CClleevveellaanndd SSttaattee UUnniivveerrssiittyy EEnnggaaggeeddSScchhoollaarrsshhiipp@@CCSSUU Mathematics Faculty Publications Mathematics and Statistics Department 1-1-2016 MMooddeelliinngg CCrroossss--CCoonnttaammiinnaattiioonn DDuurriinngg PPoouullttrryy PPrroocceessssiinngg:: DDyynnaammiiccss iinn TThhee CChhiilllleerr TTaannkk Daniel Munther Cleveland State University, [email protected] Xiaodan Sun Xi'an Jiaotong University Yanni Xiao Xi'an Jiaotong University Sanyi Tang Shaanxi Normal University Helio Shimozako York University See next page for additional authors Follow this and additional works at: https://engagedscholarship.csuohio.edu/scimath_facpub Part of the Mathematics Commons HHooww ddooeess aacccceessss ttoo tthhiiss wwoorrkk bbeenneefifitt yyoouu?? LLeett uuss kknnooww!! RReeppoossiittoorryy CCiittaattiioonn Munther, Daniel; Sun, Xiaodan; Xiao, Yanni; Tang, Sanyi; Shimozako, Helio; Wu, Jianhong; Smith, Ben A.; and Fazil, Aamir, "Modeling Cross-Contamination During Poultry Processing: Dynamics in The Chiller Tank" (2016). Mathematics Faculty Publications. 206. https://engagedscholarship.csuohio.edu/scimath_facpub/206 This Article is brought to you for free and open access by the Mathematics and Statistics Department at EngagedScholarship@CSU. It has been accepted for inclusion in Mathematics Faculty Publications by an authorized administrator of EngagedScholarship@CSU. For more information, please contact [email protected]. AAuutthhoorrss Daniel Munther, Xiaodan Sun, Yanni Xiao, Sanyi Tang, Helio Shimozako, Jianhong Wu, Ben A. Smith, and Aamir Fazil This article is available at EngagedScholarship@CSU: https://engagedscholarship.csuohio.edu/scimath_facpub/206 Modeling(cid:1)cross-contamination(cid:1)during(cid:1)poultry(cid:1)processing:(cid:1)Dynamics in(cid:1)the(cid:1)chiller(cid:1)tank Daniel(cid:1)Munther,(cid:1)Xiaodan(cid:1)Sun,(cid:1)Yanni(cid:1)Xiao,(cid:1)Sanyi(cid:1)Tang,(cid:1)Helio(cid:1)Shimozako, Jianhong(cid:1)Wu,(cid:1)Ben(cid:1)A.(cid:1)Smith,(cid:1)Aamir(cid:1)Fazil 1. Introduction contradictory results. One reason for such issues is that studies wereconductedatthelaborpilotscaleunderspecificconditions Poultry contamination by bacterial pathogens such as Salmo thatleavetheirresultsdifficulttosynthesize(Bucheretal.,2012). nella,CampylobacterandEscherichiacoliO157:H7,continuestopose In this work, part one of a series of studies, we develop a aseriousthreattopublichealthbothinCanadaandontheglobal mathematicalmodeltogaininsightintothemainmechanismsof scale. According to the World Health Organization, 25% of food chlorine decay and cross contamination during the chilling pro borne outbreaks are closely associated with cross contamination cess. This approach is important because of its ability to test eventsinvolvingdeficienthygienepractices,contaminatedequip mechanistichypothesesaswellastohelpstreamlineexperiments ment,contaminationviafoodhandlers,processing,orinadequate that would other wise be expensive both financially and tempo storage(Carrasco,Morales Rueda,&Garcia Gimenoetal.,2012).As rally.Morespecifically,modelinginformedinsightscanbeusedas processinghasbeenhighlightedasapivotaljunctureinthesupply cost effectivetoolstohelpdescribethemechanismsdrivingcross chain, both for preventing and potentially promoting cross contamination, and to establish unambiguous, quantifiable links contamination, researchers have conducted numerous studies, between processing control parameters (such as chiller water attemptingtodeterminepathogenprevalenceandconcentrationat temperature, wash time, chlorine concentration, carcass towater variousprocessingstages.However,theunderlyingmechanismsof volumeratio,etc.)andpathogenprevalenceandconcentration.In crosscontaminationarestillpoorlyunderstoodand,furthermore, turn,thequantifiedconnectionsbetweencontrolparametersand many studies evaluating the efficacy of intervention strategies pathogendynamicscanprovideinvaluableinformationintermsof during processing have presented inconsistent and even testingcontrolstrategiestokeeppathogenlevelsbelowthresholds. While our focus is the chiller process of a typical modernized Canadianpoultryinspectionprogramplant(highspeed),ourmodel canbeeasilygeneralizedtochillerprocessesinotherlocales.Also, 272 D.Muntheretal./FoodControl59(2016)271 281 the modeling framework and techniques can be modified to thetank.Thesetwoassumptionsleadtothefollowingequationfor describesimilarmechanismsintheprocessofdefeathering,evis P,thetotalkgofchickencarcassesinthetankattimet(cid:2)0(min): cerationandscalding.Wedescribethebackgroundandmodeling formulation in Section 2. In Section 3 we apply our model to P0 N εdpP; (1) generic,non pathogenicE.colicontaminationofbroilercarcasses, discuss detailed parameter estimation, and perform sensitivity where analysis. Using the results of the sensitivity analysis, we discuss 8 thresholdswithinwhichcross contaminationandchlorinecontrol >>><0; t(cid:3) 1 playalesserroleaswellaswhencross contaminationmayposea ε dp more significant risk. Also, in Section 3, we compare model pre >>>:1; t> 1: dictionsforE.colilevelsonpoultryexitingthechillertankwhen dp freechlorine(FC)inputisusedat50mg/lornotatall.Theseresults aregivenintermsofUSDAbaselinevalues.Inaddition,weexamine Notethat'isthederivativewithrespecttotimeandthefunction thedynamicsofFCinactivationviatheorganicloadinchillerred εensuresthatnocarcasseswillleavethetankbeforethe“average” water,i.e.,chillerwaterthathasbeenexposedtopoultrycarcasses, washtime1/dphaselapsed. organic material and possibly pathogens. In the final section, we Asthechickensenterandmovethroughthechillertank,they suggest some quantified rules of thumb for managing cross releasehighamountsoforganicmaterial(intheformofblood,fat, contamination issues and discuss the feasibility of developing protein,etc.)intothewater.Suchmaterialisimportantbecauseit morecomplexmodelsandofsimplifyingthecomplexityofcross alterschillerwaterchemistryaswellasmicrobialcounts(Russell, contaminationmodelsforrelativelyeasyimplementation. 2012). We represent the organic material in the chiller tank at timet>0byJ(kg).Inordertorelatethistothetotalsuspended solids(concentration),weconsiderJ/TV,whereTVisthetotaltank 2. Backgroundandchillermodel volumeinml.Forsimplicity,weassumethattheamountoforganic materialcomingintothewaterisproportionaltotheincomingrate Canadahasavarietyofpoultryprocessingoperations,ranging ofchickencarcassesN(kg/min)andthisisrepresentedbyq2(0,1). from smaller traditional type processing to state of the art, high Notethatinreality,theamountoforganicmaterialshedfromin speedoperations.Inthiswork,weconsideratypicalmodernized dividual carcasses may be independent of one another. Also, we poultry processing plant (high speed), which covers most of the assume, via the flow through the tank, that the organic material Canadianslaughterproduction(basedonpersonalcommunication with CFIA officers, which we will reference from nowon as [P]). spendsonaverage1/dpminutesinthetank.Thereforewebuildthe followingequationforJ: Essentially,ourprocessingframeworkinvolvesapoultryslaughter establishment which operates under the CFIA approved Modern J0 qN εdpJ: (2) izedPoultryInspectionProgram(MPIP);seeCFIA(2014)formore information. This perspective leads to several assumptions that guideourmodelformulation.Theseinclude(1)thetypicalweight ofacarcassis2kg;(2)thetypicalprocessingspeedis180carcasses/ 2.2. Averagemicrobialloadoncarcassesandorganicmaterialin min;(3)theaveragedwelltimeofcarcassesinthechillertankis thetank 45 min; (4) red water is not recycled, rather the set up involves freshwaterintakeatthebeginningofthechillertank,withover Oneofthekeypurposesofthemodelistounderstandthedy flowattheend;(5)amaximumof50ppm(mg/l)offreechlorine namicsoftheaveragemicrobialloadonboththepoultryandthe (FC)isadded(ifany)atthebeginningofthechillertank,andmixed organic material in the chiller tank. To do so, we represent the withincomingfreshwater;and(6)duetomodelsimplificationand average microbial load (CFU/(kg ml)) on the chicken and organic alackofdata,weassumethatorganicmatterandmicrobesdonot bind/attachtothetanksurfaces. materialinthetankattimet>0,byvpandvj,respectively.Notice Our model is built aroundtwomain types of mechanisms:(i) that the units for vp and vj are (CFU/(kg ml)) since we scale the those that involve typical processing procedures for immersion averagebacterialoadperkgbythetankvolumeTV.Formodeling purposes, it is convenient to scale by the tank volume and this chillinginhighspeedpoultryprocessingfacilitiesinCanadaand(ii) scaling should not be connected with bacterial concentration bacteria transfer, bacteria inactivation, and water chemistry dy measurementstakenfromtypicalrinseproceduresusedtoquan namics during the chilling process. Refer to Table 2 for a list of tifythemicrobialloadonapreorpost chillcarcass.Forinstance, parameterscorrespondingtotype(i)and(ii).Tobeclear,thepa theUSDAconductedstudiesusinga400mlcarcassrinseinorderto rametersinvolvedwiththeparticularprocessingassumptionsand determine E. coli levels on individual poultry carcasses during dynamics, as in (i), are what specifies our model for Canadian processingandreportedtheirresultsinunitsCFU/ml(USDA,2012). poultry programs. The mechanisms under type (ii) are general Weassumethatthechickensenterthechillerprocesswithan mechanisms that are expected in a typicallarge scale immersion averagelevelofsCFU/kg.Uponenteringthetank,acertainfraction chilling procedure that is utilized during poultry processing in ofthiscontaminationlevelinitiallyshedsintothechillerwater.Let manylocales,notjustCanada.Therefore,inthissectionaswellas this fraction be r and so 0 < r < 1. Also, as the carcasses move Section 3, where we apply our model to generic E. coli contami through the chiller tank, we suppose that continued microbial nation, data used to quantify the type (ii) mechanisms need not necessarilybeCanadian. shedding occurs at a rate bvp, where b (1/min) is the shedding parameter (i.e., the shedding rate is proportional to the current Wenowformulatethechillermodelinseveralsteps. averagecontaminationlevelonthepoultry).Inaddition,bacterial attachmentoccursviacontactbetweenacarcassandmicrobialsin 2.1. Thecarcassdynamicsandtotalsuspendedsolids the chiller water. If we let W (CFU/ml) be the microbial concen trationinthechillerwaterattimet,thenweassumethisattach Weassumethattheincomingrateofchickencarcassestothe ment occurs at a rate bW, where b (1/(kg min)) is the binding tankisN(kg/min)andthechickensspendonaverage1/dp(min)in parameter. Inadditiontoshedding/binding,weconsidertheinactivationof W' kwCW, where kw (l/(mg min)) is the inactivation rate microbesoncarcasssurfacesviafreechlorine(FC)contactduring parameter.Notice,thechangeinWattimetisproportionaltoCW, thechillerprocess.WhiletheeffectivecontactofFCwithcarcass illustratingthecommonmassactionassumptionforsuchchemical surfaces(andthereforewithmicrobesattachedtocarcasssurfaces) reactions(Deborde&vonGunten,2008).Puttingtogetherthisidea during immersion chilling is, to our knowledge, not well docu with(i) (iv),theequationforWis: mented in the literature, there are multiple studies quantifying VinaancBtirviaetsieonn, 2ra0t0e7s;oZfhmanicgr,oLbueos,iZnhsooulu,tWioanngv,ia&FCM;ilsleneer(,H2e0l1b5li)nagn&d W0 ð1þqÞsTrVNþbvjJþbvpP bPW bJW kwCW g10W(cid:5)3TV; referencestherein.Ifweletkw>0betheinactivationrateofmi (5) crobesinthechillerwater,thenwearguethattheinactivationrate viaFCofmicrobesoncarcasssurfacescanbewrittenasakw,where where the first term (1 þ q)srN/TV reflects the amount of a2(0,1).Forinstance,inthefreshproduceindustry,studieshave contaminationthatinitiallyshedsfromthepoultryintothewater concludedthatsurfacecharacteristicscanreduceeffectivecontact andthelasttermgW/10(cid:5)3TV,reflectstheconcentrationofmicrobes ofchemicalsanitizersduringwashcycleprotocols(Adams,Hartley, that exit the tank with the outflow. Notice, we assume the tank & Cox, 1989; Gil, Selma, Lo(cid:2)pez Ga(cid:2)lvez, & Allende, 2009). Since volumeTVisconstantintime,soinflow outflow.Assumingthatl carcass surfaces are irregular and this is an important factor in liters of fresh water are added to the tank per carcass and each determining contamination levels (Thomas & McMeekin, 1980), carcass on averageweighs m kg, g N(kg/min)*(1 carcass/m kg) similartotheresultsfromfreshproducestudies,FCcontactwith *l(lfreshwater/carcass) Nl/m(l/min)istheadditionrateoffresh microbes attached tocarcass surfaces should be significantlyless waterpercarcass. thanFCcontactwithmicrobesinthechillerwater.Combiningthese IntermsoftheequationforC,wehave: ideas, the average decrease of the microbial load on carcasses is gtiiovneanlbtyoathkwevppCro,dwuhcetroefwtheeascsuurmreentthmatictrhoibsiadlecloreaadseanisdptrhoepoFrC C0 c110(cid:5)g3TV dcC kcWC akcvjJC akcvpPC hJJC g10(cid:5)C3TV; concentrationC(mg/L). (6) Finally,takingintoaccountthefactthat1/dpistheaveragewash time (and assuming that the natural death rate of the microbes wherec1(mg/l)istheFCconcentrationoftheinputwater,gisas attachedtothepoultryandorganicmaterialiszero(Russell,2012)) above,andsoc1g/10(cid:5)3TV(mg/(lmin))measurestherateofincrease ourequationforvpbecomes: ofFCinthewater.ThenaturaldecayrateofFCinthetankisrep resentedbydc(1/min).Also,kc(ml/(CFUmin))reflectstherateat v0p ð1 PTrVÞsNþbW bvp dpvp akwvpC: (3) wcrhobicehsicnhltohreinweaitsero,xaikdcizisedthoerradteegartadwehdicdhuFeCtiosdineagcrtaidvaetdindguemtio In a similar manner, we can construct an equation for vj as irnaatectaivtawtinhgicmhitchroeboersgoanniccarmcaastsersiuarlfaincetsh,eantdanhkJ(1d/e(ckrgeamseins))thisethFCe follows: through chemical binding. For the terms involving kc and hJ we v0j ð1 JTrÞqsNþbW bvj dpvj akwvjC: (4) atisosnuamletoththatetphreoddeuccrteoasfethineFrCescpoenctcievnetrianttieornacattintgim“espteicsiepsr”o,paonrd V therefore kc, akc and hJ are types of second order rate constants. Noticethatthecarcasssurfacetemperaturesduringtheinitial Finally,gC/10(cid:5)3TVillustratesthelossofFCduetooutflowofwater coolingphaseofchillingmaypromotemicrobialgrowthoncarcass fromthetank. surfaces.PleaserefertoSection4forreasonsastowhywedonot includethisdynamicinEquations(3)and(4)forspecificapplica 2.4. Completemodel tiontoE.colicontamination. Puttingtogetherthesixequationsabove,ourmodelbecomes: 2.3. Dynamicsinthechillerwater J0 qN εdpJ; conTcheenttrwatoiomnainintvhaericahbillelesrwweaetexra,manindeCin(cmlugd/eL)W, t,hteheFCmciocrnocbeinal v0j ð1 JTrÞqsNþbW bvj dpvj akwvjC; tration in the water. Assuming that red water is not filtered or V recycled,bacteriadonotmultiplyinthewaterbecauseofthelow P0 N εdpP; rteemguplaetriaotnusr,e,a(cid:3)n4d(cid:4)CbaasctpeerriaClansuardviaivnalFoiondItnhsepewctaiotenrAgisenecxyp(eCcFtIeAd) v0p ð1PTrÞsNþbW bvp dpvp akwvpC; V (Ratkowsky,Olley,McMeekin, & Ball,1982;Wang & Doyle,1998; srN W wCFaItAe,r2(0d1u4e),toWshdeeeprefnodrsceosninfotuhretwhiantegrs,:e(tic).)m,(iicir)ombeicsroshbeedsiinntwoattheer W0 ð1þqÞ TV þbvjJþbvpP bPW bJW kwCW g10(cid:5)3TV; abtytaFcChainngdt(oiv)pothueltflryoworroatregaonfiwcamtearteinri/aolu,t(ioiif)tmheictraonbke.sWinhaicletivthateerde C0 c110(cid:5)g3T dcC kcWC akcvjJC akcvpP hJJC g10(cid:5)C3T : V V maybeconcentrationdifferencesalongthelengthofthetank,for (7) simplicityweassumecompletemixingforthedynamicsin(i) (iv). Observethatinjuredbacteriacellsmightalsoberepresentedinthis SeeTable1foraconcisedescriptionofthemodel(7)variables. model, given the assumption that they shed from/adhere to car cassesatthesameratesasotherintact/viablecells.However,froma 2.5. Modelproperties modelingstandpointwedonotdifferentiatebetweeninjuredand viablecells. Notethatsystem(7)ispositivelyinvariantforallnon negative, Intermsofquantifyingtheabovemechanisms,forinstance,the notidenticallyzeroinitial conditions.Thisessentiallymeans that inactivation rate for microbes in water due to FC is given by foreachsetofnon negative,notidenticallyzeroinitialconditions, Table1 (7)approachSontheorderof200e250min,whichmeansduringa Modelvariables. typical8hshift,ifthereislittlevariationintheaverageE.coliload Variable Description&units oncarcasses,i.e.szconstant,themodelpredictsthatcontamina tionlevelsinthewaterandonthepoultryinthetankwillequili J Amountoforganicmaterialintankattimet(kg) vj Averagemicrobialloadonorganicmaterialintankattime brate. Practically, this gives us mathematical justification to t(CFU/(kgml)) simplify the dynamics in the tank and consider only the equilib P Amountofpoultryintankattimet(kg) riumsolutionSof(7).However,sincetheparametersinvolvedwith vp Averagemicrobialloadonpoultryintankattimet(CFU/(kgml)) E.colicontaminationarenotpreciselyknown,wewanttounder W Microbialconcentrationinchillerwaterattimet(CFU/ml) standhowsensitiveSistothemodelparameters.Thissensitivity C FCconcentrationinchillerwaterattimet(mg/l) analysisisvitalformakinginformedconclusionsforE.colicontrol asweillustrateinthefollowingsections. Note that s may vary significantly during an 8 h shift and there exists a unique solution to(7) which stays positive in each thereforedependingoncertaintimeintervals,contaminationlevels component for all positive time (i.e. we can meaningfullyascribe inthechillerwaterandonthepoultrymayvaryinsteadofequili physical units to each component of the solution). Furthermore, suchsolutionsto(7)arenotunboundedinfinitetime.Combining brating.Realistically,then,sshoulddependontime.However,in order to build control strategies and rules of thumb for treating theseideas,weseethatsystem(7)canunambiguouslydescribethe suchcases,wefirstseekresultsthatactasareferencepoint.Thatis, variableswehaveassociatedtoacontinuouspoultrychillingpro wefirstdeterminearangeforsinwhichcross contaminationplays cesswithpotentialcross contaminationdynamics. asignificantroleandgainanunderstandingofwhichparameters Inaddition,oursystemhasapositiveequilibriumstatewhich play dominant roles in contributing to cross contamination at wedenoteby equilibrium. We do this via sensitivity analysis, assuming s is (cid:2) (cid:3) S J(cid:6);v(cid:6);P(cid:6);v(cid:6);W(cid:6);C(cid:6) : constant but randomly selected fromwithin its range. Please see j p Section 3.1 for details concerning this analysis as well as Section ThisequilibriumstateSisindependentoftimeandaccordingto 3.2.2andSection4forsituationswheresmayvaryasafunctionof timeandhowthemodel(7)canbeappliedtoquantitativemicro numerical calculations (not shown) it attracts all solutions with bialriskassessment(QMRA). non negative, not identically zero initial conditions. This means thatastimeincreases,biologicallyrelevantsolutions(asdescribed above)movecloserinvaluetoS.Theequationsforeachcoordinate 3.1. ParameterbaselineandrangeestimationforE.coli ofS,intermsofmodelparameters,aregivenbelow(notethatsome contamination areimplicitlygivenforthesakeofclarity). 3.1.1. ParametersspecifictoCanadianprocessing J(cid:6) qN (8) ReferringtoTable2,specificprocessingparametervalueswere dp obtained from personal communication with CFIA officers, as referencedby[P].RefertothebeginningofSection2forthedetails (cid:6) N ofCanadianhighspeedchillingspecifications.Notethattheother P (9) dp studiesreferencedforthesesameparametersinTable2confirmthe uniformityofsomeoftheseassumptionsforimmersionchillingin v(cid:6) v(cid:6) (cid:4) ð1 rÞdps (cid:5)þ bW(cid:6) (10) otherlocationssuchastheU.S.andBrazil. p j TV bþdpþakwC(cid:6) bþdpþakwC(cid:6) 3.1.2. AverageE.coliloadonincomingcarcasses (cid:2) (cid:3) ð1þqÞsrNþb v(cid:6)J(cid:6)þv(cid:6)P(cid:6) s:Following(Northcuttetal.,2008),wecansetabaselinevalue W(cid:6) TV j p (11) fors,theaveragemicrobialloadonincomingcarcasses(CFU/kg). bðP(cid:6)þJ(cid:6)ÞþkwC(cid:6)þ10g3TV From(Northcuttetal.,2008),therinsingproceduretoquantifythe bacterialloadonpoultrypriortothechillingprocess,indicatesthat C(cid:6) (cid:4) (cid:5)(cid:6) c1g (cid:7) trhinesaavtee.rGagiveeEn.cao1li0l0evmellfroinrsine,ctohmisintrgancasrlacatesssetosirso1u0g2h.6lyCF1U04/.m6ClFinUtohne 10(cid:5)3TV dcþkcW(cid:6)þakcv(cid:6)jJ(cid:6)þakcv(cid:6)pP(cid:6)þhJJ(cid:6)þ10g3TV theaveragecarcass.Assumingtheaveragecarcassweightis2kg, s 104.6/2z2(cid:7)104CFU/kg.Forsensitivityanalysis,weestablish (12) thefollowingrangefors,103to106CFU/kgbasedonE.colidatain Cavanietal.(2010). 3. ApplicationofmodeltoE.colicross-contaminationduring 3.1.3. ShedrateofE.colifromcarcassestochillerwater b:FromNorthcuttetal.(2008),wecanestimatebbycomparing immersionchilling thepre chillbacterialoadonacarcassandthepost chillbacteria load.Followingtherinseprocedurein(Northcuttetal.,2008),the Duetothefactthatthewehavearelativelycompletedatasetfor pre chillE.coliloadrecoveredfromanaveragesinglecarcasswas generic E. coli, both bacteria levels and transfer during industrial chillerprocesses(Cavani,Schocken Iturrino,Garcia,&deOliveira, 102.6CFU/mlandthepost chillloadwas101.1CFU/ml.Byconser 2010; Northcutt, Smith, Huezo, & Ingram, 2008; Tsai, Schade, & vationoftheE.colipopulation,andconsideringa45minaverage Molyneux,1992)andchlorineinactivation(Helbling&VanBriesen, washtime,weestimatetheshedratetobe: 2007),wetailorourmodeltoaddressthespecificdynamicsasso ln(cid:4)101:1(cid:8)102:6(cid:5) ciated to the chiller water chemistry and cross contamination of b :077 1=min: 45 broilercarcassescontaminatedwithnon pathogenicE.coli.Forthe parameterrangesspecifictothissituation,thesolutionsofsystem Intermsofarangeforb,weuse0.04to0.1.Consideringthata Table2 BaselineparametersvaluesforapplicationtoE.coli.Parameterswithreference[P],correspondtoinformationobtainedfrompersonalcommunicationwithCFIAofficers. Currentlythereisnodocumenteddataforthesereferencesmarkedwith[P].Parameterslandgareextrapolatedfromguidelinesin(CFIA,2014). Type Parameter/Reference Description Values/Units (i) TV Tankvolume 5(cid:7)107ml (Cavanietal.,2010) N Carcassprocessrate 360kg/min (Cavanietal.,2010) (Northcuttetal.,2008) 1/dp Averagewashtime 45min [P],(Northcuttetal.,2008) l Freshwater/carcass 1.7l/carcass (CFIA,2014) m Averagecarcassweight 2kg [P] g inputwaterrate 306l/min (CFIA,2014) c1 inputFCconcentration 0 50mg/l (USDA,2012;CFIA,2014) (ii) dc FCdecayrate 4.1(cid:7)10 5min 1 (Li,Gu,Qi,Ukita,&Zhao,2003) s Prechillcarcassload 2(cid:7)104CFU/kg (Northcuttetal.,2008) r Initialshedfraction 0.15 estimated q Organicmaterialfraction 0.011 (Northcuttetal.,2008) (Tsaietal.,1992) b Microbialattachmentrate 0.01(kgmin) 1 (Munther&Wu,2013) (Northcuttetal.,2008) b Microbialshedrate 0.077min 1 (Northcuttetal.,2008) a FractionforFCkillrateoncarcass 0.001 0.1 estimated kw FCkillrateinwater 216l/(mgmin) (Helbling&VanBriesen,2007) kc FCdecayrateviakilling 0.0069ml/(CFUmin) (Helbling&VanBriesen,2007) hJ FCoxidationrate 0.0017(kgmin) 1 (Tsaietal.,1992) carcass undergoes an averagechilling time of 45 min, this corre Usingtherelationshipabove,weseethatkc 0.0069.Fromthe spondstoa1to2logreductiononthepoultry. rangegivenforthecontacttimeabove,wefindthatkw2[150,300], while kc ranges from about 0.0048 to 0.0096. Performing similar calculationswithinactivationofE.coliO157:H7datafrom(Zhang 3.1.4. FCinactivationkinetics etal.,2015),wefindthatkwz276andkcz.02.Hereweusethe t.t0rha3et2kioc“±na3nt.0lodo0igk9n1wa0:(”cmFtirgvCo/aFmltU)em/H1mi0nel3l.bTCliinhFnUaagct/tmaiisnvl,adoittifVoEtana.nkcBeocrsloiien0isnt.e0asnc3ot2l(u2mtt0iimo0gn7/el)i,onwifsoeFnCgheiacvmvoeeninnctuehbtnaeyt. v(rZeahsruiaelntsgtahcearttoa5sls.l,oi2tgs0110ra5rne).gdBeu,ecfctoiaorunssieimskapcclibhcaiietryveelwydeainffife0xc.2tks5cmswo.0dit0ehl691o0uatmnpdugt/dslooafnsFoiCtt includeitinthesensitivityanalysis. ThestudyinHelblingandVanBriesen(2007)indicatesthatE.coliis IntermsofinactivationofE.colioncarcasssurfaces,weassume veryreactivewithFCandthe“contacttime”iscalculatedbyinte thattherateisakwandthelossofFCduetothisinactivation,since gratingtheFCconcentrationcurveoverthetimeinterval[t0,t3],i.e., kc is proportional to kw, is akc. While a is not precisely known, thetimeintervalittakestoreducethemicrobialconcentrationby considering the discussion in Section 2.2, weassume that it is at 103. most0.1.Forthesensitivityanalysis,weassumethatarangesfrom Considering the units of kw and kc, and the fact that it takes 0.001to0.1. 0.032mg/lofFCtoeliminate103CFU/mlofE.coliinoneminute,we calculate: 3.1.5. Organicmaterialinthechillerwater kw 3:125(cid:7)104kc: q:FromTsaietal.(1992),thetotalsuspendedsolidsinthechiller water is 0.35% (i.e. about 3500 mg/l) (this value is similar to the Also, as bacteria are organic substances, we can model their initialmeasuringstationinthetank(Northcuttetal.,2008)).With inactivationbyFCusinga secondorderratereaction(Deborde& ourtankvolumegivenbyTV 5(cid:7)107ml,thistranslatestoabout von Gunten, 2008), W' kwCW. Considering this equation on 175 kg. Using the total suspended solids to estimate the organic thetimeinterval[t0,t3],wecansolveforkwasfollows: material in the tank, J(t) should equilibrate to about 175 kg. We (cid:4) (cid:5) know that the positive equilibrium value of J qN/dp without kw lnðWZ ðt3t0CÞð=sWÞdðst3ÞÞ ln:013023 215:867: filteSriinncge.TNhaisnidmdppliaersethfiaxtedqzfro1m75o/u(4r5p*r3o6c0e)ssin0g.0a1s1s.umptions,we allowqtovaryfrom0.005to0.03asthismeansthatJvariesfrom t0 about80to490kg. 3.1.6. FCoxidationrateviaorganicmaterialintank and 3). These input values are then fed into our model (7) and hJ:Wewanttoestimatetherateofchlorinereactionwiththe linked with the correspondingoutputs for W*, v(cid:6)p and C*. Using a organicmaterialinthechillerwater.Fromourmodel,weusethe samplesizeofn 1000,wecalculatethepartialrankcorrelation followingequation: coefficients(PRCCs)correspondingtoeachparameter.Briefly,the PRCCvaluesquantifythedegreeofmonotonicitybetweenrespec C0 hJJC: tive parameters and outputs. For more details concerning this analysis, please refer to (Marino, Hogue, Ray, Krischner, & Zhao, FromTsaietal.(1992),thechillerwaterisassumedtohavethe 2008). Observe that for a complete uncertainty and sensitivity total suspended solids at equilibrium, J 3500 mg/l (or 0.35%). analysis, we should also perform an extended Fourier amplitude Assuming,asabove,TV 5(cid:7)107ml,wehavethatforlargeenough sensitivity test (eFAST), however, more relevant data for certain t, JðtÞ≡J J10T6Vmlg 175 kg. Substituting this into the model, we parameterrangesisneededinordertojustifyanextensivesensi solvetoget tivityanalysis. Fig.1 illustratesthe PRCC values using thebaseline and range CðtÞ C0e(cid:5)hJJt: (13) valuesforcorrespondingparameterscomingfromTables2and3. FromFig.1(A)and(C),wefirstnoticethatW*andv(cid:6) arestrongly ReferringtothedatainTable5ofTsaietal.(1992),weseethat influenced by s (CFU/kg), the average E. coli load opn pre chilled chlorine depletion from organic material has both a “fast” and poultry. This is a logical result, as increasing the pre chiller mi “slow”kinetic.Forourpurposes,weconsideronlythefastkineticas crobial load on the poultry will in general lead to an increase in we have a continuous flow of chlorine and organic material microbialconcentrationinthechillerwateraswellasoncarcasses enteringthechilltank.FromTsaietal.(1992),theaverageofthis duringchilling.WequantifytheaverageE.coliload(duringchill fast kinetic is 0.29/min with standard deviation 0.10. Combining ing) on the poultry, v(cid:6) (CFU/(kg ml)), as the time independent thiswiththeratein(13),leadstohJJ 0.29±.10.SinceJ 175kg, expression: p our baseline value for hJ .0017 and the range is hJ2 [0.0011,0.0022]. Note, the residual chlorine data from Tsai et al. (1992)isnotthesameasFC,however,weassumethattheresid v(cid:6) (cid:4) ð1 rÞdp (cid:5)sþ bW(cid:6) : (14) ualchlorineisproportionaltotheFCandthereforethedecayrate p TV bþdpþakwC(cid:6) bþdpþakwC(cid:6) forbothtypeswillbegivenbyhJ. Equation(14)canbeunderstoodasfollows:thefirsttermcor respondstotheE.coliloadthatremainsonthepoultryduringthe 3.1.7b.: TBoinedsintigmraatteetohfeE.bcionldiitnogproautletroyfinE.tcaonlki in suspension in the fcrhaicllteiornproofctehses.iTnhcaotmisi,n(g1E.rc)dolpi/(lToVa(dboþndtphþe paokwuClt*r)y)qthuaatndtiofieessnthoet processwatertothepoultryduringchilling,weadoptthe“trans shed during chilling and is not inactivated by FC contact with missionrate”perspectivethatiscommontodiseasemodels.Ina carcasssurfaces.ThesecondtermquantifiestheE.coliloadgained diseasemodelwithawellmixedpopulation,thisrateisbasedon via cross contamination from contaminated chiller water. Recall the number of successful contacts an infective individual makes that b is the water to carcass transmission parameter,1/b is the with the susceptible population (Brauer, 2008). For the chilling characteristictimescaleforE.colisheddingfromcarcassesintothe process,thenumberofcontactsdependson(a)thepoultrytowater chillerwater(itisproportionaltothetimeittakestoshed1e2log10 ratio,(b)theaveragedwelltimeofthepoultryinthetankand(c) CFU)and1/dpistheaveragedwelltimeofacarcassinthechiller the“path”thecarcassestakethroughthetank.Wesupposethatthe tank.Noticethat1/(akwC*)isthecharacteristictimescaleforFCto tank is 9 m long (Northcutt et al., 2008) and its volume is inactivateE.colioncarcasssurfaceswhentheFChasreachedthe TV 5(cid:7)107ml.BecausetheequilibriumamountofpoultryP* N/ equilibriumamountofC*.Combiningthesethreetimescales,we dp 16200kg,thepoultrytowaterratioisP*/TV 0.324kg/l. observe that 1/(bþdpþakwC*) is the characteristic time scale of We want to know how many liter “cubes” of contaminated cross contaminationduringan8 hshiftofcontinuousprocessing. waterthis0.324kgofpoultry“hits”asittravelsthroughthetank. Inotherwords,someoftheE.coligainedfromcross contamination Assuming the 0.324 kg poultry unit travels straight through the maybeshedorinactivatedbeforethecarcassleavesthechillertank tank, and because a liter cube of water has a side dimension of andthemodelaccountsforthisbyshorteningtheeffectivecross 0.1 m,thepoultry unit“hits”about90 cubesofwater.Therefore, contaminationtimescalefrom1/dpto1/(bþdpþakwC*). 1 kg of poultry “hits” about 270 L cubes during its 45 min trip Whilev(cid:6) issensitivetomanyparameters,referringtoFig.1(C), throughthetank.Wedescribebasfollows: we see thapt c1, a, l and q play more influential roles. In terms of 270 hits chlorineefficacy,Fig.1indicatesthatc1(theinputFCconcentration) b 45 minm; hasastrongeffectonreducingtheE.coliloadoncarcassesduring kg chilling.Thiseffectiscoupledwiththeimpactofl(theinputoffresh water/carcass)asincreasinglincreasestheadditionrateofFCtothe wheremistheprobabilityofsuccessfulE.coliattachment.Currently tank(seeformula(6)).Fromanindustrystandpoint,itisimportant wehavenodataform,butweestimateittobebetween.02%and2% to note that both c1 and l can be directly controlled during the success.SeeMunther&Wu(2013)foradiscussionontheproba chillingprocess.Additionally,fromaregulatoryperspective,c1has bilityofE.coliattachmenttolettuceduringacommercialproduce anupperbound.Considering thislimitation,wewillgiveamore wash.Inthatcontext,mz1%.Puttingtheseideastogetherindicates detaileddiscussionofFCcontrolaswellasdiscusstheroleofain thatb2[0.001,0.1]. Section 3.2.3. In terms of water usage and the parameter l, an interesting study would be to use the model (7) predictions to 3.2. Resultsfromsensitivityanalysis comparethetradeoffsbetweensimultaneouslyminimizingE.coli loadsintheredwaterandoncarcasssurfacesandthecostasso InordertounderstandhowparametervariationsaffectW*,v(cid:6) ciatedtowaterconsumption. p * andC (i.e.theequilibriumE.colilevelsinthechillerwater,onthe Fig.1(C)suggeststhatanincreaseinq(thefractionoforganic poultry and the equilibrium FC level), we use Latin hypercube material the sheds from carcasses intothewater) leads to an in samplingtobuildamatrixofparameterinputvalues(seeTables2 creaseinv(cid:6).Whileqcannotbedirectlycontrolledduringchilling,as p D.Muntheretal./FoodControl59(2016)271 281 277 Table3 Parametersandrangesforsensitivityanalysis. Parameter/Reference Baseline Range s 2(cid:7)104CFU/kg 103 106 (Cavanietal.,2010;Northcuttetal.,2008) r 0.15 0.05 0.30 Estimated q 0.011 0.005 0.03 (Northcuttetal.,2008;Tsaietal.,1992) b 0.01(kgmin) 1 0.001 0.1 (Munther&Wu,2013;Northcuttetal.,2008),estimated b 0.077min 1 0.04 0.1 (Northcuttetal.,2008) a 0.05 0.001 0.1 estimated kw 216l/(mgmin) 150 300 (Helbling&VanBriesen,2007;Zhangetal.,2015) l 1.7l/carcass 1 4 (Northcuttetal.,2008;CFIA,2014) c1 25mg/l 0 50 (USDA,2012;CFIA,2014) hJ 0.0017(kgmin) 1 0.0011 0.0022 (Tsaietal.,1992) Fig.1. PRCCvaluesforparameterswithrespectiveoutput:(A)W*,(B)C*(C)v(cid:6)pand(D)bW*/(bþdpþakwC*).Parametersthataresignificant,(p<0.05)aremarkedwithastar. in the case of c1 and l, it can be indirectly regulated during pre controlled during processing, we estimate the magnitude of the chiller processing. See Section 3.2.3 for more details on the rela fraction (1 r)dp/(TV(b þ dp þ akwC*)), the cross contamination tionship between q, FC control and pre chiller processing term, bW*/(b þ dp þ akwC*), and the E. coli level in water W* as interventions. follows:foreachfixedc1in[0,50]mg/l,weperformMonteCarlo simulationstocalculatetherespectivevaluesoftheseterms.Then, fittinganormaldistributiontotherangeofrespectiveoutputs,we 3.2.1. RulesofthumbforE.colicross contamination calculatethe95%confidenceinterval. Under what conditions should we worry about cross TheresultsareillustratedinFig.2.ExaminingFig.2(A),wesee ccinootnnottaacmmomiinnpaattaiirooinnng?teItnrhmete.rmSmiansgcneoiftcuE1qd(uethaoteifoinnthp(eu1t4fi)Fr,Cstthciotsenqrcmueensattrniaodtniottnhr)eanccasrlnoatsbeses rthanatgethseov9e5r%acpopnrfiodxeimncaeteilnyte[1rv0a(cid:5)l10fo.4r,1(01(cid:5)8.4]r)adsp/c(1TVv(abriþesdfproþma[k0w,5C0*)]). Similarly, from Fig. 2(B) and (C), the cross contamination term lesser role in ensuring that cross contamination will not be an ranges over [10(cid:5)6.75,10(cid:5)3.3] and W* ranges over [10(cid:5)4.4,10(cid:5)3], issue.Thatis,usingmaximumFCinput,comparedtozeroFCinput, respectively. These results provide quantifiable evidence that FC reducestherangeofs,withinwhichcross contaminationdictates playsasignificantroleinreducingtheE.coliloadonpoultrybothby themagnitudeofv(cid:6)p,onlybyabout1log10CFU. directlyinactivatingthebacteriaoncarcasssurfacesandbyinac tivating the bacteria in the chiller water which reduces the load 3.2.2. Cross contaminationandflocktoflocktransmission gainedviacross contamination. OuranalysisalsoindicatesthatW*issensitivetob,r,andb,and However,the questionremains,when shouldweworryabout bW*/(bþdpþakwC*)issensitivetorandb(seeFig.1(A)and(D)). cross contamination?UsingtheresultsfromFig.2(A)and(B),we ThisinformationcoupledwithourdiscussioninSection3.2.1points cancomparetheexpressionsforv(cid:6)pwhenc1 0(noFCinput)and to potential cross contamination issues as described in the whenc1 50(maximumFCinput).First,forc1 0,wehavethat following archetypal situation: Suppose chickens are processed fromavarietyoffarmsatoneparticularprocessingcenterbutfarm v(cid:6)z10(cid:5)8:4sþ10(cid:5)3:3 (15) (A), at some juncture, delivers chickens that carry a significantly p higherE.coliloadascomparedwiththechickensfromtheother * Next,forc1 50,weseethat farms.Itislikelythen,thattheE.coliloadinthechillerwater,W, will dramatically increase via the shed from farm (A) chickens * v(cid:6)z10(cid:5)10:4sþ10(cid:5)6:75 (16) duringchilling.IfthemagnitudeofW ishigherorcomparableto p themagnitudeofsfromcarcasses nowentering thechillertank, From Equation (15), if s (cid:3) 105.1, the magnitude of the cross and the chiller water has yet to be replaced, then cross contamination term plays a dominant role in determining the contamination (flockto flock) may be significant. To obtain rules overallorderofv(cid:6).Thatis,duringatypical8 hshift,ifnoFCisused, ofthumbforsuchscenarios,thatarebackedbyscientificrigorat p cross contaminationhasaprimaryeffectindeterminingtheE.coli the mechanistic scale, our model suggests the need for specific levelonpost chillerpoultrywhentheaverageE.coliloadonpre experimentstocapturethecomponentsofshedding(randb)and chillercarcassesisontheorderof5log10CFUorless. cross contamination (b) more precisely. Refer to Section 4 for a Following the same reasoning, Equation (16) indicates that if moredetaileddiscussion. s (cid:3) 103.7, then the cross contamination dynamic is again signifi cant.Inotherwords,duringatypical8 hshift,ifmaximumFCinput 3.2.3. FCcontrolandinactiviation isused,cross contaminationplaysaleadingroleindeterminingthe While our findings in Section 3.2.1 show that cross E.colilevelonpost chillerpoultrywhentheaverageE.coliloadon contaminationcanstronglyinfluencetheresultingE.coliloadon pre chillercarcassesisontheorderof4log10CFUorless.Therefore, chilled poultry, both in the presence or absence of FC input, the whileFCinputsignificantlyreducesE.colilevelsinthewaterandon results in Figs.1 and 2 demonstrate that FC input is pivotal as a poultryduringchilling,fromamanagementperspective,itplaysa control.ToquantifythiscontrolonE.colilevels,weagainconsider (A) (B) (C) (cid:237)8 (cid:237)3 (cid:237)2.8 (cid:237)3 (cid:237)3.5 (cid:237)8.5 (cid:237)3.2 )) (cid:237)4 ) *C ) w *)C (cid:237)3.4 k (cid:68) w (cid:237)4.5 +p (cid:237)9 k(cid:68) +d +p *W) (cid:237)3.6 b d (cid:237)5 d/(T(vp (cid:237)9.5 */(b+W log( (cid:237)3.8 )(cid:85) (cid:69) (cid:237)5.5 (cid:237) ( (cid:237)4 1 g ( o ( l og (cid:237)6 l (cid:237)4.2 (cid:237)10 (cid:237)6.5 (cid:237)4.4 (cid:237)10.5 (cid:237)7 (cid:237)4.6 0 50 0 50 0 50 c c c 1 1 1 Fig.2. 95%confidenceintervalfor(A)FractionofunshedE.colilevelonpoultryvsFCinputc1,(B)Levelofcross-contaminationonpoultryvsFCinputc1,and(C)E.colilevelinred waterW*vsFCinputc1.
Description: