Environmental change and human encroachment into wildlife habitats are key drivers in the emergence and transmission of zoonotic diseases (Lambin et al., 2010; Patz et al., 2004). Individual movements into different habitats influence exposure to disease vectors and animal reservoirs, determining risk and propagation of vector-borne diseases (Stoddard et al., 2013; Stoddard et al., 2009; Pindolia et al., 2012). Increased contact between these populations is theorised to drive increases of the zoonotic malaria Plasmodium knowlesi in Malaysian Borneo, now the main cause of human malaria within this region. P. knowlesi is carried by long- and pig-tailed macaques (Macaca fascicularis and M. nemestrina) and transmitted by the Anopheles leucospryphus mosquito group, both populations highly sensitive to land cover and land use change (Moyes et al., 2016). Although higher spatial overlap between people, macaques and mosquito vectors likely drives transmission, the impact of human movement and land use in determining individual infection risks is poorly understood (Imai et al., 2014).

The emergence of the zoonotic malaria Plasmodium knowlesi has been positively associated with both forest cover and historical deforestation (Fornace et al., 2016b; Shearer et al., 2016). However, out of necessity, statistical approaches to assess environmental risk factors for P. knowlesi and other infectious diseases typically evaluate relationships between disease metrics and local land cover surrounding houses or villages. While an individual may spend most of their time within the vicinity of their residence, this area does not necessarily represent where they are most likely to be exposed to a disease. This is supported by varying associations between P. knowlesi occurrence and landscape variables at different distances from households, ranging from 100 m to 5 km, likely partially due to human movement into different surrounding habitats (Fornace et al., 2016b; Brock et al., 2019). Although land cover variables describing physical terrestrial surfaces are frequently incorporated into disease models, land use is rarely quantified. Land use is commonly defined as ‘the arrangements, activities, and inputs that people undertake in certain land cover types’ (IPCC, 2000). Places with similar types of land cover may be used very differently, with the activities and frequencies with which people visit these places determining the spatial distribution of disease (Lambin et al., 2010).

Mathematical modelling studies have revealed the importance of spatial variation in contact rates due to the movement of individuals through heterogeneous environments with varying transmission intensity (Acevedo et al., 2015). A multi-species transmission model of P. knowlesi highlighted the role of mixing patterns between populations in different ecological settings in determining the basic reproductive rate and subsequent modelling studies illustrate the sensitivity of this disease system to population densities of both people and wildlife hosts (Imai et al., 2014; Yakob et al., 2018). However, although mechanistic models have been extended to explore the potential importance of these heterogeneities in disease dynamics, there are inherent constraints on model complexity and most models make simplistic assumptions about the habitat uses of different populations.

Empirical data on human population movement is increasingly available, allowing assessment of the impact of mobility on infectious disease dispersion and risks (Pindolia et al., 2012). On larger spatial scales, mobile phone data have revealed the role of human migration in the transmission of infectious diseases such as malaria, dengue and rubella (Chang et al., 2019; Wesolowski et al., 2015a; Wesolowski et al., 2015b).Although this data can provide insights into long range movements, spatial resolution of this data is limited, particularly in areas with poor or no mobile coverage, such as forested areas (Wesolowski et al., 2016). Alternatively, the advent of low-cost GPS tracking devices allows quantification of fine-scale movements, demonstrating marked heterogeneity in individual movement and risk behaviours (Stoddard et al., 2013; Vazquez-Prokopec et al., 2013). Combining these data with detailed data on land cover and vector dynamics can provide new insights into how landscapes affect P. knowlesi transmission.

Previous studies of P. knowlesi have relied on questionnaire surveys, identifying self-reported travel to nearby plantations and forest areas as a risk factor for P. knowlesi and other malaria infections (e.g. Grigg et al., 2017; Singh et al., 2004; Yasuoka and Levins, 2007). However, the resultant spatial range and frequency of these movements remain unknown and the definition of different habitat types is entirely subjective. Further, little is known about differences in local movement patterns in different demographic groups. While infections in male adults have been linked to forest and plantation work, it is unknown whether infections reported in women and young children are likely to arise from exposure to similar environments (Barber et al., 2012). The main mosquito vector in this area, An. balabacensis, is primarily exophagic and has been identified in farm, forest and village areas near houses (Wong et al., 2015; Manin et al., 2016). Macaque populations are reported in close proximity to human settlements and molecular and modelling studies suggest transmission remains primarily zoonotic in this area (Imai et al., 2014; Lee et al., 2011; Chua et al., 2017). A case control study detected higher abundances of An. balabacensis near P. knowlesi case housesholds, suggesting the possibility of peri-domestic transmission (Manin et al., 2016). Understanding the importance of these habitats is essential to effectively target intervention strategies and predict impacts of future environmental changes.

Key questions remain about where individuals are likely to be exposed to P. knowlesi and how landscape determines risk. Functional ecology approaches allow the distribution of different populations to be modelled based on biological resources and relate transmission to landscape and environmental factors (Hartemink et al., 2015). Within wildlife ecology, numerous methods have been developed to estimate utilisation distributions (UDs), the probability of an individual or species being within a specific location during the sampling period (Papworth et al., 2012). Although these methods traditionally rely on kernel density smoothing, kernel density estimates may not actually reflect time individuals spend in a specific location if there is substantial missing data or irregular time intervals. Alternatively, biased random bridges (BRBs) improve on these methods by estimating the utilisation distribution as a time-ordered series of points, taking advantage of the autocorrelated nature of GPS tracks to bias movement predictions towards subsequent locations in a time series (Benhamou, 2011). This allows for interpolation of missing values and adjustment for spatial error to estimate utilisation distributions representing both the intensity (mean residence time per visit) and frequency of individual visits to specific locations. By integrating these estimates of individual space use with detailed spatial and environmental data in a Bayesian framework, fine-scale patterns of human land use can be predicted and overlaid with spatiotemporal models of mosquito distribution. This allows exploration of how landscape composition, as well as configuration and connectivity between habitats, impacts human exposure to P. knowlesi and other vector-borne and zoonotic diseases.

Focusing on one aspect of land use, human movement and time spent within different land cover types, we explored the role of heterogeneity in local space use on disease exposure. Rolling cross-sectional GPS tracking surveys were conducted in two study areas with on-going P. knowlesi transmission in Northern Sabah, Malaysia (Matunggong and Limbuak; Fornace et al., 2018). We aimed to characterise local movement patterns and identify individuals and locations associated with increased P. knowlesi exposure risks by: 1. analysing individual movement patterns and developing predictive maps of human space use relative to spatial and environmental factors, 2. modelling biting rates of the main vector An. balabacensis, and 3. assessing exposure risks for P. knowlesi based on predicted mosquito and human densities (Figure 1) Integrating these three approaches allowed a uniquely spatially explicit examination of disease risk.

Figure 1

Download asset Open asset

Between February 2014 and May 2016, 285 consenting people participated in the GPS tracking study with 243 included in the final analysis including 109 in Limbuak and 134 in Matunggong (Table 1). The most commonly reported occupation was farm or plantation work (n = 73), primarily conducted within the immediate vicinity of the house. A total of 3,424,913 GPS points were collected, representing 6,319,885 person-minutes of sampling time. Median sampling duration was 16.27 days (IQR 13.72–19.97), with points recorded for a median of 59.1% (IQR: 46.9–71.1%) of the sampling duration. Maximum distances travelled ranged from no travel outside the house to 116 km, with a median distance travelled of 1.8 km. Utilisation distributions (UDs), the probability of an individual being in a location in space within a given time (Figure 3), varied by gender and occupation. Individuals at the more rural Limbuak site covered larger distances (Table 2), with the largest distances covered by individuals reporting primary occupations of fishing (n = 5) and office work (n = 9). Although substantial differences were reported in all movements (24 hr sampling) between seasons, no seasonal differences were observed in human movements during peak Anopheles biting times (6pm-6am).

Figure 3

Download asset Open asset

For both study areas, we developed models of community space use during peak mosquito biting hours (6pm – 6am), in the form of resource utilisation functions, predictions of time- and space-specific UDs on the basis of spatial and environmental variables (Papworth et al., 2012). Between 6pm – 6am, human space use (UDs) was mostly predictable and negatively correlated with distance from the individual’s house, other houses, roads and slope. The AUC for presence/absence models was 0.936 for Matunggong and 0.938 for Limbuak and RMSE for the overall model was 0.0073 and 0.0043 for Matunggong and Limbuak, respectively. While individuals were more likely to use areas further away from forests in the Matunggong site, human space use was positively correlated with proximity to forests in the Limbuak site (Table 3). Despite marked differences between different demographic groups and seasons observed during 24 hr movements, these factors did not improve the predictive power of the model for movements between 6pm and 6am.

Between August 2013 and December 2015, 4814 An. balabacensis were caught from 328 sampling nights in 155 unique locations. The median biting rate was 2.1 bites per night per person, ranging from 0 to 28 bites per person per night (Figure 4). Despite monthly variation, including temporal autocorrelation did not improve model fit (Table 4). Although no associations were identified between land classification and vector density in this site, models identified positive relationships with enhanced vegetation indices (EVI) and negative associations with distance to forest and human population density (Table 5). Of 1524 mosquitoes tested for Plasmodium sporozoites, the median sporozoite rate was 0.24% (95% CI: 0.09–0.58%).

Figure 4

Download asset Open asset

For individuals included in the GPS tracking study in Matunggong, where both human movement and entomology data were available, we calculated exposure risks as a derived quantity from utilisation distributions and mosquito biting rate models. Exposure varied markedly between individuals, with an overall 150-fold difference in predicted mean probabilities of infected bites per night (range: 0.00005–0.0078) (Table 6). No clear differences were observed between genders, age groups or occupations of individuals sampled and there was no association between risk and distance travelled.

Using the resource utilisation function with demographic and spatial data for all individuals in Matunggong, we predicted community-wide space use and estimated exposure to infected mosquitoes (Figure 5). The predicted number of person nights per grid cell for the entire community ranged from 0 to 12.79 (median: 0.01, IQR: 0.0004–0.99), with the mean probability of a community member exposed to an infected bite per grid cell of 0.00082 (IQR: 0.00001, 0.00050). Although over 43% of the study site is forest and relatively high biting rates were predicted in forests during the study period (mean: 1.94, range: 0.04–12.59), this habitat was rarely used by people in the evenings, with less than 8% of predicted person-nights in forests. Models only based on mosquito biting rates and not including human space use predicted 42% of infectious bites occurred in forested areas and only 8.6% of bites occurring within 100 m of houses (Figure 5C). In contrast, when space use patterns are included, over 91% of predicted infected bites were predicted within 500 m of houses (Figure 5D). Highest exposure risks were consistently found near forest edges and in close proximity to households, despite spatial and temporal heterogeneity and model uncertainty (Figure 4).

Figure 5

Download asset Open asset

This study highlights the importance of human space use in different land cover types in determining exposure to zoonotic and vector-borne diseases such as P. knowlesi. Although P. knowlesi has previously been associated with forest exposure (e.g. Grigg et al., 2017) and higher biting rates have been reported in forest interiors (Wong et al., 2015), this novel approach incorporating both mosquito and human space use data provides a new perspective on peri-domestic transmission, with more than 90% of infectious bites predicted in areas surrounding households at forest edges. This study additionally demonstrates the utility of ecological methods to understand human movement and identify geographical areas associated with higher contact with disease vectors.

Within these communities, local movement patterns during peak vector times were largely predictable and could be explained by spatial and environmental factors. However, despite this finding, there was substantial variation in predicted exposure between individuals as a result of heterogeneity in habitats used. No significant differences in exposure were predicted between men and women, with individuals with high exposure risks identified across occupational and age groups. Although this finding differs from clinical reports, a comprehensive survey within this community identified equal proportions of men and women exposed to P. knowlesi as evidenced by specific antibody responses and data on asymptomatic infections suggests higher numbers of non-clinical infections in women (Fornace et al., 2018; Fornace et al., 2016a). While infrequent events or long-range movements (such as hunting trips) may contribute to these differences in clinical cases and may not have been captured within this two-week sampling period within the study site, this analysis highlights the importance of routine movements into local environments in shaping exposure risks.

This improved understanding of how local human land use is related to exposure risk has important implications for surveillance and control programmes. Malaria control programmes often rely on interventions within the house, such as insecticide treated bednets and indoor residual spraying; however, movements outside during peak biting times illustrate the importance of also targeting outdoor transmission. The identification of areas where exposure is likely to occur can further be used to refine interventions; for example, although insecticide treated hammocks have been proposed for deep forest environments, larval source management may be more appropriate to target environments in close proximity to houses. Although initial P. knowlesi cases were primarily identified in adult men living and working in forests (Singh et al., 2004), this study illustrates the potential importance of peri-domestic habitats in transmission and provides quantitative insight on mixing between people and infected mosquitoes in forest fringe areas. As Malaysia moves towards malaria elimination, surveillance systems are incorporating novel focal investigation methods, including monitoring changes in local land use and populations at risk (Bahagian Kawalan Penyakit, Kementerian Kesihatan Malaysia, 2016). In additional to routine vector surveillance, this study highlights the need to incorporate measures of human space when defining risk zones.

Even with the large and highly detailed movement dataset analysed, this study was limited by the availability of mosquito data; as human landing catch data were assembled from other studies, there was not uniform spatial and temporal coverage of the study site increasing uncertainty. The limited mosquito data availability precluded development of mosquito biting rate models for Limbuak and other outlying islands. An additional limitation to estimating mosquito biting rates was the difficulty obtaining spatially and temporally resolute remote sensing data for predictors due to high cloud cover (Weiss et al., 2015). As few positive mosquitoes were identified, uniform estimates of sporozoite rates based on available data were used across the Matunggong site; if further data were available, these models could be refined to incorporate estimates of human and macaque density, mosquito biting preferences in different habitats and infection levels in all hosts (Yakob et al., 2010). Additionally, as this study was designed to quantitatively estimate time spent in different landscapes, further studies could explore other aspects of land use, such as the purposes of travel, activities undertaken or practices used to modify or management land cover.

Despite these limitations, this is the first large-scale study to utilise GPS tracking data and ecological methods to create fine-scale maps of exposure risk. This study highlights the importance of incorporating heterogenous patterns of human space use into disease models, as the majority of human exposure may occur in areas with lower vector biting rates but greater probabilities of human use. Further, results quantitatively illustrate the importance of forest edges and local habitat in P. knowlesi transmission and can inform understanding of other zoonotic and vector-borne diseases.

Data on human subjects is not available due to ethical restrictions around sharing identifiable information. All other data is publicly available with relevant links or publications included. Code to reproduce this analysis is available on GitHub or as supplementary information.

1. 1. NASA LP DAAC

   (2015) NASA Landsat Science

   Landsat 8 Operational Land Imager.

   https://landsat.gsfc.nasa.gov/landsat-8/
2. 1. NASA LP DAAC

   (2015) ASTER

   Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model.

   https://doi.org/10.5067/ASTER/ASTGTM.003
3. 1. DAAC NL

   (2015) MODIS

   MODIS/ Terra Vegetation Indices 16-Day L3 Global 250m Grid SIN V006.

   https://doi.org/10.5067/MODIS/MOD13Q1.006
4. 1. NASA TRMM

   (2015) Tropical Rainfall Measurement Mission

   Daily TRMM and other satellites precipitation product (3B42 V6 derived).

   https://doi.org/10.5067/TRMM/TMPA/DAY/7

1. 1. Acevedo MA
   2. Prosper O
   3. Lopiano K
   4. Ruktanonchai N
   5. Caughlin TT
   6. Martcheva M
   7. Osenberg CW
   8. Smith DL

   (2015) Spatial heterogeneity, host movement and mosquito-borne disease transmission

   PLOS ONE 10:e0127552.

   https://doi.org/10.1371/journal.pone.0127552

   • PubMed
   • Google Scholar
2. Report

   1. Bahagian Kawalan Penyakit, Kementerian Kesihatan Malaysia

   (2016)

    Garis Panduan Pencegahan Malaria Re-Introduction Di Malaysia

   Putra Jaya: Kementerian Kesihatan Malaysia.

   • Google Scholar
3. 1. Barber BE
   2. William T
   3. Dhararaj P
   4. Anderios F
   5. Grigg MJ
   6. Yeo TW
   7. Anstey NM

   (2012) Epidemiology of Plasmodium knowlesi malaria in north-east sabah, Malaysia: family clusters and wide age distribution

   Malaria Journal 11:401.

   https://doi.org/10.1186/1475-2875-11-401

   • PubMed
   • Google Scholar
4. 1. Barbet-Massin M
   2. Jiguet F
   3. Albert CH
   4. Thuiller W

   (2012) Selecting pseudo-absences for species distribution models: how, where and how many?

   Methods in Ecology and Evolution 3:327–338.

   https://doi.org/10.1111/j.2041-210X.2011.00172.x

   • Google Scholar
5. 1. Benhamou S

   (2011) Dynamic approach to space and habitat use based on biased random bridges

   PLOS ONE 6:e14592.

   https://doi.org/10.1371/journal.pone.0014592

   • PubMed
   • Google Scholar
6. Book

   1. Blangiardo M
   2. Spatial CM

   (2015) Spatio-Temporal Bayesian Models with R-INLA

   Chichester, United Kingdom: John Wiley & Sons, Ltd.

   https://doi.org/10.1002/9781118950203

   • Google Scholar
7. 1. Brock PM
   2. Fornace KM
   3. Grigg MJ
   4. Anstey NM
   5. William T
   6. Cox J

   (2019) Predictive analysis across spatial scales links zoonotic malaria to deforestation

   Proceedings of the Royal Society B 2019:286.

   https://doi.org/10.1098/rspb.2018.2351

   • Google Scholar
8. 1. Chang HH
   2. Wesolowski A
   3. Sinha I
   4. Jacob CG
   5. Mahmud A
   6. Uddin D
   7. Zaman SI
   8. Hossain MA
   9. Faiz MA
   10. Ghose A
   11. Sayeed AA
   12. Rahman MR
   13. Islam A
   14. Karim MJ
   15. Rezwan MK
   16. Shamsuzzaman AKM
   17. Jhora ST
   18. Aktaruzzaman MM
   19. Drury E
   20. Gonçalves S
   21. Kekre M
   22. Dhorda M
   23. Vongpromek R
   24. Miotto O
   25. Engø-Monsen K
   26. Kwiatkowski D
   27. Maude RJ
   28. Buckee C

   (2019) Mapping imported malaria in Bangladesh using parasite genetic and human mobility data

   eLife 8:e43481.

   https://doi.org/10.7554/eLife.43481

   • PubMed
   • Google Scholar
9. 1. Chua TH
   2. Manin BO
   3. Daim S
   4. Vythilingam I
   5. Drakeley C

   (2017) Phylogenetic analysis of simian Plasmodium spp. infecting Anopheles balabacensis baisas in sabah, Malaysia

   PLOS Neglected Tropical Diseases 11:e0005991.

   https://doi.org/10.1371/journal.pntd.0005991

   • PubMed
   • Google Scholar
10. Data

    1. Didan K

    (authors) (2015) MOD13Q1 MODIS/Terra vegetation indices 16-Day L3 global 250m SIN grid V006

    NASA EOSDIS Land Processes DAAC.

    https://doi.org/10.5067/MODIS/MOD13Q1.006
11. 1. Fornace KM
    2. Nuin NA
    3. Betson M
    4. Grigg MJ
    5. William T
    6. Anstey NM
    7. Yeo TW
    8. Cox J
    9. Ying LT
    10. Drakeley CJ

    (2016a) Asymptomatic and submicroscopic carriage of Plasmodium knowlesi malaria in household and community members of clinical cases in sabah, Malaysia

    Journal of Infectious Diseases 213:784–787.

    https://doi.org/10.1093/infdis/jiv475

    • PubMed
    • Google Scholar
12. 1. Fornace KM
    2. Abidin TR
    3. Alexander N
    4. Brock P
    5. Grigg MJ
    6. Murphy A
    7. William T
    8. Menon J
    9. Drakeley CJ
    10. Cox J

    (2016b) Association between landscape factors and spatial patterns of Plasmodium knowlesi infections in sabah, Malaysia

    Emerging Infectious Diseases 22:201–209.

    https://doi.org/10.3201/eid2202.150656

    • PubMed
    • Google Scholar
13. 1. Fornace KM
    2. Herman LS
    3. Abidin TR
    4. Chua TH
    5. Daim S
    6. Lorenzo PJ
    7. Grignard L
    8. Nuin NA
    9. Ying LT
    10. Grigg MJ
    11. William T
    12. Espino F
    13. Cox J
    14. Tetteh KKA
    15. Drakeley CJ

    (2018) Exposure and infection to Plasmodium knowlesi in case study communities in northern sabah, Malaysia and Palawan, the philippines

    PLOS Neglected Tropical Diseases 12:e0006432.

    https://doi.org/10.1371/journal.pntd.0006432

    • PubMed
    • Google Scholar
14. 1. Grigg MJ
    2. Cox J
    3. William T
    4. Jelip J
    5. Fornace KM
    6. Brock PM
    7. von Seidlein L
    8. Barber BE
    9. Anstey NM
    10. Yeo TW
    11. Drakeley CJ

    (2017) Individual-level factors associated with the risk of acquiring human Plasmodium knowlesi malaria in Malaysia: a case-control study

    The Lancet Planetary Health 1:e97–e104.

    https://doi.org/10.1016/S2542-5196(17)30031-1

    • PubMed
    • Google Scholar
15. 1. Hartemink N
    2. Vanwambeke SO
    3. Purse BV
    4. Gilbert M
    5. Van Dyck H

    (2015) Towards a resource-based habitat approach for spatial modelling of vector-borne disease risks

    Biological Reviews 90:1151–1162.

    https://doi.org/10.1111/brv.12149

    • PubMed
    • Google Scholar
16. Book

    1. Held L
    2. Schrodle B
    3. Posterior RH

    (2010) Cross-validatory Predictive Checks: A Comparison of MCMC and INLA

    In: Kneib T, Tutz G, editors. Statistical Modelling and Regression Structures. Physica-Verlag HD. pp. 91–110.

    https://doi.org/10.1007/978-3-7908-2413-1_6

    • Google Scholar
17. 1. Hooten MB
    2. Hanks EM
    3. Johnson DS
    4. Alldredge MW

    (2013) Reconciling resource utilization and resource selection functions

    Journal of Animal Ecology 82:1146–1154.

    https://doi.org/10.1111/1365-2656.12080

    • PubMed
    • Google Scholar
18. 1. Imai N
    2. White MT
    3. Ghani AC
    4. Drakeley CJ

    (2014) Transmission and control of Plasmodium knowlesi: a mathematical modelling study

    PLOS Neglected Tropical Diseases 8:e2978.

    https://doi.org/10.1371/journal.pntd.0002978

    • PubMed
    • Google Scholar
19. Book

    1. IPCC

    (2000)

    Land Use, Land-Use Change, and Forestry

    Watson RT, Noble IR, Bolin B, Ravindranath NH, Verado DJ, Dokken DJ, editors. UK: Cambridge University Press.

    • Google Scholar
20. 1. Johnston R
    2. Brady HE

    (2002) The rolling cross-section design

    Electoral Studies 21:283–295.

    https://doi.org/10.1016/S0261-3794(01)00022-1

    • Google Scholar
21. 1. Lambin EF
    2. Tran A
    3. Vanwambeke SO
    4. Linard C
    5. Soti V

    (2010) Pathogenic landscapes: interactions between land, people, disease vectors, and their animal hosts

    International Journal of Health Geographics 9:54.

    https://doi.org/10.1186/1476-072X-9-54

    • PubMed
    • Google Scholar
22. 1. Lee KS
    2. Divis PC
    3. Zakaria SK
    4. Matusop A
    5. Julin RA
    6. Conway DJ
    7. Cox-Singh J
    8. Singh B

    (2011) Plasmodium knowlesi: reservoir hosts and tracking the emergence in humans and macaques

    PLOS Pathogens 7:e1002015.

    https://doi.org/10.1371/journal.ppat.1002015

    • PubMed
    • Google Scholar
23. 1. Lindgren F
    2. Rue H
    3. Lindström J

    (2011) An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach

    Journal of the Royal Statistical Society: Series B 73:423–498.

    https://doi.org/10.1111/j.1467-9868.2011.00777.x

    • Google Scholar
24. 1. Lindgren F
    2. Rue H

    (2015) Bayesian spatial modelling with R-INLA

    Journal of Statistical Software 63:1–25.

    https://doi.org/10.18637/jss.v063.i19

    • Google Scholar
25. 1. Manin BO
    2. Ferguson HM
    3. Vythilingam I
    4. Fornace K
    5. William T
    6. Torr SJ
    7. Drakeley C
    8. Chua TH

    (2016) Investigating the contribution of Peri-domestic transmission to risk of zoonotic malaria infection in humans

    PLOS Neglected Tropical Diseases 10:e0005064.

    https://doi.org/10.1371/journal.pntd.0005064

    • PubMed
    • Google Scholar
26. 1. Moyes CL
    2. Shearer FM
    3. Huang Z
    4. Wiebe A
    5. Gibson HS
    6. Nijman V
    7. Mohd-Azlan J
    8. Brodie JF
    9. Malaivijitnond S
    10. Linkie M
    11. Samejima H
    12. O'Brien TG
    13. Trainor CR
    14. Hamada Y
    15. Giordano AJ
    16. Kinnaird MF
    17. Elyazar IR
    18. Sinka ME
    19. Vythilingam I
    20. Bangs MJ
    21. Pigott DM
    22. Weiss DJ
    23. Golding N
    24. Hay SI

    (2016) Predicting the geographical distributions of the macaque hosts and mosquito vectors of Plasmodium knowlesi malaria in forested and non-forested Areas

    Parasites & Vectors 9:242.

    https://doi.org/10.1186/s13071-016-1527-0

    • PubMed
    • Google Scholar
27. 1. Papworth SK
    2. Bunnefeld N
    3. Slocombe K
    4. Milner-Gulland EJ

    (2012) Movement ecology of human resource users: using net squared displacement, biased random bridges and resource utilization functions to quantify hunter and gatherer behaviour

    Methods in Ecology and Evolution 3:584–594.

    https://doi.org/10.1111/j.2041-210X.2012.00189.x

    • Google Scholar
28. 1. Patz JA
    2. Daszak P
    3. Tabor GM
    4. Aguirre AA
    5. Pearl M
    6. Epstein J
    7. Wolfe ND
    8. Kilpatrick AM
    9. Foufopoulos J
    10. Molyneux D
    11. Bradley DJ
    12. Working Group on Land Use Change and Disease Emergence

    (2004) Unhealthy landscapes: policy recommendations on land use change and infectious disease emergence

    Environmental Health Perspectives 112:1092–1098.

    https://doi.org/10.1289/ehp.6877

    • PubMed
    • Google Scholar
29. 1. Pindolia DK
    2. Garcia AJ
    3. Wesolowski A
    4. Smith DL
    5. Buckee CO
    6. Noor AM
    7. Snow RW
    8. Tatem AJ

    (2012) Human movement data for malaria control and elimination strategic planning

    Malaria Journal 11:205.

    https://doi.org/10.1186/1475-2875-11-205

    • PubMed
    • Google Scholar
30. 1. Rue H
    2. Martino S
    3. Chopin N

    (2009) Approximate bayesian inference for latent gaussian models by using integrated nested Laplace approximations

    Journal of the Royal Statistical Society 2:319–392.

    https://doi.org/10.1111/j.1467-9868.2008.00700.x

    • Google Scholar
31. 1. Sadykova D
    2. Scott BE
    3. De Dominicis M
    4. Wakelin SL
    5. Sadykov A
    6. Wolf J

    (2017) Bayesian joint models with INLA exploring marine mobile predator-prey and competitor species habitat overlap

    Ecology and Evolution 7:5212–5226.

    https://doi.org/10.1002/ece3.3081

    • PubMed
    • Google Scholar
32. 1. Sh N
    2. Homathevi R
    3. Chua TH

    (2016)

    Mosquitoes of kudat: species composition and their medical importance (Diptera: culicidae)

    Serangga 21:149–162.

    • Google Scholar
33. 1. Shearer FM
    2. Huang Z
    3. Weiss DJ
    4. Wiebe A
    5. Gibson HS
    6. Battle KE
    7. Pigott DM
    8. Brady OJ
    9. Putaporntip C
    10. Jongwutiwes S
    11. Lau YL
    12. Manske M
    13. Amato R
    14. Elyazar IR
    15. Vythilingam I
    16. Bhatt S
    17. Gething PW
    18. Singh B
    19. Golding N
    20. Hay SI
    21. Moyes CL

    (2016) Estimating geographical variation in the risk of zoonotic Plasmodium knowlesi infection in countries eliminating malaria

    PLOS Neglected Tropical Diseases 10:e0004915.

    https://doi.org/10.1371/journal.pntd.0004915

    • PubMed
    • Google Scholar
34. 1. Singh B
    2. Sung LK
    3. Matusop A
    4. Radhakrishnan A
    5. Shamsul SSG
    6. Cox-Singh J
    7. Thomas A
    8. Conway DJ

    (2004) A large focus of naturally acquired Plasmodium knowlesi infections in human beings

    The Lancet 363:1017–1024.

    https://doi.org/10.1016/S0140-6736(04)15836-4

    • Google Scholar
35. 1. Stoddard ST
    2. Morrison AC
    3. Vazquez-Prokopec GM
    4. Paz Soldan V
    5. Kochel TJ
    6. Kitron U
    7. Elder JP
    8. Scott TW

    (2009) The role of human movement in the transmission of vector-borne pathogens

    PLOS Neglected Tropical Diseases 3:e481.

    https://doi.org/10.1371/journal.pntd.0000481

    • PubMed
    • Google Scholar
36. 1. Stoddard ST
    2. Forshey BM
    3. Morrison AC
    4. Paz-Soldan VA
    5. Vazquez-Prokopec GM
    6. Astete H
    7. Reiner RC
    8. Vilcarromero S
    9. Elder JP
    10. Halsey ES
    11. Kochel TJ
    12. Kitron U
    13. Scott TW

    (2013) House-to-house human movement drives dengue virus transmission

    PNAS 110:994–999.

    https://doi.org/10.1073/pnas.1213349110

    • PubMed
    • Google Scholar
37. 1. Vazquez-Prokopec GM
    2. Bisanzio D
    3. Stoddard ST
    4. Paz-Soldan V
    5. Morrison AC
    6. Elder JP
    7. Ramirez-Paredes J
    8. Halsey ES
    9. Kochel TJ
    10. Scott TW
    11. Kitron U

    (2013) Using GPS technology to quantify human mobility, dynamic contacts and infectious disease dynamics in a resource-poor urban environment

    PLOS ONE 8:e58802.

    https://doi.org/10.1371/journal.pone.0058802

    • PubMed
    • Google Scholar
38. 1. Weiss DJ
    2. Mappin B
    3. Dalrymple U
    4. Bhatt S
    5. Cameron E
    6. Hay SI
    7. Gething PW

    (2015) Re-examining environmental correlates of Plasmodium falciparum malaria endemicity: a data-intensive variable selection approach

    Malaria Journal 14:68.

    https://doi.org/10.1186/s12936-015-0574-x

    • PubMed
    • Google Scholar
39. 1. Wesolowski A
    2. Qureshi T
    3. Boni MF
    4. Sundsøy PR
    5. Johansson MA
    6. Rasheed SB
    7. Engø-Monsen K
    8. Buckee CO

    (2015a) Impact of human mobility on the emergence of dengue epidemics in Pakistan

    PNAS 112:11887–11892.

    https://doi.org/10.1073/pnas.1504964112

    • Google Scholar
40. 1. Wesolowski A
    2. Metcalf CJ
    3. Eagle N
    4. Kombich J
    5. Grenfell BT
    6. Bjørnstad ON
    7. Lessler J
    8. Tatem AJ
    9. Buckee CO

    (2015b) Quantifying seasonal population fluxes driving rubella transmission dynamics using mobile phone data

    PNAS 112:11114–11119.

    https://doi.org/10.1073/pnas.1423542112

    • PubMed
    • Google Scholar
41. 1. Wesolowski A
    2. Buckee CO
    3. Engø-Monsen K
    4. Metcalf CJE

    (2016) Connecting mobility to infectious diseases: the promise and limits of mobile phone data

    Journal of Infectious Diseases 214:S414–S420.

    https://doi.org/10.1093/infdis/jiw273

    • PubMed
    • Google Scholar
42. 1. Wong ML
    2. Chua TH
    3. Leong CS
    4. Khaw LT
    5. Fornace K
    6. Wan-Sulaiman WY
    7. William T
    8. Drakeley C
    9. Ferguson HM
    10. Vythilingam I

    (2015) Seasonal and spatial dynamics of the primary vector of Plasmodium knowlesi within a major transmission focus in sabah, Malaysia

    PLOS Neglected Tropical Diseases 9:e0004135.

    https://doi.org/10.1371/journal.pntd.0004135

    • PubMed
    • Google Scholar
43. 1. Yakob L
    2. Bonsall MB
    3. Yan G

    (2010) Modelling knowlesi malaria transmission in humans: vector preference and host competence

    Malaria Journal 9:329.

    https://doi.org/10.1186/1475-2875-9-329

    • PubMed
    • Google Scholar
44. 1. Yakob L
    2. Lloyd AL
    3. Kao RR
    4. Ferguson HM
    5. Brock PM
    6. Drakeley C
    7. Bonsall MB

    (2018) Plasmodium knowlesi invasion following spread by infected mosquitoes, macaques and humans

    Parasitology 145:101–110.

    https://doi.org/10.1017/S0031182016002456

    • PubMed
    • Google Scholar
45. 1. Yasuoka J
    2. Levins R

    (2007) Impact of deforestation and agricultural development on anopheline ecology and malaria epidemiology

    The American Journal of Tropical Medicine and Hygiene 76:450–460.

    https://doi.org/10.4269/ajtmh.2007.76.450

    • PubMed
    • Google Scholar

Thank you for submitting your article "Local human movement patterns and land use impact exposure to zoonotic malaria in Malaysian Borneo" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Ben Cooper as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Neil Ferguson as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Nick Ruktanonchai (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This study builds on previous work establishing environmental risk factors for human infections with the zoonotic malaria Plasmodium knowlesi and uses human movement data obtained by GPS tracking, spatiotemporal models of mosquito distribution, and spatial land use data to provide a new perspective on P. knowlesi transmission. The analysis predicts that more than 90% of infectious bites will occur in areas surrounding households at forest edges.

Essential revisions:

The reviewers were agreed that this work potentially represents a significant advance to the P. knowlesi literature. However, there were concerns that some aspects of the work were not clearly motivated, and considerable attention to both the Materials and methods and Results is needed to ensure the work is clearly described, accessible to a broad audience and repeatable. Specifically, the following three points were considered essential revisions:

1) Provide a null model for comparison to illustrate added value of data.

Also, clarify the value of the BRB model over simpler approaches such as kernel smoothing or linear interpolation.

2) Fully describe the results about different land use patterns.

3) Clarify multiple aspects of methods and improve reporting.

A detailed list of specific points that need to be clarified can be found in the individual reviews below.

Reviewer #1:

This paper describes some interesting data concerning human movement patterns and there has been some sophisticated statistical modelling. However, I only have a superficial knowledge of the zoonotic malaria literature and, after carefully reading the manuscript, feel that I am not in a position to judge whether or not this work represents a substantial step forward in understanding the epidemiology of P. knowlesi.

1) I couldn't follow the model specification in the subsection “Human space use”:

e.g. these points were not clear

"we randomly sampled equal numbers of absences within the study site". I don't know what this means. Is this sampling from the data? What are the "equal numbers" equal to? These "absences" were sampled, but what was done with the samples? I may be missing something very obvious here, and perhaps this will be familiar to people working with similar models, but I could only begin to guess what is being done here and why.

Is "occurrence" the same thing as "presence" or something else? It hasn't been made clear what "occurrence" means here.

The first equation suggests "occurrence" for individual j in cell i is defined to be 1 if utilisation of that cell by the same individual is greater than 0. How does this relate (if at all) to the earlier statement that "absence" was defined as a grid cell with a UD of less than 0.00001?

I'm confused by the second equation as well, as the z_ij values are defined in terms of the y_ij values in the first equation, but the y_ij values are defined in terms of the z_ij values.

2) –Subsection “Exposure to infected vectors”, last paragraph. I couldn't make sense of this section. Specific questions:

i) r_it, is a risk and so on the interval [0,1]. However, nothing in the first equation of this paragraph bounds the calculated r_it to this interval. Is the implicit assumption that r_it values will always be much less than 1 so that there is no need to worry about these bounds?

ii) How are a_jt and Z_j quantified, and how is uncertainty in these quantities accounted for?

Reviewer #2:

In 'Local human movement patterns and land use impact exposure to zoonotic malaria in Malaysian Borneo', the authors analyze a unique set of data to understand the role of human movement and land use patterns on the risk of exposure to P. knowlesi in Malaysia. P. knowlesi has unique ecological and epidemiological characteristics which makes the interdisciplinary approach (by capturing data about different environmental and human behavioral factors) particularly relevant. The authors use statistical models fit to GPS logger data from individuals, mosquito biting data, and geographic data to identify locations with the highest risk of P. knowlesi exposure. The primary result is that the majority of risk is nearby houses or forested areas, a useful finding that would be strengthened if more background about current public health interventions was provided. Given this result, there are points in the text where it appears that the manuscript was written for other results – based on these results it does not appear that the very detailed nature of daily movement patterns (which are quite variable) was not relevant to risk (since night time movement patterns were fairly consistent across groups). This is a useful and valuable finding, but often the text does not appear to emphasize the actual results (which should not be dismissed) presented.

Overall, there are three primary concerns about the current manuscript.

1) The authors do not provide a null model for comparison. Undoubtably, these data were time consuming and expensive to collect and it would not be feasible to collect these data systematically. However, the authors do not provide a reasonable null model as a point of comparison to clearly illustrate the added value of their data. For example, if you assumed that individuals spent 85%, 90%, 95%, etc. percentage of their time at their residence location instead of using the detailed GPS loggers, are the results substantially different? Similarly for the mosquito biting data, if you assumed that biting increased with forest cover, then how different are the results compared to the detailed analysis the authors provided? Since the final results are in terms of where there is the highest risk, it would be useful to see a comparison of where do these additional data and analyses add value by identifying locations of high risk that would have been missed otherwise. Without this type of comparison, which should be within the scope of the current article, it is difficult to assess how these results are different from a simpler model and set of assumptions. In general, it would be helpful if the authors highlighted where it would have been assumed that there was high risk if not for this analysis.

2) The authors do not fully describe their results about different land use patterns. Throughout the manuscript, the authors note that geographic variables about land use may not accurately reflect how individuals actually use the land. This is an important point and a useful topic to study for spatial analyses of infectious diseases that utilize these types of data. However, it is not clear exactly how their results translate to which types of patterns of use actually correspond to the geographic variables. For example, do certain geographic covariates correspond to different travel behavior? And in which instances do the different land use variables seem to better approximate similar travel behavior (within variable heterogeneity of use)?

3) Finally, it is unclear if the authors took into account or adjusted for their sampling scheme and the level of interpolation applied. Are there measures of uncertainty that could be applied to the analysis, perhaps informed by a null model? In addition, the total number of person nights appears to be highly correlated with the placement of the houses – which would suggest a spatial sampling bias, however there are a number of areas in red up to hundreds of kilometers from the houses. To what extent are these areas with a high number of person nights a function of the statistical smoothing or other covariates versus the raw data? It is understandable if the raw GPS coordinates are not made publicly available, but without additional information about the actual data and a more through presentation of these data, it is unclear to what extent these results are reflective of the actual travel patterns.

Reviewer #3:

Overall, this was a very well-written and interesting paper. The authors have done a generally good job writing up a pretty comprehensive set of analyses, and it represents an important step forward for modeling the spatial interaction of human movement patterns and vector borne disease transmission.

Essential revisions:

The reviewers were agreed that this work potentially represents a significant advance to the P. knowlesi literature. However, there were concerns that some aspects of the work were not clearly motivated, and considerable attention to both the Materials and methods and Results is needed to ensure the work is clearly described, accessible to a broad audience and repeatable. Specifically, the following three points were considered essential revisions:

1) Provide a null model for comparison to illustrate added value of data.

We can appreciate this request and have included further details of a model of mosquito biting and infection rates not incorporating human movement to address this concern in the reviewer specific comments below (response to reviewer 2, point 1).

Also, clarify the value of the BRB model over simpler approaches such as kernel smoothing or linear interpolation.

We have included substantially more background on the BRB model in the Introduction (Introduction, sixth paragraph), including advantages compared with kernel density smoothing. Specifically, this includes using the GPS points as a time-ordered series of points to more accurately estimate space use when there are missing data or irregular time intervals.

2) Fully describe the results about different land use patterns.

We have clarified the difference between land use and land cover and fully described the aspect of land use we have focused on (i.e. time spent in different land cover types) (Introduction, second and last paragraphs). The full results of this analysis are included in the predictive movement model (Table 2) and we have described additional land use topics which could be addressed in the Discussion (e.g. behaviours or activities in different locations) in the Discussion (last paragraph). We have additionally included a map of the study sites and land cover (Figure 2). Full details on classification of land cover are included in Fornace et. Al., 2018.

3) Clarify multiple aspects of methods and improve reporting.

We have substantially revised the methods and reporting throughout the manuscript. Specific points have been included below.

Reviewer #1:

1) I couldn't follow the model specification in the subsection “Human space use”:

We thank the reviewer for these comments and we have substantially revised the model description to clarify this (Materials and methods). We appreciate the time taken to review these methods in detail.

e.g. these points were not clear

"we randomly sampled equal numbers of absences within the study site". I don't know what this means. Is this sampling from the data? What are the "equal numbers" equal to? These "absences" were sampled, but what was done with the samples? I may be missing something very obvious here, and perhaps this will be familiar to people working with similar models, but I could only begin to guess what is being done here and why.

We have reworded this section to clarify these issues in the last paragraph of the subsection “Human space use”. This model includes two components: 1) a binomial model of whether or not the grid cell was visited, with presence defined as UD > 0.00001 and absence defined as UD < 0.00001; and 2) a γ model of the proportion of time spent at a grid cell if it was visited by an individual.

To fit the binomial component of this model, we needed to define presences and absences (i.e. places visited and not visited). As the UD is a probability distribution, it does not reach 0 and we considered all grid cells having a UD of less than 0.00001 as absences, i.e. we considered this value as negligible. To fit a binomial model of grid cells an individual visited, we included all grid cells visited (UD > 0.00001) and randomly subsampled an equal number of grid cells not visited. Within species distribution models, equal numbers of absences (or pseudoabsences) are commonly sampled to improve model predictive ability and we have included a reference describing this. This is described in the last paragraph of the subsection “Human space use”.

Is "occurrence" the same thing as "presence" or something else? It hasn't been made clear what "occurrence" means here.

We apologise that this was unclear and we have reworded this to be “presence” throughout. Specifically, this refers to the probability of grid cells being visited by an individual during the sampling period.

The first equation suggests "occurrence" for individual j in cell i is defined to be 1 if utilisation of that cell by the same individual is greater than 0. How does this relate (if at all) to the earlier statement that "absence" was defined as a grid cell with a UD of less than 0.00001?

Please see our second response to point #1 above.

I'm confused by the second equation as well, as the z_ij values are defined in terms of the y_ij values in the first equation, but the y_ij values are defined in terms of the z_ij values.

We have edited the symbols used for clarity and have revised the hurdle method description to clarify the utilisation distribution probability of a grid cell is conditional of an individual having visited this grid cell (subsection “Human space use”, last paragraph). We have also edited all equation symbols used for consistency (Materials and methods).

2) Subsection “Exposure to infected vectors”, last paragraph. I couldn't make sense of this section. Specific questions:

i) r_it, is a risk and so on the interval [0,1]. However, nothing in the first equation of this paragraph bounds the calculated r_it to this interval. Is the implicit assumption that r_it values will always be much less than 1 so that there is no need to worry about these bounds?

We have revised the wording around this to clarify that this refers to an exposure risk, specifically referring to the number of infectious bites received by an individual or the total of infectious bites occurring within a particular location (grid cell). As such, this quantity could theoretically range from 0 to infinity (subsection “Exposure to infected vectors”, last paragraph).

ii) How are a_jt and Z_j quantified, and how is uncertainty in these quantities accounted for?

These were calculated as derived quantities from the 3 components: human space use (y), mosquito biting rates (m) and infection rate (c) (Figure 1). The notation used around this was not clear in the earlier draft so we have edited to ensure consistency. In addition, we have included the details of the mosquito sporozoite data used to estimate proportion of infectious mosquitoes (–subsections “Human space use” and “Exposure to infected vectors”). As these numbers were very low, it was not possible to model variation in infection rates across time and space and we have included this as a limitation in the Discussion (Discussion, last paragraph).

Reviewer #2:

In 'Local human movement patterns and land use impact exposure to zoonotic malaria in Malaysian Borneo', the authors analyze a unique set of data to understand the role of human movement and land use patterns on the risk of exposure to P. knowlesi in Malaysia. […] This is a useful and valuable finding, but often the text does not appear to emphasize the actual results (which should not be dismissed) presented.

We can appreciate the reviewer’s point and have included additional results to clarify. Specifically, while most of the night time movements are predictable (results described in Table 2 and Results), there were still substantial variations in the predicted numbers of infectious bites (Table 5 and Results).

Overall, there are three primary concerns about the current manuscript.

1) The authors do not provide a null model for comparison. Undoubtably, these data were time consuming and expensive to collect and it would not be feasible to collect these data systematically. However, the authors do not provide a reasonable null model as a point of comparison to clearly illustrate the added value of their data. For example, if you assumed that individuals spent 85%, 90%, 95%, etc. percentage of their time at their residence location instead of using the detailed GPS loggers, are the results substantially different? Similarly for the mosquito biting data, if you assumed that biting increased with forest cover, then how different are the results compared to the detailed analysis the authors provided? Since the final results are in terms of where there is the highest risk, it would be useful to see a comparison of where do these additional data and analyses add value by identifying locations of high risk that would have been missed otherwise. Without this type of comparison, which should be within the scope of the current article, it is difficult to assess how these results are different from a simpler model and set of assumptions. In general, it would be helpful if the authors highlighted where it would have been assumed that there was high risk if not for this analysis.

We can appreciate the reviewer’s point about the need for comparison with a null model. In choosing such a model, we do not feel it would be valid to compare with estimates of risk assuming individuals spend X% of time around the house; while these might be reasonable assumptions to make based on this data, these assumptions cannot be made without collecting this type of data and the purpose of this study was to examine the distribution of these movements precisely because these data do not exist in the context of zoonotic malaria. We have included a sentence in the Discussion highlighting the need to consider measures human space when defining risk areas (third paragraph).

Rather, to highlight the importance of including movement, we have included additional data about where infectious mosquito bites would be expected based only on mosquito distribution data, serving as a null model (Results, last paragraph). This is additionally illustrated in Figure 5, where areas with higher mosquito biting rates (5C) do not correspond to areas with higher probabilities of human use (5D). This model highlights the difference in areas which would be identified using only standard vector surveillance vs. areas identified by also including human movement. Human landing catch data on mosquito biting rates are a core component of surveillance programmes and used to guide interventions. By also considering human space use, this highlights importance of areas with lower mosquito biting rates but much higher probabilities of human presence.

2) The authors do not fully describe their results about different land use patterns. Throughout the manuscript, the authors note that geographic variables about land use may not accurately reflect how individuals actually use the land. This is an important point and a useful topic to study for spatial analyses of infectious diseases that utilize these types of data. However, it is not clear exactly how their results translate to which types of patterns of use actually correspond to the geographic variables. For example, do certain geographic covariates correspond to different travel behavior? And in which instances do the different land use variables seem to better approximate similar travel behavior (within variable heterogeneity of use)?

Land use is commonly defined as the “total of arrangements, activities and inputs that people undertake in a certain land cover type” (IPCC, 2000). Within this broad definition, we have specifically focused on one aspect of land use, human use and movement into different environments (Introduction, last paragraph). This is a core component of this analysis and led to the development of a predictive model including the probability individuals would visit a specific location and, if the location was visited, the proportion of time spent at this location (Materials and methods, subsection “Human space use”). To develop this model, we specifically evaluated a number of geographic covariates, including distance to own house, distance to closest house, distance to roads, land cover class, distance to forest edge, elevation and slope (subsection “Human space use”, last paragraph). The full results of the final model and influence on space use are included in Table 2.

This model specifically describes one aspect of travel behaviour, the probability of visiting a location and the time spent in this location. If, by travel behaviour, this reviewer is referring to the activities undertaken by individual in a specific location, this is beyond the scope of this study as GPS data is specifically used to quantitatively measure the locations visited and the time spent in these locations. Behaviours in different environments may be better assessed using qualitative methods or detailed travel diaries on activities undertaken within different areas.

To clarify any confusion between land cover (the physical terrestrial surface) and land use (human activities undertaken on these surfaces), we have edited the Introduction and included the IPCC reference on the definition. We have also clarified that we are specifically focusing on one specific aspect of land use which is most relevant for infection dynamics (Introduction, second and last paragraphs). We have additionally added a sentence in the Discussion about the potential for future behavioural studies to explore other aspects of land use (Discussion, fourth paragraph) and have edited the wording to further clarify.

3) Finally, it is unclear if the authors took into account or adjusted for their sampling scheme and the level of interpolation applied. Are there measures of uncertainty that could be applied to the analysis, perhaps informed by a null model?

Yes, we did adjust for the sampling scheme. Although a total of 285 randomly selected individuals were sampled (roughly 20% of the populations over the age of 8 within these sites), predictive models used data for all individuals residing within these communities. We have substantially revised the Materials and methods section to clarify these points, particularly in relation to the characteristics of the community and individuals sampled for GPS tracking (Table 1, Figure 2). The level and method of interpolation used to estimate utilisation distributions are described in the sixth paragraph of the Introduction and Materials and methods subsection “Human space use”), with source code included as Source code 1. This includes the parameters used for smoothing and interpolating missing data.

In addition, the total number of person nights appears to be highly correlated with the placement of the houses – which would suggest a spatial sampling bias, however there are a number of areas in red up to hundreds of kilometers from the houses. To what extent are these areas with a high number of person nights a function of the statistical smoothing or other covariates versus the raw data? It is understandable if the raw GPS coordinates are not made publicly available, but without additional information about the actual data and a more through presentation of these data, it is unclear to what extent these results are reflective of the actual travel patterns.

We have included additional details on the demographic characteristics of the individuals included within the GPS tracking study and the study site population used for predictive models (Table 1). To address concerns about spatial bias, we have additionally included a map of the two study sites, with locations of households included in the GPS tracking study and survey (Figure 2). Further details on the individual household locations and GPS points could not be included due to ethics restrictions.

The prediction of high numbers of person nights spent in close proximity to houses is not an artefact of sampling bias but rather because of the effects of the covariates. One of the main predictors of human movement between 6pm and 6am is Euclidean distance from the individuals own house, followed by distance from other houses and roads. The predictive model was limited to areas within 5km of the households included in the study site (subsection “Human space use”, last paragraph) for all individuals aged 8 and over residing in this community. We have included an additional line in the Discussion that this does not allow assessment of long range or infrequent movements (third paragraph).

Regarding concerns about spatial smoothing, we have included further detail on using biased random bridges to estimate utilisation distributions (Introduction, sixth paragraph) and the specific parameters used (subsection “Human space use”), with an example script to run this analysis included in Source code 1. We used a minimum smoothing parameter of 30m to account for the resolution of habitat data (land cover and other predictors). The effects of these parameters have also been discussed in referenced literature (Benhamou, 2011).

Author details

1. Kimberly M Fornace

   1. Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, United Kingdom
   2. Centre on Climate Change and Planetary Health, London School of Hygiene and Tropical Medicine, London, United Kingdom

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   For correspondence

   Kimberly.Fornace@lshtm.ac.uk

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-5484-241X

2. Neal Alexander

   Department of Infectious Disease Epidemiology, London School of Hygiene and Tropical Medicine, London, United Kingdom

   Contribution

   Validation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

3. Tommy R Abidin

   Department of Pathobiology and Medical Diagnostics, Faculty of Medicine and Health Sciences, Universiti Malaysia Sabah, Kota Kinabalu, Malaysia

   Contribution

   Data curation, Investigation

   Competing interests

   No competing interests declared

4. Paddy M Brock

   Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow, United Kingdom

   Contribution

   Formal analysis, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

5. Tock H Chua

   Department of Pathobiology and Medical Diagnostics, Faculty of Medicine and Health Sciences, Universiti Malaysia Sabah, Kota Kinabalu, Malaysia

   Contribution

   Data curation, Writing—review and editing, Interpretation of mosquito data

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8984-8723

6. Indra Vythilingam

   Parasitology Department, Faculty of Medicine, University of Malaya, Kuala Lumpur, Malaysia

   Contribution

   Data curation, Writing—review and editing, Interpretation and analysis of mosquito data

   Competing interests

   No competing interests declared

7. Heather M Ferguson

   Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow, United Kingdom

   Contribution

   Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

8. Benny O Manin

   Department of Pathobiology and Medical Diagnostics, Faculty of Medicine and Health Sciences, Universiti Malaysia Sabah, Kota Kinabalu, Malaysia

   Contribution

   Data curation, Investigation, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-0726-6146

9. Meng L Wong

   Parasitology Department, Faculty of Medicine, University of Malaya, Kuala Lumpur, Malaysia

   Contribution

   Data curation, Investigation, Collection and analysis of mosquito data

   Competing interests

   No competing interests declared

10. Sui H Ng

    Department of Pathobiology and Medical Diagnostics, Faculty of Medicine and Health Sciences, Universiti Malaysia Sabah, Kota Kinabalu, Malaysia

    Contribution

    Data curation, Investigation, Collection and analysis of mosquito data

    Competing interests

    No competing interests declared

11. Jon Cox

    Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, United Kingdom

    Contribution

    Conceptualization, Funding acquisition, Methodology, Writing—review and editing

    Competing interests

    No competing interests declared

12. Chris Drakeley

    Faculty of Infectious and Tropical Diseases, London School of Hygiene and Tropical Medicine, London, United Kingdom

    Contribution

    Conceptualization, Funding acquisition, Writing—review and editing

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0003-4863-075X

• 1,742

  Page views

• 261

  Downloads

• 34

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.