33 \n42 Abs showed similar significant associations as the Ag model for sex, age and tree coverage. \n43 After accounting for the effect of covariates, the radii of the clusters were larger for Bm14 \n44 and Bm33 Abs compared to Ag and Wb123 Ab. The predictive maps showed that Ab-\n45 positivity was more widespread across the territory, while Ag-positivity was more confined \n46 to villages in the north-west of the main island. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n3\n47\n48 Conclusion\n49 The findings may facilitate more specific targeting of post-MDA surveillance activities by \n50 prioritising those areas at higher risk of ongoing transmission.\n51\n52 Key words\n53 Wuchereria bancrofti; mass drug administration surveillance; environment; geographic \n54 information systems; Pacific Islands; geostatistics; predictive model\n55\n56\n57\n58\n59\n60\n61\n62\n63\n64\n65\n66\n67\n68\n69\n70\n71\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n4\n72 Author summary\n73 The Global Programme to Eliminate Lymphatic filariasis (LF) aims to interrupt transmission \n74 by implementing mass drug administration (MDA) of antifilarial drugs in endemic areas; and \n75 to alleviate suffering of those affected through improved morbidity management and \n76 disability prevention. Significant progress has been made in the global efforts to eliminate \n77 LF. One of the main challenges faced by most LF-endemic countries that have implemented \n78 MDA is to effectively undertake post-validation surveillance to identify residual hotspots of \n79 ongoing transmission. American Samoa conducted seven rounds of MDA for LF between \n80 2000 and 2006. Subsequently, the territory passed transmission assessment surveys in \n81 February 2011 (TAS-1) and April 2015 (TAS-2). However, the territory failed TAS-3 in \n82 September 2016, indicating resurgence. We implemented a Bayesian geostatistical analysis to \n83 predict LF prevalence estimates for American Samoa and examined the geographical \n84 distribution of the infection using sociodemographic and environmental factors.  Our \n85 observations indicate that there are still areas with high prevalence of LF in the territory, \n86 particularly in the north-west of the main island of Tutuila. Bayesian geostatistical \n87 approaches have a promising role in guiding programmatic decision making by facilitating \n88 more specific targeting of post-MDA surveillance activities and prioritising those areas at \n89 higher risk of ongoing transmission.\n90\n91\n92\n93\n94\n95\n96\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n5\n97 Introduction\n98 Lymphatic filariasis (LF) is a vector-borne parasitic disease caused by three species of filarial \n99 worms – Wuchereria bancrofti , Brugia malayi, and B. timori (1). The presence of adult \n100 worms in the lymphatic vessels leads to damage of the lymphatic system, causing clinical \n101 disease characterised by lymphoedema of the limbs or genitals, such as elephantiasis and \n102 scrotal hydrocoeles (1). LF is one of the leading causes of chronic disability worldwide, being \n103 responsible for over 5 million disability-adjusted life years before the implementation of \n104 elimination strategies against the infection (2, 3).\n105 In 1997, the World Health Organization (WHO) targeted LF for global elimination as a \n106 public health problem by 2020 (4). Subsequently, WHO launched the Global Programme to \n107 Eliminate Lymphatic Filariasis (GPELF) in 2000 that included two strategies: first, the \n108 implementation of mass drug administration (MDA) to interrupt the community-level \n109 transmission of LF, and second, management and prevention of morbidity and disability for \n110 people with chronic complications (5). By 2019, 72 countries were still considered endemic \n111 by the GPELF and 50 still required MDA (6). A number of countries have already achieved \n112 validation of LF elimination as a public health problem after intensive community-based \n113 MDA programs (including Cambodia, The Cook Islands, Egypt, Kiribati, Malawi, Maldives, \n114 Marshall Islands, Niue, Palau, Sri Lanka, Thailand, Togo, Tonga, Vanuatu, Viet Nam, Wallis \n115 and Futuna, and Yemen) (6). Some countries have stopped MDA and are under surveillance \n116 to determine if LF elimination criteria have been met. One of the main challenges faced by \n117 most LF-endemic countries that have implemented MDA is to effectively undertake post-\n118 MDA and post-validation surveillance (7). \n119\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n6\n120 American Samoa successfully completed seven rounds of MDA with a single dose of \n121 diethylcarbamazine (DEC) and albendazole from 2000-2006. Subsequently, the territory \n122 passed the WHO-recommended school-based transmission assessment surveys (TAS) \n123 conducted in 2011 (TAS-1) and 2015 (TAS-2) with crude prevalences of Ag-positive of 0.2% \n124 (95% confidence interval (CI) 0.0 to 0.8%) and 0.1% (95% CI 0.0 to 0.7%), respectively (8, \n125 9). Despite this achievement, the territory failed TAS-3 in 2016 with an adjusted Ag \n126 prevalence of 0.7% (95% CI 0.3 to 1.8%), higher than the threshold and the recommended \n127 upper confidence limit of 1% (10). The findings in TAS-3 suggested potential resurgence of \n128 LF in the territory and were confirmed by a community-based survey conducted in the same \n129 year with an Ag prevalence of 6.2% (95%CI 4.5 to 8.6%) in individuals aged >8years (10). \n130 Evidence from others studies conducted in the territory in 2010 and 2014 also suggested \n131 ongoing LF transmission and the potential persistence of residual hotspots (11-14).\n132\n133 WHO recommends conducting follow-up surveys of nearby households of Ag-positive \n134 children identified through TAS to complement post-MDA surveillance (15). However, the \n135 recommendations are vague and lack a clear threshold for triggering a programmatic response \n136 (16). As LF prevalence decreases, the ability of the diagnostic methods, particularly in the \n137 TAS, to detect areas with residual transmission is also limited (7). This limitation is of \n138 particular importance in areas where the geographical distribution of LF has been \n139 demonstrated to be highly heterogeneous (17). In American Samoa, a recent study confirmed \n140 clustering of the infection in areas that were previously suspected as hotspots in 2010 and \n141 2014 (Fagali’i village in the far north-west of Tutuila island, and also in the Ili’Ili-Vaitogi-\n142 Futiga area that is located on the south coast) and identified other potential areas where there \n143 is still potential residual infection (11, 18). Therefore, strategies for identifying foci of \n144 infection in low-prevalence settings are crucial in the context of the LF elimination efforts, \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n7\n145 both from the perspective of targeting communities for MDA and also for understanding the \n146 future of the post-MDA surveillance needs.\n147\n148 W. bancrofti, B. malayi, and B. timori require two hosts to complete its life cycle, the human \n149 and the mosquito hosts. Therefore, sociodemographic, economic and environmental factors \n150 that act at different spatial scales have the potential to influence the transmission pathways of \n151 the parasites (19). The clustered distribution of LF has been associated with landscape \n152 characteristics and climatic factors in several LF-endemic areas including the Pacific Islands \n153 (20, 21). Bayesian model-based geostatistics combining socio-demographic and \n154 environmental data with infection prevalence data proven to be able to predict disease \n155 distribution in areas with scarce information (22-24). Hence, understanding how \n156 environmental and sociodemographic factors interact to determine parasite transmission is \n157 essential for the design and implementation of effective elimination strategies against LF.\n158\n159 The aim of this study was to identify areas where there is potential residual transmission of \n160 W. bancrofti in American Samoa and produce LF predictive prevalence maps that can be \n161 used to help guide and target future LF elimination strategies. A Bayesian model-based \n162 geostatistics approach was used to: (i) assess and quantify the associations between LF \n163 infection markers and sociodemographic and environmental factors at the household level \n164 and (ii) develop spatial prediction of prevalence estimates of LF in American Samoa using \n165 different infection markers – Ag and antibodies (Abs) against Wb123, Bm14, Bm33.\n166\n167 Methods\n168 1. Ethical considerations\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n8\n169 Ethical approval for the 2016 field survey was obtained from the American Samoa \n170 Institutional Review Board and the Human Research Ethics Committee at the Australian \n171 National University (protocol number 2016/482) and the University of Queensland \n172 (2021/HE000896). After explaining the purpose and procedures of the survey, all adults and \n173 parents/guardians of the minors (<18 years) who agreed to participate were asked to sign an \n174 informed written consent form. Full details of local collaborations and official permissions to \n175 visit villages have been previously described (10).\n176  \n177 2. Study area\n178 American Samoa is a United States territory in the South-central Pacific located \n179 approximately between latitudes 11° North and 15° South and longitudes 168° East and 172° \n180 West (Fig 1). The total land area of the territory is 200 km 2 and comprises five inhabited \n181 volcanic islands Tutuila, Aunu’u, Ofu, Olosega and Ta’ū, and two remote coral atolls \n182 (Swains Island and Rose Atoll). In 2010, the population of American Samoa was 55,519, the \n183 majority of whom (95%) lived in Tutuila, the largest island (198.9 km 2), where the capital \n184 Pago Pago is located.\n185\n186 American Samoa lies in the tropical savanna climate zone characterized by alternate wet \n187 (October to May) and dry (June to September) seasons (25). Temperatures vary slightly \n188 between the hottest period (December to April), when the average is approximately 31°C, \n189 and the coolest period (June to August), when the average is 29 °C. The annual average \n190 rainfall ranges from 3000 to 6000 mm, with 70% occurring during the hot and wet season. \n191 The average elevation is 482 meters (m) with the highest point being Lata Mountain on the \n192 island of Ta’ū (970 m). \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n9\n193\n194 Fig 1.  Map of American Samoa and the distribution of the built-up areas in the main \n195 island of Tutuila\n196\n197 3. Data from community survey of lymphatic filariasis in 2016\n198 Data on LF infection markers, Ag and Wb123, Bm14 and Bm33 Abs, were obtained from a \n199 two-stage equal probability cluster survey conducted in American Samoa in 2016. Full details \n200 about survey design and sampling methods have been previously reported (10). Briefly, 30 \n201 primary sampling units (PSUs) were randomly selected from a total of 70 villages/village \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n10\n202 segments/village groups, that were defined based on a population size of less than 2000. Two \n203 villages that were previously identified and confirmed as LF hotspots in 2010 and 2014, \n204 respectively, were also added to the survey as PSUs (11). Within each PSU, a population \n205 proportionate sampling method was implemented to randomly select households from a geo-\n206 referenced list of buildings obtained from the American Samoa Department of Commerce \n207 (26). In total, the survey included 32 PSUs (across 30 villages) and 754 households. A \n208 household member was defined as an individual who considered the selected house as their \n209 principal place of residence or who slept in that house the previous night. All consenting \n210 household members aged ≥8 years were surveyed and blood samples were tested for \n211 circulating filarial Ag using the Alere TM Filariasis Test Strip (FTS) (Abbott, Scarborough, \n212 ME) (27) and for Wb123, Bm14 and Bm33 Abs using multiplex bead assays (MBA) (28).\n213\n214 Standardised electronic questionnaires were administered by bilingual field research \n215 assistants (in Samoan or English based on each participant’s preference). The demographics \n216 data collected included sex, age and work location. Work location was categorised as indoor, \n217 outdoor, tuna cannery (largest private employer in American Samoa), and other (including \n218 mixed indoor/outdoor, unemployed, retired or unknown). \n219\n220 4. Geospatial data sources\n221 We downloaded and assembled spatial and environmental data that have been found to be \n222 associated with the geographical distribution of LF in other endemic regions (23, 29-32). The \n223 boundary administrative maps and the covariate data consider for the analyses were derived \n224 from the following datasets: \n225\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n11\n226 i) Village boundaries and buildings. A map of village boundaries and the distribution of \n227 all buildings in the territory were downloaded from the Fagatele Bay National Marine \n228 Sanctuary GIS data archive website (33).\n229 ii) Coastline and streams. The American Samoa coastline and network of streams \n230 covering the entire territory were extracted in a shapefile format from the Fagatele \n231 Bay National Marine Sanctuary GIS data archive website (33).\n232 iii) Population density. Data on population density for 2010/2011 were downloaded from \n233 the Pacific Data Hub website (34). A grid (i.e. raster surface) was available for \n234 American Samoa at the resolution of 100 m. \n235 iv) Elevation. Data were obtained in a GeoTIFF format at the spatial resolution of 10 m \n236 from the United States Geological Survey (USGS) 10-m Digital Elevation Model \n237 (DEM): American Samoa: Tutuila (35).\n238 v) Rainfall. Average monthly rainfall for 2016 were downloaded from the Pacific \n239 Environment Data Portal (36) in a raster format  at the spatial resolution of 1 km. \n240 There was limited availability of spatial monthly rainfall datasets for the years prior to \n241 the survey. Therefore, the monthly rainfall layers from 2016 were used based on the \n242 assessment of the representativeness of the ten-year period prior the survey \n243 (Supplementary material).\n244 vi) Land surface temperature. Satellite sensor data on land surface temperatures from the \n245 Moderate Resolution Imaging Spectroradiometer (MODIS) satellite were obtained \n246 from the USGS Earth Explorer website (37). These data were downloaded at 1 km \n247 resolution for every eight days from January 1 to December 31 2016. \n248 vii) Land use/land cover map. Data were derived at 10m resolution from the Sentinel-2 \n249 Global Land Use/Land Cover (LULC) Timeseries produced by Impact Observatory, \n250 Microsoft, and the Environmental Systems Research Institute (Esri) (38).\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n12\n251\n252 5. Covariate data download and processing\n253 The geo-referenced data sets that included the locations of the surveyed households, the \n254 covariates and the boundary map of American Samoa were imported into ArcGIS version \n255 10.7.1 (39) to extract data (measured on a continuous scale) for the territory.  The \n256 geographical distributions of the covariates are shown in Fig 2.\n257\n258  Elevation estimates for the territory were extracted in meters (m) above sea level.\n259  A layer of the distance between each household location and the nearest coastline was \n260 developed (in m) using the Euclidean Distance Tool.\n261  The Euclidean Distance Tool was also used to estimate the distance (in m) between \n262 each household location and the nearest permanent surface stream. \n263  The monthly rainfall (mm) datasets were used to estimate the annual average rainfall \n264 and rainfall of the driest (August) and wettest (December) months in 2016. \n265  Annual average temperature and temperature of the hottest (December) and coolest \n266 (July) months in 2016 were estimated from the fortnightly temperature layers. \n267  The global LULC cover map with 11 LULC classes was used to generate four separate \n268 rasters for the LULC categories that cover the territory of American Samoa: crops, \n269 rangelands, trees, and built/urban area (Table 1).\n270\n271 Table 1. Land cover class definitions\nLand Use/Land cover \nClass\nDescription\nCrops Human planted/plotted cereals, grasses, and crops not at tree height. \nExamples: corn, wheat, soy, fallow plots of structured land.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n13\nRangeland Open areas covered in homogenous grasses with little to no taller \nvegetation; wild cereals and grasses with no obvious human plotting \n(i.e., not a plotted field). Examples: natural meadows and fields with \nsparse to no tree cover, open savanna with few to no trees, \nparks/golf courses/lawns, pastures. Mix of small clusters of plants \nor single plants dispersed on a landscape that shows exposed soil or \nrock; scrub-filled clearings within dense forests that are clearly not \ntaller than trees; examples: moderate to sparse cover of bushes, \nshrubs and tufts of grass, savannas with very sparse grasses, trees or \nother plants.\nTrees Any significant clustering of tall (~15 feet or higher) dense \nvegetation, typically with a closed or dense canopy. Examples: \nwooded vegetation, clusters of dense tall vegetation within \nsavannas, plantations, swamp or mangroves (dense/tall vegetation \nwith ephemeral water or canopy too thick to detect water \nunderneath).\nBuilt/Urban Human made structures; major road and rail networks; large \nhomogenous impervious surfaces including parking structures, \noffice buildings and residential housing. Examples: houses, dense \nvillages / towns / cities, paved roads, asphalt.\n272\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n14\n273\n274 Fig 2. The geographical distributions of the covariates in American Samoa (a) \n275 Population density (people/m2), (b) Annual average rainfall (mm), (c) Land cover, (d) \n276 streams and coastline, (e) temperature (K x 0.02) and (f) Elevation (m).\n277\n278 6. Buffer zones \n279 The GPS locations of the surveyed households were used to delineate a buffer zone of 20 m \n280 around the households in ArcGIS (39). The buffer size was selected to represent an \n281 approximate distance within which the participants would spend extensive periods of time, \n282 and therefore have greatest exposure to the environmental conditions withing the buffers \n283 (40). For each surveyed location, the data extracted within the buffer zone included the \n284 spatial mean values of population density, distance to coastline and streams, elevation, annual \n285 average rainfall, rainfall in the wettest (December) and driest (August) months in 2016, \n286 annual average temperature, and temperature of hottest (January) and coolest (July) months \n287 in the same year. Each of the four land cover classes covering the American Samoa territory \n288 were summarised as percentages of area within the 20 m buffer.\n289\n290 7. Descriptive analyses\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n15\n291 For each infection marker and the covariates, summary statistics were calculated in R \n292 software R-4.0.3 (41). Crude prevalence of Ag, and Wb123, Bm14 and Bm33 Abs were \n293 estimated and mapped at the village level, and binomial exact methods were applied to \n294 estimate 95% confidence intervals (95% CI). Of note, in all subsequent analyses data were \n295 examined at the individual level and the respective household locations. \n296\n297 8. Variable selection \n298 Collinearity between covariates was assessed using Spearman’s correlation. Non-spatial \n299 univariate logistic regression models were developed using R software R-4.0.3  (41) to \n300 examine the association of each LF infection marker (outcome variables) with the \n301 sociodemographic and environmental factors (covariates). For the strongly correlated \n302 covariates (Spearman’s correlation coefficient ρ > 0.9), the ones with the highest value of \n303 Akaike Information Criterion (AIC) in the univariate regression models was excluded. For \n304 each infection marker, multivariate logistic regression models were developed incorporating \n305 the remaining covariates. From these models, covariates were sequentially removed to assess \n306 AIC and p-values. The models with the lowest AIC were selected for further analyses and \n307 covariables with p <0.05 were retained.\n308\n309 9. Multivariable non-spatial and spatial regression models\n310 Bayesian geostatistical multivariate regression models were fitted using the OpenBUGS \n311 software version 3.2.3 rev 1012 (42). For each infection marker, separate logistic regression \n312 models were developed based on the binary outcome of the laboratory results. First, Bayesian \n313 geostatistical models were developed with the sociodemographic and environmental \n314 covariates as fixed effects but without considering the spatial dependence of the data. Then, \n315 Bayesian geostatistical models for each infection marker were fitted using Markov chain \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n16\n316 Monte Carlo (MCMC) methods. The MCMC approach was selected for the geostatistical \n317 analyses because it allows the model to incorporate spatial dependence in both the infection \n318 and covariate data, and also enables full representation of uncertainty in model outputs (43). \n319\n320 The deviance information criterion (DIC) statistic was calculated to assess if the inclusion of \n321 spatial dependence in the data improved the fit of the models. Low DIC values indicate a \n322 better fit. Covariates in the models were considered statistically significant if the 95% \n323 credible intervals (95% CrI) of the estimated odds ratios (OR) excluded 1. \n324\n325 The mathematical notation of the spatial model is provided below, and contains all of the \n326 components of the non-spatial model. Assuming a Bernoulli-distributed dependent variable, \n327 Yij, corresponding to the results of the infection markers (0=negative, 1=positive) of the ith \n328 participant (I = 1. . .2,671) the j th location (j= 1. . .736), the model structures were as follows: \n329\n330 𝑌𝑖𝑗 ~ 𝐵𝑒𝑟𝑛(𝑝𝑖𝑗) \n331\nlogit (𝑝𝑖𝑗)\n=  𝛼 + 𝛾 𝗑 age 𝑖 + δ × female𝑖 + ε × outdoor𝑖 + η × mixed𝑖 + ρ × starkist𝑖 + ν\n× student𝑖 + θ × other𝑖 + σ × unemployed𝑖 +\n𝑧\n𝑧=1\n𝛽𝑧  𝗑 λ𝑧𝑗 +  𝑠𝑗\n332\n333 where α is the intercept, γ and δ are coefficients for age and females, and ε, η, ρ, ν, θ and σ are \n334 coefficient for the occupation categories. β is a matrix of z coefficients, λ is a matrix of z \n335 environmental variables and population density, and sj a geostatistical random effect. The \n336 correlation structure of the geostatistical random effect was assumed to be an exponential \n337 function of the distance between points: \n338\n339 𝑓(𝑑𝑘𝑙;ϕ) = exp[–ϕ𝑑𝑘𝑙]\n340\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n17\n341 where d kl are the distances between pairs of points k and l, and ϕ is the rate of decline of \n342 spatial correlation per unit of distance. A normal distribution was used for the priors for the \n343 intercept and the coefficients (mean = 0 and precision, the inverse of variance, = 1 × 10 –3), \n344 whereas a uniform distribution was specified for ϕ (with upper and lower bounds s= 0.03 and \n345 100; the lower bound set to ensure spatial correlation at the maximum separating distance \n346 between survey locations was <0.5). A non-informative gamma distribution was used to \n347 specify the priors for the precision (shape and scale parameters = 0.001, 0.001).\n348\n349 A burn-in of 1,000 iterations were run first and discarded. Sets of 10,000 iterations were then \n350 run and examined for convergence. Convergence was assessed by visual inspection of history \n351 and density plots and by examining autocorrelation of the model parameters. In each model, \n352 convergence was achieved for all variables at approximately 30,000 iterations. The last \n353 10,000 values from the posterior distributions of each model parameter were recorded. The \n354 rate of decay of spatial correlation between locations (ϕ) with distance and the variance of the \n355 spatial structured random effect (σ 2) were also stored. \n356\n357 10. Predicted prevalence of lymphatic filariasis \n358 To predict LF prevalence at unsampled locations, a regular 150 m × 150 m grid was overlaid \n359 on a map of American Samoa to extract the average environmental data for each grid cell. \n360 The predicted probabilities at the unsampled locations were estimated using the \n361 spatial.unipred function in OpenBUGS. The function applies the model equation at each \n362 unsampled location using the covariates values extracted for them and the distance between \n363 those locations and the surveyed locations. Bayesian kriging was applied in ArcGIS to \n364 generate smooth risk maps of the posterior distributions of predicted prevalence of each LF \n365 infection marker.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n18\n366\n367 Results\n368 1. Sample description and sample site locations\n369 The final dataset used for analyses included 754 households in 736 unique locations (some \n370 households shared the same building structure) from 32 PSUs in 30 villages. The total \n371 number of participants was 2,671 with a mean age of 33.5 years (range 8–93), and 54.7% (n \n372 = 1462) were female. Figs 3a and 3b show the locations of sampled villages and the \n373 geographical distribution of the survey locations, respectively. The highest overall crude \n374 prevalence was observed for Bm33 Ab (45.6%, 95% CI 43.7−47.5%), followed by Wb123 \n375 Ab (25.6%, 95% CI 24.0− 27.3%), Bm14 Ab (13.1%, 95% CI 11.8−14.4%) and Ag (5.1, \n376 95% CI 4.2−5.9%). At the village level, Fagali'i (n=81) and Fagamalo (n=13), located in the \n377 far north-west of Tutuila Island, consistently showed high overall crude prevalence of all \n378 infection markers. A detailed description of the Ag and Ab results has been presented \n379 elsewhere (7, 10, 18). Figs 3c, 3d, 3e and 3f display the observed geographical distributions \n380 of the prevalence of Ag and Wb123, Bm14 and Bm33 Abs, respectively, by village. The \n381 maps confirm that villages with high prevalence of Bm33 Ab were more widespread across \n382 the territory, while the distribution of villages with high prevalence of Ag, Wb123 and Bm14 \n383 Ab was more confined to the north-west of the territory.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n19\n384\n385 Fig 3. a) Distribution of the sampled villages on the island of Tutuila, American Samoa \n386 2016, b) surveyed household locations, and village-level seroprevalence of c) Ag, d) \n387 Wb123 Ab, e) Bm14 Ab and f) Bm33 Ab.\n388\n389 2. Variable selection and univariate regression analyses\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n20\n390 The descriptive statistics and maps of the covariates considered for the analyses are presented \n391 in Table 2 and Supplementary Fig 1, respectively. Because temperature data were not \n392 available for large areas of American Samoa, this covariate was excluded from analyses. We \n393 identified four pairs of variables with Spearman’s rank >0.9 that were assessed with the \n394 univariate regression models. After comparing the AIC of the stepwise multivariate logistic \n395 regression models, the selected variables for the Bayesian non-spatial and spatial analyses \n396 included: sex, age, work location, population density, elevation, rainfall in the wettest month \n397 (December), distance to streams, cropland, tree coverage and urban areas.\n398\n399 Table 2. Descriptive statistics of environmental covariates within 20 m buffers of \n400 surveyed household locations in American Samoa in 2016\nVariable Mean Median Standard \ndeviation \nMinimum Maximum\nPopulation density (people/m2) 2.47 2.26 2.06 0.01 12.53\nElevation (m) 77.72 43.96 92.99 0 479.86\nDistance to the coastline (m) 1081.97 644.25 1117.78 3.06 4371.57\nDistance to streams (m) 358.25 151.12 434.39 0.32 1825.00\nAverage annual rainfall (mm) 3469.69 3374.57 698.28 2095.10 4630.68\nRainfall in the driest month - \nAugust (mm)\n232.46 232.97 51.99 125.27 324.61\nRainfall in the wettest month - \nDecember (mm)\n381.14 369.73 60.28 255.96 528.90\nLand cover\n   Cropland (%) 0.01 0 0.03 0 0.91\n   Rangeland (%)  0.85 0 6.83 0 0.99\n   Tree coverage (%) 5.14 0 19.93 0 99.00\n   Urban (%) 76.36 100 34.38 0 100.00\n401\n402\n403 3. Bayesian non-spatial and spatial models\n404 For all infection markers, the best-fit model included the spatial random effect. Tables 3 and \n405 4 show the odds ratios (ORs) and 95% CrI from the Bayesian non-spatial and spatial models \n406 for Ag and Wb123, Bm14 and Bm33 Abs, respectively. \n407\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n21\n408 3.1 Multivariate non-spatial and spatial models for Ag\n409 The DICs of the models of Ag with and without accounting for spatial correlation were 830.3 \n410 and 1122.3, respectively. In the spatial model, females had a 29.6% (95% CrI: 16.0–41.1%) \n411 lower risk of being Ag-positive than males. There was a 1.4% (95% CrI: 0.02–2.7%) \n412 decrease in the odds of Ag positivity for every year of age. Tree coverage was also positively \n413 associated with Ag-positivity, with an estimated increase of 0.3% (95% CrI: 0.06–0.6%) in \n414 the odds of Ag-positivity for each 1% increase in the extent of tree coverage in the 20 m \n415 buffers.\n416 After accounting for the effect of the statistically significant variables, the variance of the \n417 spatially structured random effect was 1.64 (0.55 to 4.92). The values of the decay parameter \n418 for spatial correlation (ϕ), was 86.03. This means that, after accounting for the effect of \n419 covariates, the radius of the clusters was approximately 3.9 km. (ϕ is measured in decimal \n420 degrees, therefore, the cluster size is calculated dividing 3 by ϕ; at the equator, one decimal \n421 degree is approximately 111 km).\n422\n423 3.2 Multivariate geostatistical model for Wb123, Bm14 and Bm33 Abs\n424 The DICs of the models of positivity for Wb123 Ab with and without accounting for spatial \n425 correlation were 2707.9 and 2861.1, respectively. In the spatial model, females had a 52.6% \n426 (95% CrI: 41.4–61.4%) lower risk of Wb123 Ab positivity than males. There was also an \n427 estimated increase of 2.4% (95% CrI: 1.9%–2.9%) in Wb123 Ab-positivity for every year of \n428 age (Table 4). Also, there was an increase in prevalence of being positive for Wb123 Ab of \n429 145.2% (95% CrI: 67.1–268.4%) and 46.9% (95% CrI: 16.3–83.3%) for tuna cannery \n430 workers and those who work in other locations (excluding outdoors and tuna cannery), \n431 respectively, compared to indoor workers. Additionally, there was a significant increase of \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n22\n432 0.3% (95% CrI: 0.02–0.6%) in the prevalence of Wb123 Ab-positivity for each 1% increase \n433 in the coverage of trees in the 20 m buffers.\n434\n435 The spatial Bm14 Ab model had a DIC of 1896.6, while the model without the spatial \n436 component had a DIC of 1898.6. In the spatial model, there was a decrease in the prevalence \n437 of Bm14 Ab of 53.8% (95% CrI: 40.3–64.5%) for females compared to males. Age was also \n438 as significant covariate with an increase in the prevalence of Bm14 Ab of 2.8% (95% CrI: \n439 2.2–3.5%) per every year of age. The prevalence of positive for Bm14 Ab was higher for \n440 those who worked outdoor and tuna cannery locations compare to those working indoors. The \n441 increase in the prevalence was 221.2% (95% CrI: 55.2–558.6%) and 96.3% (95% CrI: 24.3–\n442 225.4%), respectively. Tree coverage had a significant positive association with positivity for \n443 Bm14 Ab, with an estimated increase of 0.7% (95% CrI: 0.4–0.9%) in Bm14 positivity for \n444 each 1% increase in tree coverage in the 20 m buffer area. Population density had a \n445 significant negative association with Bm14 Ab prevalence, with a decrease of 12.0% (95% \n446 CrI: 2.7–21.0%) for every person/m 2.\n447\n448 The spatial model for Bm33 Ab also had a lower DIC, 3492.4, compared with the nonspatial \n449 model, 3505.6. Similar to all the other infection markers, the decrease in prevalence of Bm33 \n450 Ab was 28% (95% CrI: 14.3–39.1%) in females compared to males, and the increase per \n451 every year of age was 2.4% (95% CrI: 1.9–2.9%). Also, workers in outdoor and tuna cannery \n452 locations had an increase of 119.1% (95% CrI: 1.3–406.8%) and 74.8% (95% CrI: 18.1–\n453 165.5%) compared to workers in indoor areas. The prevalence of Bm33 positivity was found \n454 to increase by 0.3% (95% CrI: 0.07–0.6%) with a 1% increase in the extent of tree coverage \n455 in the 20 m buffers. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n23\n456\n457 In the model of Wb123 Ab the variance of the spatially structured random effect was 1.1 (0.5 \n458 to 2.1) and in the models of Bm14 and Bm33 these parameters were 0.002 (0.001 to 0.004) \n459 and 0.9 (0.5 to 1.4), respectively, meaning that the residual spatial variation was higher for \n460 the model of Wb123 Ab. The value of the decay parameter for spatial correlation (ϕ) was \n461 83.4 for Wb123 Ab, 73.7 for Bm14 Ab, and 78.3 for Bm33 Ab. These estimates indicate that \n462 after accounting for the effect of covariates, the radii of the clusters were approximately 3.9, \n463 4.5 and 4.2 km, respectively. \n464\n465\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n24\nTable 3 Odds ratios (ORs) and 95% CrI from the Bayesian non-spatial models for Ag and Wb123, Bm14 and Bm33 antibodies in a \ncommunity survey in American Samoa in 2016\nModel Participants\nN (%)\nAntigen Wb123 antibody Bm14 antibody Bm33 antibody \nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nTotal samples 2671 2671 2671 2671 2671\nTotal positives, N (%) - 135 (5.05) 684 (25.60) 350 (13.10) 1219 (45.63)\nGender\n  Male\n  Female\n1209 (45.26)\n1462 (54.74)\nRef\n0.58\n(0.49 to 0.68)\nRef\n0.49\n(0.41 to 0.59)\nRef\n0.46\n(0.36 to 0.59)\nRef\n0.72\n(0.61 to 0.85)\nAge (per year) - 0.97\n(0.96 to 0.98)\n1.02\n(1.02 to 1.03)\n1.03\n(1.02 to 1.04)\n1.02\n(1.02 to 1.03)\nWork location\n  Indoor 727 (27.22) Ref Ref Ref Ref\n  Outdoor 40 (1.50) 1.01\n(0.83 to 1.22)\n2.78\n(1.41 to 5.47)\n3.24\n(1.56 to 6.68)\n2.87\n(1.39 to 6.10)\n  Tuna cannery 131 (4.90) 0.99\n(0.82 to 0.19)\n2.21\n(1.46 to 3.32)\n1.94\n(1.18 to 3.15)\n1.69\n(1.14 to 2.51)\n  Others 1773 (66.40) 0.51\n(0.44 to 0.60)\n1.50\n(1.18 to 1.91)\n1.27\n(0.94 to 1.73)\n1.10\n(0.89 to 1.35)\nPopulation density \n(people/m2)\n- 0.89\n(0.81 to 0.98)\n0.95\n(0.88 to 1.02)\n0.88\n(0.79 to 0.97)\n0.95\n(0.89 to 1.02)\nElevation (m) - 1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\nDistance to streams (m) - 1.00\n(1.00 to 1.00)\n0.99\n(0.99 to 0.99)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\nRainfall in the wettest month \n- December (mm)\n- 1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\nLand Cover\n  Cropland (%) - 1.00 1.04 0.96 1.03\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n25\n(0.94 to 1.05) (0.99 to 1.11) (0.82 to 1.05) (0.98 to 1.11)\n  Trees (%) - 1.01\n(1.00 to 1.01)\n1.01\n(1.00 to 1.01)\n1.01\n(1.00 to 1.01)\n1.01\n(1.00 to 1.01)\n  Built/Urban (%) - 0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\nDIC 1122.32 2861.15 1898.63 3505.57\nFootnotes: ORs, Odds ratios; 95% CrI, 95% credible interval; DIC, deviance information criterion. Statistically significant ORs are highlighted in blue.\nTable 4 Odd ratios (ORs) and 95% CrI from the Bayesian geostatistical models for Ag and Wb123, Bm14 and Bm33 antibodies in the \ncommunity survey in American Samoa in 2016\nModel Participants\nN (%)\nAntigen positive Wb123 antibody \npositive\nBm14 antibody \npositive\nBm33 antibody \npositive\nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nORs, posterior mean\n(95% CrI)\nTotal samples 2671 2671 2671 2671 2671\nTotal positives, N (%) - 135 (5.05) 684 (25.60) 350 (13.10) 1219 (45.63)\nGender\n  Male\n  Female\n1209 (45.26)\n1462 (54.74)\nRef\n0.70\n(0.59 to 0.84)\nRef\n0.47\n(0.39 to 0.59)\nRef\n0.46\n(0.35 to 0.60)\nRef\n0.72\n(0.61 to 0.86)\nAge (per year) - 1.01\n(1.00 to 1.03)\n1.02\n(1.02 to 1.03)\n1.03\n(1.02 to 1.04)\n1.02\n(1.02 to 1.03)\nWork location\n  Indoor 727 (27.22) Ref Ref Ref Ref\n  Outdoor 40 (1.50) 1.01\n(0.83 to 1.23)\n1.53\n(0.70 to 3.18)\n3.21\n(1.55 to 6.59)\n2.19\n(1.01 to 5.07)\n  Tuna cannery 131 (4.90) 1.05\n(0.87 to 1.27)\n2.45\n(1.67 to 3.68)\n1.96\n(1.24 to 3.25)\n1.75\n(1.18 to 2.66)\n  Others 1773 (66.40) 0.86 1.47 1.28 1.08\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n26\n(0.70 to 1.05) (1.16 to 1.83) (0.96 to 1.69) (0.89 to 1.32)\nPopulation density (people/m2) - 0.96\n(0.85 to 1.08)\n0.99\n(0.90 to 1.08)\n0.88\n(0.79 to 0.97)\n0.95\n(0.88 to 1.03)\nElevation (m) - 1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\nDistance to streams (m) - 1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\nRainfall in the wettest month - \nDecember (mm)\n- 1.00\n(0.99 to 1.01)\n1.00\n(0.99 to 1.01)\n1.00\n(0.99 to 1.00)\n1.00\n(0.99 to 1.00)\nLand Cover\n  Cropland (%) - 0.99\n(0.93 to 1.04)\n1.04\n(0.97 to 1.10)\n0.96\n(0.80 to 1.05)\n1.04\n(0.98 to 1.11)\n  Trees (%) - 1.00\n(0.99 to 1.00)\n1.00\n(1.00 to 1.01)\n1.01\n(1.00 to 1.01)\n1.01\n(1.00 to 1.01)\n  Built/Urban (%) - 1.00\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\n0.99\n(0.99 to 1.00)\nHeterogeneity structured 1.64\n(0.55 to 4.92)\n1.12\n(0.54 to 2.12)\n0.002\n(0.001 to 0.005)\n0.87\n(0.54 to 1.44)\nϕ (Decay of spatial correlation) 86.03\n(54.95 to 97.37)\n83.4\n(51.32 to 99.36)\n73.74\n(42,51 to 98.37)\n78.30\n(47.77 to 98.59)\nDeviance Information Criterion 830.28 2707.95 1896.61 3492.37\nFootnotes: ORs, Odds ratios; 95% CrI, 95% credible interval; DIC, deviance information criterion. Statistically significant ORs are highlighted in blue.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n27\n3.3 Spatial predictions\nMaps of the mean and standard deviation (SD) of the posterior distributions of predicted \nprevalence of each of the LF infection markers are shown in Fig 4. The highest predicted \nprevalence of all infection makers (≥0.61%) was mainly confined in the north-west part, an \narea that corresponds largely to the coastal villages of Fagali'i and Fagamalo. There were also \npredicted residual foci of high prevalence of all Abs (higher for Bm33 and Wb123 Abs than \nBm14) in the southwest part of Tutuila, in areas that belong to Vaitogi and Futiga villages \n(0.21% and 0.49%), and high predicted prevalence estimates of Wb123 and Bm33 Ab in the \nwestern part of Tafuna village. High prevalence of Bm33 Ab covered larger areas compared \nto the other infection markers (≥0.21%), with higher prevalence estimates in confined areas \nin the north-west (≥ 0.61%), south-west (≥0.51%), the north-east (≥0.51%) and the central \npart around the Pago Pago area (≥0.41%). The maps of the posterior SDs demonstrate that the \nlevel of uncertainty was higher in inhabited areas in the north that were predominantly \ncovered by trees (Figs 2 and 4).\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n28\nFig 4. Spatial distribution of predicted prevalence and standard deviations of Ag (a and \nb), Wb123 Ab (c and d), Bm14 Ab (e and f), and Bm33 Ab (g and h) in American Samoa \n2016.  Note that the scale for SD for Bm14 Ab is different to the others maps.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n29\nDiscussion\nIn this study, we conducted a Bayesian geostatistical analysis of LF infection markers at the \nhousehold level and produced predictive prevalence maps for American Samoa in 2016. In \naddition, this study examined potential sociodemographic and environmental factors that may \ninfluence the geographical distribution of LF in the territory. To our knowledge, this is the \nfirst time that the distributions of LF infection markers have been examined at such high \nspatial resolution to predict prevalence. Our results suggest that there are still areas with high \nprevalence of LF infection markers (including Ag) in American Samoa, particularly in the \nnorth-west of the main island of Tutuila. Also, we found that there are sociodemographic and \nenvironmental factors that may underly the geographical distribution of LF and potentially \ncontribute to persistent transmission.  These predicted prevalence estimates of LF infection \nmarkers may help maximise the effectiveness of post-intervention surveillance by \ncontributing to the identification of areas with highest probability of residual transmission \n(7).\nThe results showed that the predicted prevalence of Ag, Wb123, Bm14 and Bm33 Abs \ndiffered geographically across the territory. Areas around Fagali'i and Fagamalo villages in \nthe north-west had the highest predicted prevalence of all infection markers. Also, in the \nsouth, high predicted prevalence, particularly for Bm33 Ab, were observed in localised areas \nin Vaitogi, Futiga and Tafuna villages. These findings concurred with the results of a \nprevious research conducted in the territory that found significant spatial dependency for all \ninfection markers, and confirmed the presence of LF clusters and hotspots in the north-west, \nsouth and central part of Tutuila (18). The high-risk area in the far north-west was also \npreviously identified as potential hotspot of residual infection in cross-sectional surveys \nconducted in American Samoa in 2010, 2014 and 2016 (10, 11, 13). In this study, the cluster \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n30\nsizes for all infection markers were larger compared to the previous findings (11, 18). This \ndiscrepancy in cluster size may be explained by the implementation of different spatial \nmethods, and also by the incorporation of sociodemographic and environmental covariates \ninto the geostatistical models (noting that the cluster size is in the residual component). These \ncovariates may be associated with heterogeneous exposure to mosquito bites. In areas where \nthe parasite is transmitted predominantly by night-biting mosquitos, clustering of infection \naround household locations can be expected and has been demonstrated  (29, 32, 44). The \nresults of this study support recent evidence that the home environment may be also an \nimportant area for exposure in LF-endemic regions where W. Bancrofti is transmitted by the \nday-biting mosquito, Ae. Polynesiensis. (45, 46). The cluster size also suggests that \ntransmission may be occurring not only around households, but also in surrounding areas \nwhere the household members are likely to frequent (such as bus stops, schools and \nworkplaces). In Samoa, a multilevel hierarchical modelling found that the intraclass \ncorrelation coefficients for Ag-positive individuals was higher at households (0.46) compared \nto primary sampling units (0.18) and regions (0.01) (46). The timely identification of these \nsmall pockets of residual infection can be used to prioritise further interventions to reduce the \nrisk of LF recrudescence or resurgence in the territory.\nThe predictive models developed for the different LF infection markers can help characterise \nthe spatial patterns of serological responses to LF in American Samoa. In W. bancrofti \nendemic areas, WHO recommends the use of Ag testing to assess the impact of the MDA and \ndetermine when the elimination targets have been reached (47). However, there is increasing \nevidence that suggests that the use of Ag alone for post-MDA surveillance may not be \nsufficiently sensitive to detect residual infection (7, 48, 49). Therefore, antifilarial Ab testing \nis currently been examined as an alternative or complementary method of diagnosing LF in \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n31\npost-MDA surveillance surveys (48-51). However, the dynamics of the Ab responses post-\ninfection and post-treatment are still not well understood (52). In this study, the geographical \ndistribution of the predicted prevalence of Wb123 and Bm14 Abs were more clustered \ncompared to the widespread distribution of positive Bm33 Ab responses. This finding \nsuggests that Bm33 Ab may not be the best indicator to identify areas of ongoing W. \nbancrofti transmission but may be used to provide information about levels of historical \nexposure and infection. Studies that have monitored the development of antifilarial immunity \nin LF endemic areas have shown that Bm33 Ab can be detected more than one year before \nthe other Ab responses, and can decrease after MDA (50). Additional longitudinal studies are \nrequired to help monitor how the stage of the infection and magnitude of the immunological \nresponses determine the spatial patterns of antifilarial Abs. Such information will have \nimplications for the selection of the most suited LF diagnostic tools in low prevalence and \npost-MDA settings.\nThere were consistent associations between the infection markers and the sociodemographic \nvariables included in the models. The observed differences among females and males and the \npositive association with age is most likely to be exposure-related. These findings support \nwhat has been observed previously in the territory and in most LF-endemic areas (11, 45, 53). \nMales spend more time working outdoors compared to females. However, it has also been \nsuggested that immunological and hormonal gender differences may account for the lower \ninfection rates in females (54). There was a consistent positive association between all Abs \nand tuna cannery workers. Also, a positive association between Bm14 and Bm33 Abs and \nindividuals working in outdoor locations. In 2013, An average of 17.6% of the total \nemployed population in the territory worked in tuna cannery which is the largest non-\ngovernment employer in American Samoa (55). The nature and time of day when this work \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n32\ntakes place may increase the risk of the exposure to mosquitoes. Higher prevalence of Wb123 \nAb was also previously observed in tuna cannery workers in the territory but no associations \nbetween Ab responses has been identified with other occupational groups (11). \nThe spatial models for all infection makers indicated that there was a positive association \nbetween the prevalence of LF and the extent of tree coverage in the 20 m buffers. This \nfinding supports the hypothesis that the tree coverage may impact mosquito population \ndynamics and behaviours (56). Tree canopy may sustain W. Bancrofti life cycle in high \ntemperature areas by facilitating the survival of mosquitos that move in response to food \nsupply (14). Most of American Samoa is steep, with approximately half of the area covered \nby rainforest (14, 57). Trees primarily cover most areas in the northern part of the territory \n(Fig 2) where the highest prevalence of LF was observed. Rainfall has been shown to be \nassociated with high prevalence of LF in several endemic countries where the infection is \ntransmitted by different vectors (20, 23, 31, 58, 59). No associations between the prevalence \nof infection markers and rainfall were found in this study. This finding was unexpected and \ndeserves further investigation. These findings raise the need for high-quality spatial \nenvironmental datasets that can be used in further studies to determine the association of LF \nand other potential environmental drivers. \nThe strengths of this study include the availability of data at the household level that allowed \nus to assess the geographical distribution of LF in American Samoa at a small spatial scale. In \nthis way, it was possible to explore the home environment as an exposure area of importance. \nThe study also developed geostatistical models for different infection markers that may be used \nas baseline information to characterise the spatial patterns of the antifilarial Ab responses in \nthe long-term. Besides the predicted prevalence maps, the spatial models developed here also \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n33\nprovided outputs to determine the associated uncertainty of the prevalence estimates (60). The \nmaps of the SD (uncertainty) highlight the areas where predictions were imprecise and that \nneed to be explored in future studies. \nThe limitations of the study include the lack of high-quality spatial environmental datasets for \nthe territory. As a result, it was not possible to include covariates such as temperature, that has \nconsistently been associated positively with LF (20, 30, 31). Also, the rainfall data used in the \nstudy was only available in a spatial format for the year 2016. Based on the assessment \npresented in the supplementary files, data were found to be representative of the average \nrainfall estimates for the ten-year period prior the survey (most likely time period of potential \nexposure). Despite this limitation, we believe that our results provide valuable information \nabout the potential sociodemographic and environmental factors that may be influencing the \ndistribution of the infection in American Samoa.\nIn this study, the Bayesian geostatistical models incorporating sociodemographic and \nenvironmental covariates showed that the predicted prevalence of LF was not homogeneous \nin American Samoa. Small-scale spatial variation in LF prevalence was observed which \nindicates that there is scope for further spatial analyses to help inform spatially-targeted \ninterventions in American Samoa. Areas of priority for further study include the north and \nsouth-western part of the territory. Also, longitudinal monitoring of the prevalence of Ag, \nWb123, Bm14 and Bm33 Abs would be useful to better understand the dynamics and \npotential use of different LF infection markers to inform and support the ongoing post-MDA \nsurveillance efforts. \nFunding\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n34\nThis study was supported by the Coalition for Operational Research on Neglected Tropical \nDiseases (COR-NTD), which is funded at The Task Force for Global Health primarily by the \nBill & Melinda Gates Foundation [OPP1053230], the United Kingdom Department for \nInternational Development, and by the United States Agency for International Development \nthrough its Neglected Tropical Diseases Program. CLL was supported by Australian National \nHealth and Medical Research Council Fellowships (APP1193826).\nAuthor’s contributions\nAMC and CLL developed the study conception and design. Analyses were performed by \nAMCR and CLL. AMCR and CLL drafted the manuscript. All authors helped in the \ninterpretation of results and critically reviewed the manuscript.\nDeclaration of interests\nWe declare no competing interests.\nData sharing\nThe data used in the present study are available from the corresponding author on\nreasonable request.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n35\nReferences\n1. World Health Organization. Lymphatic Filariasis Fact Sheet [Online] 2020 [Available \nfrom: https://www.who.int/news-room/fact-sheets/detail/lymphatic-filariasis \n2. Murray CJ, Vos T, Lozano R, Naghavi M, Flaxman AD, Michaud C, et al. Disability-\nadjusted life years (DALYs) for 291 diseases and injuries in 21 regions, 1990-2010: a \nsystematic analysis for the Global Burden of Disease Study 2010. Lancet. \n2012;380(9859):2197-223.\n3. Ton TGN, Mackenzie C, Molyneux DH. The burden of mental health in lymphatic \nfilariasis. Infectious Diseases of Poverty. 2015;4(1).\n4. Ottesen EA, Hooper PJ, Bradley M, Biswas G. The Global Programme to Eliminate \nLymphatic Filariasis: Health Impact after 8 Years. PLoS Neglected Tropical Diseases. \n2008;2(10):e317.\n5. Ramaiah KD, Ottesen EA. Progress and Impact of 13 Years of the Global Programme \nto Eliminate Lymphatic Filariasis on Reducing the Burden of Filarial Disease. PLoS Neglected \nTropical Diseases. 2014;8(11):e3319.\n6. World Health Organization. Global programme to eliminate lymphatic filariasis: \nprogress report, 2019. 2020.\n7. Lau CL, Sheel M, Gass K, Fuimaono S, David MC, Won KY, et al. Potential strategies \nfor strengthening surveillance of lymphatic filariasis in American Samoa after mass drug \nadministration: Reducing ‘number needed to test’by targeting older age groups, hotspots, and \nhousehold members of infected persons. PLOS Neglected Tropical Diseases. \n2020;14(12):e0008916.\n8. Lau CL, Won KY, Becker L, Magalhaes RJS, Fuimaono S, Melrose W, et al. \nSeroprevalence and Spatial Epidemiology of Lymphatic Filariasis in American Samoa after \nSuccessful Mass Drug Administration. PLoS Neglected Tropical Diseases. 2014;8(11).\n9. Lau CL, Won KY, Lammie PJ, Graves PM. Lymphatic Filariasis Elimination in \nAmerican Samoa: Evaluation of Molecular Xenomonitoring as a Surveillance Tool in the \nEndgame. PLoS Negl Trop Dis. 2016;10(11):e0005108.\n10. Sheel M, Sheridan S, Gass K, Won K, Fuimaono S, Kirk M, et al. Identifying residual \ntransmission of lymphatic filariasis after mass drug administration: Comparing school-based \nversus community-based surveillance-American Samoa, 2016. PLoS neglected tropical \ndiseases. 2018;12(7):e0006583.\n11. Lau CL, Won KY, Becker L, Magalhaes RJS, Fuimaono S, Melrose W, et al. \nSeroprevalence and spatial epidemiology of lymphatic filariasis in American Samoa after \nsuccessful mass drug administration. PLoS Negl Trop Dis. 2014;8(11):e3297.\n12. Lau CL, Won KY, Lammie PJ, Graves PM. Lymphatic filariasis elimination in \nAmerican Samoa: evaluation of molecular xenomonitoring as a surveillance tool in the \nendgame. PLoS neglected tropical diseases. 2016;10(11):e0005108.\n13. Lau CL, Sheridan S, Ryan S, Roineau M, Andreosso A, Fuimaono S, et al. Detecting \nand confirming residual hotspots of lymphatic filariasis transmission in American Samoa 8 \nyears after stopping mass drug administration. PLoS neglected tropical diseases. \n2017;11(9):e0005914.\n14. Schmaedick MA, Koppel AL, Pilotte N, Torres M, Williams SA, Dobson SL, et al. \nMolecular xenomonitoring using mosquitoes to map lymphatic filariasis after mass drug \nadministration in American Samoa. PLoS Negl Trop Dis. 2014;8(8):e3087.\n15. World Health O. Monitoring and epidemiological assessment of mass drug \nadministration in the global programme to eliminate lymphatic filariasis : a manual for national \nelimination programmes. Geneva: World Health Organization; 2011.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n36\n16. World Health Organization. Validation of elimination of lymphatic filariasis as a public \nhealth problem 2017 [Available from: \nhttps://www.who.int/publications/i/item/9789241511957.\n17. Irvine MA, Kazura JW, Hollingsworth TD, Reimer LJ. Understanding heterogeneities \nin mosquito-bite exposure and infection distributions for the elimination of lymphatic filariasis. \nProc Biol Sci. 2018;285(1871).\n18. Wangdi K, Sheel M, Fuimaono S, Graves PM, Lau CL. Lymphatic filariasis in 2016 in \nAmerican Samoa: Identifying clustering and hotspots using non-spatial and three spatial \nanalytical methods. PLOS Neglected Tropical Diseases. 2022;16(3):e0010262.\n19. Eneanya OA, Cano J, Dorigatti I, Anagbogu I, Okoronkwo C, Garske T, et al. \nEnvironmental suitability for lymphatic filariasis in Nigeria. Parasites &amp; Vectors. \n2018;11(1).\n20. Cano J, Rebollo MP, Golding N, Pullan RL, Crellen T, Soler A, et al. The global \ndistribution and transmission limits of lymphatic filariasis: past and present. Parasites & \nvectors. 2014;7(1):1-19.\n21. Cuenco KT, Ottesen EA, Williams SA, Nutman TB, Steel C. Heritable factors play a \nmajor role in determining host responses to Wuchereria bancrofti infection in an isolated South \nPacific island population. The Journal of infectious diseases. 2009;200(8):1271-8.\n22. Moraga P, Cano J, Baggaley RF, Gyapong JO, Njenga SM, Nikolay B, et al. Modelling \nthe distribution and transmission intensity of lymphatic filariasis in sub-Saharan Africa prior \nto scaling up interventions: integrated use of geostatistical and mathematical modelling. \nParasites & vectors. 2015;8(1):1-16.\n23. Stensgaard A-S, Vounatsou P, Onapa AW, Simonsen PE, Pedersen EM, Rahbek C, et \nal. Bayesian geostatistical modelling of malaria and lymphatic filariasis infections in Uganda: \npredictors of risk and geographical patterns of co-endemicity. Malaria journal. 2011;10(1):1-\n15.\n24. Fronterre C, Amoah B, Giorgi E, Stanton MC, Diggle PJ. Design and analysis of \nelimination surveys for neglected tropical diseases. The Journal of infectious diseases. \n2020;221(Supplement_5):S554-S60.\n25. National Oceanic and Atmospheric Administration. National Weather Service 2022 \n[Available from: https://www.weather.gov/wrh/Climate?wfo=ppg.\n26. American Samoa GIS User Group ASGDoC. Resource Center  [Available from: \nhttps://www.doc.as.gov/resource-center.\n27. Abbott. ALERE™ FILARIASIS TEST STRIP  [Available from: \nhttps://www.globalpointofcare.abbott/en/product-details/alere-filariasis-test-\nstrip.html#:~:text=Alere%E2%84%A2%20Filariasis%20Test%20Strip,filarial%20antigen%2\n0from%20Wuchereria%20bancrofti.\n28. Won K, Sambou S, Barry A, Robinson K, Jaye M, Sanneh B, et al. Use of antibody \ntools to provide serologic evidence of elimination of lymphatic filariasis in the gambia. The \nAmerican journal of tropical medicine and hygiene. 2018;98(1):15-20.\n29. Boyd HA, Waller LA, Flanders WD, Beach MJ, Sivilus JS, Lovince R, et al. \nCommunity-and individual-level determinants of Wuchereria bancrofti infection in Leogane \nCommune, Haiti. The American journal of tropical medicine and hygiene. 2004;70(3):266-72.\n30. Eneanya OA, Cano J, Dorigatti I, Anagbogu I, Okoronkwo C, Garske T, et al. \nEnvironmental suitability for lymphatic filariasis in Nigeria. Parasites & vectors. 2018;11(1):1-\n13.\n31. Mwase ET, Stensgaard A-S, Nsakashalo-Senkwe M, Mubila L, Mwansa J, Songolo P, \net al. Mapping the geographical distribution of lymphatic filariasis in Zambia. PLoS neglected \ntropical diseases. 2014;8(2):e2714.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n37\n32. Kwarteng EVS, Andam-Akorful SA, Kwarteng A, Asare D-CB, Quaye-Ballard JA, \nOsei FB, et al. Spatial variation in lymphatic filariasis risk factors of hotspot zones in Ghana. \nBMC public health. 2021;21(1):1-13.\n33. Fagatele Bay National Marine Sanctuary. GIS Data Archive 2022 [Available from: \nhttps://dusk.geo.orst.edu/djl/samoa/.\n34. Pacific Data Hub. American Samoa Populatin Grid 2020 [Available from: \nhttps://pacificdata.org/data/dataset/asm_population_grid_2020.\n35. U.S. Geological Survey (USGS). USGS 10-m Digital Elevation Model (DEM): \nAmerican Samoa: Tutuila. Distributed by the Pacific Islands Ocean Observing System \n(PacIOOS) 2015 [Available from: http://pacioos.org/metadata/usgs_dem_10m_tutuila.html.\n36. Pacific Environmental Data Portal. Rainfall data American Samoa 2017 [Available \nfrom: https://pacific-data.sprep.org/.\n37. The National Aeronautics and Space Administration (NASA). Land Surface \nTemperature and Emissivity (LST&E) (MOD11A2) 2016 [Available from: \nhttps://lpdaac.usgs.gov/products/mod11a2v006/.\n38. Karra K, Kontgis C, Statman-Weil Z, Mazzariello JC, Mathis M, Brumby SP, editors. \nGlobal land use/land cover with Sentinel 2 and deep learning. 2021 IEEE international \ngeoscience and remote sensing symposium IGARSS; 2021: IEEE.\n39. ESRI: Environmental Systems Research Institute. ArcGIS Software version 10.7.1. \nRedlands. California.2019.\n40. McLure A, Graves PM, Lau C, Shaw C, Glass K. Modelling lymphatic filariasis \nelimination in American Samoa: GEOFIL predicts need for new targets and six rounds of mass \ndrug administration. Epidemics. 2022:100591.\n41. RStudio Team. RStudio: Integrated Development for R. RStudio, PBC, Boston, MA \n2020.\n42. Group. MoOPM. OpenBUGS version 3.2.3 rev 1012. 2014.\n43. Best N, Richardson S, Thomson A. A comparison of Bayesian spatial models for \ndisease mapping. Statistical methods in medical research. 2005;14(1):35-59.\n44. Drexler N, Washington CH, Lovegrove M, Grady C, Milord MD, Streit T, et al. \nSecondary mapping of lymphatic filariasis in Haiti-definition of transmission foci in low-\nprevalence settings. 2012.\n45. Joseph H, Moloney J, Maiava F, McClintock S, Lammie P, Melrose W. First evidence \nof spatial clustering of lymphatic filariasis in an Aedes polynesiensis endemic area. Acta \ntropica. 2011;120:S39-S47.\n46. Lau CL, Meder K, Mayfield HJ, Kearns T, McPherson B, Naseri T, et al. Lymphatic \nfilariasis epidemiology in Samoa in 2018: Geographic clustering and higher antigen prevalence \nin older age groups. PLoS Neglected Tropical Diseases. 2020;14(12):e0008927.\n47. World Health Organization. Global programme to eliminate lymphatic filariasis: \nmonitoring and epidemiological assessment of mass drug administration. World Health \nOrganization, Geneva, Switzerland. 2011.\n48. Won KY, Robinson K, Hamlin KL, Tufa J, Seespesara M, Wiegand RE, et al. \nComparison of antigen and antibody responses in repeat lymphatic filariasis transmission \nassessment surveys in American Samoa. PLoS neglected tropical diseases. \n2018;12(3):e0006347-e.\n49. Cadavid Restrepo AM, Gass K, Won KY, Sheel M, Robinson K, Graves PM, et al. \nPotential use of antibodies to provide an earlier indication of lymphatic filariasis resurgence in \npost–mass drug ad ministration surveillance in American Samoa. International Journal of \nInfectious Diseases. 2022;117:378-86.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n38\n50. Hamlin KL, Moss DM, Priest JW, Roberts J, Kubofcik J, Gass K, et al. Longitudinal \nmonitoring of the development of antifilarial antibodies and acquisition of Wuchereria \nbancrofti in a highly endemic area of Haiti. PLoS Negl Trop Dis. 2012;6(12):e1941.\n51. Weil GJ, Curtis KC, Fischer PU, Won KY, Lammie PJ, Joseph H, et al. A multicenter \nevaluation of a new antibody test kit for lymphatic filariasis employing recombinant Brugia \nmalayi antigen Bm-14. Acta tropica. 2011;120:S19-S22.\n52. Steel C, Kubofcik J, Ottesen EA, Nutman TB. Antibody to the filarial antigen Wb123 \nreflects reduced transmission and decreased exposure in children born following single mass \ndrug administration (MDA). PLoS Negl Trop Dis. 2012;6(12):e1940.\n53. Ichimori K, Tupuimalagi-Toelupe P, Iosia VT, Graves PM. Wuchereria bancrofti \nfilariasis control in Samoa before PacELF (Pacific Programme to Eliminate Lymphatic \nFilariasis). Tropical Medicine and Health. 2007;35(3):261-9.\n54. Witt C, Ottesen EA. Lymphatic filariasis: an infection of childhood. Tropical Medicine \n& International Health. 2001;6(8):582-606.\n55. Commerce ASGDo. American Samoa Statistical Yearbook 2013. American Samoa \nGovernment Department of Commerce Statistics Department Pago …; 2015.\n56. Kelly-Hope L, Paulo R, Thomas B, Brito M, Unnasch TR, Molyneux D. Loa loa vectors \nChrysops spp.: perspectives on research, distribution, bionomics, and implications for \nelimination of lymphatic filariasis and onchocerciasis. Parasites & vectors. 2017;10(1):1-15.\n57. Nakamura S. Soil survey of American Samoa. Department of Agriculture, Soil \nConservation Service in cooperation with the Government of American Samoa. 1984.\n58. Grziwotz F, Strauß JF, Hsieh C-h, Telschow A. Empirical dynamic modelling identifies \ndifferent responses of Aedes Polynesiensis subpopulations to natural environmental variables. \nScientific reports. 2018;8(1):1-10.\n59. Palaniyandi M, Anand P, Maniyosai R. Spatial cognition: a geospatial analysis of vector \nborne disease transmission and the environment, using remote sensing and GIS. Information \nSystems. 2014;1:52.\n60. Diggle PJ, Tawn JA, Moyeed RA. Model ‐based geostatistics. Journal of the Royal \nStatistical Society: Series C (Applied Statistics). 1998;47(3):299-350.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint \n\n39\nSupporting information captions\nS1 table. Average monthly rainfall (mm) in the Pago Pago area, in American Samoa from \n2000-2020 (Data extracted from the National Weather\nS1 Figure. Average monthly rainfall (mm) for the period 2000-2020 and average montly \nrainfall (mm) in 2016 in American Samoa. The driest (August) and wettest (December) months \nin 2016 were representative of the average rainfall in the respective months in previous 20 \nyears\nS2 Figure. Average annual rainfall (mm) for the period 2000-2020. The horizontal line \nindicates the average in the previous 10 years. Total rainfall in 2016 was representative of \naverage rainfall in the previous 20 years.\n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted September 26, 2022. ; https://doi.org/10.1101/2022.09.26.22280353doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}