learning and mapping lyme disease patient trajectories ... · 12/23/2017 · comorbidities and...
TRANSCRIPT
LearningandMappingLymeDiseasePatientTrajectoriesfromElectronicMedicalDataforStratificationofDiseaseRiskandTherapeuticResponse
OsamuIchikawa1,2*,BenjaminS.Glicksberg1,3*,BrianKidd1,LiLi1#,JoelT.Dudley1#
1. DepartmentofGeneticsandGenomicSciences,InstituteforNextGenerationHealthcare,IcahnSchoolofMedicineatMountSinai,770LexingtonAve.,NewYork,NY10065.
2. DrugResearchDivision,SumitomoDainipponPharma.Co.Ltd.,3-1-98Kasugade-naka,Konohana-ku,Osaka,554-0022,Japan.
3. InstituteforComputationalHealthSciences,UniversityofCalifornia,55016thSt,SanFrancisco,California,94158.
*Theseauthorscontributedequallytothiswork
#Correspondenceandrequestsformaterialsshouldbeaddressto
Dr.LiLi([email protected])orDr.JoelDudley([email protected])
InstituteforNextGenerationHealthcare
IcahnSchoolofMedicineatMountSinai
770LexingtonAve.
15thFloor
NewYork,NY10065,USA
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
2
ABSTRACT(250words)
Background
Lymedisease(LD)isanepidemic,tick-borneillnesswithapproximately329,000incidencesdiagnosedeachyearinUnitedStates.Long-termuseofantibioticsisassociatedwithseriouscomplications,includingpost-treatmentLymediseasesyndrome(PTLDS).ThelandscapeofcomorbiditiesandhealthtrajectoriesassociatedwithLDandassociatedtreatmentsisnotfullyunderstood.Consequently,thereisanurgentneedtoimproveclinicalmanagementofLDbasedonamorepreciseunderstandingofdiseaseandpatientstratification.
Methods
Weusedaprecisionmedicinemachine-learningapproachbasedonhigh-dimensionalelectronicmedicalrecords(EMRs)tocharacterizetheheterogeneouscomorbiditiesinaLDpopulationanddevelopsystematicpredictivemodelsforidentifyingmedicationsthatinfluencetheriskofsubsequentcomorbidities.
Findings
Weidentified3,16,and17comorbiditiesatbroaddiseasecategoriesassociatedwithLDwithin2,5,and10yearsofdiagnosis,respectively.AthigherresolutionofICD-9levels,wepinpointedspecificco-morbiddiseasesonatimescalethatmatchedthesymptomsassociatedwithPTLDS.Weidentified7,30,and35medicationsthatinfluencedtherisksofthereportedcomorbiditieswithin2,5,and10years,respectively.Thesemedicationsincludedsixpreviouslyassociatedwiththeidentifiedcomorbiditiesand29newfindings.Forinstance,thefirst-lineantibioticdoxycyclineexhibitedaconsistentlyprotectiveeffectfortypicalsymptomsofLD,including‘backacheNotOtherwiseSpecified(NOS)’and‘chronicrhinitis’,butconsistentlyincreasedtheriskof‘cataractNOS’,‘tearfilminsufficiencyNOS’,and‘nocturia’.
Interpretation
OurapproachandfindingssuggestnewhypothesesforprecisionmedicinetreatmentsregimensanddrugrepurposingopportunitiestailoredtothephenotypicprofilesofLDpatients.
Funding
TheSteven&AlexandraCohenFoundation
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
3
INTRODUCTION Lymedisease(LD)isavector-borne,infectiousdiseasecausedbythebacteriumBorreliaburgdorferithatistransmittedtohumansthroughtickbites.AccordingtotheUSCentersforDiseaseControlandPrevention(CDC),around329,000LDcasesoccurannuallyanditbecomesamajorUSpublichealthproblemthatcausessubstantialuseofhealthcareresources.LDismostprevalentintheNortheastandupperMidwest,and95%ofallconfirmedcasesin2015werereportedin14states1.ThesymptomologyofLDisheterogeneous,althoughsomegeneralpatternshaveemerged.ThefirstmanifestationofLDisoftenanexpandingannularlesion,callederythemamigrans,nearthebitelocation,butthissignispresentinonly70–80%ofpatients2.Thelengthoftimefortherashtooccur,alongwiththecharacteristicsoftherash(e.g.,compositionandsize)canalsovary3.Otherclinicalfeaturesthatoftenarise,singlyorincombination,includefever,pain,arthritis,myopericarditis,neurologicalsymptoms(e.g.,facialnervepalsy),andsatelliterashes.Oneexplanationforthisvariabilityisthatthegenotypeofthetickitselfmightaffectaspectsofpathogenesis,suchastheprobabilityofhematogenousdissemination4,5.TheneurologicalmanifestationsinLD,reportedin3–12%ofpatients,areofgreatestconcern6.Thesephenomena,collectivelycalledneuroborreliosis,areoftenassociatedwithintensepainthatcanmanifesteithersoonafterinfectionormuchlater,frommonthstoyearsafterward. AccurateandprecisediagnosesofLDpresentseveralchallenges.Typically,laboratorytestingofLDfollowsidentificationofcutaneousmanifestationsfromvisualinspectionbutthesemanifestationsarenotalwayspresent.Currentguidelinesrecommendserologictesting,atwo-phaseprocessconsistingofanenzymeimmunoassaywithin30daysofsymptomonset,followedbyWesternblotafter30daysfromsymptomoccurs7,8iftheearlytestisactive.Eventogether,thisdiagnosticstrategyhaspoorsensitivity,particularlyduringtheacutephase,withfalse-negativeratesofupto50%9.Otherlaboratorymethodsarespecificforparticularmanifestations,e.g.,testingofCSFforcentralnervoussysteminvolvement.RecentworkhasshownthatincorporationofdatafromvariouswearabledevicescandetectearlysignsofLDandassociatedinflammatoryresponses.Forexample,variationsinperipheralcapillaryoxygensaturation(SpO2),amarkerassociatedwithphysiologicalmacro-phenotypessuchasfatigue,canbemeasuredbyportablebiosensorstofacilitatemoreaccurateandrapidLDdiagnosisfromvariationsinthesemeasurementsandcouldbeeconomicallyfeasibleforwidespreaduseinthefuture.Currently,however,cliniciansstillhavetorelyontraditionalmeasurestodiagnosepatients10.Furthermore,comorbidconditionscaninterferewithbothdiagnosisandtreatment.Forinstance,otherinfectionscanbeconcurrentlytransmittedwithLD11,makingdifferentialdiagnosisevenmoredifficultandsometimesrequiringspecialized,alternativetreatmentstrategies.ManystudieshaveattemptedtodevelopmethodsfordifferentiatingLDfromothersimilarsyndromes,e.g.,septicarthritisvs.LDofthekneeinchildren12. Followingsuccessfuldiagnosis,LDismostcommonlytreatedwithantibioticssuchasdoxycycline,amoxicillin,cefuroximeaxetil,andceftriaxone.Althoughthesemedicationshave
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
4
highcurerates(~90%)13,theyareassociatedwithseriouscomplicationsandadverseevents,especiallyunderprolongeduse14,153,16.Notablyinthisregard,onestudyshowedthatcertainfirst-linetreatments,specificallyintravenousceftriaxonefollowedbydoxycyclineforchronicLDpatientswerenoteffectivecomparedtoplaceboforcingdiscontinuationofthetrial17.AnotherstudyreportedthatrepeatedIVceftriaxonetreatmentforLymeencephalopathyresultedinonlyminorcognitiveimprovements,withhighratesofrelapseofcognitivesymptoms18.ThesefindingssuggestthatunknownfactorsareresponsibleforthehighvariabilityoftreatmentoutcomesforpatientswithLD.Additionally,upto20%oftreatedpatientsdeveloppost-treatmentLymediseasesyndrome(PTLDS),inwhichlingeringsymptomssuchasfatigue,pain,orjointandmuscleacheslastformonthsorevenyears.Thecausesandfrequenciesofthesesymptomsremainunclear,andtheissueisfurtherconfoundedbythepresenceofconcurrentdiseases. SeveralrecentstudiesrevealtheuncharacterizedcomplexityofdiseasecourseandtreatmentresponseinLD.Onestudyfoundcertainfirst-linetreatments,specificallyintravenousceftriaxonefollowedbydoxyclycline,forchronicLDpatientswerenoteffectivecomparedtoplaceboandwereforcedtodiscontinuetheirstudy17.AnotherstudyassessedtheeffectsofrepeatedIVceftriaxonetreatmentforLymeencephalopathyandfoundonlyminorcognitiveimprovementswithhighratesofrelapseincognitionissues18.Therefore,itseemsthatthereareotherfactorsatplaythatmayexplainthehighvariabilityoftreatmentoutcomesforpatientswithLD.Itisdifficulttodisentangletowhatextentgiventreatmentresponsesanddiseasesequelaeareduetodifferencesinindividualimmuneresponses,patientcharacteristics,diseaseburden,andtreatmenttiming,ortothemedicationsthemselves.Indeed,itisverylikelythatresponseandoutcomedependonacomplexinterplaybetweenthesefactors,makingclinicians'jobsextremelydifficult.Toaddressthediversesymptomology,imperfectdiagnosticstrategies,andvariabletreatmentoutcomesofLD,comprehensivestudydesignsarerequired.Forexample,aninvestigationofriskfactorsforLDinfection,suchasbehavioralandenvironmentalriskfactors,revealedthatLD–positiveserologyissignificantlyassociatedwithclinicalanddemographicfeaturessuchaspreviousself-reportedLDdiagnosisandage,behavioralfactorssuchaswearingprotectiveclothing,andgeographic/environmentalfactorssuchasshrubedgedensityinpropertylocation19.Anotherstudyevaluatedtheriskstoindividualsbasedongeographicalfeaturessuchasthesuitabilityofthelocalhabitatforticks20. Althoughtheaforementionedstudieshaveprovidedagreatdealofusefulinformation,thevariabilityinglobalriskprofilesforLDpathogenesisremainsincompletelyunderstood,andthereisanunmetneedforpersonalizedtreatmentrecommendationsthattakeintoaccountindividualcharacteristicssuchasdemographicsanddiseaseburden.ElectronicMedicalRecords(EMRs)fromhospitalscontainawealthoflongitudinal,patient-leveldataencompassingpriorhistoryofprescriptionsanddiseasediagnoses,alongwithclinicaloutcomes,thatcanbeexploitedtoinvestigatetheseissuesinadata-drivenfashion.TodatetherehasbeennosystematicanalysisofLDusingEMRdata,particularlyfromahospitalwithinahigh-riskstate.In
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
5
thisstudy,weleverageanEMRdatasetrepresentingoverfivemillionuniquepatientsofdiverseracialandethnicbackgroundscollectedfromalargeacademicmedicalcenterinNewYorkCity.Althoughnotitselflocatedadjacenttoawoodedarea,MSHcaterstopatientsfromalloverthestate.NewYorkisoneoftheaforementioned14statesreportingthevastmajorityofLDcases,andin2017hadanincidencerateof16.4per100,000individuals(CDC),oneofthehighestinthecountry. WehypothesizedthatEMRdatafromMSHcouldprovidearichframeworkforstudyingtheheterogeneityofLymemanifestation,aswellasthequalityandefficacyoftreatment.Usingvariousstate-of-the-artstatisticalandmachinelearningmethods,wesoughttoidentifypatternsofclinicaloutcomesinordertohelpphysiciansdeveloptreatmentstrategiestailoredtopatients'diseaseprofiles.Specifically,weinvestigatedhowdemographicandclinicalfactorsaffectLDmanifestationsandclinicaloutcomesinthecontextofvarioustreatments.OurmaingoalsweretoidentifycomorbiditiesassociatedwithLDanddevelopasystematicpredictivemodelforidentifyingmedicationsthatinfluencetheriskoftheseconditions.Asanalternativetoone-size-fits-allstrategiesfortreatingLD,ourmethodologyfacilitatesdirectingtreatmentrecommendationsandidentifyingpossiblerepurposingopportunitiestailoredtothephenotypicprofilesofLymepatients.Ourstudyisthefirstdata-drivenefforttoprioritizemedicationsforLDbasedonanindividual’sphenotypeprofile.WeidentifiedLyme-associatedcomorbiditiesatthelevelofbroaddiseasecategories,pinpointedspecificco-morbiddiseasesassociatedwithLDovertime,andusedmachinelearningtopredictmedicationsthatinfluencetherisksofthesecomorbidities. WeexpectthatthenovelframeworkandfindingsfromthisstudycansupportcurrentandfutureeffortstodeveloppersonalizedtreatmentstrategiesforpatientswithLD,includingeffortstoprovidephysicianswithabroaderevidentiaryfoundationonwhichtobasetheirtreatmentrecommendations(e.g.,selectionofantibiotics)basedonindividualpatients'diseaseriskprofiles.Additionally,morepreciseknowledgeofpredictedadverseeventswouldfacilitateimprovementsinmonitoringandmanagementstrategies.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
6
METHODSPatientpopulationandstandardizationofclinicalterminologyPatientcohortWeutilizedElectronicMedicalRecords(EMRs)fromtheMountSinaiDataWarehouse(MSDW),thelargestcomprehensiveEMRsysteminNewYorkCity,whichincludesdatafromaraciallyandethnicallydiversepatientbase.Since2000,morethan4.5millionuniquepatientrecordshavebeendocumentedinthissystem.DiseasediagnosesareencodedasInternationalClassificationofDiseases,9thRevision(ICD-9)billingcodes,whichhavebeenusedextensivelyinEMR-relatedanalyses21.Inthisstudy,weretrievedrecordsfromallpatientsdiagnosedwithLymedisease(LD)withtheICD-9code088.81(n=2,134).Werestrictedthedatatorecordsoccurringbetween2000and2015,allowingforupto15-yearfollow-up.InaccordancewithHealthInsurancePortabilityandAccountability(HIPAA)andProtectedHealthInformation(PHI)guidelines,theagesofthesepatientswerecensoredat18and90.Finally,weonlykeptdatafrompatientswithdefinedage,self-reportedsex,andself-reportedrace/ethnicity(referredtoas“race”inthismanuscript).Forthetotalof1,767Lymepatients,therewere930females(52.6%)and837males(47.4%),withanaverageageof47.8±19.7.Theracialbreakdownofthecohortisasfollows:1,201Caucasian(70.0%),49African-American(2.8%),34Hispanic/Latino(1.9%),and483Others(27.3%).Forthesepatients,wealsoretrievedallotheravailableclinicalvariablesfromEMR,includingprescriptionsandotherdiseasediagnoses.Inaddition,weretrievedIgM/IgGlabmeasurementspertinenttoLD.IgMorIgGWesternblotlabswereavailablefor28patientsafteroratthetimeofdiagnosis(45dayswindowofdiagnosis).Ofthose,89%patients(25/28)werereportedaseitherIgMorIgGWesternblot–positive,confirmingtruepositivityforLyme.Theremaining11%ofpatients(3/28)werereportedasIgM-negative,butallwerepositiveforp23,anantigenspecifictoLyme.Intotal,wecompiled3,936diseasesdiagnosisand5,723prescriptions.Weprovideaschematicofourstudydesign,approach,andpatientselectioncriteriainFigure1.ClinicalsourcesandtermstandardizationWecategorizeddiseasesusingtheClinicalClassificationsSoftware(CCS)forICD-9diagnosiscodes,developedbyAHRQ22,whichaggregatesandcharacterizesmorethan14,000ICD-9codesintobroadercoherentdiseasecategories.ThisstrategyhelpstoavoidsamplesizelimitationasaresultofusingICD-9codesalone.Forcategorization,weusedthe‘Single-LevelDiagnosis’(CCS-single)level,whichhasatotalof283differentcategories.WestandardizedmedicationdatabymappingtotheRxNormontology23.Specifically,wemappedthesetermstoingredientcodes,yielding793normalizedmedications.Toensuretherobustnessofouranalyses,werequiredasamplesizeof>20patientsforcalculationsofthesignificanceofdiseasedirectionalityanddisease–medicationassociation.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
7
MachinelearningmethodsandanalysisDiseasepairtemporaldirectionalityForallpatientswithLD,wefirstassesseddisease-pairconnectivitypatternsforcomorbiddiseases.Specifically,wedeterminedwhetherthemembersofeachpairexhibitedasignificantpatternintheirtemporalorder,e.g.,whetheroneprecededtheothermoreoftenthanexpectedbychance.WeperformedacumulativebinomialprobabilitytesttoassessthetemporalorderingoftheassociationsbetweenLymeandallotherdiseases,assuminga50%probabilityofeithertooccurbeforetheother.Weperformedthefollowinganalysisonbothbroadandnarrowdiseasecategories.Atthebroaderlevel,weanalyzedrepresentativeCCS-single-levelcategoriesbecausethisstrategycouldenhancesignalsthatmightbelostduetosmallsamplesizeattheICD-9level.Second,weperformedtheanalysisusingstandardICD-9codesinordertodetectassociationsatahigherresolutionforcertaincodesthatmaybemoreprevalent.Becausethesecomorbidconditionscanbeeitherchronicoracute,weperformedseveraliterationsofthisanalysisoverdifferenttimewindows,specifically2,5,and10years.Foreachtimewindow,werestrictedcollectionofinformationforthecomorbiddiseasesinbothtemporaldirections,relativetothedateoffirstLymediagnosis.Forthe2-yearwindow,forexample,weonlycollecteddiseasedataforeachpatient2yearsbeforeand2yearsafterthedateofLymediagnosis.FortheCCS-single-andICD-9–levelanalyses,weperformed275and3,639testsforeachwindow,respectively.Last,todeterminewhetherdiseasepairswithsignificanttemporaldirectionalitywerealsosignificantlycomorbid,weperformedalogisticregressionforeachpaircontrollingforage,sex,andself-reportedrace.Theoutcomevariableinthismodelwasthediseasethatwasshowntooccuraftertheotherinthetemporalanalysis(significantinthebinomialassessment).Definitionofoutcomesandcovariatesinthemachine-learningmodelTodiscoverriskfactorsornewtherapeuticoptionsforLDsequelae,wefocusedonthenewonsetofdiseasecomorbiditiesmorethan7daysafterthediagnosisofLD.Ofthe1,767LDpatientsintheoverallcohort,wesystematicallyassessedthecomorbiditiesandmedicationassociationsfor1,196patientswhowerefollowedupformorethan7daysandhadatleastoneprescriptionrecordinMSH’sEMRsystem.Likeourdisease-pairtemporaldirectionalityanalysis,wesettimewindowsof2,5,and10years.Foreachpatient,wecollectedthediseasesdiagnosedwithin2,5,or10yearsaftertheirfirstLymediagnosisdate.Wealsoretrievedmedicationsprescribedwithin1yearpriortoand2,5,or10yearsafterthefirstLymediagnosis.OutcomecomorbiditiesweredefinedbyICD-9codeandcategorizedusingCCS-single-levelDiagnosisterms.Featureselection
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
8
Weconsideredmanydiseasevariables,codedbyCCS-single-levelcategories,andmedicationvariables,whichweremappedtoRxNormingredientcodes.Accordingly,weadoptedafeatureselectionmethod,penalizedlogisticregressionwiththeadaptiveLASSO(Eq.1),toidentifyvariablesofthehighestrelevancethatassociatedwithensuingcomorbiditiesfollowingLDdiagnosis.TheadaptiveLASSOisanextensionofthetraditionalLASSO24thatusescoefficient-specificweights25.TheadaptiveLASSOestimatorselectsthezerocoefficientsofthetrueparametersareestimatedaszerowithprobabilitytendingtoone,whichiscalledsparsityproperty.Andthenon-zerocomponentsareestimatedasifthetruesparsemodelwereknownaprior,whichisasymptoticallynormal26.Letℒ!(𝛽;𝑌,𝑋)bethenegativelog-likelihoodparametrizedbyβforasampleofsizen.TheadaptiveLASSOestimatorisdefinedas:𝛽 = 𝑎𝑟𝑔𝑚𝑖𝑛! {ℒ! 𝛽;𝑌,𝑋 + 𝜆! 𝜔! 𝛽!!
!!! } (1)where𝜔! = |𝛽!|!!isacoefficientspecificweightsvector,and𝜆!isaregularizationparameter.Wesetthepositiveconstantγas1accordingtoZouetal.25,andobtained𝛽bythemaximumlikelihoodestimateofRidgeregression.The𝜆!valueforminimumAUCwaschosenby10-foldcrossvalidation.WeusedtheRpackageglmnet27forthesepenalizedregressions.LogisticregressionmodelWeusedoddsratio(OR)fromlogisticregression(Eq.2)toassesstheriskoffuturecomorbidityprogressiononeachmedicationtaken(i.e.eitherincreasedriskorprotectiveeffect).WeanalyzedthepairsofoutcomediseasecomorbidityandthemedicationsthatwereselectedbytheadaptiveLASSO.Inthismodel,weadjustedforage,sex,self-reportedrace,andthefollow-uptimeframe.
𝑙𝑜𝑔 !!!!
= 𝛽! + 𝛽!𝑚𝑒𝑑𝑖𝑐𝑎𝑡𝑖𝑜𝑛 + 𝛽!𝑎𝑔𝑒 + 𝛽!𝑔𝑒𝑛𝑑𝑒𝑟 + 𝛽!𝑟𝑎𝑐𝑒 + 𝛽!𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑟𝑒𝑑 𝑝𝑒𝑟𝑖𝑜𝑑(2)
wherePistheprobabilityofadisease,medicationisabinaryvariable,ageisacontinuousparameter, genderisabinaryvariable(Female/Male);raceisacategoricalvariable(Caucasian,AfricanAmerican,Hispanic/Latino,orOther),andobservedperiodisacontinuousparameter.𝛽coefficientsforeachcovariaterepresenttheeffectsizewhencontrollingforallothers.PropensityscorematchingTocontrolforpotentialconfoundingfactorsduetoimbalancesofclinicalcharacteristics,notlimitedtoageandgender,weanalyzedthetemporaleffectsofmedicationsafterthepropensityscorematchingtoselectanappropriatecontrolcohortforthetargetedcasecohort28.Thus,wecreatedcomparablecohorts,consistingofgroupstreatedoruntreatedwitha
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
9
targetedmedication,basedonasetofcovariatesatthebaselinetimepoint,i.e.,timezeroforeachpatient.Thebaselinetimepointwasdefinedasthefirstprescriptiondayofthetargetedmedicationor7daysafterLDdiagnosis,whicheverwaslater,becauseweobserveddiseasecomorbiditiesformorethan7daysafterLDdiagnosis.Thepropensityscoresoftargetedprescriptionswerepredictedbyalogisticregressionmodel,includingothersignificantmedicationsanddiseaseconfoundersselectedbytheadaptiveLASSOwitha10-yeartimewindow,withpatientdemographicsascovariates.Eachpatientprescribedagivenmedicationwasmatchedtoacorrespondingcomparisonpatient(1:1ratio)bynearest-neighbormatching.Forinstance,weanalyzedassociationbetweendoxycyclineand‘backacheNotOtherwiseSpecified(NOS)'(ICD-9code:724.5),‘chronicrhinitis'(472.0),‘tearfilminsufficiency(insuffic)NOS(375.15)’,and‘cataractNOS(366.9)’,andbetweenamoxicillinand‘acuteupperrespiratoryinfection(URI)NOS(465.9)’.Atotalof328,330,358,370,and115subjectswereselectedforeachmedication-comorbiditypairinthepropensityscore-matchedtreated/untreatedgroup.TheRpackageMatchIt29wasusedforpropensityscorematching.SurvivalanalysisWegeneratedsurvivalcurvesbytheKaplan–Meiermethodandexamineddifferencesinsurvivalamongsubgroupsbythelog-ranktest,withpropensityscorematchingofcasesandcontrols.WecalculatedhazardratiosusingCoxproportionalhazardsmodels:ℎ 𝑡 = ℎ! 𝑡 exp (𝛽!𝑚𝑒𝑑𝑖𝑐𝑎𝑡𝑖𝑜𝑛)(3)whereℎ 𝑡 istheexpectedhazardattimet,ℎ! 𝑡 isthebaselinehazard,andmedicationisabinaryvariable.WeverifiedtheproportionalhazardsassumptionbyconfirmingthatSchoenfeldresidualsareindependentoftime(Schoenfeldtestp>0.1).WeusedtheRpackagessurvivalandsurvminerforthesurvivalanalysis.RoleofthefundingsourceThefunderofthestudyhadnoroleinstudydesign,datacollection,dataanalysis,datainterpretation,orwritingofthemanuscript.Thecorrespondingauthorshadfullaccesstoallthedatainthisstudyandhadfinalresponsibilityforthedecisiontosubmitforpublication.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
10
RESULTSIdentifiedcomorbiditiesassociatedwithLymedisease,groupedasbroaderdiseasecategoriesWeassessedthetemporalorderingoftheassociationsbetweenLymedisease(LD)andotherdiseases;specifically,wesoughttodeterminewhetheragivencomorbiditytendedtooccurbeforeorafterdiagnosisofLD.Werestrictedouranalysistodiseaseswithreporteddatesandonlyincludedthefirstreportedencounterofadiagnosis,resultingin41,713disease-diseasepairs,andanalyzedtheirtemporalorderingatthepatientlevel.Specifically,foreachcomorbiddiseasepair(i.e.LDandanotherdiseasecategory),wetabulatedthenumberofpatientswithbothdiseasesandassessedwhichdiseaseinthepairoccurredfirst,oriftheyoccurredatthesametime,basedonthevisitdates.Overall,outofthe275Lyme-comorbiditycombinationsforalltimewindows,21werenominallysignificant,with5diseasesoccurringpriortoLDand16occurringafter(p<0.1;Table1a).Forthe2-yearwindow,weidentifiedthreediseasecategoriessignificantlyassociatedwithLD,withtwopriorandoneafter;forthe5-yearwindow,16categories,withfourpriorand12after;andforthe10-yearwindow,17categorieswithfivepriorand12after(Table1a).WereconfirmedsomepreviouslyreportedLymecomorbidities,including‘nutritionaldeficiencies’[p=0.069,probability(prob)=0.54at5years;p=0.035,prob=0.55at10years]30,31,‘visiondefect’(p=0.099,prob=0.57at5years;p=0.076,prob=0.57at10years)32 33,and‘disorderoflipidmetabolism’(p=0.036,prob=0.54at5years;p=0.076,prob=0.57at10years) 34.Additionally,weidentifiedseveraldiseasecomorbiditiesnotpreviouslyreported,including‘cataract’(p=0.0037,prob=0.65at5years;p=0.033,prob=0.64at10years),‘acutebronchitis’(p=0.094,prob=0.62at2years;p=0.022,prob=0.64at5years;p=0.034,prob=0.63at10years),and‘nonmalignantbreastconditions’(p=0.057,prob=0.58at5years;p=0.066,prob=0.58at10years).AcompletelistofdiseasecategoriesisshowninTable1a.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
11
Time
(year) Direction Disease(CCS)
Lyme
First
Disease
First
Same
Time Pvalue Prob.
2 - Coronaryatherosclerosisandotherheartdisease 18 58 17 1.10E-02 0.62
2 - Administrative/socialadmission 41 66 9 8.17E-02 0.57
2 + Acutebronchitis 23 11 3 9.39E-02 0.62
5 - HIVinfection 2 21 9 5.51E-02 0.66
5 - Coronaryatherosclerosisandotherheartdisease 38 80 17 1.92E-02 0.59
5 - Administrative/socialadmission 55 88 9 3.09E-02 0.58
5 - Disordersoflipidmetabolism 117 269 111 3.63E-02 0.54
5 + Openwoundsofextremities 24 6 0 7.15E-04 0.8
5 + Openwoundsofhead;neck;andtrunk 19 9 0 4.36E-02 0.68
5 + Fractureoflowerlimb 16 8 0 7.58E-02 0.67
5 + Cataract 57 21 10 3.67E-03 0.65
5 + Fractureofupperlimb 18 9 1 9.25E-02 0.64
5 + Acutebronchitis 36 17 3 2.20E-02 0.64
5 + Analandrectalconditions 26 11 5 8.21E-02 0.62
5 + Nonmalignantbreastconditions 60 36 7 5.72E-02 0.58
5 + Othereyedisorders 75 41 14 4.76E-02 0.58
5 + Neoplasmsofunspecifiednatureoruncertainbehavior 60 39 6 8.58E-02 0.57
5 + Blindnessandvisiondefects 58 33 11 9.89E-02 0.57
5 + Nutritionaldeficiencies 219 141 47 6.85E-02 0.54
10 - HIVinfection 2 25 9 1.44E-02 0.69
10 - Coronaryatherosclerosisandotherheartdisease 44 91 17 9.18E-03 0.6
10 - Administrative/socialadmission 61 97 9 2.20E-02 0.58
10 - Disordersoflipidmetabolism 121 301 111 1.59E-03 0.56
10 - Essentialhypertension 96 271 124 1.20E-02 0.55
10 + Openwoundsofextremities 28 7 0 2.54E-04 0.8
10 + Maintenancechemotherapy;radiotherapy 14 5 1 5.77E-02 0.7
10 + Cataract 64 26 10 3.32E-03 0.64
10 + Occlusionorstenosisofprecerebralarteries 22 10 3 8.77E-02 0.63
10 + Acutebronchitis 37 19 3 3.37E-02 0.63
10 + Poisoningbyothermedicationsanddrugs 24 7 8 9.98E-02 0.62
10 + Analandrectalconditions 29 14 5 9.67E-02 0.6
10 + Othereyedisorders 87 43 14 7.69E-03 0.6
10 + Nonmalignantbreastconditions 65 41 7 6.60E-02 0.58
10 + Blindnessandvisiondefects 63 36 11 7.62E-02 0.57
10 +
Inflammation;infectionofeye(exceptthatcausedbytuberculosis
orsexuallytransmitteddisease) 78 39 20 6.19E-02 0.57
10 + Nutritionaldeficiencies 228 143 47 3.51E-02 0.55
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
12
Time
(year)
Direction
Disease(CCS)
ICD9
Disease(IC
D9)
Lyme
First
Disease
First
Same
Time
pvalue
(binom
ial)
Prob
.(binom
ial)
pvalue
(logistic
regressio
n)
OR(lo
gistic
regressio
n)
5-
HIVinfection
V08
HIVpo
sitiveNOS
021
8
1.21
E-02
0.72
1.80
E-05
2.17
5-
Administrativ
e/social
admiss
ion
V20.2
Routinchildhealth
exam
16
49
43.18
E-04
0.71
1.31
E-02
2.08
5
+Nutritiona
ldeficiencies
268.9
VitaminDdeficiencyNOS
203
124
36
1.37
E-02
0.56
1.20
E-17
75.64
5
+Ca
taract
366.9
CataractNOS
40
18
74.08
E-02
0.62
2.90
E-30
4.27
5
+Ac
utebron
chitis
466.0
Acutebron
chitis
36
16
31.50
E-02
0.65
4.02
E-21
3.56
5+
Non
maligna
ntbreast
cond
ition
s79
3.80
Unspe
cifie
dab
norm
al
mam
mog
ram
28
12
08.29
E-03
0.70
8.92
E-10
2.61
10
-
HIVinfection
V08
HIVpo
sitiveNOS
024
8
3.50
E-03
0.75
1.80
E-05
2.17
10
-Disorderso
flipid
metab
olism
27
2.0
Purehypercholesterolem
84
159
50
8.04
E-02
0.54
1.60
E-30
2.22
10
-Disorderso
flipid
metab
olism
27
2.4
Hype
rlipide
miaNEC
/NOS
97
187
64
9.01
E-02
0.54
1.06
E-18
1.78
10
-
Essentialhypertension
401.9
Hype
rten
sionNOS
96
247
120
8.16
E-02
0.53
9.83
E-08
1.39
10
-Ad
ministrativ
e/social
admiss
ion
V20.2
Routinchildhealth
exam
17
54
48.82
E-05
0.72
1.31
E-02
2.08
10
+
Nutritiona
ldeficiencies
268.9
VitaminDdeficiencyNOS
211
124
36
4.67
E-03
0.57
1.20
E-17
75.64
10
+
Cataract
366.16
Senilenuclearcataract
17
21
1.29
E-03
0.85
2.94
E-03
1.97
10
+
Cataract
366.9
CataractNOS
45
21
73.02
E-02
0.62
2.90
E-30
4.27
10
+
Infla
mmation;infectionof
eye(excep
ttha
tcau
sedby
tube
rculosisorse
xually
tran
smitted
disease)
373.00
Blep
haritisNOS
33
21
1.14
E-07
0.92
2.27
E-14
3.79
10
+
Othereyediso
rders
375.15
Tearfilm
insufficNOS
24
11
13.26
E-02
0.67
4.44
E-12
3.36
10
+
Acutebron
chitis
466.0
Acutebron
chitis
37
17
31.66
E-02
0.65
4.02
E-21
3.56
10
+Non
maligna
ntbreast
cond
ition
s79
3.80
Unspe
cifie
dab
norm
al
mam
mog
ram
31
12
02.70
E-03
0.72
8.92
E-10
2.61
Table1(a;top
)Dise
asesassociatedwith
Lym
e,analyzedasCCS-single-levelcategories(pvalue<0.1).A
totalof2
75dise
ases
weretested
.(b;bottom)D
iseasesassociatedwith
Lym
e,byICD-9category.A
totalof3
,639
dise
asesweretested
.Thistab
le
includ
esonlytheICD-9diseasesth
atareclassified
intoth
eCC
S-sin
gle-levelcategoriessho
wnin(a
),an
dafulllisto
fassociatio
ns
isprovided
inSup
plem
entaryTable1.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
13
HighlightedspecificknownornoveldiseasesassociatedwithLD,analyzedathigherresolutionAlthoughtheCCS-single-levelcategorieswerehelpfulinidentifyingdiseasegroupsofrelevancefromabroaderperspective,wealsoperformedthesameanalysisatahighergranularity(Table1bandS.Table1).Tothisend,usingtheICD-9codes,wesoughttodeterminewhichspecificdiseasesdrovethesignalandwhetherthesignalstillpersisted.Atotalof3,639Lyme–comorbiditycombinationswereanalyzedusingtheICD-9codes.Atthe2-yearwindow,fivepairswerenominallysignificant(p<0.1duetotherelativelysmallsamplesize,N≥20patientsforLymefirstonset),withfourpriortoLDandoneafter(S.Table1).Forthe5-yearwindow,wefound53significantassociations,with49priortoLDand4afterwards.Forthe10-yearwindow,wefound75significantassociations,with67priortoLDdiagnosisand8after.ThesignificanceofalldiseasecategoriessignificantlyassociatedwithLDthatweidentifiedinthepreviousanalysispersisted,includingthefourdiseasesthatsignificantlyoccurredpriortoLD:‘purehypercholesterolemia'(p=0.080,prob=0.54at10years),‘hyperlipidemiaNEC/NOS’(p=0.090,prob=0.54at10years),‘hypertensionNOS'(p=0.082,prob=0.53at10years),and‘coronaryatherosclerosis(athero)NOS'(p=0.022,prob=0.61at5years;p=0.075,prob=0.62at10years).Ninesequelaediseases,namely‘vitaminDdeficiencyNOS'(p=0.014,prob=0.56at5years;p=0.0047,prob=0.57at10years),‘cataractNOS'(p=0.041,prob=0.62at5years;p=0.030,prob=0.62at10years),‘senilenuclearcataract'(p=0.0013,prob=0.85at10years),‘tearfilminsufficiency(insuffic)NOS'(p=0.033,prob=0.67at10years),‘acutebronchitis'(p=0.015,prob=0.65at5years;p=0.017,prob=0.65at10years),‘blepharitisNOS'(p=1.1E-7,prob=0.92at10years),‘unspecifiedabnormalmammogram'(p=0.0083,prob=0.70at5years;p=0.0027,prob=0.72at10years),‘HIVpositiveNOS'(p=0.012,prob=0.72at5years;p=0.0035,prob=0.75at10years)’and‘routinechildhealthexam'(p=3.2E-4,prob=0.71at5years;p=8.8E-5,prob=0.72at10years)drovethesignalfromthebroaddiseasecategories(Table1b).Additionally,wereportedLDcomorbiditiesifothersignificantICD-9codesassociatedwithLDwithalistofcomorbiditiesarenotwidelyknown,including‘insomniaNotElsewhereClassifiable(NEC)'(p=0.042,prob=0.59at5years;p=0.045,prob=0.58at10years),‘obstructivesleepapnea'(p=0.085,prob=0.60at5years;p=0.044,prob=0.62at10years),‘cervicalgia'(p=0.073,prob=0.59at5years;p=0.041,prob=0.60at10years),and‘dysuria'(p=0.096,prob=0.59at5years;p=0.034,prob=0.62at10years). WeconfirmedthatthelargemajorityofthecomorbiditypairsweresignificantlyassociatedwithLDwithconcordantdirectionalitybyadjustingage,gender,andracebylogisticregression(p<0.1).WeprovideacompletelistofICD-9leveldiseaseassociationsthatpassedoursignificancethresholdinbothanalysesSupplementalTable1.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
14
MedicationspredictedtomodulateriskofsubsequentcomorbiditiesinLDpatients,analyzedasbroaderdiseasecategoriesToinvestigatehowvariousprescribedmedicationsinfluencedtheriskofsubsequentdiseasepathogenesis,wefocusedoncomorbiditieswithonsetafterthefirstdiagnosisofLD.UsingtheadaptiveLASSOmethodologyandalogisticregressionmodel,weinvestigatedallmedicationsprescribedtoLDpatientspriortothecomorbidities.Wefound3,12,and18medicationsassociatedwithdiseasecomorbidities,classifiedbyCCS-single-levelcategories,within2,5,and10yearsafterLymediagnosis,respectively(S.Table2,Figure2a,Figure2b).Fourmedication–Lymecomorbidityassociationsweresupportedbypublishedstudies35-41,andweconfirmedthatthesemedicationsmodulatedtherisksofLymecomorbidities,includingfluticasone–‘cataract’(adjustedOR=1.94,p=0.072at5years;adjustedOR=2.01,p=0.033at10years)hydrochlorothiazide–‘neoplasmsofunspecifiednatureoruncertainbehavior’(adjustedOR=2.23,p=0.031at5years;adjustedOR=2.48,p=0.0092at10years),metformin–‘nutritionaldeficiencies’(adjustedOR=2.05,p=0.097at10years),andesomeprazole–‘nutritionaldeficiencies’(adjustedOR=1.75,p=0.093at10years).Fiveantibiotics,doxycycline,azithromycin,levofloxacin,clavulanate,andmupirocin,andoneantiviraldrug,valacyclovir,werepredictedtomodulatetheriskofsubsequentcomorbidities.Doxycycline,afirst-lineantibioticthatwasthemostprescribedantibioticinourEMRforpatientswithLD(39%,N=553),wasassociatedwithanelevatedriskofeyedisorders,including‘cataract’(adjustedOR=2.05,p=0.092at2years;adjustedOR=1.70,p=0.067at10years),‘blindnessandvisiondisorders’(adjustedOR=2.05,p=0.016atthe5years;adjustedOR=1.95,p=0.019at10years),and‘othereyedisorders’(adjustedOR=1.81,p=0.024)(Figure2a).Inregardto‘nutritionaldeficiencies',elevenmedicationswerepredictedtoberiskfactorsandthreetobeprotective.Amongthe11riskfactormedicationsweretwoantibiotics,levofloxacin(adjustedOR=2.26,p=0.0093at5years;adjustOR=2.77,p=7.0E-4at10years)andclavulanate(adjustedOR=1.64,p=0.094at10years),andoneantiviralprophylactic,valacyclovir(adjustedOR=2.56,p=0.014at5years;adjustedOR=2.58,p=0.011at10years).Additionally,twopainrelievers,diclofenacandhydrocodone,andoneanti-allergymedication,azelastine,werealsoriskfactorsfor‘nutritionaldeficiencies'(respectively:OR=3.08,p=0.0015at5years/adjustedOR=3.02,p=0.0014at10years;adjustedOR=1.90,p=0.034at10years;andadjustedOR=3.44,p=0.0037/adjustedOR=2.95,p=0.0094).Interestingly,wecouldidentifynewtherapeuticoptionsfortheLDadjunctivetreatment.Threemedications,propofol,docusate,andheparin,consistentlyexhibitedaprotectiveeffectagainst‘nutritionaldeficiencies’at5and10yearsafterLD(respectively:adjustedOR=0.32,p=6.6E-4at5years/adjustedOR=0.43,p=0.0044at10years;adjustedOR=0.37,p=0.0039at5years/adjustedOR=0.46,p=0.014at10years;andadjustedOR=0.44,p=0.024at5years/adjustedOR=0.41,p=0.014at10years)(Figure2).Inaddition,acetaminophenexhibitedaprotectiveeffectattheearlystage(2yearspost-Lyme)withadjustedOR=0.44,p=0.0069(S.Table2).
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
15
MedicationspredictedtomodulateriskofsubsequentcomorbiditiesinLDpatients,analyzedattheICD-9levelInthehigher-resolutionanalysisusingICD-9codes,weidentified7,22,and31medicationsthatweresignificantlyassociatedwiththediseasecomorbiditiesat2,5,and10yearspost-Lyme(S.Table3,Figure3aand3b).Amongthesewerepreviouslyreportedriskassociations42-45:forinstance,steroidprednisonewasariskfor‘paininlimb’withadjustedOR=2.16,p=0.030at5years/adjustedOR=2.49,p=0.0063at10years,andciprofloxacinwasariskfor‘jointpain-shoulder(shlder)’withadjustedOR=4.42,p=8.9E-5at5years/adjustedOR=4.39,p=3.1E-5at10years.Inaddition,fiveofthesideeffectsforfourmedicationswerereportedintheSIDERdatabase46,47.Twosteroids,fluticasoneandmometasone,andonepainreliever,hydrocodone,wereassociatedwithincreasedriskfor‘acuteupperrespiratoryinfection(URI)NOS’incomparisonwiththeplacebogroup(Stable3)andwererediscoveredinourstudy(respectively:adjustedOR=2.92,p=5.2E-5at5years/adjustedOR=3.44,p=4.7E-7at10years;adjustedOR=2.86,p=0.0028at10years;adjustedOR=4.01,p=4.1E-5at5years/adjustedOR=4.45,p=1.8E-6at10years).Wealsoreconfirmedtheriskassociationsbetweenfluticasoneand‘chronicrhinitis’(adjustedOR=4.70,p=1.4E-4at2years/adjustedOR=4.78,p=6.5E-7at5years/adjustedOR=4.86,p=4.4E-8at10years)anddiclofenacand‘paininlimb’(adjustedOR=3.43,p=0.0011at10years). DoxycyclineexhibitedaconsistentlyprotectiveeffectagainsttypicalsymptomsofLD,including‘backacheNOS’(adjustedOR=0.44,p=0.018at5years/adjustedOR=0.50,p=0.035at10years)and‘chronicrhinitis’(adjustedOR=0.48,p=0.036at5years/adjustedOR=0.48,p=0.024at10years)(Figure3a,3b).Furthermore,sevenantibiotics,doxycycline,amoxicillin,azithromycin,ciprofloxacin,levofloxacin,mupirocin,andsulfamethoxazole,andoneantiviraldrug,valacyclovir,modulatedtheriskofsubsequentcomorbidities.Doxycyclineconsistentlyincreasedtheriskof‘cataractNOS’(adjustedOR=2.57,p=0.053at2years/adjustedOR=1.89,p=0.058at10years),‘tearfilminsufficNOS’(adjustedOR=2.64,p=0.042at5years/adjustedOR=2.37,p=0.050at10years),and‘nocturia’(adjustedOR=3.46,p=0.010at2years)(Figure3b).Amoxicillin,anotherantibioticrecommendedforLD,increasedtheriskof‘acuteURINOS’(adjustedOR=3.01,p=6.5E-4at2years/adjustedOR=2.41,p=8.4E-4at5years/adjustedOR=2.60,p=1.3E-4at10years).Furthermore,azithromycinwasassociatedwithanincreasedriskofurinary-relateddiseasessuchas‘nocturia’(adjustedOR=4.62,p=3.5E-5at5years/adjustedOR=4.90,p=1.3E-5at10years)and‘urinary(urin)tractinfectionNOS’(adjustedOR=2.18,p=0.0095at5years),aswellasrespiratorydiseasessuchas‘acuteURINOS’(adjustedOR=8.63,p=3.9E-12at2years/adjustedOR=7.55,p=1.0E-16at5years/adjustedOR=7.2,p=1.4E-17at10years),‘acutebronchitis’(adjustedOR=5.15,p=5.9E-5at5years/adjustedOR=4.75,p=1.1E-4at10years),and‘chronicrhinitis’(adjustedOR=2.99,p=7.2E-4at5years;adjustedOR=3.07,p=1.5E-4at10years).Ciprofloxacin,afluoroquinoloneantibiotic,increasedriskof‘nocturia’(adjustedOR=7.19,p=5.9E-6at5years/adjustedOR=6.92,p=7.0E-6at10years)and‘jointpain-shlder’(adjustedOR=4.42,p=8.9E-5at5years/adjustedOR=6.92and
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
16
p=7.0E-6at10years).Mupirocin,anantibioticusedtotreatskininfection,increasedriskofskindisorders,including‘solardermatitisNEC’(adjustedOR=12.93,p=6.9E-10at5years/adjustedOR=13.96,p=3.0E-11at10years)and‘scar&fibrosisofskin’(adjustedOR=11.52,p=6.9E-9at5years/adjustedOR=12.46,p=1.0E-9at10years)(Figure3a,3b).‘VitaminDdeficiencyNOS’,commoninpatientswithpersistentLD48,isaspecificformofnutritionaldeficiency,acomorbidityidentifiedearlieratthebroader(CCS-single)level(Figure2a,2b).Severalmedicationsincreasedtheriskofthiscondition,threeat5yearspost-Lymeandfiveat10years.Theseincludedtwoanti-infectivedrugs,levofloxacin(adjustedOR=2.68,p=0.0012at10years)andvalacyclovir(adjustedOR=1.95,p=0.087at5years/adjustedOR=2.01,p=0.065at10years),azelastine(adjustedOR=3.37,p=0.0035at5years/adjustedOR=2.92,p=0.0085at10years),diclofenac(adjustedOR=2.75,p=0.0032at5years/adjustedOR=2.75,p=0.0026at10years),andsimvastatin(adjustedOR=1.77,p=0.083at10years).Ontheotherhand,fourmedicationsprotectedagainst‘vitaminDdeficiencyNOS’:docusate(adjustedOR=0.33,p=0.0015at10years),propofol(adjustedOR=0.37,p=0.0015at10years),fentanyl(adjustedOR=0.46,p=0.012at5years),andacetaminophen(adjustedOR=0.39,p=0.0025at2years)(Figure3a,3b).Respiratorydisease(Figure3b)isacomplicationfrequentlyreportedafterLD49.Weidentified11medicationsthatincreasedriskfortheseconditionsandtwothatexhibitedprotectiveeffects.InadditiontothethreemedicationsreportedinSIDERdatabaseandamoxicillinandazithromycinabove,themedicationsthatconferredincreasedriskfor‘acuteURINOS’wereanantibiotic,levofloxacin(adjustedOR=3.18,p=9.1E-4at10years),asteroid,methylprednisolone(adjustedOR=2.14,p=0.027at5years/adjustedOR=2.31,p=0.0077at10years),cyclobenzaprine(adjustedOR=3.21,p=0.0025at5years/adjustedOR=2.91,p=0.0037at10years),homatropine(adjustedOR=7.16,p=7.3E-7at10years),atorvastatin(adjustedOR=2.67,p=0.0017at10years),andaspirin(adjustedOR=1.80,p=0.073at10years).Bycontrast,acetaminophen(adjustedOR=0.55,p=0.025at5years/adjustedOR=0.62,p=0.050at10years),andoxycodone(adjustedOR=0.49/p=0.021at10years)wereassociatedwithprotectiveeffectsagainstthisdisease.MedicationsthatmodulateLDpathophysiologyondifferenttimescalesWeidentified16medicationsassociatedwithdiseasecomorbiditieswithin5yearspost-LD,81%(13/16)ofwhichoverlappedwiththoseidentifiedasassociated10yearspost-LymebytheCCS-single-levelcategorization.Specifically,fiveoutofsixanti-infectivedrugs,doxycycline,azithromycin,levofloxacin,mupirocin,andvalacyclovir,appearedinbothtimeframes.Moreover,22medicationswereassociatedwiththeICD-9–leveldiseasecomorbiditieswithin5yearsafterLyme,95%(21/22)ofwhichwerealsoidentifiedinthe10-yearpost-Lymeanalysis.Amongthose21medications,fiveareantibiotics(doxycycline,amoxicillin,azithromycin,ciprofloxacin,mupirocin)andoneisanantiviraldrug(valacyclovir).Thefourmedications
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
17
associatedwithcomorbiditiesexclusivelyinthe5yearspost-Lyme,clavulanate,gabapentin,midazolam,andfentanyl,mayimpactrelativelyearlyLymecomorbidities(S.Figure1).Atotalof17medicationsoverlappedbetweentheCCS-singleandICD-9levelsineitherthe5-yearor10-yeartimewindows.Fiveofthemwereanti-infectivedrugs,namelydoxycycline,azithromycin,levofloxacin,mupirocin,andvalacyclovir.Inthe5-yeartimewindow,16medicationswereassociatedwithcomorbiditiesclassifiedbyCCS-single-levelcategory,ofwhich50%(8/16)werealsoidentifiedattheICD-9level.At10yearspost-Lymediagnosis,weidentified21significantassociationsbetweenmedicationsandcomorbidities,ofwhich81%(17/21)wereconsistentwiththoseidentifiedattheICD-9level.Survivalanalysisoffirst-linemedicationsinpropensity-matchedpopulationsBythecross-sectionalanalysisdescribedabove,wedemonstratedthatcertainmedicationsincreasedriskorprotectedagainstdiseasecomorbiditiesinpatientswithLD.Athigherresolution(i.e.,usingICD-9codes)with10-yearfollowup,wefoundthatdoxycycline,themostcommonlyusedantibioticfortreatmentofLD13,protectedagainst‘backacheNOS’and‘chronicrhinitis’,butincreasedriskof‘tearfilminsufficNOS’and‘cataractNOS’.Anothercommonlyusedantibiotic,amoxicillin,wasassociatedwithelevatedriskof‘acuteURINOS’. Priortopropensityscorematching,weidentifiedsignificantdifferencesinthedistributionsofdemographicandclinicalcharacteristicsbetweenthedoxycycline/amoxicillin-treatedanduntreatedgroupsbefore.Thedoxycycline-treatedgroupwassignificantlyolderthantheuntreatedgroup(P<0.007),whereastheamoxicillin-treatedgroupwassignificantlyyoungerthantheuntreatedgroup(P=8.7E-4).Inaddition,doxycyclinewasprescribedmorefrequentlytomalethanfemalepatients(P<0.03).Thedoxycycline/amoxicillin-treatedgroupshadhigherprevalencesofcertainpre-existingcomorbiditiesandahigherprescriptionrateofparticularmedicationsthantheuntreatedgroups(S.Table4).Moreover,boththedoxycycline/amoxicillintreatedgroupshadhigherpropensityscoresthanthecorrespondinguntreatedgroups(P<0.001).Toclarifythelongitudinaleffectsofdoxycyclineandamoxicillin,weanalyzedtheseassociationsbypropensity-score-matchedsurvivalanalyses.Afterpropensityscorematching,thecontrolcohortswerewellbalancedwiththetreatedgroupsintermsofobservedcovariates(S.Table4).Thisanalysisrevealedthattheriskof‘backacheNOS’(Figure4a)and‘chronicrhinitis’(Figure4b)wassignificantlylowerinthedoxycycline-treatedcohortthanintheuntreatedcohort(HR=0.42,p=0.020;HR=0.49,p=0.040,respectively;Table3).Furthermore,Kaplan-Meiercurvesdemonstratedthatthecumulativeprobabilitiesofremainingfreefrom‘cataractsNOS’and‘tearfilminsufficNOS’wereloweramongdoxycycline-treatedpatients(p=0.0672and0.0608,respectively;Figures4cand4d).Coxregressionanalysissupportedastatisticallysignificantassociationbetweendoxycyclineusageandincreasedriskofboth‘cataractNOS’and‘tearfilminsufficNOS’(HR=1.90,p=0.072;HR=2.65,p=0.071).Ontheotherhand,patientsprescribedamoxicillinhadsignificantlyhigherhazardratiosfor‘acuteURINOS’(HR=2.26,p=0.0091;Figure
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
18
4e).Therefore,theeffectsofdoxycyclineandamoxicillinrevealedbythecross-sectionalanalysiswereconfirmedbysurvivalanalysesusingthepropensityscore-matchedcohort(Table3).
Medication Disease(ICD9) ICD9Pvalue(LogRank)
HazardRatio(90%CI) Pvalue(Cox)
DoxycyclineTearfilminsufficNOS 375.15 6.08E-02 2.65(1.09-6.45) 7.13E-02
Doxycycline CataractNOS 366.9 6.72E-02 1.9(1.06-3.42) 7.18E-02Doxycycline Chronicrhinitis 472.0 3.60E-02 0.49(0.28-0.87) 3.99E-02Doxycycline BackacheNOS 724.5 1.67E-02 0.42(0.23-0.78) 2.03E-02Amoxicillin AcuteURINOS 465.9 7.41E-03 2.26(1.35-3.78) 9.13E-03
Table2Survivalanalysesoffirst-linetherapeuticsforLymediseaseusingapropensity-score-matchedcohort.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
19
DISCUSSIONProperdiagnosis,treatment,andmanagementofLymeDisease(LD)aredifficultforavarietyofreasons.Inparticular,thecomplexinterplaybetweenvarioustreatmentsandcurrentclinicalstatus,includingdiseaseburden,canleadtoawiderangeofsequelae.Thisstudyrepresentsthefirstdata-drivenefforttoidentifyclinicalfactorsthataffecttreatmentofLDpatientsusinglarge-scaleEMRdata.EMRsystemscontaininformationpertainingtopatients'healthovertime,includingtreatmentsadministeredandclinicaloutcomes.Thus,oneadvantageofusingthesedataincomparisonwithprospectiveclinicaltrialstudiesistheavailabilityoflongitudinaldataspanningmorethan10years,increasingthelikelihoodofcapturinglong-termeffects.Weutilizedstatisticalandmachinelearninganalysestoidentifyassociationsbetweenmedications,includingfirst-linetreatments,andcomorbiditypathogeneses.Incontrasttoone-size-fits-allstrategies,ourapproachmayfacilitatethepersonalizationoftreatmentregimensbasedontheclinicalprofilesofaffectedindividuals.Thisstrategictransitionisessentialinlightofthetremendousvariabilityinefficacyofantibioticsandtheadverseeventsassociatedwiththesetreatments. WefirstidentifiedallsignificantcomorbiditiesofpatientswithLDbeforeandaftertheirLymeinfections.Next,weappliedmachinelearningmodelstoassesstheeffectofmedicationtreatmentontheriskofdevelopingsubsequentconditions.Ouranalysesidentifiedknownassociationsbetweenmedicationsandspecificdiseasecomorbidityoutcomes,andalsodiscoveredconnectionsbetweendrugsandLDthatcouldfacilitateprecisionmedicineaimedattailoringtreatmentstoaffectedindividuals'Lymesymptomprofile. Ouranalysisidentifiedco-morbidconditionsthatweretypicallypresentbeforeLDinfection.Althoughthisdiseaseclearlyrequirescontactwiththebacteria,certainphysiologicalpropertiesmakeindividualsmoreorlesssusceptibletoinfection.Forexample,wefoundthatindividualscategorizedashaving‘disordersoflipidmetabolism’weremorelikelyhaveLDinfectioninthefuture.Borreliaburgdorferirequirescholesterolforgrowth;researchershavefoundthatapolipoproteinE(apoE)-deficientandlow-densitylipoproteinreceptor(LDLR)-deficientmice,whichhavehighlevelsamountsofserumcholesterol,aremoresusceptiblethanwild-typemicetopathogenesisinducedbythisbacterium34.Additionally,patientswithhypercholesterolemiacouldincreasesusceptibilitytotriggertickbiteforthisvectorbornediseaseduetobodyheat,CO2,andmoisturewhicharekeyattractantssimilarlytomosquitoes50 51,butthisisbeyondthescopeofthisanalysisandneedstobefurtherinvestigated.Additionally,weidentifiedanassociationbetweenHIVinfectionandLD.Specifically,ourdatashowedthatpatientswithHIValmostexclusivelydevelopedLDsubsequenttotheHIVdiagnosis,suggestingthatimmunesystemalterationincreasestheriskofLD.AhandfulofcasereportshaveindicatedthatHIV-positiveimmunocompromisedpatientsdevelopmoresevereLymecomplicationsfollowinginfection52-54.Althoughthespecificimmunologicalmechanisms
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
20
drivingthisconnectionremainunclear,itseemsreasonabletospeculatethatimmunosuppressionplaysanimportantrole55.WealsoidentifiedconditionsthataremorelikelytobepresentafteraLDdiagnosisthanbeforehand,consistentwiththepossibilitythatthesediseasesaresideeffectsorcomplicationsarisingfromLymeinfection.Manyoftheseassociationsarewelldocumented,enhancingourconfidenceinourresults.Specifically,wefoundthatthecategories‘nutritionaldeficiencies’,‘cataract’,‘acutebronchitis’,‘othereyedisorders’,and‘Inflammation;infectionofeye(exceptthatcausedbytuberculosisorsexuallytransmitteddisease)’aremorelikelytooccurafterLymeinfectionthanbeforehandorbychance. Theresultsofthisanalysisfeedintoourdrug-comorbidityassociationsnetworkandcanbeusedtoinformtreatmentregimens.Theassociationsidentifiedforfirst-lineantibiotictreatmentsofLDhavethemoststraightforwardpotentialapplication.TheseagentsactbykillingB.burgdorferi,andthuspreventthedevelopmentofmanycomplicationsassociatedwithprolongedexposuretothebacteria.However,whileantibioticsarethemosteffectivefirst-linetreatmentsforLD,theirefficaciesarenonethelesslimited;moreover,somesymptomsmaypersistnotwithstandingtheuseofantibiotics,andlong-termexposuretotheseagentsrisksadditionalcomplications.Indeed,itispossiblethatevenacuteuse(usuallyamonth)ofthesetreatmentsisassociatedwithlong-termcomplicationsofLDthatyettobedetermined.Evenwhentreated,upto20%ofpatientsdevelopPost-TreatmentLymeDiseaseSyndrome(PTLDS),inwhichsymptomsincludingfatigueormusclepainlastformonthsoryears.AlthoughtheetiologyofPTLDSisnotyetknown,bettertailoringoftreatmentstrategiestoanindividual’sphenotypicprofilecouldpreventormodulatetheriskofdevelopingthesesymptoms.WeidentifiedanumberofcomorbiditiesmatchingthesymptomsalignedwithPTLDS,includingchronicpain,chronicrhinitis(Figures3aand3b).Notablyinthisregard,wealsofoundthatusageofsteroidmedicationsincreasestheriskofmanysymptomscommontoPTLDS.Consistentwiththis,corticosteroiduseisassociatedwithpooroutcomesforLDpatients43.Inparticular,wefoundthatprednisonewasassociatedwithelevatedriskfor‘backacheNOS’,‘paininlimb’and‘otherabnormalglucose’,definedattheICD-9level.Useoffluticasonewasassociatedwithelevatedriskfor‘chronicrhinitis’,‘postnasaldrip’,‘cervicalgia’,and‘acuteURINOS’;mometasonewasalsoassociatedwithelevatedriskfor‘acuteURINOS’;andmethylprednisolonewaslinkedtoincreasedriskof‘chronicrhinitis’,‘backacheNOS’,and‘acuteURINOS’aswell.Steroids,whichsuppresspatients’immunesystems,mightbeparticularlyharmfultoLDpatients,allowingthebacteriatogrow,ratherthanattackingtheinfections.ThesefindingssuggestthatsteroiduseshouldbelimitedinLDpatients,andthatpatientsexposedtothesedrugsshouldbemonitoredcarefullyforcomplications. Otherfindingsfromourdrug-comorbiditynetworkmightfacilitatepersonalizationoftreatmentregimens,withmorefavorableclinicaloutcomesforpatients.Forexample,severalanti-infectiousdrugs,painrelieverssuchasdiclofenacandhydrocodone,andtheanti-allergymedicationazelastinewerealsoassociatedwithhigherratesof‘nutritionaldeficiencies’,
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
21
suggestingthatphysiciansshouldconsiderrecommendingvitaminsupplementsforpatientsreceivingthesetreatments. Doxycyclineisalreadyassociatedwitharangeofsideeffects,includingpain,increasedpressureinsidetheskull56,andgastrointestinalinjury57.Thenuancesoftheseassociationsarenotwellunderstood.Ontheotherhand,wealsofoundthatdoxycyclineusewasassociatedwithlowerriskof‘backacheNOS’andlowerratesof‘chronicrhinitis’;thelatteristobeexpected,asitisacommonsymptomofchronicLD11.Thismedicationincreasedtheriskofmanyeye-relatedissues,suchas‘cataractNOS’,withthe2-and10-yeartimewindows,suggestingthatitmightexertbothshort-termandlong-termsideeffects.Wealsofoundassociationsbetweenthismedicationandelevatedriskfor‘tearfilminsufficNOS’and‘blindnessandvisiondisorders’. Thedevelopmentofrespiratory-relatedcomplicationsisamajorconcernforpatientsinfectedwithLD.Inonecasestudy,secondaryadultrespiratorydistresssyndromecausedthedeathofapatientaffectedwithLDduringthecourseofher2-monthtreatment58.Thepatientdidnotrespondtoconventionaltreatments,includingantibiotics,salicylates,andsteroids.Inourstudy,weidentifiedmedicationsthatareassociatedwithelevatedriskofrespiratory-relateddiseases.Specifically,wefoundthattheantibioticsamoxicillin,levofloxacin,andazithromycinallconferredincreasedriskof‘acuteURINOS’.Inadditiontotheknownrisksoftwosteroids,prednisolone,andmometasone,reportedintheSIDERdatabase(Stable3),wefoundthatanothersteroid,methylprednisolone,wasalsoassociatedwithincreasedriskforthisdisease.Whileweareunabletoinfercausationfromouranalyses,theassociationsweidentifiedwillhopefullyinformphysiciansoftherisksandencouragethemtotaketheappropriateprophylacticmeasures.ParticularattentionshouldbepaidtoLymepatientswithrespiratorycomplicationsbecausetheirimmunesystemsarealreadyweakenedfromLymeinfection,andcertainmedications(suchasamoxicillin)areineffectiveattreatingtheseinfections59.Thisstudyhadseverallimitations.Oneissueistherelativelowsamplesize,whichisaconsequenceoftherarityofthisdiseasealthoughourhospitalhasthelargestEMRsysteminNYC.BasedontheavailableELISAandWesternblotlabtestsspecificforIgMandIgG,wefoundagreatconcordanceforpositiveserologyinpatientswithLymeICDcodes,whichenhancedourconfidenceofidentifyingtruepositiveLDpatients.Basedonlimitedavailabilityofthesedata,however,wehadtousetheICD-9codealonetoselectthepatientcohortinthisstudy.Astickbitesaremostlikelytooccurinsurroundingrurallocationsinwhichforestsarepresent,manypatientsmaybeinitiallydiagnosedinadifferentfacility,andthencometoMSHforfollow-uptreatment.AnotherlimitationisrelatedtothecloseproximityofMSHtoothermedicalcentersinthearea.Specifically,patientsmayseektreatmentatothernearbyhospitals,resultinginthelossofvaluableinformationfromourEMRsystem.Finally,becausewedonothaveaccesstopatients'historicEMRdatafromoutsideofMSH,ourtemporalanalysesmaynotaccuratelycapturethetruetimelineofacquisitionofdiseasecomorbidities.Wearecurrentlyperforming
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
22
anexternalreplicationanalysisatanotheracademicmedicalcenter,andtheresultsofthiseffortmaybolsterourconclusions.Additionally,weareapplyingthefindingsfromourcurrentstudyinordertomodelexplicit,optimaltreatmentrecommendationsatthepatientlevel.Fromthiswork,wehopetoenhancenotonlythesuccessratesoftreatmentofLD,butalsotofacilitatepreemptivestrategiesformanaginghigh-riskensuingconditions.OurstudyisthefirsttoinvestigateacomprehensiveandraciallydiverseEMRwiththeaimofdiscoveringthedetailedclinicalprofilesofpatientsbeforeandafterdiagnosisofLD.Weidentifiedalistofmedications,includingantibioticsrecommendedfortreatment,whichrepresentpossibleriskfactorsforchronicLDorPTLDS.Inaddition,wehopetoinvestigatethecontributionsofgenomicsandgeneticvariantstodifferencespathophysiology.Ourpredictivemedication–comorbiditymodelsprovideanevidence-basedapproachfortreatmentregimensthattakesintoaccountriskforLymecomorbidities,withtheultimategoalofguidingprecisionmedicinebasedontheindividualclinicalphenotypesofpatientswithLD.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
23
CONTRIBUTORSLLandJDwereresponsibleforinitialstudydesign.LLwasresponsiblefordatacollection.OI,BG,andLLwereresponsibleforstudyimplementation,interpretingthedata,andongoingmanagement.OIandBGwereresponsibleforstatisticalanalysis,predictivemodels,literaturereview,andgeneratingtablesandfigures.OI,BG,LL,andJDandwroteandeditedmanuscript.BKprovidedHIVinsightsandedits.LLandJDsupervisedthestudy.DECLARATIONOFINTERESTSDr.DudleyhasreceivedconsultingfeesorhonorariafromJanssenPharmaceuticals,GlaxoSmithKline,AstraZeneca,andHoffman-LaRoche;isascientificadvisortoLAMTherapeutics;andholdsequityinNuMediiInc.,AyasdiInc.,andOntomics,Inc.Dr.IchikawaisanemployeeofSumitomoDainipponPharmaCo.,Ltd.Therestofauthorsdeclarenocompetinginterests.ACKNOWLEDGEMENTSThisstudywasfundedbytheSteven&AlexandraCohenFoundation.WethankSaviGloweandFranciscoSantiagoforadministrativesupport.WethankDr.ChristopherPatilforprovidingscientificedits.WethankMountSinaiDataWarehouseforsupportingdataaccess.REFERENCES1. Nelson CA, Saha S, Kugeler KJ, et al. Incidence of Clinician-Diagnosed Lyme Disease,
United States, 2005-2010. Emerg Infect Dis 2015; 21(9): 1625-31.
2. Steere AC. Lyme disease. N Engl J Med 2001; 345(2): 115-25.
3. Shapiro ED. Clinical practice. Lyme disease. N Engl J Med 2014; 370(18): 1724-31.
4. Wormser GP, Brisson D, Liveris D, et al. Borrelia burgdorferi genotype predicts the
capacity for hematogenous dissemination during early Lyme disease. J Infect Dis 2008;
198(9): 1358-64.
5. Hanincova K, Mukherjee P, Ogden NH, et al. Multilocus sequence typing of Borrelia
burgdorferi suggests existence of lineages with differential pathogenic properties in humans.
PLoS One 2013; 8(9): e73066.
6. Kuehn BM. CDC estimates 300,000 US cases of Lyme disease annually. JAMA 2013;
310(11): 1110.
7. Engstrom SM, Shoop E, Johnson RC. Immunoblot interpretation criteria for
serodiagnosis of early Lyme disease. J Clin Microbiol 1995; 33(2): 419-27.
8. Dressler F, Whalen JA, Reinhardt BN, Steere AC. Western blotting in the
serodiagnosis of Lyme disease. J Infect Dis 1993; 167(2): 392-400.
9. Wormser GP, Nowakowski J, Nadelman RB, Visintainer P, Levin A, Aguero-Rosenfeld
ME. Impact of clinical variables on Borrelia burgdorferi-specific antibody seropositivity in
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
24
acute-phase sera from patients in North America with culture-confirmed early Lyme disease.
Clin Vaccine Immunol 2008; 15(10): 1519-22.
10. Li X, Dunn J, Salins D, et al. Digital Health: Tracking Physiomes and Activity Using
Wearable Biosensors Reveals Useful Health-Related Information. PLoS Biol 2017; 15(1):
e2001402.
11. Berghoff W. Chronic Lyme Disease and Co-infections: Differential Diagnosis. Open
Neurol J 2012; 6: 158-78.
12. Baldwin KD, Brusalis CM, Nduaguba AM, Sankar WN. Predictive Factors for
Differentiating Between Septic Arthritis and Lyme Disease of the Knee in Children. J Bone
Joint Surg Am 2016; 98(9): 721-8.
13. Wormser GP, Dattwyler RJ, Shapiro ED, et al. The clinical assessment, treatment,
and prevention of lyme disease, human granulocytic anaplasmosis, and babesiosis: clinical
practice guidelines by the Infectious Diseases Society of America. Clin Infect Dis 2006;
43(9): 1089-134.
14. Feder HM, Jr., Johnson BJ, O'Connell S, et al. A critical appraisal of "chronic Lyme
disease". N Engl J Med 2007; 357(14): 1422-30.
15. Halperin JJ. Prolonged Lyme disease treatment: enough is enough. Neurology 2008;
70(13): 986-7.
16. Patel R, Grogg KL, Edwards WD, Wright AJ, Schwenk NM. Death from inappropriate
therapy for Lyme disease. Clin Infect Dis 2000; 31(4): 1107-9.
17. Klempner MS, Hu LT, Evans J, et al. Two controlled trials of antibiotic treatment in
patients with persistent symptoms and a history of Lyme disease. N Engl J Med 2001;
345(2): 85-92.
18. Fallon BA, Keilp JG, Corbera KM, et al. A randomized, placebo-controlled trial of
repeated IV antibiotic therapy for Lyme encephalopathy. Neurology 2008; 70(13): 992-
1003.
19. Finch C, Al-Damluji MS, Krause PJ, et al. Integrated assessment of behavioral and
environmental risk factors for Lyme disease infection on Block Island, Rhode Island. PLoS
One 2014; 9(1): e84758.
20. Guerra M, Walker E, Jones C, et al. Predicting the risk of Lyme disease: habitat
suitability for Ixodes scapularis in the north central United States. Emerg Infect Dis 2002;
8(3): 289-97.
21. Denny JC, Bastarache L, Ritchie MD, et al. Systematic comparison of phenome-wide
association study of electronic medical record data and genome-wide association study
data. Nat Biotechnol 2013; 31(12): 1102-10.
22. Healthcare Cost and Utilization Project (HCUP). HCUP Clinical Classifications Software
(CCS) for ICD-9-CM. . Agency for Healthcare Research and Quality; 2017.
23. Nelson SJ, Zeng K, Kilbourne J, Powell T, Moore R. Normalized names for clinical
drugs: RxNorm at 6 years. J Am Med Inform Assoc 2011; 18(4): 441-8.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
25
24. Tibshirani R. Regression shrinkage and selection via the lasso. Journal of the Royal
Statistical Society Series B (Methodological) 1996: 267-88.
25. Zou H, Zhang HH. On the Adaptive Elastic-Net with a Diverging Number of
Parameters. Ann Stat 2009; 37(4): 1733-51.
26. Fan J, Li R. Variable selection via nonconcave penalized likelihood and its oracle
properties. Journal of the American statistical Association 2001; 96(456): 1348-60.
27. Friedman J, Hastie T, Tibshirani R. Regularization Paths for Generalized Linear
Models via Coordinate Descent. J Stat Softw 2010; 33(1): 1-22.
28. Little RJ, Rubin DB. Causal effects in clinical and epidemiological studies via potential
outcomes: concepts and analytical approaches. Annu Rev Public Health 2000; 21: 121-45.
29. Ho DE, Imai K, King G, Stuart EA. MatchIt: nonparametric preprocessing for
parametric causal inference. Journal of Statistical Software 2011; 42(8): 1-28.
30. Cantorna MT, Hayes CE. Vitamin A deficiency exacerbates murine Lyme arthritis. J
Infect Dis 1996; 174(4): 747-51.
31. Cristea V, Crisan M. Lyme disease with magnesium deficiency. Magnes Res 2003;
16(4): 287-9.
32. Ormerod LD, Puklin JE, Giles CL. Chronic Propionibacterium acnes endophthalmitis as
a cause of intermediate uveitis. Ocul Immunol Inflamm 1997; 5(1): 67-8.
33. Heinz C, Schoonbrood S, Heiligenhaus A. Intermediate uveitis in children and young
adults: differences in clinical course, associations and visual outcome. Br J Ophthalmol
2014; 98(8): 1107-11.
34. Toledo A, Monzon JD, Coleman JL, Garcia-Monco JC, Benach JL.
Hypercholesterolemia and ApoE deficiency result in severe infection with Lyme disease and
relapsing-fever Borrelia. Proc Natl Acad Sci U S A 2015; 112(17): 5491-6.
35. Wang JJ, Rochtchina E, Tan AG, Cumming RG, Leeder SR, Mitchell P. Use of inhaled
and oral corticosteroids and the long-term risk of cataract. Ophthalmology 2009; 116(4):
652-7.
36. Pottegard A, Hallas J, Olesen M, et al. Hydrochlorothiazide use is strongly associated
with risk of lip cancer. J Intern Med 2017; 282(4): 322-31.
37. Friedman GD, Udaltsova N, Chan J, Quesenberry CP, Jr., Habel LA. Screening
pharmaceuticals for possible carcinogenic effects: initial positive results for drugs not
previously screened. Cancer Causes Control 2009; 20(10): 1821-35.
38. Niafar M, Hai F, Porhomayon J, Nader ND. The role of metformin on vitamin B12
deficiency: a meta-analysis review. Intern Emerg Med 2015; 10(1): 93-102.
39. Ho JM, Juurlink DN. Considerations when prescribing trimethoprim-sulfamethoxazole.
CMAJ 2011; 183(16): 1851-8.
40. Audebert A, Zittoun J, Zittoun R, Bousser J. [Macrocytic animia due to the
combination trimethoprim-sulfamethoxazole]. Sem Hop 1975; 51(4): 243-5.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
26
41. Heidelbaugh JJ. Proton pump inhibitors and risk of vitamin and mineral deficiency:
evidence and clinical implications. Ther Adv Drug Saf 2013; 4(3): 125-33.
42. Straubinger RK, Straubinger AF, Summers BA, Jacobson RH. Status of Borrelia
burgdorferi infection after antibiotic treatment and the effects of corticosteroids: An
experimental study. J Infect Dis 2000; 181(3): 1069-81.
43. Jowett N, Gaudin RA, Banks CA, Hadlock TA. Steroid use in Lyme disease-associated
facial palsy is associated with worse long-term outcomes. Laryngoscope 2017; 127(6):
1451-8.
44. Stahlmann R, Lode H. Safety considerations of fluoroquinolones in the elderly: an
update. Drugs Aging 2010; 27(3): 193-209.
45. Gultuna S, Koklu S, Arhan M, Aydin F, Mesci P, Uskudar O. Ciprofloxacin induced
tendinitis. J Clin Rheumatol 2009; 15(4): 201-2.
46. Kuhn M, Campillos M, Letunic I, Jensen LJ, Bork P. A side effect resource to capture
phenotypic effects of drugs. Mol Syst Biol 2010; 6: 343.
47. Kuhn M, Letunic I, Jensen LJ, Bork P. The SIDER database of drugs and side effects.
Nucleic Acids Res 2016; 44(D1): D1075-9.
48. Donta ST. Issues in the diagnosis and treatment of lyme disease. Open Neurol J
2012; 6: 140-5.
49. Abbott RA, Hammans S, Margarson M, Aji BM. Diaphragmatic paralysis and
respiratory failure as a complication of Lyme disease. J Neurol Neurosurg Psychiatry 2005;
76(9): 1306-7.
50. Enserink M. What mosquitoes want: secrets of host attraction. Science 2002;
298(5591): 90-2.
51. Kim YM, Kim S, Cheong HK, Ahn B, Choi K. Effects of heat wave on body
temperature and blood pressure in the poor and elderly. Environ Health Toxicol 2012; 27:
e2012013.
52. Gugliotta JL, Goethert HK, Berardi VP, Telford SR, 3rd. Meningoencephalitis from
Borrelia miyamotoi in an immunocompromised patient. N Engl J Med 2013; 368(3): 240-5.
53. Bremell D, Sall C, Gisslen M, Hagberg L. Lyme neuroborreliosis in HIV-1 positive men
successfully treated with oral doxycycline: a case series and literature review. J Med Case
Rep 2011; 5: 465.
54. van Burgel ND, Oosterloo M, Kroon FP, van Dam AP. Severe course of Lyme
neuroborreliosis in an HIV-1 positive patient; case report and review of the literature. BMC
Neurol 2010; 10: 117.
55. Elsner RA, Hastey CJ, Olsen KJ, Baumgarth N. Suppression of Long-Lived Humoral
Immunity Following Borrelia burgdorferi Infection. PLoS Pathog 2015; 11(7): e1004976.
56. Lochhead J, Elston JS. Doxycycline induced intracranial hypertension. BMJ 2003;
326(7390): 641-2.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
27
57. Affolter K, Samowitz W, Boynton K, Kelly ED. Doxycycline-induced gastrointestinal
injury. Hum Pathol 2017; 66: 212-5.
58. Kirsch M, Ruben FL, Steere AC, Duray PH, Norden CW, Winkelstein A. Fatal adult
respiratory distress syndrome in a patient with Lyme disease. JAMA 1988; 259(18): 2737-
9.
59. Little P, Stuart B, Moore M, et al. Amoxicillin for acute lower-respiratory-tract
infection in primary care when pneumonia is not suspected: a 12-country, randomised,
placebo-controlled trial. Lancet Infect Dis 2013; 13(2): 123-9.
60. Wei WQ, Mosley JD, Bastarache L, Denny JC. Validation and enhancement of a
computable medication indication resource (MEDI) using a large practice-based dataset.
AMIA Annu Symp Proc 2013; 2013: 1448-56.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
28
FigurelegendsFigure1Workflowofthestudy,outliningstepsfromdataorganizationtostatisticalmethodologies.Figure2Medication–Lymediseasecomorbiditynetwork,analyzedbyCCS-single-levelcategories,intimewindowsof5years(a)and10years(b).Significantassociationsbetweenmedications(cyan)andcomorbidities(magenta)areconnectedbyredorbluelines(p<0.1).Redlinesindicateriskassociations(OR>1),andbluelinesindicateprotectiveassociations(OR<1).Medicationsandindications(green)wereconnectedbasedoninformationinthepublicknowledgebaseMEDI60.Figure3Medication–LymediseasecomorbiditynetworkattheICD-9levelsintimewindowsof5years(a)and10years(b).Significantassociationsbetweenmedications(cyan)andcomorbidities(magenta)areconnectedbyredorbluelines(p<0.1).Redlinesindicateriskassociations(OR>1),andbluelinesindicateprotectiveassociations(OR<1).Figure4Kaplan–Meierplotofpropensity-score-matchedsurvivalanalysis(a)doxycycline–‘backacheNOS’(ICD-9code:724.5),(b)doxycycline–‘chronicrhinitis’(472.0),(c)doxycycline–‘cataractNOS’(366.9),(d)doxycycline–‘tearfilminsufficNOS’(375.15),and(e)amoxicillin–‘acuteURINOS’(465.9).Supplementaryfigure1Venndiagramofthemedicationsthatsignificantlyassociatedwithatleastonediseasecomorbidityinthe5-and10-yeartimewindows.(a)CCS-single-levelcategories.(b)ICD-9level.SupplementarytableS1AlldiseasesassociatedwithLyme,byICD-9category(pvalue<0.1).SupplementarytableS3Medicationspredictedtomodulateriskofdiseasecomorbidities,byCCS-single-levelcategory(pvalue<0.1).SupplementarytableS4Medicationspredictedtomodulateriskofdiseasecomorbidities,byICD-9category(pvalue<0.1).SupplementarytableS5Thebalanceofcovariatesbeforeandafterpropensityscorematching.
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
Mount Sinai HospitalEMR
Lyme Disease Cohort
Clinical Classification Software (CCS)RxNorm
n = 1,767
Disease diagnoses (Dx)Medication prescriptions (Rx)
Adaptive Lasso
Lyme
Feature Selection
Logisitc Regression
Cross-sectional Analysis
Dx Temporal Directionality
Propensity Score Matching
Temporal Analysis
Survival Analysis
Time Window
Filtering
A
with medication data>7 days follow-up n = 1,196
BICD-9 = 088.81
n = 2,134
Diagnosed year ≥ 2000n = 1,787
Defined age, sex, racen = 1,767
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
CataractBlindness and vision defects Nutritional deficiencies
Neoplasms of unspecified nature or
uncertain behavior Other eye disorders Nonmalignant breast
conditions Acute bronchitis
Diseases of the nervous system and sense
organs
Diseases of the respiratory system
Diseases of the circulatory system
Diseases of the digestive system
Diseases of the genitourinary system
Injury and poisoning
Diseases of the skin and subcutaneous tissue
Diseases of the musculoskeletal system and connective tissue
Infectious and parasitic diseases
DoxycyclineValacyclovirHeparin Levofloxacin DiclofenacClavulanate GabapentinAzelastineDocusate Propofol Midazolam Prednisone Fluticasone Azithromycin Hydrochlorothiazide Mupirocin
Neoplasms
Symptoms; signs; and ill−defined conditions
and factors influencing health status
Mental illness
Residual codes; unclassified; all e codes
Endocrine; nutritional; and metabolic diseases and immunity disorders Diseases of the blood
and blood−forming organs
-
+
Logarithm of the odds ratio
Common drugs with ICD-9 results*
Indication of drug
Lyme comorbidity
Drug
Anti infectious drug
* ***** **
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
Azelastine Tuberculin Folic acid Metformin DoxycyclineHydrocodoneEsomeprazole
Neoplasms of unspecified nature or
uncertain behavior
Inflammation; infection of eye (except that
caused by tuberculosis or sexually
transmitteddisease)
CataractNutritional deficiencies Blindness and vision defects Acute bronchitis
Diseases of the respiratory system
Diseases of the skin and subcutaneous tissue
Infectious and parasitic diseases
Diseases of the genitourinary system
Diseases of the musculoskeletal system and connective tissue
Diseases of the circulatory system
Diseases of the digestive system
LevofloxacinDocusateHeparin Valacyclovir DiclofenacPropofol Fluticasone Prednisone Acetaminophen Mometasone Albuterol Azithromycin Hydrochlorothiazide Mupirocin
Symptoms; signs; and ill−defined conditions
and factors influencing health status
Injury and poisoningDiseases of the nervous
system and sense organs
Mental illness
Endocrine; nutritional; and metabolic diseases and immunity disorders
Residual codes; unclassified; all e codes
Neoplasms
Diseases of the blood and blood−forming
organs
-
+
Logarithm of the odds ratio
Common drugs with ICD-9 results
** **
*
* ** **** *** ***
Indication of drug
Lyme comorbidity
Drug
Anti infectious drug
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
Albuterol
Hydrocodone Amoxicillin
Joint pain−shlder
Postnasal drip
Acute URI NOSExt asthma w/o stat asth
Acetaminophen
Cyclobenzaprine
Pain in limbUrinary frequency
Mupirocin
InsomniavaTretinoin Solar dermatitis NEC
Scar & fibrosis of skin Zolpidem
Abn blood chemistry NEC
Alprazolam
Atorvastatin
Other abnormal glucose
Backache NOS
Prednisone
Cervicalgia
Fluticasone
Nocturia
Doxycycline Urin tract infection NOSAcute bronchitis
Methylprednisolone
Tear film insuffic NOS
Chronic rhinitisAzithromycin
Ciprofloxacin
Esomeprazole
Amlodipine
DiclofenacFentanyl
Valacyclovir
Vitamin D deficiency NOS
Azelastine
-
+
Logarithm of the odds ratio
Lyme comorbidity
Drug
Anti infectious drug
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
Aspirin
Atorvastatin
Acute bronchitis
Nocturia
Cyclobenzaprine
Acetaminophen
Homatropine
Abn blood chemistry NEC
Amoxicillin
Postnasal drip
Azithromycin
Mometasone
Acute URI NOSHydrocodone
Levofloxacin Oxycodone
Ext asthma w/o stat asth
Esomeprazole
Sulfamethoxazole
Urin tract infection NOS
Joint pain−shlder
Urinary frequency
Ciprofloxacin
Tretinoin Scar & fibrosis of skin
Alprazolam ZolpidemSolar dermatitis NEC Mupirocin
Insomnia NEC
Pain in limb
Azelastine
Diclofenac
Other abnormal glucose
Prednisone Albuterol
Hydrochlorothiazide
Vitamin D deficiency NOS
Simvastatin
Fluticasone
Amlodipine
Backache NOS
Tear film insuffic NOS
Doxycycline Cervicalgia
Methylprednisolone
Cataract NOS
Chronic rhinitis
Chronic pain NECValacyclovir
Propofol
Docusate
Dysuria
-
+
Logarithm of the odds ratio
Lyme comorbidity
Drug
Anti infectious drug
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint
Doxycycline –
Doxycycline +
Doxycycline –
Doxycycline +
Doxycycline –
Doxycycline +
Doxycycline –
Doxycycline +
Amoxicillin –
Amoxicillin +
a b
c d
e
.CC-BY-NC-ND 4.0 International licensenot certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (which wasthis version posted December 23, 2017. . https://doi.org/10.1101/239020doi: bioRxiv preprint