Towards integrated surveillance of zoonoses: spatiotemporal joint modeling of rodent population data and human tularemia cases in Finland

RESEARCH ARTICLE Open Access Towa ds in eg a ed su eillance o zoonoses: spa io empo al join modeling o oden popula ion da a and human ula emia cases in Finland C. Ro ejanap ase 1* , A. Lawson 2 , H. Rossow 3 , J. Sane 4 , O. Hui u 5 , H. Hen onen 5 and V. J. Del Rio Vilas 6 Abs ac Backg ound: The e a e an inc easing numbe o geo-coded in o ma ion s eams a ailable which could imp o e public heal h su eillance accu acy and e iciency when p ope ly in eg a ed. Speci ically, o zoono ic diseases, knowledge o spa ial and empo al pa e ns o animal hos dis ibu ion can be used o aise awa eness o human isk and enhance ea ly p edic ion accu acy o human incidence. Me hods: To his end, we de elop a spa io empo al join modeling amewo k o in eg a e human case da a and animal hos da a o o e a modeling al e na i e o combining mul iple su eillance da a s eams in a no el way. A case s udy is p o ided o spa io empo al modeling o human ula emia incidence and oden popula ion da a om Finnish heal h ca e dis ic s du ing yea s 1995–2012. Resul s: Spa ial and empo al in o ma ion o oden abundance was shown o be use ul in p edic ing human cases and in imp o ing ula emia isk es ima es in 40 and 75% o heal h ca e dis ic s, espec i ely. The human ela i e isk es ima es’s anda d de ia ion wi h oden ’s in o ma ion inco po a ed a e smalle han hose om he model ha has only human incidence. Conclusions: These esul s suppo he in eg a ion o oden popula ion a iables o educe he unce ain y o ula emia isk es ima es. Howe e , mo e in o ma ion on se e al co a ia es such as en i onmen al, beha io al, and socio-economic ac o s can be in es iga ed u he o deepe unde s and he zoono ic ela ionship. Keywo ds: Su eillance in eg a ion, Join diseases modeling, Zoonoses, Tula emia, Finland Backg ound Disease isk mapping is impo an o he unde s anding o he spa ial epidemiology o in ec ious diseases. In mos cases and e en o mul i-hos diseases such as zoonoses, isk es i- ma ion has been conduc ed in a uni a ia e ashion based on humancaseda aalone.Modelingo mul i a ia eheal h da a, in o med by mul iple s eams o geo-coded in o ma- ion, allows obse a ion o concu en pa e ns among da a s eams and condi ioning on one ano he . As a esul , mul i- a ia e me hods can deli e g ea e s a is ical powe and lead o mo e p ecise isk es ima ion and enhanced e en de ec ion. Speci ically, o zoono ic diseases, knowledge o spa ial and empo al pa e ns o he animal hos could in- o m incidence in humans. In eg a ion o da a and analyses, whe he o popula ion o heal h ela ed a iables, has been sugges ed o imp o e zoo- noses su eillance accu acy and e iciency [1,2]. In eg a ion appea s mo e easible o endemic zoonoses, and o hose wi h domes ica ed animals as sou ce, gi en he likely g ea e a ailabili y o animal heal h da a. Fo zoonoses wi h a non-domes ica ed animal sou ce (e.g. syl a ic yellow e e , u- la emia), a ailabili y o animal heal h da a is likely o be a lim- i ing ac o owa ds in eg a ion and al e na i e animal da a sou ces mus be sough . Tula emia is an in ec ious disease caused by an in acellu- la bac e ium, F ancisella ula ensis.Thediseaseisendemic * Co espondence: [email p o ec ed] 1 Depa men o T opical Hygiene, Facul y o T opical Medicine, Mahidol Uni e si y, Ra cha hewi, Bangkok 10400, Thailand Full lis o au ho in o ma ion is a ailable a he end o he a icle © The Au ho (s). 2018 Open Access This a icle is dis ibu ed unde he e ms o he C ea i e Commons A ibu ion 4.0 In e na ional License (h p://c ea i ecommons.o g/licenses/by/4.0/), which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons license, and indica e i changes we e made. The C ea i e Commons Public Domain Dedica ion wai e (h p://c ea i ecommons.o g/publicdomain/ze o/1.0/) applies o he da a made a ailable in his a icle, unless o he wise s a ed. Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 h ps://doi.o g/10.1186/s12874-018-0532-8 in No h Ame ica and pa s o Eu ope, wi h ecu en ou - b eaks in Sweden and Finland [3,4]. F ancisella ula ensis has a wide ange o hos s wi h ansmission mos commonly ia a h opod ec o s [5]. Roden s could play a ole in he zoono ic ansmission o he disease a e indings o a ela- ionship be ween ole popula ion cycles and human ula - emia incidence in Finland [6] and Sweden [7]. Speci ically in Finland, oden popula ion dynamics displayed a spa io em- po al ela ionship wi h human ula emia cases, such ha human case numbe s peaked one yea a e peak oden densi ies [6]. Simila indings, om s udies o ula emia ou - b eaks, indica e ha high oden densi ies migh ela e o occu ences in humans [8–11]. This wo k explo es he applica ion o spa io empo al join models o concu en animal and human geo- e e enced da a sou ces in an e o o explain possible pa e ns be- ween he dis ibu ion and/o he abundance o he animal hos and human disease.. The p oposed me hodology is e alua ed on i s pe o mance in p edic ing human disease isk and imp o ing isk es ima ion in a case s udy o ula - emia human incidence and oden popula ion da a in Finland. No only ou me hod con ains me hodological no - el y wi h po en ial applica ions in spa ial epidemiology, i also helps o e eal a disease pa e n in he case s udy which was no conside ed in p e ious s udies. Me hods Da a sou ces A comple e desc ip ion o he oden popula ion and human ula emia incidence da a is a ailable om ea lie epo s [6]. B ie ly, da a on oden popula ion le els, p edominan ly bank oles (Myodes gla eolus) and ield oles (Mic o us ag es is), we e collec ed ac oss Finland by he Na u al Resou ces Ins i- u e Finland and ca ego ized in o h ee popula ion le els: de- cline, inc ease, and peak [12]. Human ula emia cases we e epo ed as labo a o y-con i med o he Na ional In ec ious Disease Regis e , kep by he Na ional Ins i u e o Heal h and Wel a e o Finland. Bo h human cases and oden da a we e agg ega ed in o 20 Finnish heal hca e dis ic s o e he pe iod 1995–2012 [6]. An indica o o quan i y he in luence o o- den popula ion le els on human incidence is de eloped o each heal h dis ic . Plo s o human cases and bina y oden s a us o he 20 Finnish heal h dis ic s in he pe iod 1995– 2012 show a one-yea lagged inc ease in he numbe o hu- man ula emia cases a e oden popula ion peaks o ce ain dis ic s and yea s (Fig. 1). A simila pa e n was ound be- ween human cases and he ca ego ical oden popula ion s a- us (Fig. 2). These suppo he choice o spa io empo al models which will be de eloped in he nex sec ion. S a is ical me hodology Wep oposeaBayesian amewo k ojoin lyanalyze oden popula ion s a us and human case incidence. We assume ha he human cases a e associa ed wi h oden s’s a us h ough a la en s uc u e. Le human cases (coun s), h i ,a heal h dis ic iand ime ollow a Poisson dis ibu ion wi h mean = e i θ i whe e e i is he expec ed numbe o human cases in he i h heal h dis ic (p esumably cons an ac oss he yea s) and θ i is he ela i e isk a he i h heal h dis ic and yea The e a e a numbe o ways o calcula e he ex- pec ed a e. In his pape , he expec ed a es, e i , a e calcu- la ed as he a e age case coun a a ea io e he ime pe iod as ei¼P hi Twhe e Tis he leng h o s udy pe iod Fig. 1 Human cases (solid line) and 2-le el oden s a us (do ) o he 20 Finnish heal h dis ic s o e yea s 1995–2012 Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 2 o 8 (T= 8 yea s).Human popula ion da a was ob ained om he Finnish Popula ion Regis e Cen e [13]. Howe e , he popula ion a ia ion be ween egions was ound o be lim- i ed and combined wi h he low o e all a e o he disease, i was decided ha a ime a e aged a e would be app o- p ia e in his case. A simple app oach o join ly model human incidence and oden popula ion da a is o conside a 2-le el o- den popula ion indica o as in [6]. As a bina y a iable, we deno e i = 0 i he numbe o oden s declined and i = 1 i he numbe o oden s is a peak o inc eased. The 2-class oden s a us is hen assumed o ollow a Be noulli dis ibu ion wi h pa ame e p i being he p ob- abili y o oden o he i h heal h dis ic and yea .To speci y he pa ame e s in he join likelihoods, θ i o humans and p i o oden s, linea p edic o s a e decom- posed addi i ely in o spa ial, and space- ime in e ac ion andom e ec s as ollows α ,α h a e he o e all mean le els o oden and hu- man espec i ely, and assumed o ha e ze o-mean Gaussian p io dis ibu ions. The la en a iables uh i;u i; h i; ia e included o model non- empo al backg ound a ia ion wi h spa ial and non-spa ial p io dis ibu ions. The spa ial s uc u e o uh i;u i ollow an in insic condi- ional au o eg essi e (ICAR) [14] model and he non-spa ial dis ibu ion o h i; iis assumed o be ze o-mean Gaussian p io dis ibu ion. A Gaussian dis- ibu ion wi h ze o mean is assumed o λh ;λ a =1 and an au o eg essi e p io dis ibu ion is assumed o λh ;λ a > 1 which allows o a ype o nonpa ame ic empo al e ec . δh i ;δ i ep esen he empo al end o each heal h dis ic a yea . We also assume ha he space- ime andom e ec s o human and oden a e p o- po ional wi h one-yea lag. This is suppo ed by he inding sugges ed in [6] ha 1-yea empo al lag e ec humani ¼hi Poisson eiθi ðÞ log θi ðÞ¼αhþuh iþ h iþλh þδh i δh i ¼βiδ i −1 βi¼βp i i −1þβn i1− i −1 ðÞ oden i ¼ i Be noulli pi ðÞ logi pi ðÞ¼α þu iþ iþus kþ s kþλ þδ i αhN0;τ−1 αh  ;α N0;τ−1 α  δh i N0;τ−1 δh  ;δ i N0;τ−1 δ  λh Nλh −1;τ−1 λh  ;λ Nλ −1;τ−1 λ  λh 1N0;τ−1 λh  ;λ 1N0;τ−1 λ  uh iICAR τ−1 uh  ; h iN0;τ−1 h  u iICAR τ−1 u  ; iN0;τ−1  us kICAR τ−1 us  ; s kN0;τ−1 s  βp iN0;τ−1 βp  ;βn iN0;τ−1 βn  τ−1=2 Uni o m 0;10ðÞ Fig. 2 Human cases (solid line) and 3-le el oden s a us (do ed line) o he 20 Finnish heal h dis ic s o e yea s 1995–2012 Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 3 o 8 can be bene icial o p edic human ula emia ou b eaks. To model unobse ed ecological e ec s associa ed wi h ole cycles a he i e bo eal zones in Finland (Sou he n Finland, Sou hwes e n and Inland Finland, Eas e n Finland, No he n Finland, and Lapland) [13], wo add- i ional pa ame e s us k; s ka e included as he spa ial and non-spa ial con ex ual a iables o egional s a e le el k. δ i ;βp i;βn ia e assumed o ollow a ze o-mean Gaussian p io dis ibu ion and he uni o m dis ibu ion on (0,10) is used o model all s anda d de ia ion pa ame e s [15]. Al hough he e is some e idence ha a 2-le el oden s a us could ha e a lagged p edic i e abili y on human ula emia [6], we also wan o ex end ou conside a ion o include he o iginal h ee oden le els in he join modeling and assume a ca ego ical likelihood o oden popula ion s a us. A speci ica ion o mul iple ca ego ies o oden s a us can be de ined as humani ¼hi Poisson eiθi ðÞ log θi ðÞ¼αhþuh iþ h iþλh þδh i δh i ¼βi; i −1δ i −1;i −1 oden i ¼ i Ca ego ical p1;i ;p2;i ;p3;i  pj;i ¼ exp μj;i  P3 j¼1exp μj;i  μj;i ¼α þu iþ iþus kþ s kþλ þδ j;i whe e j= 1 indica es he declining oden popula ion le el, j= 2 is he inc easing le el, and j= 3 is he oden ’s popula ion a peak, and δ j;i Nð0;τ−1 δ Þ. The o he p io dis ibu ions o he andom e ec s e ms a e assumed he same as he 2-le el model. We u he assume ha he space- ime andom e ec s o human and oden a e p opo ional wi h one-yea lag. Tha is δh i αδ i −1;i −1, i.e. δh i ¼βi; i −1δ i −1;i −1whe e βi; i −1 is he p opo ional pa ame e . This speci ica ion is de- eloped o examine he lagged e ec o oden s a us, δ i −1;i −1, a le el j= i -1 in heal h dis ic ia ime -1on he numbe o human cases h ough he space- ime in e ac ion e m,δh i . Model e alua ion To e alua e he models, we use wo goodness o i mea- su es: he De iance In o ma ion C i e ion (DIC) [16,17] whe e he e ec i e numbe o pa ame e s is es ima ed in e ms o de iance’s a iance, and he Wa anabe-Akaike in o ma ion c i e ion (WAIC) [18–20]. Bo h measu es a e compu ed unde he likelihood o human da a o compa e he models’ i wi h and wi hou he con ex ual a iables. We also compa e he pos e io s anda d de ia- ions o θ i om he wo models (bina y and ca ego ical) wi h he oden da a, wi h hose om a model based only human da a, o examine he bene i s o inco po a ing animal da a in su eillance. Resul s a e ob- ained om 10,000 pos e io samples using WinBUGS so wa e a e a bu n-in pe iod o 10,000 d aws. To as- sess he mixing o pos e io sample s, we adop Gelman’sR bs a is ics p oposed in [18,21] o mul iple chain con e gence and con e ged chains should ha e he alue o R bapp oxima ely 1. Resul s The his og ams o R bes ima es o θ i om pos e io sam- ple unde he models a e displayed in Fig. 3. The R bes i- ma es unde all he models a e app oxima ely o less han 1.005 which indica es he chains con e ge o a pos- e io dis ibu ion. Table 1displays he DIC and he measu es o model pe o mance. We compa ed he models (bina y s. poly omous) on hei DIC and WAIC alues unde he human likelihood. The bina y model wi hou he con ex ual e ec shows he smalles DIC and WAIC alues and hence i can be conside ed as ha - ing he bes i o he da a. To p o ide e idence o he bene i s om inco po a ing oden in o ma ion in o he model, he pos e io es ima es o s anda d de ia ion (SD) o θ i om bo h bina y and poly omous models a e compa ed and p esen ed in Table 2 o each a ea o e he ime pe iod. The SD es ima es o θ i wi h oden ’s in- o ma ion inco po a ed a e smalle han hose om he model ha has only human incidence. This esul sup- po s he in eg a ion o oden popula ion a iables o educe he unce ain y o ula emia isk es ima es. To help assess he p edic i e pe o mance o oden abundance da a on he occu ence o human cases, we de elop a me ic on he 0 o 1 scale, namely he deg ee o posi i e indica o (DP) o each heal h dis ic as DPi¼expðβp iÞ expðβp iÞþ1:The DP indica o is de i ed om he bina y model wi hou he con ex ual ac o s as his was he model wi h he lowes DIC and WAIC. High alues o DP (close o 1) would indica e heal h dis ic s wi h a high numbe o human ula emia cases in he cu en yea , gi en inc easing o a peak oden popu- la ions in he pas yea . DP can be in e p e ed in a simila ashion o he sensi i i y o a diagnos ic es . High alues o DP would sugges good p edic i e alue o he oden da a on he occu ence o human ula - emia cases. Yea -speci ic modeling was no conside ed due o insu icien da a. Thus posi i e dependence alues ep esen an a e aged e ec o oden popula- ions on human incidence ac oss all he yea s (Table 3). The e a e 8 (40%) heal h dis ic s, mos ly on he wes and sou h o he coun y, wi h mean DP alues la ge han 0.8 and he lowe 95% c edible in e al abo e 0.5 (Fig. 4). Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 4 o 8 Discussion O he modi ica ions a e possible. Fo example, in join modeling human and oden da a we could conside hu- man cases being dependen on oden popula ion h ough p i as a ca ego ical co a ia e. Fo spa ial uni , smalle o adminis a i e a eas di e en om heal h dis- ic s may ha e led o mo e disc imina o y indings (o no gi en inc eased noise om smalle uni s). Simila ly, mo e in o ma i e da a on oden popula ion le els, e.g. oden densi ies as in o he s udies [11] migh ha e e- sul ed in mo e desc ip i e models. Roden popula ions luc ua e wi h a highly a ying ampli ude, which means ha abundances may a y subs an ially om one peak o he nex , e en wi hin he same egions [12]. I ula - emia ansmission o humans is a phase and densi y dependen p ocess, as i mos likely is [6], a ia ion in ole abundance du ing successi e peaks may educe he p edic i e alue o models employed he e. Subsequen s udies mus explo e he inco po a ion o da a on he p ecise loca ion o he oden apping si es h ough he use o some o m o in e pola ion [22]. In addi ion o hos popula ion changes, en i onmen al ac o s also seem o impac on he occu ence o ula emia ou - b eaks [23,24]. Inco po a ion o e idence on mosqui o dis ibu ion (no pu posely cap u ed a his momen in Finland), ain all and wa e bodies in o ou models would be s aigh o wa d. Conclusions Space- ime p oximi y and con ac pa e ns be ween humans and animals play a cen al ole in in ec ion isk. In his esea ch, we a emp ed o assess he su eillance ele ance o egula ly collec ed oden popula ion da a o i) imp o e ula emia isk es ima es and ii) in o m Table 1 DIC (pD) and WAIC (pWAIC) co esponding o he human likelihood o he model compa ison Roden s a us Con ex ual e ec DIC pD WAIC pWAIC Bina y Wi h 4269.945 1115.127 3463.587 557.5636 Wi hou 4262.807 1110.623 3458.687 555.3114 Poly omous Wi h 4276.408 1123.747 3465.682 561.8733 Wi hou 4324.426 1170.232 3482.997 585.1162 Fig. 3 His og ams o ^ Rs a is ics o θ i om pos e io sample unde models Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 5 o 8 ea ly p edic ion o human ula emia cases in Finland. To ha e ec , we de eloped bina y and poly omous a chi- ec u es o join ly model human incidence and oden s a us wi h one-yea lag. Ou esul s e u ned a he e o- geneous pic u e bu o many heal h dis ic s oden popula ion s a us was ele an o he occu ence o hu- man ula emia cases. We ha e shown ha he inco po - a ion o oden popula ion da a led o an imp o emen in he accu acy o human isk es ima es in 15 (75%) heal h dis ic s, compa ed o models only conside ing human ula emia cases. Fu he mo e, ou pu posely buil indica o (DP) showed ha in 8 (40%) o he heal h dis ic s, inc easing and a -peak oden popula ions we e obus p edic o s o human ula emia cases in he ollowing yea . Howe e , o ew dis ic s (e.g. Länsi-- Pohja and Kainuu) whe e he model based only on hu- man da a leads o mo e p ecise es ima es o incidence, hese a eas also ha e he low alues o posi i e indica o (DP) wi h he co esponding c edible in e als c ossing 0.5. This sugges s ha he dis ibu ion o zoono ic pa h- ogens in animal and human popula ions spa ially a ies as we assumed acco ding o local bio ic and abio ic de- e minan s. To conduc u he in es iga ion, we need mo e in o ma ion on se e al co a ia es such as en i on- men al, beha io al, and socio-economic ac o s. How- e e , he pla o m p oposed in his esea ch can acili a e in iden i ica ion o geog aphical a eas ha a e po en ially sui able o ansmission. Ou objec i e was o de elop di e en models o combine mul iple da a sou ces al eady a ailable, on animals and humans, o be e in o m he occu ence o zoonoses. The p esen wo k shows he u iliza ion o animal popula ion da a (in he absence o animal heal h- ela ed da a on oden s) o in o m human isk. Al hough di e en model pa ame e iza ions and, in pa icula , e idence on o he pu a i e p edic o s, as epo ed elsewhe e [25], could con ibu e u he Table 2 S anda d de ia ion o θ i calcula ed om pos e io sample s o each a ea o e he ime pe iod o he models Heal h dis ic Bina y Ca ego ical Only Human Wi h con ex ual e ec Wi hou con ex ual e ec Wi h con ex ual e ec Wi hou con ex ual e ec Sou hwes Finland 1.0216 1.0193 1.1402 1.1569 1.3604 Sa akun a 3.6330 3.6124 3.5493 3.5371 3.6365 Kan a-Häme 0.7336 0.7354 0.7632 0.7689 1.0398 Pi kanmaa 3.2779 3.2862 3.2114 3.2071 3.4309 Päijä -Häme 0.9032 0.9011 0.9682 1.0048 1.0475 Kymenlaakso 2.4761 2.4859 2.5092 2.4819 2.7240 Sou h Ka elia 0.5381 0.5268 0.5841 0.6060 0.6480 Sou he n Sa onia 0.7490 0.7518 0.7448 0.7722 0.8808 Eas e n Sa onia 0.7536 0.7551 0.7024 0.6833 0.7279 No h Ka elia 0.5753 0.5695 0.5475 0.5732 0.5982 No he n Sa onia 1.5263 1.5273 1.4859 1.5119 1.6492 Cen al Finland 5.5499 5.5361 5.5562 5.6234 6.0488 Sou h Bo hnia 4.7209 4.7507 4.7506 4.7505 4.7708 Vaasa 2.1997 2.2187 2.2057 2.1912 2.1572 Cen al Bo hnia 3.2251 3.2411 3.2267 3.1927 3.1619 No h Bo hnia 7.7036 7.6829 7.6786 7.6660 7.5354 Kainuu 0.1984 0.1957 0.2255 0.2468 0.2596 Länsi-Pohja 0.7437 0.7415 0.7377 0.7399 0.6897 Lapland 0.4929 0.4905 0.4979 0.5332 0.6323 Helsinki and Uusimaa 3.3256 3.3243 3.3982 3.4146 3.8419 A e age o e all a eas 2.2174 2.2176 2.2242 2.2331 2.3421 Table 3 The mean alues and 95% c edible in e als (C I) o DP unde he bina y oden model wi hou he con ex ual e ec o 20 heal h dis ic s Heal h dis ic Lowe 95% C I Mean Uppe 95% C I Sou hwes Finland 0.0129 0.4974 0.9881 Sa akun a 0.8447 0.9686 1.0000 Kan a-Häme 0.0019 0.5099 0.9985 Pi kanmaa 0.7794 0.9428 0.9996 Päijä -Häme 0.0037 0.4921 0.9951 Kymenlaakso 0.0000 0.5057 1.0000 Sou h Ka elia 0.0004 0.5031 0.9996 Sou he n Sa onia 0.0001 0.5010 1.0000 Eas e n Sa onia 0.9407 0.9929 1.0000 No h Ka elia 0.0000 0.5005 1.0000 No he n Sa onia 0.0006 0.3543 0.9965 Cen al Finland 0.9485 0.9931 1.0000 Sou h Bo hnia 0.9276 0.9901 1.0000 Vaasa 0.9621 0.9953 1.0000 Cen al Bo hnia 0.9755 0.9973 1.0000 No h Bo hnia 0.8998 0.9832 1.0000 Kainuu 0.0004 0.5060 0.9992 Länsi-Pohja 0.0000 0.5789 1.0000 Lapland 0.0013 0.5064 0.9993 Helsinki and Uusimaa 0.0347 0.4788 0.9568 Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 6 o 8 e idence o be e in o m human isk, ou p oposed me hodology demons a es i s abili y o quan i y he associa ion be ween he oden s a us and human incidence. This is po en ially use ul in p edic ion o human ou b eak when pu in he heal h-policy pe spec i e. Abb e ia ions C I: C edible in e al; DIC: De iance In o ma ion C i e ion; DP: Deg ee o posi i e indica o ; ICAR: In insic condi ional au o eg essi e model; SD: S anda d de ia ion; WAIC: Wa anabe-Akaike in o ma ion c i e ion Acknowledgemen s We would like o hank he e iewe s o commen s ha g ea ly imp o ed he manusc ip . Funding This esea ch is pa ially suppo ed by he Facul y o T opical Medicine, Mahidol Uni e si y. The unding body had no ole in he design o analysis o he s udy, in e p e a ion o esul s, o w i ing o he manusc ip . A ailabili y o da a and ma e ials The da a om he Na ional Ins i u e o Heal h and Wel a e o Finland is a ailable on public domain and upon eques . The da a om he Na u al Resou ces Ins i u e is a ailable o om he ep esen ing au ho s upon eques . Au ho s’con ibu ions Au ho s CR, AL and VJDRV designed he s udy wi h c i ical e iew om HR. CR pe o med he s a is ical analyses in consul a ion wi h AL, VJDRV, HR, JS, OH and HH. CR d a ed he pape wi h inpu om AL and VJDRV. HR, JS, OH and HH we e esponsible o c i ical e ision and imp o emen s o he manusc ip . All au ho s ead and app o ed he inal manusc ip . E hics app o al and consen o pa icipa e Su eillance da a om he Na ional In ec ious Disease Regis e is agg ega e, de-iden i ed da a o public use and he seconda y analyses a e no subjec ed o e hical app o al p ocess, acco ding o he Finnish In ec ious Disease Ac (Sec ion 42). Consen o publica ion No Applicable. Compe ing in e es s The au ho s decla e ha hey ha e no compe ing in e es s. Publishe ’sNo e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. Au ho de ails 1 Depa men o T opical Hygiene, Facul y o T opical Medicine, Mahidol Uni e si y, Ra cha hewi, Bangkok 10400, Thailand. 2 Depa men o Public Heal h Sciences, Medical Uni e si y o Sou h Ca olina, Cha les on, SC 29425, USA. 3 Depa men o Ve e ina y Biosciences, Facul y o Ve e ina y Medicine, Uni e si y o Helsinki, Helsinki, Finland. 4 Na ional Ins i u e o Heal h and Wel a e, Helsinki, Finland. 5 Na u al Resou ces Ins i u e Finland, Helsinki, Finland. 6 School o Ve e ina y Medicine, Uni e si y o Su ey, Guild o d, UK. Recei ed: 28 July 2017 Accep ed: 27 June 2018 Re e ences 1. V bo a L, e al. U ili y o algo i hms o he analysis o in eg a ed Salmonel la su eillance da a. Epidemiol In ec . 2016;144(10):2165–75. 2. Wend A, K eienb ock L, Campe A. Join use o dispa a e da a o he su eillance o Zoonoses: a easibili y s udy o a one heal h app oach in Ge many. Zoonoses Public Heal h. 2016;63(7):503–14. 3. Rossow H, e al. Risk ac o s o pneumonic and ulce oglandula ula aemia in Finland: a popula ion-based case-con ol s udy. Epidemiol In ec . 2014; 142(10):2207–16. 4. Eliasson H, e al. The 2000 ula emia ou b eak: a case-con ol s udy o isk ac o s in disease-endemic and eme gen a eas, Sweden. Eme g In ec Dis. 2002;8(9):956–60. 5. Dennis DT, e al. Tula emia as a biological weapon: medical and public heal h managemen . Jama. 2001;285(21):2763–73. 6. Rossow H, e al. Incidence and se op e alence o ula aemia in Finland, 1995 o 2013: egional epidemics wi h cyclic pa e n. Eu o Su eill. 2015;20(33). h ps://doi.o g/10.2807/1560-7917.ES2015.20.33.21209 7. Tä n ik A, Sands öm G, Sjös ed A. Epidemiological analysis o ula emia in Sweden 1931–1993. FEMS Immunol Med Mic obiol. 1996;13(3):201–4. 8. Rein jes R, e al. Tula emia ou b eak in es iga ion in Koso o: case con ol and en i onmen al s udies. Eme g In ec Dis. 2002;8(1):69–73. 9. Allue M, e al. Tula aemia ou b eak in Cas illa y León, Spain, 2007: an upda e. Eu o Su eill. 2008;13(32) 10. G unow, R., e al., Su eillance o ula aemia in Koso o*, 2001 o 2010.2012. 11. Luque-La ena JJ, e al. Tula emia ou b eaks and common ole (Mic o us a alis) i up i e popula ion dynamics in no hwes e n Spain, 1997–2014. Vec o Bo ne Zoono ic Dis. 2015;15(9):568–70. 12. Ko pela K, e al. Nonlinea e ec s o clima e on bo eal oden dynamics: mild win e s do no nega e high-ampli ude cycles. Glob Chang Biol. 2013; 19(3):697–710. 13. Sane J, e al. Regional di e ences in long- e m cycles and seasonali y o Puumala i us in ec ions, Finland, 1995–2014. Epidemiol In ec . 2016;144(13): 2883–2888. 14. Besag J. Spa ial in e ac ion and he s a is ical analysis o la ice sys ems. J R S a Soc Se B Me hodol. 1974:192–236. Fig. 4 Map o mean o DP unde he bina y oden model wi hou he con ex ual e ec o each heal h dis ic Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 7 o 8 15. Gelman A. P io dis ibu ions o a iance pa ame e s in hie a chical models (commen on a icle by B owne and D ape ). Bayesian Anal. 2006;1(3):515–34. 16. Spiegelhal e DJ, e al. Bayesian measu es o model complexi y and i . J R S a Soc Se ies B S a Me hodol. 2002;64(4):583–639. 17. Celeux G, e al. De iance in o ma ion c i e ia o missing da a models. Bayesian Anal. 2006;1(4):651–73. 18. Gelman A, e al. Bayesian da a analysis, ol. 2. USA: Chapman & Hall/CRC Boca Ra on, FL; 2014. 19. Gelman A, Hwang J, Veh a i A. Unde s anding p edic i e in o ma ion c i e ia o Bayesian models. S a Compu . 2014;24(6):997–1016. 20. Wa anabe S. A widely applicable Bayesian in o ma ion c i e ion. J Mach Lea n Res. 2013;14(Ma ):867–97. 21. B ooks SP, Gelman A. Gene al me hods o moni o ing con e gence o i e a i e simula ions. J Compu G aph S a . 1998;7(4):434–55. 22. Diggle PJ, Menezes R, Su Tl. Geos a is ical in e ence unde p e e en ial sampling. J R S a Soc: Se C: Appl S a . 2010;59(2):191–232. 23. Fai h S, e al. G ow h condi ions and en i onmen al ac o s impac ae osoliza ion bu no i ulence o F ancisella ula ensis in ec ion in mice. F on Cell In ec Mic obiol. 2012;2:126. 24. Leblebicioglu H, e al. Ou b eak o ula emia: a case–con ol s udy and en i onmen al in es iga ion in Tu key. In J In ec Dis. 2008;12(3):265–9. 25. A iza-Miguel J, e al. Molecula in es iga ion o ula emia ou b eaks, Spain, 1997–2008. Eme g In ec Dis. 2014;20(5):754. Ro ejanap ase e al. BMC Medical Resea ch Me hodology (2018) 18:72 Page 8 o 8