Pindolia, D.K., Garcia, AJ., Huang, Z.

12 Pages·2013·3.11 MB·English

by  Deepa K Pindolia

Pindoliaetal.MalariaJournal2013,12:397
http://www.malariajournal.com/content/12/1/397
RESEARCH Open Access
The demographics of human and malaria
movement and migration patterns in East Africa
Deepa K Pindolia1,2,3*, Andres J Garcia1,2, Zhuojie Huang1,2,4,5, David L Smith6,7,8, Victor A Alegana3,9,
Abdisalan M Noor3,10, Robert W Snow3,10 and Andrew J Tatem7,9
Abstract
Introduction: The quantificationof parasite movements can provide valuable information for control strategy
planning across all transmission intensities.Mobile parasite carrying individuals can instigate transmission in
receptive areas, spread drug resistant strains and reduce the effectivenessof control strategies.The identification of
mobile demographic groups, their routes of travel and how these movements connectdiffering transmission zones,
potentially enables limited resources for interventions to be efficiently targeted over space, time and populations.
Methods: Nationalpopulation censuses and household surveys provide individual-levelmigration, travel, and other
datarelevant forunderstandingmalaria movement patterns. Togetherwithexistingspatially referencedmalaria data
and mathematical models, network analysis techniques were used to quantify thedemographics ofhumanand
malaria movementpatternsinKenya,Uganda and Tanzania. Movement networks were developed based on con-
nectivity and magnitudes offlow within each country and compared to assess relative differences between regions
and demographic groups. Additional malaria-relevant characteristics, such as short-term travel and bed netuse,
were also examined.
Results: Patterns of human and malaria movements varied between demographic groups, within country regions
and between countries. Migration rates were highest in 20–30 year olds in allthreecountries, but when accounting
for malaria prevalence, movements in the10–20 year age group became more important. Different age and sex
groups also exhibited substantial variations interms of the most likely sources, sinksand routes of migration and
malaria movement, as well as risk factors for infection, such as short-term travel and bed netuse.
Conclusion: Census and survey data, together with spatially referenced malaria data, GIS and network analysis
tools, can be valuablefor identifying,mapping and quantifyingregionalconnectivities and the mobility of different
demographic groups. Demographically-stratified HPM and malaria movement estimates can provide quantitative
evidence to inform the design ofmore efficient intervention and surveillance strategies that are targeted to specific
regions and population groups.
Background areas risks reintroduction and resurgence in malaria-free
Increased investment in malaria control and international receptiveareas,andhasunderminedeliminationeffortsin
donor support in recent years has led to reductions in the past [5-8]. In non-elimination settings, understanding
transmission,morbidityandmortalityinmanymalariaen- the patterns of parasite dispersal from local hotspots of
demic parts of the world [1-4]. The movement of malaria transmission can aid the design of additional targeted
parasites, primarily through the movement of infected controlbyidentifyingboththeregionswhereimportedin-
humans,isimportantforsuccessfulinterventionstrategies fectionsoriginateandwheretheymaycontributesubstan-
across the full range of transmission intensities. Human tially to transmission [9]. Finally, HPM patterns have
population movement (HPM) from higher transmission driventhespreadofdrugresistantparasitestrains[10,11].
Strategic control and elimination plans should therefore
*Correspondence:[email protected] be built on a strong evidence base including information
1EmergingPathogensInstitute,UniversityofFlorida,Gainesville,FL,USA on HPM and likely parasite movement volumes and
2DepartmentofGeography,UniversityofFlorida,Gainesville,FL,USA
routes [9]. Moreover, identifying key demographic groups
Fulllistofauthorinformationisavailableattheendofthearticle
©2013Pindoliaetal.;licenseeBioMedCentralLtd.ThisisanopenaccessarticledistributedunderthetermsoftheCreative
CommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,and
reproductioninanymedium,providedtheoriginalworkisproperlycited.
Pindoliaetal.MalariaJournal2013,12:397 Page2of12
http://www.malariajournal.com/content/12/1/397
mostlikelytocarryinfectionscanprovideusefulinforma- scales and demographic groups. By combining these data
tionfortailoredandtargetedinterventionandsurveillance with P. falciparum transmission maps and mathematical
efforts[12]. models, the demographic groups most likelyto move and
A variety of data types, statistical analyses and math- carry infection were explored, and likely sources, sinks,
ematical models have been used to quantify HPM pat- connectivity and importation routes of infection-carrying
terns [13,14] and specific HPM patterns relevant for individuals compared between demographic groups.
malaria dynamics [15-20] at different spatial scales. Finally, within group heterogeneities in short-term travel
National surveillance data, such as hospital patient re- and bed net use were assessed to further illustrate the
cords, that provide individual-level travel history and heterogeneities in travel and risk patterns that exist, and
demographic data have also been used to directly quantify toidentifyhigh-riskmalariamovementgroups.
features of imported malaria cases [21]. However, surveil-
lance data is likely to miss asymptomatic parasite carriers Methods
and non-health seeking cases [22], and comprehensiveand Data
reliable surveillance systems to detect imported cases are Humanpopulationmovement(HPM)data
generally under developed in low-income countries. In Migration and movement data from national household
these settings, directly estimating malaria movement has survey data and national statistical bureaus for Kenya,
primarily been based on travel history data from selected Uganda and Tanzania were obtained (Additional file 1).
population groups or geographic areas, with travel studied Individual-level census and survey data that included
asapossibleriskfactorforinfection[23,24].Recentlyhow- HPM data (migration and short-term travel-related ques-
ever, the availability of various HPM data types, high reso- tions), demographic descriptions, rural-urban stratifica-
lution spatially-referenced Plasmodium falciparum and tions and malaria-relevant records (such as bed net use)
Plasmodium vivax malaria metric data [25,26], mathemat- were obtained. The available census and survey datasets
ical models [27-30] and computational toolshaveprovided differedintermsofsamplesizes,representedpopulations,
an alternative approach to indirectly measure malaria migration and travel questions asked, rural and urban
movements [18]. Airline passenger networks and P. falcip- location records, demographics captured and malaria-
arummalariatransmissionmapsbeenusedtomodellarge- relevant variables recorded. However, datasets were simi-
scale malaria movements [31,32]. Novel study methods lar inthewaythatHPM,demographiccharacteristicsand
basedonmobilephoneusagedatacombinedwithP.falcip- rural-urban status weredefined.HPM was defined in two
arum malaria transmission maps, for example, have begun ways.First,anindividualorigin-destinationspecificmigra-
to tackle HPM and malaria movement dynamics at other tion(describedasaflowbetweenfirstand/orsecond-level
scales[19],suchasinZanzibarislandandatanationallevel administrativeboundarieswithinacountry)wasidentified
in Kenya [9,33]. Demographic and socioeconomic break- ifthe previous residence location of anindividualdiffered
downs of HPM, personal malaria protection and motiva- fromcurrentresidencelocation.Second,short-termtravel
tions for HPM have been reviewed in the context of per individual was identified if the individual spent time
malaria control and elimination [18]. However, detailed away from their current place of residence. Demographic
comparisons of high-risk demographic HPM groups have characteristics were described by age and gender. Rural-
notbeenundertakenatanationalorregionallevel,despite urban status was defined for each individual based on the
their importance in understanding malaria movement and rural-urbanstatusofresidencehouseholds.Individualbed
refiningquantitativeevidenceforguidingpolicydecisions. netusewasassessedbasedonwhetheranindividualslept
National population and housing census data can pro- underabednetthenightbeforethedatacollectiondate.
vide valuable individual-level records for quantifying
migration, travel and connectivity at a national scale, Nationalcensusmigrationmicro-data
that have been shown to correlate strongly to finer tem- Census micro-data, a systematically selected subset of
poral scale HPM [34], and have been used to analyse countrywide national housing and population census
migration patterns for many years [35-37]. Census and data obtained from the Integrated Public Use Microdata
national household surveys also provide individual-level Series (IPUMS) [39], was obtained for all three coun-
data for demographic and socioeconomic characteristics, tries. The census micro-data for Kenya was a 5% sample
motivations for travel and use of bed nets, which if ana- from its 1999 census, for Uganda a 10% sample from its
lysed in a systematic way, can be used to illustrate rela- 2002 census and for Tanzania a 10% sample from its
tive HPMvariations within the population [38]. Here we 2002 census. The data samples included individual-level
collate HPM datasets from Kenya, Uganda and Tanzania data for all questions included in each census. Questions
and use them with network analysis techniques and Geo- about migration, migrants, and their demographics were
graphicalInformationSystems(GIS)toolstodescribeand extracted from the most recent IPUMS data available for
examine HPM patterns across differing spatiotemporal each country. IPUMS migration data for Kenya were
Pindoliaetal.MalariaJournal2013,12:397 Page3of12
http://www.malariajournal.com/content/12/1/397
obtainedfromcurrentresidencedistrictscomparedtoresi- (origin was defined by current administrative unit of
dencedistricts1yearpriortothecensus(69administrative residence). The datasets were organized into categories
level 2 units). For Uganda, migration data were obtained that represented: type of dataset used to quantify it (cen-
fromcurrentresidencedistrictscomparedtopreviousresi- sus data or survey data), country that the dataset repre-
dence districts (56 administrative level 2 units) and num- sented, data collection time (for example, month/year in
ber of years in current residence districts. For Tanzania, which data was collected), spatial-temporal category of
migration data were obtained at a lower spatial resolution HPMrecords(migrationorshort-termtravel).Migration
than Kenya and Uganda. Previous region (25 administra- data (stratified by age, gender, rural-urban status of des-
tivelevel1units)ofresidencewascomparedtocurrentre- tination) was obtained from IPUMS samples for each
gion of residence to describe migration. Individual-level country and used to undertake between country com-
demographic records (age and gender) and rural/urban parisons. As the definition of a migrant differed between
status of current residence locations were available for all countries, patterns of relative differences within coun-
migrantsinallthreecensusmicro-datasamples. tries were compared, rather than absolute value compar-
isons (Additional file 2). For Kenya, the IPUMS data was
Migrationandshort-termtraveldatafromnational supplemented with household survey data (KIHBS) to
householdsurveys compare short-term travel and bed net use immigrant
National household survey datasets with individual-level and non-migrant populations, stratified by age and gen-
HPM, demographic and bed net use data were obtained der. Comparisons were further verified using simple lin-
for Kenya. The Kenya Integrated Household Budget earregressions(Additionalfile3).
Survey (KIHBS) 2004/2005 data was obtained from the
Kenya National Bureau of Statistics. Individual-level HPMandmalariamovementnetworks
migration data included current district of residence I-PUMS migration data were used to construct stratified
(same 69 districts as the Kenya 1999 census), district of migrant flow networks, with origin and destination ad-
birth and rural/urban status of current and previous ministrative units for migrant flows as nodes in each
districts. For each individual in the survey, information network, as previously developed by Tatem et al. [41].
on age and gender were included. Migration was defined Migrant flows were stratified based on age and gender
using current district of residence and district of birth, characteristics of individuals and rural/urban status of
and further stratified by rural/urban location and demo- their destination. The stratified origin-destination mi-
graphic characteristics. Individual records on the num- grant flow data were then used to develop three types of
ber of cumulative months each individual spent away stratified directed migrant networks, based on attributes
from home in the past 12 months were used as a proxy assigned to each network edge: (i) un-weighted migrant
foranindividualengaginginshort-termtravel(>1month networks, where directional migrant flow between each
away). Other malaria relevant data extracted included origin-destination pair existed if at least one individual
bed netuseinmigrant andnon-migrant groups. HPM was recorded, (ii) weighted migrant networks,
where the magnitude of directional migrant flows be-
Malariadata tween each origin-destination pair formed the edge
Country-level malaria transmission maps for Kenya, weights of each network and (iii) malaria movement
Uganda and Tanzania were obtained from the 2010 glo- networks, where the edge weight represented direc-
bal P. falciparum endemicity maps (with P. falciparum tional migrant flows weighted by the mean age-specific
parasite rate standardized for 2-10 year olds (PfPR ), PfPR at the origin. Migrants from higher endemicity
2-10
for 1×1 km pixels) from the Malaria Atlas Project (MAP) areas had a larger weighting than migrants from lower
[25]. Previously developed mathematical models were endemicity areas, based on their current age. Larger
used to estimateage-specific PfPR for each administrative weights represented high likelihoods of malaria move-
unitbasedonthemeanPfPR estimatesperadministra- mentbetweenlocationpairs.
2-10
tive unit [29]. Administrative units were grouped into 3 Network analysis tools in the R igraph package [42]
control-relevant endemicity classes: 0>PfPR ≤5, 5> were used to identify and compare key features and
2-10
PfPR ≤40andPfPR >40[40]. properties of each stratified network and make relative
2-10 2-10
comparisons between countries and demographic groups.
Analysis Local and global network measures were used to quantify
Census and survey HPM data were extracted and stored and compare structural characteristics of networks and of
in a standardized format representing origin-destination nodeswithinnetworks[43].Localnetworkmeasureswere
migrant flows (the origins and destinations were defined usedtoexaminenode-specific(location-specific)centrality
according to the respective administrative units available withineachcountry-level networkand make comparisons
in each dataset) and origin-specific short-term flows between locations within a specific network included
Pindoliaetal.MalariaJournal2013,12:397 Page4of12
http://www.malariajournal.com/content/12/1/397
measures for un-weighted networks (in, out and total multiplying each origin-specific migrant inflow by age-
node degree, which represented the number of inward/ specific PfPR at the origin, for each destination (Index 2).
outward/total connections to node) and weighted net- Locationswithlargemigrantinflowsfromlowendemicity
works (in/out/total strength, which represented the num- origins were comparable to locations with fewer migrant
ber of inward/outward/total flows to a node). Local inflows from high endemicity origins. The most likely
network measures were summarized for each stratified sources and sinks of HPM and malaria movements were
network to examine and compare global structure and mapped and compared between age groups to illustrate
characteristicsofnetworks. relative differences between demographic groups in each
Migration patterns (connectivity), represented by un- country. 10–20 and 20–30 year age groups were con-
weighted directional networks, were compared between trastedindetailtoillustratesimilaritiesanddifferences.
demographic and rural-urban groups using mean degree Index 1: Outgoing malaria-weighted flows from dis-
(meannumberofperlocationconnections)foreachnet- trictj,toallotherdistrictsi
work. Weighted migration networks were compared to
agePfPRj(cid:1)Σn ageHPMij
assessrelativedifferencesinmigrationmagnitudes,using j¼1
mean strength of the HPM flow networks (mean num- Index 2: Incoming malaria-weighted flows to district i,
ber of migrant flows per location). Malaria movement
from allotherdistrictsj
networks were compared to assess relative differences in
likelymalariamovementmagnitudes,usingmeanstrength Σn agePfPRi(cid:1)ageHPMij
i¼1
of the malaria movement networks (mean number of
malaria-weighted migrant flows per location). Weighted
HPMandmalaria-weightedflows
migrant networks were compared to weighted malaria
Origin-destination specific HPM flow networks, stratified
movement networks to illustrate changes in the relative
by age, gender and rural/urban status were represented as
strengthoftieswithinstratifiednetworks, togiveanindi-
directional network flow maps. The network maps were
cation of the demographic groups most likely to carry in-
developedtoshowwhetherincomingmigrantsandmalaria
fections. As migration data differed between countries
movements originated from specific parts of the country
(Additional file 2), mean network measures were stan-
or whether origins were dispersed between various places.
dardizedbytherespectivetotalforeachmeasuretomake
The age-stratified flow maps were overlaid on categorized
comparisons between countries. Non-parametric multiple
mean age-specific PfPR maps to illustrate flows between
comparison Kruskal-Wallis tests were used to compare
differenttransmissionzones[40]ineachcountry.Toillus-
standardized mean network connectivity, mean HPM
trate the most important age-specific malaria movement
strength and mean malaria movement between age,
routes in each country, origin-destination location pairs
gender, rural-urban groups for each country and box-
were ranked by weighting HPM flow by the mean age-
plots were used to address within age group variations
specific PfPR at the origin, for each origin-destination pair
(Additionalfile4).
(Index 3). Origin-destination pairs with the largest origin
endemicity and the largest flows between them were as-
HPMandmalaria-weightedsourcesandsinks
Most likely ‘sources’ (districts/regions more likely to ex- sumed to be the most likely routes for infection flow. The
port migrants and malaria movements),‘sinks’ (districts/ 10 most important HPM and malaria movement flow
routesforeachcountrywerecomparedtoillustraterelative
regions more likely to import migrants and malaria
differences in malaria movementbetweenagegroups. The
movements) and routes of migrant flows and malaria
10–20and20–30yearagegroupswereagaincontrastedin
movements,stratifiedbyageandgenderofmigrantsand
detailtoillustratesimilaritiesanddifferences.
rural-urban status of destinations were identified.
Index 3: Ranking directional origin-destination spe-
Migrant sources and sinks were identified by the num-
cific HPMasrelevantfor likely malaria movement
ber of outward and inward migrants per administrative
location, respectively. Outward malaria movement was (cid:1)
agePfPRi ageHPMij
estimated by weighting migrant flows by origin endem-
icity. Weighted estimates were obtained by multiplying
age-specific outflows by age-specific mean PfPR for each Short-termtravelandbednetuseimmigrantand
origin (Index 1). Locations with low mean endemicity non-migrantpopulations
but large migrant outflows were comparable to high Short-term travel and bed net use, stratified by age and
mean endemicity origins with fewer migrant outflows if gender, were compared between migrant and non-
index values were similar. Low index values represented migrant populations in Kenya, by first plotting the data,
administrative units with low endemicity and few HPM then verifying the differences using linear regression
outflows. Incoming malaria movement was obtained by (Additional file 3). Short-term travel by immigrants was
Pindoliaetal.MalariaJournal2013,12:397 Page5of12
http://www.malariajournal.com/content/12/1/397
compared between age and gender groups, stratified by between countries, age groups, genders and urban/rural
rural-urban status to assess travel and bed net use in settings(Figure1,Additionalfiles2and4).
high-risk migrant groups. Short-term travel was then When examining the differences in connectivity of mi-
compared between migrant and non-migrant groups. grant networks between age groups (measured using
Individual bed net use was compared within migrant mean network degree), the 20–30 year old age group
groups and between immigrants and non-migrants, as had the highest values (Figure 1, 1st row of graphs), il-
stratifiedbyagegroupandgender. lustrating that this age group was likely to migrate be-
All raw data sets were extracted, reformatted and tween the largest variety locations within each country.
stored usingMicrosoftExcel,MicrosoftAccessandR. Further heterogeneity in connectivity was seen when
HPM data was extracted and organized using Micro- networkswerestratifiedbygender. Forolderagegroups,
soft Excel, Microsoft Access SQL queries and R. Malaria connectivity was higher amongst males, and for younger
endemicity data was analysed in ArcGIS and R. Network age groups, female migrant networks were more con-
methods were applied using the igraph package in R. nected. Age and gender connectivity differences were not
Multiple comparison Kruskal-Wallis tests were imple- uniformbetweencountries.Forexample,withintheyoun-
mented inR[44]. ger age groups, <5 years olds were shown to have higher
connectivity compared to 5–10 year olds in Kenya, indi-
cating that young children may be likely to migrate with
Results their parents, which was not evident in the Uganda and
HPMandmalariamovementnetworks Tanzania data. Rural-urban differences in connectivity
Migrant and malaria networks in East Africa showed vari- were also revealed, showing that migrants currently living
ousdifferencesandsimilaritiesinpatternsandmagnitudes in rural areas had significantly higher connectivity than
Kenya Tanzania Uganda
0.06
C
0.04 on
n
e
ctivity
0.02
0.075 Group
e
Network.Measur 00..002550 HPMmagnitude ffmmeeaammlleeaa__lleeru__urrurbuararblnaaln
0.000
0.10 M
a
la
ria
M
o
0.05 vem
e
n
t
0.00
5 0 0 0 0 0 0 + 5 0 0 0 0 0 0 + 5 0 0 0 0 0 0 +
< 1 2 3 4 5 6 0 < 1 2 3 4 5 6 0 < 1 2 3 4 5 6 0
− − − − − − 6 − − − − − − 6 − − − − − − 6
5 0 0 0 0 0 5 0 0 0 0 0 5 0 0 0 0 0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Age
Figure1Comparingpatternsandmagnitudesofhumanpopulationmovement(HPM)andmalariamovementinEastAfrica,basedon
meandegreeandnetworkstrengthmeasures,foreachage,genderandrural-urbanstratifiednetwork.Meandegreewasusedto
measuredifferencesinHPMconnectivityandstrengthwasusedtomeasuremagnitudesofHPMflowandlikelymalariamovement.
Pindoliaetal.MalariaJournal2013,12:397 Page6of12
http://www.malariajournal.com/content/12/1/397
migrants living in urban areas in Uganda (based on the movementsourceswerealsoclosetotheLakeVictoriare-
Kruskal-Wallis test - Additional file 4). However, in gioninUganda.Tanzaniashowedadifferentpatternfrom
Kenya and Tanzania, urban resident migrant connectiv- Uganda and Kenya, with sources and sinks of HPM and
itywashigher. malaria movement distributed across the country. Some
Magnitudes of migrant flows (measured using mean districts ranked high as HPM sources/sinks, but were not
network strength) were highest in the 20–30 year old as important for malaria movement. For example, in
age group (Figure 1, 2nd row of graphs), showing similar Tanzania, northern regions were likely sources of HPM
results to connectivity. With more female internal mi- butunlikelysourcesofimportedinfections.
grants than males in all three countries, the overall The spatial distribution of most likely sources and
patterns in the region were dominated by females. sinks of HPM and malaria movement also differed be-
Gender-specific and rural-urban migration flow hetero- tween demographic groups (10–20 and 20–30 year age
geneities were also seen. In Uganda, overall magnitudes groups compared in Figure 2). Some districts/regions
of migrant flows were higher for rural migrants for all were important in both importing and exporting HPM
age groups, however gender differences were only seen and malaria, and this also differed between age groups.
in the 20–30 year age group. For Kenya and Tanzania, For example, Nairobi was both a source and sink of
differences between rural and urban migrants were only HPM(according toIPUMS1999data,16.5%ofNairobi’s
seen for the 10–20 and 20–30 year old age groups and population had moved from another district) and sink
not foryoungeroroldermigrants. for malaria movement in both 10–20 and 20–30 year old
When migrant networks were weighted by origin age- age groups. Mombasa, the second largest city in Kenya,
specific mean PfPR, peak flow magnitudes shifted to the was a sink for HPM and malaria movement for both age
10–20 year old age group for all age, gender and rural- groups,andwasalsoa sourceforHPMinthe20–30year
urban stratifications, except Ugandan and Tanzanian old age group. Unlike in Kenya, Dar es Salaam, the com-
urban males (Figure 1, 3rd row of graphs). As with con- mercial capital of Tanzania, was a sink and a source for
nectivity and migration magnitudes, malaria movements both HPM and imported infections, and the region in
differed between gender and rural-urban stratifications. which Dodoma, the national capital, is located was only a
In Kenya, females were more likely to import malaria source of HPM and malaria movement and not a sink.
compared to males for ages less than 20 years. This Similarly,Kampala,Uganda’scapitalcity,wasasourceand
meant that in general, origins of female migrants youn- sinkforbothHPMandmalariamovement.Overall,HPM
ger than 20 years old had higher mean age-specific PfPR source/sink patterns were different between the 10–20
than origins of male migrants in the same age group. In and 20–30 year old age groups, however, malaria move-
Tanzania, urban migrants aged 10–40 years had larger ment patterns were more similar between age groups for
malaria-weighted flows than rural-residing migrants. UgandaandTanzania,butnotKenya,atanationalscale.
However, in Uganda, magnitudes of malaria-weighted
flows were always higher in rural-residing migrants than HPMandmalaria-weightedflows
urban-residingmigrants. The most common routes for HPM flows and the most
Statistically significant differences between mean con- likelyroutesformalariamovementweredifferentbetween
nectivity and mean strength of each stratified network, demographic groups for all three countries (Figure 3). In
based on the Kruskal-Wallis test, were detected for age, Kenya, Nairobi was a major sink of HPM and malaria
gender and rural-urban stratified networks for all three movements whilst the Lake Victoria region was a major
countries.Within age group variation in connectivity, mi- source (Figure 2). The flow maps showed that origins of
gration magnitudes and relative malaria movements were HPM into Nairobi were likely to be various parts of the
also detected and differed between countries (Additional country, and this differed between age groups, however,
file4). malaria movements primarily originated in the Lake
Victoria region for both age groups(Figure 3). In Uganda,
SourceandsinksofHPMandmalariamovement KampalawasbothasourceandsinkforHPMandmalaria,
Within each country, some districts/regions were more however top ranking origin-destination specific HPM and
likely to be sources, whilst others were more likely to be malaria movement flows were into Kampala from sur-
sinksformigrantflowsandmalariamovements(Figure2). rounding districts for both age groups. In Tanzania, both
The overall spatial distribution of sources and sinks dif- HPM and likely malaria movement routes occurred over
feredbetweenHPMandmalariamovementsandbetween largedistances(relativetoKenyaandUganda).Theregion
countries. In Kenya, HPM sources were more spread out in which Dodoma is located was a major source of HPM
in the southern part of the country, however, sources of andmalariamovement(Figure2),withthelargestmigrant
malaria movement were concentrated around the Lake outflows to central and western parts of the country for
Victoria and western region. As seen for Kenya, malaria bothagegroups.Formalariamovementsinthe20–30year
Pindoliaetal.MalariaJournal2013,12:397 Page7of12
http://www.malariajournal.com/content/12/1/397
Figure2Comparingage-stratifiedhumanpopulationmovement(HPM)andlikelymalariamovementsourcesandsinksinEastAfrica.
Thetenmostlikelyage-specificsourcesofHPMandlikelymalariamovements,whichrepresentdistricts/regionsthatweremostlikelytoexport
HPMandmalaria,arecolouredinred.Thetenmostlikelyage-specificsinksofHPMandlikelymalariamovement,whichrepresentdistricts/re-
gionsthatweremostlikelytoimportHPMandmalaria,arecolouredinblue.Thetenmostlikelydistricts/regionsthatwerelikelytobothexport
andimportHPMandmalariaarecolouredinyellow.
old agegrouphowever, Dodomawasa top rankingsource Differences were also seen in travel patterns when age-
of malaria movements specifically to the northern region stratified migrants were further stratified by rural-urban
intheLakeVictoriaarea,whereMwanzacityislocated. status of previous (origin) residence (Additional file 3).
Bednetusagewashigherinfemaleimmigrantscompared
Short-termtravelandbednetuseimmigrantandnon- to males for younger age groups, and higher for males in
migrantpopulations older age groups. Bed net use in the 10–20 year old age
Short-term travel and bed net use differed within mi- group was relatively low compared to other age groups,
grant groups and between immigrants and non-migrants further emphasizing the importance of this age group in
(Figure 4, Additional file 3). Within migrant groups, the malaria movement. When comparing migrants to non-
highest proportions of travellers were <5 years old, migrants, migrants were more likely to engage in short-
followed by 10–20 year olds. In general, short-term travel term travel (recorded as having been away from normal
in younger immigrant females (<5 and 5–10 year olds) residence atleast once)thannon-migrants forall age and
was more likely than in immigrant males, but for older gendergroups.Thelargestdifferencesinshort-termtravel
age groups short-term travel was more likely in males. It betweenmigrantsandnon-migrantswereseeninchildren
is important to emphasize that the 10–20 year old age (<5,5–10and 10–20year agegroups) for both malesand
group was estimated to have the highest likelihoods of females. Immigrant groups, those of<20 years weremore
malaria movement compared to other age groups, along likely to travel than older age groups, however in non-
with relatively high likelihoods of short-term travel. migrant groups, 20–30 year olds were most likely to
Pindoliaetal.MalariaJournal2013,12:397 Page8of12
http://www.malariajournal.com/content/12/1/397
Figure3Comparingage-stratifiedhumanpopulationmovement(HPM)andlikelymalariamovementflowroutesinEastAfrica.Arrows
representdirectionalHPMflowsandlikelymalariamovementsbetweendistricts/regions.Flowsareoverlaidoncategorizedage-specificmean
PfPRmapstoillustrateHPMbetweentransmissionzones.
engage in short-term travel. Differences were also seen populations and areas where imported infections are
in bed net use between migrants and non-migrants likely [12,46]. Identifying high-risk immigrants may also
(Additionalfile3). pin-pointwheredrugresistancestrainsmayariseorspread
[11] and enable more efficient targeting of effective anti-
Discussion malarial treatments. In areas of heterogeneous transmis-
The quantification and analysis of HPM can be import- sion risk, local hotspots can provide specific targets for
ant for successful planning of both malaria control and strategic intervention deployment [47,48]. Within these
elimination [45]. As shown here, patterns and magni- hotspots, identifying high-risk demographic groups likely
tudes of HPM and individual infection rates differ be- to import and export infections and how they connect
tween demographic groups [29], and further differences othertransmissionzonesacrossacountrymayenablefur-
in individual behaviors, such as bed net use and short- therrefinementofinterventiontargetsanddevelopmentof
term travel [18], lead to variation in likelihoods of mal- cost-efficient attack strategies. For residents of low trans-
aria movement between these sub-population groups. mission areas traveling to hotspots, for example, in the
Quantifying these differences allows identification of key contextof boardingschoolchildren,providingprophylaxis
demographic groups, and the likely sources, sinks and before children travel to higher endemicity home loca-
routesofinfectioninflow[9],whichenablesthedevelop- tions, insecticide spraying in dormitories and provision
ment of more stringent surveillance systems, through of bed nets may be adequate measures. On a larger
prioritizing data collection and targeting resources to scale, education programmes targeted at high-risk
Pindoliaetal.MalariaJournal2013,12:397 Page9of12
http://www.malariajournal.com/content/12/1/397
50
40
30 B
e
d
n
20 et
Migrant Female
10
e
g Migrant Male
a
nt 0
e Non−migrant Female
c
er
P Non−migrant Male
30
Total
T
20 ra
v
e
l
10
0
5 0 0 0 0 0 0 s
< 1 2 3 4 5 6 u
− − − − − − pl
5 0 0 0 0 0 0
0 1 2 3 4 5 6
Age
Figure4Comparingbednetuseandshort-termtravelbetweenmigrantsandnon-migrants,stratifiedbyageandgenderinKenya.
Foreachagegroup,theproportionofindividualsthatsleptunderabednetthepreviousnight,outofallindividualsenumerated,wasobtained
fromtheKenyaIntegratedHouseholdBudgetSurvey(KIHBS).Similarly,foreachagegroup,theproportionoftravellersfromallindividuals
enumeratedwasobtained.
mobile populations or regions identified here may also infections through short-term travel. With Uganda’s
ensure bed net usage and treatment seeking rates be- populationandgeographybeinglargelyrural,andmajor
comehigherinthesekeygroups. urban centers being smaller compared to Kenya and
Due to varying infection rates and the malaria endem- Tanzania[51],ruralHPMandmalariamovementsdom-
icity at the place of origin between age groups [29], inated urban ones. Implications of imported infections
those age groups that exhibit higher movement rates in rural compared to urban areas differ, for example rural
may not be those with the highest infection prevalence areasmaygenerallyhavehigherreceptivity,increasingthe
(Figure 1). Hereitwasshown thatingeneral, thehighest likelihood of onward transmission [52]. By stratifying
rates of HPM are in the 20–30 year age group, as ex- HPM and malaria movement by origin-destination de-
pected with high rates of adult rural–urban migration in scriptions,areasthataremostlikelytoimportandexport
low income nations [49], however, likely malaria move- migrantsandsubsequentmalariamovementscanbeiden-
ments were highest in 10–20 year olds (Figure 1). With tified(Figure2).Moreover,migrationmaps(Figure3)can
the East African culture of boarding school attendance helptoidentifydifferencesinconnectivitybetweendemo-
[50],adolescentsarelikelytomigratebetweentransmis- graphic groups. In Kenya, high levels of malaria move-
sion zones, and engage in short-term travel to visit ment connectivity is seen between Nairobi and the Lake
families thereafter (Figure 4). Short-term movements Victoria region, matching previous findings [9]. Further
between different transmission zones are likely to be stratification of these movements, such as by rural-urban
more relevant for estimating numbers of imported in- and demographic descriptions, can help define high-risk
fections [9,19], with imported infections into receptive groups important for urban malaria control. For example,
areas threatening local transmission. Migratory moves adultmalemigrantsinNairobimayworkonconstruction
have however been shown to correlate strongly with sites where pools of stagnant water provide environments
shorter term connectivity [34], therefore, quantifying for mosquito breeding [53] and imported infections may
relative differences in demographically stratified migra- instigate outbreaks. Defining such high-risk populations
torymoves, as undertakenhere, providesa good indica- may become important in the future if successful control
tionoftherelativemagnitudesanddirectionsofshorter leads to low transmission, driving epidemiological shifts
term HPM patterns. However, these results also showed that make adult males more important than traditionally
that bed net use was higher immigrant populations, vulnerable pregnant women and children. These alterna-
which may reduce onward transmission of imported tive high-risk groups that may become a priority when
Pindoliaetal.MalariaJournal2013,12:397 Page10of12
http://www.malariajournal.com/content/12/1/397
aiming for elimination have recently been termed ‘hot- populations, such as durations of stay and access to
pops’[12,54]. healthcare within the 10–20 year old age group, would
While theresults outlined here point to clear patterns enable refined movement estimates within specific high-
andtrends,arangeofuncertaintiesstillremain.HPMis risk demographic groups to be made [19,30]. Specifically
the most difficult component to measure in the demo- for migrant groups, respondent-driven sampling has been
graphic equation [55], and data on it directly captures shown to be an effective technique for tracing frequency
only long temporal scales of human movement, which oftraveltohomelocations[56].
may not be the dominant type of movement for the
carriage of infections [15]. Nevertheless, it is strongly Conclusion
indicative of shorter temporal scale movements across With funding expected to decline in the near future and
sub-national spatial scales [34], and thus does provide a the need for cost-effective intervention strategies [57],
valuable indicator of connectivity amongst different novel research methods that provide input to evidence-
demographic groups and regions. Census data does not based decision-making are required. Here we have pre-
allow for origin-destination specific within administra- sented approaches that build on readily available datasets
tive boundary HPM to be estimated. Therefore, even toquantifythe variationsin relative connectivity and mo-
with high-resolution malaria maps, infection heteroge- bility across countries and between demographic group-
neities at small scales cannot be assessed. Additionally, ings. Through linkage with spatial malaria datasets, these
differently posed migration questions and differences in outputs can be translated into quantitative estimates of
missing data samples across census datasets limit the malaria parasite movement routes, sources, sinks and
use of census data in providing relative comparisons of rates,whichareimportantforunderstandingtransmission
HPM and malaria movement between countries. In the dynamicsanddesigningeffectiveandcost-efficientcontrol
surveydatausedhere,short-termtraveldatadidnot in- strategies.
cludedestinationsoftravelordurationsofstay,whichif
availablecouldbeusedwithexistingmathematicalmodels Additional files
toestimateindividualprobabilitiesofinfectionacquisition
and numbers of imported infections [19,30,33], providing Additionalfile1:EastAfricandatasetsformalaria-relevantHPMfor
Kenya,UgandaandTanzania.
absolute rather than relative measures. Further temporal
Additionalfile2:Comparingconnectivitysimilaritiesand
descriptions may also highlight seasonal variations in
differencesbetweencountries.
travelpatterns, whichifassociatedwithmalariadata, may
Additionalfile3:Short-termtravelimmigrantgroups,stratifiedby
illustrate seasonal differences in malaria movement. It is rural/urbanstatusanddistrictofcurrentresidenceinKenyaand
important to note however, that limitations do not only regressionmodelstostatisticallycomparetravelandbednetuse
betweenagegroups,genderandmigrationstatus.
exist in mobility and movement data, but also in the
Additionalfile4:Kruskal-Wallistestresultsandboxplotsto
biological knowledge of malaria infection acquisition
illustratevariationswithinagegroups.
[33], theuseofPfPR as ameasureforendemicityacross
transmission intensities [40] and demographic groups,
Competinginterests
and thedifficulty in measuring innate transmission risk,
Theauthorsdeclarethattheyhavenocompetinginterests.
orreceptivity,ofanarea.
Future work will examine the potential of linking the Authors’contributions
data sources and analyses presented here with other DKPdidtheliteraturesearch,identifieddatasets,carriedouttheanalysisand
wrotethefirstdraftofthemanuscript.AJGandZHcontributedtothe
mobility datasets, such as mobile phone usage data [9],
analysisofthemanuscript.DLScontributedtotheanalysisandreviewofthe
to estimate demographically-specific malaria movement manuscript.VAAcontributedtothedatacompilation.AMNandRWS
rates at high spatial and temporal resolutions. In most contributedtothereviewofthemanuscript.AJTcontributedtothewriting,
analysisandreviewofthemanuscript.Allauthorsreadandapprovedthe
countries, censuses and household surveys are under-
finalversionofthemanuscript.
taken regularly, with the data being often made freely
available at aggregated levels, meaning that the methods Acknowledgements
used here can be applied in other malarious regions. To AJT&DLSacknowledgefundingsupportfromtheEmergingPathogens
Institute,UniversityofFlorida,theRAPIDDprogramoftheScienceand
expand the value of census and survey HPM data within
TechnologyDirectorate,DepartmentofHomelandSecurity,andtheFogarty
regions such as East Africa, extracting and analyzing InternationalCenter,NationalInstitutesofHealth,andarealsosupportedby
cross-border migration and travel-related data is import- grantsfromNIH/NIAID(U19AI089674)andtheBillandMelindaGates
Foundation(#49446and#1032350).DLSacknowledgesfundingsupport
ant for a more comprehensive assessment of movement
fromBloombergFamilyFoundation.RWSissupportedbytheWellcome
relevant for malaria control, elimination and drug resist- TrustasPrincipalResearchFellow(#079080).AMNissupportedbya
ance scenario planning [18], and future work will focus WellcomeTrustIntermediateResearchFellowship(##095127).BothRWSand
AMNarealsosupportedbyWellcomeTrustMajorOverseasProgramme
onthisaspect.Developingsub-populationsurveystoob-
granttotheKEMRI/WellcomeTrustResearchProgramme(#092654).The
tain more detailed travel histories in high-risk migrant fundershadnoroleinstudydesign,datacollectionandanalysis,decisionto