C-reactive protein and albumin kinetics after antibiotic therapy in community-acquired bloodstream infection

C-reactive

protein

and

albumin

kinetics

after

antibiotic

therapy

in

community-acquired

bloodstream

infection

Pedro

Póvoa

,

Olav

Sivertsen

Garvik

,

Pernille

Just

Vinholt

,

Court

Pedersen

,

Thøger

Gorm

Jensen

,

Hans

Jørn

Kolmos

,

Annmarie

Touborg

Lassen

,

Kim

Oren

Gradel

*

a_{PolyvalentIntensiveCareUnit,HospitaldeSãoFranciscoXavier,CHLO,EstradadoFortedoAltodoDuque,1449-005Lisbon,Portugal} b

NOVAMedicalSchool,CHRC,NewUniversityofLisbon,CampodosMártiresdaPátria,1169-056Lisbon,Portugal

c

CenterforClinicalEpidemiology,OdenseUniversityHospital,andResearchUnitofClinicalEpidemiology,DepartmentofClinicalResearch,Universityof SouthernDenmark,Kløvervænget30,Entrance216,GroundFloor,5000OdenseC,Denmark

d

DepartmentofClinicalBiochemistryandPharmacology,OdenseUniversityHospital,Sdr.Boulevard29,Entrance40,5000OdenseC,Denmark

e

DepartmentofInfectiousDiseases,OdenseUniversityHospital,Sdr.Boulevard29,Entrance20,5000OdenseC,Denmark

f_{DepartmentofClinicalMicrobiology,OdenseUniversityHospital,J.B.WinsløwsVej21,2ndFloor,5000OdenseC,Denmark} g_{DepartmentofEmergencyMedicine,OdenseUniversityHospital,Kløvervænget25,Entrance63-65,5000OdenseC,Denmark}

ARTICLE INFO Articlehistory:

Received28August2019

Receivedinrevisedform17March2020 Accepted25March2020 Keywords: Community-acquiredbloodstream infections C-reactiveprotein Albumin Mortality Prognosis ABSTRACT

Objectives: We assessed C-reactive protein (CRP) and plasma albumin (PA) kinetics to evaluate community-acquiredbloodstreaminfection(CA-BSI)patients’1-yearoutcomes.

Methods:Population-basedstudy,withCRPandPAmeasurementsonday1(D1)andD4.RelativeCRP variations in relationto D1 CRPvaluewere evaluated(CRP-ratio). Patients were classified asfast response,slowresponse,non-response,andbiphasicresponse.

Results:Atotalof935patientswereincluded.AtD4,theCRP-ratiowaslowerinsurvivorsonD365in comparisonwithD4–D30non-survivorsandD30–D365non-survivors(p<0.001).Incomparisonwith fastresponsepatients,non-responseandbiphasicresponsepatientshad2.74and5.29increasedrisk, respectively,ofdeathinD4–D30and2.77and3.16increasedrisk,respectively,ofdeathinD31–D365.PA levelsremainedroughlyunchangedfromD1–D4,butlowerD1PApredictedhighershortandlong-term mortality(p<0.001).ThediscriminativeperformanceoftheCRP-ratioandD1PAtoidentifypatientswith poorshortandlong-termmortalityafteradjustmentswasacceptable(AUROC=0.79).

Conclusions:SerialCRPmeasurementsatD1andD4afterCA-BSIisclinicallyusefultoidentifypatients withpooroutcome.IndividualpatternsofCRP-ratioresponsewithPAatD1furtherrefineourabilityof predictingshortorlong-termmortality.

©2020TheAuthor(s).PublishedbyElsevierLtdonbehalfofInternationalSocietyforInfectiousDiseases. ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(

http://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction

Community-acquiredbloodstreaminfection(CA-BSI)isa well-defined infectious clinical entity that is associated with high morbidity and mortality, particularly in patients with previous comorbidities(Gradelet al.,2013).Inaddition,CA-BSIoutcome depends markedly on the timing and adequacy of empiric antibiotictherapy(Póvoaetal.,2005b;Kumar,2014).

ThediagnosisofCA-BSIisbasedonacombinationofclinical, laboratoryandmicrobiologycriteria(Grossetal.,1994).Currently, theassessmentofpatientresponsetoantibiotictherapyrelieson theresolutionofthesamecriteriaasusedfordiagnosis,e.g.fever and leukocytosis, which are very unspecific and also poorly sensitive(Halmetal.,1998;Kaukonenetal.,2018).Toovercome theselimitations,cliniciansfrequentlyusebiomarkerkinetics,in particularC-reactiveprotein(CRP)andprocalcitonin(Póvoaetal., 2011,2016).

For47ventilator-associatedpneumoniapatientsand44CA-BSI patients, we found that the patterns of CRP-ratio response to antibiotictherapyweremarkedlyassociatedwithmortalityand clinicalcourse,asearlyasday4(Póvoaetal.,2005a,2005b).Butits impactonlong-termoutcomeshasneverbeenevaluated.

Wehavepreviouslyshownthatplasmaalbumin(PA)decreases significantlyjustbeforetheCA-BSIdiagnosis(Gradeletal.,2018b).

*Correspondingauthorat:CenterforClinicalEpidemiology,OdenseUniversity Hospital,andResearchUnitofClinicalEpidemiology,InstituteofClinicalResearch, UniversityofSouthernDenmark,Kløvervænget30,Entrance216,GroundFloor, 5000OdenseC,Denmark.

E-mailaddress:kim.gradel@rsyd.dk(K.O.Gradel).

https://doi.org/10.1016/j.ijid.2020.03.063

1201-9712/©2020TheAuthor(s).PublishedbyElsevierLtdonbehalfofInternationalSocietyforInfectiousDiseases.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

ContentslistsavailableatScienceDirect

International

Journal

of

Infectious

Diseases

Asfarasweareaware,thevalueofPAkineticshasneverbeen studiedintheassessmentofCA-BSIresponsetoantibiotictherapy. The Danish Observational Registry of Infectious Syndromes (DORIS)isapopulation-basedresearchdatabasecomprisingahigh numberofCA-BSIpatientsaswellastheirdatabeforeandafterthe referenceCA-BSIepisode,namelybiochemistryandmicrobiologic data(Gradeletal.,2013,2018b).

ThisgaveustheopportunitytoassessCRPandPAkineticsin response to antibiotic therapy in order to identify patients’ outcomes,and tovalidate theconcept of patternsof CRP-ratio responsetoantibiotics.

Materialsandmethods Setting

TheDanishnationalhealthsystem,coveringbothprimaryand hospitalcare,istaxfinancedandconsequentlyfreeofchargefor theindividualpatient.Allresidentshaveauniquecivilregistration numberusedforallhealthcontactsandlinkagebetweenhealth administrativeregistries(Schmidtetal.,2014).Theadmissionofall residentswithacuteillnesses froma well-defined geographical regiontoahospitalpromptedapopulation-basedstudybasedon datafromregistries(vitalstatus,diagnoses,laboratorydata)and the medical records (sepsis severity and CNS dysfunction on admission)in theDORISresearchdatabase(Gradel etal., 2013, 2018b). Sepsis was categorized as severe sepsis/septicshock if organ dysfunction or hypotension (systolic blood pressure <90mmHg)occurred. Becausedata onhypoperfusionwerenot validwe didnot distinguish between severe sepsis and septic shock(Gradeletal.,2013).

Studycohort

The DORIS research database has been described in detail elsewhere(Gradeletal.,2013).Inbrief,itcomprisedalladults(>14 y)residinginFunenCountywithafirstepisodeofCA-BSI,inthe period2000–2008,a total of N=2785patients. For thepresent study,thecohortcomprisedthe2418patientswith1 measure-mentofCRPandPAonthesamedaywithinatimeperiodfrom30 daysbeforethrough30daysaftertheCA-BSIdate(Gradeletal., 2018b). A CA-BSI was defined as BSI occurring <3 days after hospitaladmissionandwithoutinpatientcontactinthepreceding 7days.WehadalltheCA-BSIpatients’CRPandPAmeasurements from 2000 through 2010 (Gradel et al., 2018b). Patients with unrealisticallylowlevelsofPA(<11g/L)atthedayoftheCA-BSI diagnosiswereexcluded(N=3,0.12%).TheCRPvaluesbelowthe limitofdetection(<10mg/Lin2000–2003,N=324[2.0%];<5mg/L in2004–2009,N=259[1.6%])wererandomlyallocatedavaluein therange0–9mg/L(forCRP<10mg/L)or0–4mg/L(forCRP<5mg/ L),asdescribedelsewhere(Gradeletal.,2018b).

We refer to Figure 1 for derivation of the studycohort. To capturethemostrelevantmeasurement,weinitiallyretrievedthe highestCRPandthelowestPAvaluewithinthe_first24hofthe CA-BSIdiagnosisdate(designatedD1).Thecohortofthepresentstudy comprised935patientsaliveonD4withCRP/PAmeasuredonD1 andonD4,theirCRPonD1was>50mg/L(Chanetal.,2004;Gaini etal.,2006;Knudtzenetal.,2014),andtheirvitalstatuscouldbe assessedasfromD1.

AnalysesofCRPandPAlevels

All analyses were performed by the Department of Clinical Biochemistry and Pharmacology, Odense University Hospital (OUH) and resultswere recorded in the Netlab (Medasys S.A., Littau,Switzerland)database.BothCRPandPAweremeasuredon

ModularP1(Roche,Mannheim,Germany),CRPusingan immune-turbidimetricprincipleandPAbyuseofabromocresolgreen dye-bindingmethod.Allspecimendatesrefertodateofdrawofblood specimens.

Definitions

DuetothebiologicalpropertiesofCRP(Vigushinetal.,1993; Póvoa,2002),weassessedtherelativeCRPvariations(CRP-ratio). TheCRP-ratiowascalculatedonD4inrelationtotheCRPvalueon D1.

Patients were classi_fied according to an individual pattern, previouslydefined(Póvoaetal.,2005a),ofaCRP-ratioresponseto antibiotictherapy:fastresponse–whentheCRP-ratioatD4was <0.4oftheD1CRP;slowresponse–characterizedbyacontinuous, butslow,decreaseoftheCRP-ratio,whichwas0.4and<0.8on D4;non-response–whentheCRP-ratioalwaysremained0.8of theD1value;biphasicresponse–characterizedbyaninitial CRP-ratiodecreasetolevels<0.8oftheD1CRP,followedbyasecondary risetovalues0.8.

For PA,noindividualpatterns afteraninfectionhavetoour knowledgebeenreportedintheliterature.Becausetheone-time levelofPAonD1wasastrongprognosticmortalitypredictorinour studycohort(Magnussenetal.,2016)weincorporatedthisinthe model.

Statisticalanalysis

Initially,wecomputedcontingencytableswithcategoricaland continuousvariablesofbaselinecharacteristics.

Wedeterminedtwooutcomes,D4–D30(shortterm)andD31– D365(longterm)mortality.

BoxplotsofCRPandPAwerecomputedonD1,D2,D3,andD4in relationtowhetherpatientsdiedonD4–D30,diedonD31–D365, orwerealiveonD365(excludingpatientscensoredbeforeD365as theirresultswereinsignificantlydifferentfrompatientsaliveon D365). We performed a nonparametric test for trend across ordered groups (Cuzick,1985)for the three patientsgroups in relationtotheirmeanCRPandPAlevelsoneachofthedaysD1,D2, D3,andD4.WereiteratedtheboxplotsbycomputingratiosonD2, D3,andD4incomparisontoD1andperformedlinearregression analysestodeducewhethertheratiosdifferedfromD1.

WefurthercomputedKaplan–MeiercurvescoveringD4–D365, ofthefourCRP-ratioresponsepatternsandPAquartilesonD1and usedthelog-ranktesttoassesswhetherthecurvesdiffered.

The differences between hazard ratios in Cox proportional hazards regression analyses and odds ratios (ORs) in logistic regression analyses were insignificant as there was almost complete follow-up (data not shown). Hence, we applied the latter,withD4–D30andD31–D365mortalityasoutcomes,asthis enabled the subsequent computation of receiver operating characteristic (ROC) curves and the areaunder these (AUROC). Initially,weperformedthreemodels,onewiththefourCRP-ratio responsepatterns,onewithPAonD1,andoneinwhichthe CRP-ratio response patterns were combined with PA on D1. We reiterated thesemodels bythe amendmentof gender, age,the Charlson comorbidityindex(0,1–2,>2points)(Charlsonet al., 1987),themainbacterialgroups(mono-microbialGram-positive, mono-microbialGram-negative,poly-microbial),and numberof organdysfunctions(0,1,>1).Forallmodels,wereportedORsand AUROCs with 95% confidence intervals (CIs) and we evaluated whetherAUROCsdifferedsignificantlyfromeachother(DeLong etal.,1988).

Alltwo-sidedp-values<0.05wereconsideredsignificant. TheStatasoftware(v.14.2,StataCorp,TX,USA)wasusedforall analyses.

Results

Baselinecharacteristics

The baseline clinical characteristics at the day of CA-BSI diagnosis are presented in Table 1. At CA-BSI diagnosis,18.8% presentedwithsepsis,58.7%withseveresepsisorsepticshock, and65.0%withoneormoreorgandysfunctions.Anumberof147 (15.7%)and197(21.1%)patientsdiedonD4–D30andD31–D365, respectively,whereas60(6.4%)patientswiththeirCA-BSIbetween 31May2008and31December2008werecensoredbeforeD365as theywerealiveonthelatestvitalstatusdate(31May2009). KineticsofCRPlevels

Regardlessofpatientgroup(aliveonD365,deadD4–D30,dead D31–D365),CRPlevelsdecreasedfromD1 toD4andtheywere significantlyloweronD3andD4incomparisontoD1(allp0.01) (Figure2,upperpanel).Therewerenodifferencesbetweenthe threepatients’groupsonD1(p=0.23)orD2(p=0.39).OnD3and D4, the CRP levels differed significantly; survivors on D365 presented the lowest CRP levels followed by D31–D365 non-survivors, with the highest levels in D4–D30 non-survivors (p0.01).PatientscensoredbeforeD365,allofwhomwerealive

atleast154daysaftertheirCA-BSI,didnotdifferfrompatients aliveonD365(datanotshown).

KineticsofPAlevels

PAlevelsdidnotdifferonD2,D3,orD4incomparisontoD1(all p0.29)(Figure2,lowerpanel).TheyroseslightlyonD4,except forpatientsdyingonD4–D30,whichwasunchangedin compari-sontoD3levels.Onalldays,therewerecleartrendsofdecreasing PA levels over the three patient groups (p<10 4_{). Patients}

censoredbeforeD365didnotdifferfrompatientsaliveonD365 (datanotshown).

MortalitycharacteristicsinrelationtoCRP-ratioresponsepatterns andPAquartilesonD1

Among the 342patients with a fast CRP-ratio pattern, 222 (64.9%)werealiveonD365(Table2).Thepercentageofpatients aliveonD365declinedoverslowresponse,non-response,and bi-phasicresponsepatterns(55.3%,41.5%,and39.7%, respectively). Therewas nocleartrendfordeathinD31–D365orforpatients censoredbeforeD365,whereasaclearincreasingtrendwasseen fordeathinD4–D30,from10.2%inthefastCRP-ratiopatternto 33.3%inthebiphasicCRP-ratiopattern.

ForPAquartilesonD1,anincreasingpercentagewasaliveon D365asthequartileincreased(from42.7%inthelowestto70.1%in thehighestquartile,Table2).ItwasespeciallydeathsonD4–D30 that contributed to this, especially in the two lower quartiles where20.4%and28.6%died,incontrasttothetwohigherquartiles (PA30g/L)wherenomorethan10.6%died.

Kaplan–Meiermortalitycurves

The increasing mortality from fast response, over slow response, non-response, and biphasic response, as depicted in

Table 2, was confirmed in the Kaplan–Meier mortality curves (Figure3,upperpanel).Likewise,thehighermortalityparallelto lowerPAquartileonD1wasseen(Figure3,lowerpanel). Logisticregressionanalyses

ForbothD4–D30andD31–D365mortality,ORs(95%CIs)inthe unadjustedandadjustedmodelsdidnotdiffermaterially(Table3). Likewise,theORsofthepatternsofCRP-ratioresponsesorPAon D1changedlittlewhenbothofthesewereincludedinthemodels. TheCRP-ratioresponsepatternshowedacleartrendofincreasing ORsfromfasttobiphasicresponseforbothmortalityoutcomes. ForPAonD1withD4–D30mortalityastheoutcome,theORwas 0.90, whereas it was 0.94 (adjusted) or 0.95 (unadjusted) with D31–D365mortalityastheoutcome.

AUROCsforthelogisticregressionmodels

Regardless of outcome, the CRP-ratio response patterns contributedrelatively littletotheAUROC(0.61for D4–D30 and 0.57 for D31–D365mortality) (Table4).PAon D1rendered an AUROCof0.70fortheD4–D30mortality,whereasitsAUROCof 0.59contributedlesstothepredictionofD31–D365mortality.The combination of the CRP-ratio response pattern and PA on D1 increasedtheAUROCs0.04incomparisontothemodelswithonly PAonD1.Theamendmentof covariatesinthefullmodel (age, gender, comorbidity, bacteria, number of organ dysfunctions) increasedtheAUROCsby0.05–0.15,withthehighestAUROCof 0.79 for the complete model with the D4–D30 mortality as outcome.

Table1

Baselinepatientcharacteristics(n=935).

Text Number(%)a

Age,years

Mean,SD 66.7,15.9

Females 429(45.9)

Charlsoncomorbidityindex

0points 194(20.8) 1–2points 396(42.4) >2points 345(36.9) Sepsisseverity Nosepsis 35(3.7) Possiblysepsis 116(12.4) Sepsis 176(18.8)

Severesepsisorsepticshock 549(58.7) Organdysfunctionwithoutsepsis 59(6.3) Numberoforgandysfunctions

0 327(34.9) 1 337(36.0) 2 271(29.0) Microbiologicalisolates Mono-microbialGram-positive 365(39.0) Staphylococcusaureus 150(16.0) Streptococcuspneumoniae 126(13.5) Streptococci,other 66(7.1) Enterococcusfaecalis 23(2.5) Mono-microbialGram-negative 474(50.7) Escherichiacoli 327(35.0) Klebsiellaspp. 37(4.0) Pseudomonasaeruginosa 34(3.6) Other 76(8.1) Poly-microbial 96(10.3) Vitalstatusdata

Deadday4–30 147(15.7) Deadday31–365 197(21.1) Aliveonday365 531(56.8) Censoredbeforeday365b

60(6.4) Admittedto,ondayofCA-BSI

Surgery 97(10.4)

Medical 651(69.6)

Intensivecareunit 58(6.2) Oncology/hematology 123(13.2)

Unknown 6(0.6)

a

Exceptfor“age,years”,cf.text.

b

Allaliveon day154.SD,standard deviation; CA-BSI, community-acquired bloodstreaminfection.

Discussion

In thepresent study we evaluated theshort and long-term mortalityinalargecohortofCA-BSIpatients,asassessedbyserial measurementsofCRPandPA.InpatientsaliveonD4wefoundthat theabsoluteandrelativechangesofCRPweremarkedlyhigherin survivorsandtheidentificationoftheindividualCRP-ratiopattern ofresponsewasindependentlyassociatedbothwithD4–D30and D31–D365mortality.Inaddition,wefoundthatlowerPAonD1 independentlypredictedhigherD4_–D30andD31_–D365mortality. Thereareseveralstudiesthatassesstheclinicalresponseby evaluatingthecourseofCRPafterprescriptionofantibiotics(Póvoa etal.,2005a,2005b;Coelhoetal.,2007;Brunsetal.,2008;Lisboa

et al., 2008; Moreno et al., 2010; Póvoa et al., 2011, 2017). Altogether,survivorspresentaD3/D4CRParound60%oftheinitial value,whereas it remains roughlyunchanged in non-survivors. However,concerningBSItheevaluationofCRPkineticshasonly beendonepreviouslyina pilotstudywitha smallsample size (N=44)(Póvoaetal.,2005b).

In ourstudy, witha much largercohort ofBSI (N=935), all community-acquired, CRP (absolute and relative changes) pre-sentedthesamecourse.AmongpatientsaliveonD4,thosewith goodshortandlong-termmortalitywasjustbelow46%oftheD1 levelwhereasitwas58%forthosedyinguptoD30(AUROC=0.61). We subsequently classified our patients according to the individualpatternsofCRP-ratioresponseaspublishedelsewhere

Figure2.BoxplotsofC-reactiveprotein(CRP)andplasmaalbumin(PA)onday1,2,3,and4,forpatientsaliveonday365,deadinday31–365,anddeadinday4–30.Absolute levelsintheleftcolumnandratiosinrelationtoday1intherightcolumn.Forratios,medianlevelsareshowninthetop.ForCRPratios,ahorizontallineisdepictedfor0.4and 0.8.

Table2

MortalityinrelationtoC-reactiveprotein(CRP)ratioresponsepatternandplasmaalbuminquartileonday1.

Text Total Aliveonday365 Censoredbeforeday365 Deadday31–365 Deadday4–30 CRP-ratioresponsepattern

Fastresponse 342 222(64.9) 24(7.0) 61(17.8) 35(10.2) Slowresponse 465 257(55.3) 27(5.8) 104(22.4) 77(16.6) Non-response 65 27(41.5) 7(10.8) 17(26.2) 14(21.5) Biphasicresponse 63 25(39.7) 2(3.2) 15(23.8) 21(33.3) PlasmaalbuminquartileonD1

11–24g/L 206 88(42.7) 8(3.9) 51(24.8) 59(28.6) 25–29g/L 216 112(51.9) 15(6.9) 45(20.8) 44(20.4) 30–33g/L 235 136(57.9) 17(7.2) 57(24.3) 25(10.6) 34–46g/L 278 195(70.1) 20(7.2) 44(15.8) 19(6.8) Total 935 531(56.8) 60(6.4) 197(21.1) 147(15.7)

(Póvoa et al., 2005a). Several studies, comprising different infectionsandclinicalsettings,haveassessedtheprognosticvalue ofindividualpatternsofCRP-ratioresponsetoantibiotictherapy (Póvoaetal.,2005a,2005b;Coelhoetal.,2007;Morenoetal.,2010; Póvoaetal.,2011,2017;Rabelloetal.,2017).Inallthesestudies, patientswitha fastresponse pattern had much lowerhospital mortalitythanthosepresentingwithnon-responseorabiphasic response.Butnoneofthosestudiesassessedtheimpactof CRP-ratiopatternsonlong-termmortality.

Inthepresentstudywefoundthatamongpatientsaliveon D4 the patterns of CRP-ratio response were associated with both short and long-term mortality, even after adjusting for clinically important confounders (Table 3). In comparison with those having a fast response pattern, patients with a non-response ora biphasicresponsepatternshada2.5and 5.0 increased risk, respectively, of death in D4–D30 and a 2.5 and 3.0 increased risk, respectively, of death in D31–D365.

Severalstudiesshowedthatthesurvivorsofsepsispresenting persistent low-grade inflammation, assessed by elevations in circulatinginterleukin(IL)6,IL-10,orCRPatthetimeofhospital dischargeareassociatedwithworselong-termoutcomes(Kellum etal.,2007;Yendeetal.,2008,2014,2019).Theresultsofourstudy go somewhat beyond that as we show that a persistent inflammationmeasuredasearlyasD4,assessedbyCRP-ratio,is stronglycorrelatedwithbothshortandlongtermmortality.

ThebiologicalpropertiesofPAdifferfromthoseofCRP,suchas lessvariationinlevelsanda half-lifeofabout20days( Franch-Arcas,2001),incontrastto19–20hforCRP(Vigushinetal.,1993). Asaresult,theabsenceofmarkedchangesofPAlevelstillD4was anexpectedresult.Similarly,itwasnotfeasibletoextrapolatethe abovefourCRP-ratioresponse patternstoPA.Preliminarily,we derivedaPApatternbasedononlydecreases,onlyincreases,or both, betweenD1–D4, but these had no prognosticprediction, especiallynotwhenPAonD1wasincorporatedinthemodels(data notshown).

The strong prognostic prediction of a bad outcome, both mortalityandmorbidity,inrelationtoa lowerPAlevelis well-known(Vincentetal.,2003;LevittandLevitt,2016).Inthepresent study,wedividedpatientsinquartilesofPAonD1and besides showinganimpactontheD4–D30mortality,thiseffectalsohad markedinfluenceonthelong-termmortality.

Finally,weassessedtheabilityoftheCRP-ratioandPAonD1to identifypatientswithpoorshortandlong-termmortality.After adjustingforgender,age,Charlson comorbidityindex,themain bacterialgroups,andnumberoforgandysfunctions,thecomplete modelpresentedanacceptablediscriminativeperformancewith anAUROCof0.79(HosmerandLemeshow,2000).

Toourknowledge,ourstudyisthefirsttohavecombinedCRP andPAasprognosticpredictorsinBSIpatients.Thecombineduse ofCRPandPAhasshowninnumerouspublicationstobegooda predictor of cancer-related mortality. Especially the Glasgow Prognostic Score (reviewedin McMillan, 2013)and the CRP/PA ratio(reviewedinXuetal.,2017)havegainedmomentum.Afew studies havecombined CRP and PAas prognosticpredictors in patientsthatmayresembleBSIpatientsmorethancancerpatients. For334patientsadmittedtoanICUwithsepsisorsepticshock, Ranzani et al. showed that the CRP/PA ratio at admission or dischargepredicted90-daymortality(Ranzanietal.,2013).In424 CA pneumonia patients, the amendmentof initialCRP and PA levelstothepneumoniaseverityindeximprovedthepredictability of 28-daymortality, withanAUROCof 0.81, similartoour full model(Leeetal.,2011).

Ourstudyhasseveralstrengths.Tothebestofourknowledge thisisthestudywiththelargestsamplesizeofCA-BSIpatients assessingserialmeasurementsofCRPandPA.Secondly,ourstudy is population-based and included clinically important data for comorbidityas wellas thepatients_’ septicconditionsand vital organdysfunctionsaroundthetimeof theCA-BSIepisode.And finally,forthefirsttimetheimpactoftheCRP-ratiopatternsonthe long-termmortalitywasassessed.

Therearealsoseverallimitationsthatshouldbeacknowledged. Firstly,it isa retrospectivestudyalthoughthedatapersewere prospectively collected from registries and medical records. Secondly,wehadnoinformationaboutthenutritionalstatus.This isaconcerninrelationtowhetherPAismainlyanutritionalor inflammatorymarker.Mostreviews,includingrecentones,imply thatPAisapoornutritionalmarker(Fuhrman,2002;Leeetal.,2015;

Table3

Logisticregressionanalyses.

Model 4–30daymortality 31–365daymortality Unadjusted Adjusteda

Unadjusted Adjusteda

C-reactiveproteinratioresponse

Fast 1(reference) 1(reference) 1(reference) 1(reference) Slow 1.74(1.14–2.67)b 1.75(1.13–2.73) 1.47(1.02–2.12)b 1.45(0.98–2.13) None 2.41(1.21–4.79) 2.74(1.32–5.68) 2.29(1.17–4.48) 2.77(1.34–5.74) Biphasic 4.39(2.34–8.23) 5.29(2.67–10.5) 2.18(1.08–4.40) 3.16(1.46–6.84) PlasmaalbuminonD1 0.90(0.87–0.92) 0.90(0.87–0.92) 0.95(0.93–0.98) 0.94(0.92–0.97) C-reactiveproteinratioresponse

Fast 1(reference) 1(reference) 1(reference) 1(reference) Slow 1.71(1.10–2.65) 1.68(1.06–2.64) 1.50(1.04–2.17) 1.46(0.99–2.16) None 3.22(1.56–6.67) 3.00(1.40–6.41) 2.55(1.29–5.07) 2.90(1.38–6.08) Biphasic 3.56(1.84–6.92) 4.29(2.11–8.71) 2.06(1.00–4.18) 2.95(1.35–6.45) PlasmaalbuminonD1 0.90(0.87–0.93) 0.90(0.87–0.93) 0.95(0.93–0.98) 0.95(0.92–0.97)

a

Adjustedforgender,age,theCharlsoncomorbidityindex(0,1–2,>2points),themainbacterialgroups(mono-microbialgram-positive,mono-microbialgram-negative, poly-microbial),andnumberoforgandysfunctions(0,1,>1).

b

Oddsratio(95%confidenceinterval).

Table4

Discriminationanalyses,basedonthelogisticregressionanalyses.

Model AUROC(95%CI)a

4–30daymortality 31–365daymortality (1)C-reactiveproteinratioresponse 0.61(0.56–0.65)A

0.57(0.53–0.61)A

(2)C-reactiveproteinratioresponse,fullmodelb

0.73(0.69–0.77)BCD

0.72(0.68–0.76)B

(3)Plasmaalbuminonday1 0.70(0.65–0.74)C

0.59(0.55–0.64)A

(4)Plasmaalbuminonday1,fullmodel 0.77(0.73–0.81)D

0.72(0.68–0.76)B

(1)+3 0.74(0.69–0.78)BD _{0.63(0.58–0.67)}C

(2)+3 0.79(0.75–0.82)E _{0.74(0.70–0.78)}B

a_{Areaunderthereceiveroperatingcharacteristiccurve(95%confidenceinterval).} b

Thefullmodelamendsthefollowingcovariates:gender,age,theCharlsoncomorbidityindex(0,1–2,>2points),themainbacterialgroups(mono-microbial gram-positive,mono-microbialgram-negative,poly-microbial),andnumberoforgandysfunctions(0,1,>1).Uppercaseletters:AUROCformodelsaresignificantlydifferent (p<0.05)iflettersaredifferent.

Levitt and Levitt,2016)and thesuddendecline of thePAlevel aroundtheCA-BSIaswellasthehighinversecorrelationwiththe CRPlevelasseeninourstudycohort(Gradeletal.,2018b)indicate thathypoalbuminemiaismainlyaninflammatorymarker.Thirdly, although we had microbiology results, we had no antibiotic sensitivitytests.Asaresult,wecannotevaluatetheadequacyof empiricantibiotictherapy.Fourthly,wehadnodataonfluidtherapy and fluid balance, including whether it contained albumin or plasma. However, neither albumin nor plasma are standard treatments to sepsis of BSI patients in OUH and our data on albuminvs.hemoglobinlevelsaroundthetimeofCA-BSIindicate thatsuchdatawould notalter theinterpretationof theresults (Gradeletal.,2018b).Fifthly,roughly50%oftheCA-BSIpatients wereexcludedfromourstudyforthreemainreasonsthatimpede thecalculationoftheD4CRP-ratio;theCRPwasnotmeasuredonD1 andD4,deathoccurredbeforeD4,andtheCRPlevelswerelowon D1.Thelackofdataisawell-recognizedlimitationofstudiesusing real-life data. Different patients present different numbers of specimensandhighlyvariabletimeintervalsbetweenspecimens. However,wehaveshownthatthetrajectoriesofCRPandPAlevels aroundtheCA-BSI(betweenday 30andday30)didnotdifferin relationtothenumberofspecimensfromeachpatient(Gradeletal., 2018a).Therefore,thiswouldenableustoperformlongitudinal analysesofreal-lifedatainourpopulation-baseddatabase.Forthe samereason,wecouldnotalsocalculateCRP-ratioforpatientswho diedbeforeD4.Wedoknow,however,thatthemortalityofBSI patientswithhighinitialCRPlevelsissignificantlyhigher(Gradel etal.,2011).Anotherimportantcharacteristicofabiomarkerfor longitudinalevaluationisamplitudeofvariation.Ifthevaluesare veryloworwithin“normal”rangeonD1,itisnotpossibletoassess changesovertimesincesomevariationscouldbetheresultofthe within-patientvariationrelatedtothemeasurementmethodology (Póvoaetal.,2015).Sixthly,accordingtotheCRP-ratiopatternsof response,at leastthree measurements areneeded toclassifya biphasicpattern.Asaresult,patientswithonlytwomeasurements couldhavebeenmisclassified.However,theD4CRP-ratioisabove 80%inbothabiphasicandanon-responsepatternandbothwere associatedwithapoorprognosis(Póvoaetal.,2005a,2005b,2017). Finally,someofourCA-BSIsarehealthcare-associatedaccordingto widelyaccepteddefinitions(Friedmanetal.,2002)butwebelieve thishaslittleimpactonthepatho-physiologicalpropertiesofthe studiedbiomarkers.

Conclusion

SerialCRPmeasurementsafterCA-BSIdiagnosisareclinically usefulasearly asD4toidentifypatientswithapooroutcome. Besides, the identification of individual pattern of CRP-ratio responsefurtherrefines ourabilityof prognostication eitherof short or long-term mortality. The addition of D1 PA further increasesthisability.Patternsassociatedwithpersistentin flam-mationareassociatedwithahighershortorlong-termmortality. Takentogethertheseresultsconfirmthatserialmeasurements ofbiomarkers,likeCRP,couldbeusednotonlyforprognostication but also in antimicrobial stewardship programs (Borges et al., 2019),butfurtherstudiesareneeded.

Funding

Thisresearchdidnotreceiveanyspecificgrantfromfunding agenciesinthepublic,commercial,ornot-for-profitsectors. Ethicsapproval

AccordingtotheGeneralDataProtectionRegulationoftheEU, implemented in Denmark in April 2018, no approval from the

DanishDataProtectionAgencyisrequired.Approvalbyanethics committee or consent from participants is not required for registry-basedresearchinDenmark.

Declarationsofinterest None.

References

BorgesI,SantanaM,LanaA,MartinsL,ColosimoE,OliveiraC,etal.Useofa C-reactiveprotein-basedprotocoltoguidethedurationofantibiotictherapyin criticallyillpatients:arandomizedcontrolledtrial.In:VincentJL,editor.39th internationalsymposiumonintensivecareandemergencymedicine.Brussels, Belgium..p.37–8.

BrunsAH,OosterheertJJ,HakE,HoepelmanAI.UsefulnessofconsecutiveC-reactive proteinmeasurementsinfollow-upofseverecommunity-acquiredpneumonia. EurRespirJ2008;32(3):726–32.

ChanYL,TsengCP,TsayPK,ChangSS,ChiuTF,ChenJC.Procalcitoninasamarkerof bacterialinfectionintheemergencydepartment:anobservationalstudy.Crit Care2004;8(1):R12–20.

Charlson ME,PompeiP,AlesKL,MacKenzieCR. Anewmethodofclassifying prognosticcomorbidityinlongitudinalstudies:developmentandvalidation.J ChronicDis1987;40(5):373–83.

CoelhoL,PóvoaP,AlmeidaE,FernandesA,MealhaR,MoreiraP,etal.Usefulnessof C-reactiveproteininmonitoringtheseverecommunity-acquiredpneumonia clinicalcourse.CritCare2007;11(4):R92.

CuzickJ.AWilcoxon-typetestfortrend.StatMed1985;4(1):87–90.

DeLongER,DeLongDM,Clarke-PearsonDL.Comparingtheareasundertwoormore correlatedreceiveroperatingcharacteristiccurves:anonparametricapproach. Biometrics1988;44(3):837–45.

Franch-ArcasG.Themeaningofhypoalbuminaemiainclinicalpractice.ClinNutr 2001;20(3):265–9.

FriedmanND,KayeKS,StoutJE,McGarrySA,TrivetteSL,BriggsJP,etal.Health care-associatedbloodstreaminfectionsinadults:areasontochangetheaccepted definition of community-acquired infections. Ann Intern Med 2002;137 (10):791–7.

Fuhrman MP. The albumin-nutrition connection: separating myth from fact. Nutrition2002;18(2):199–200.

GainiS,KoldkjaerOG,PedersenC,PedersenSS.Procalcitonin, lipopolysaccharide-bindingprotein,interleukin-6andC-reactiveproteinincommunity-acquired infectionsandsepsis:aprospectivestudy.CritCare2006;10(2):R53.

Gradel KO,Thomsen RW,Lundbye-ChristensenS,Nielsen H,SchønheyderHC. BaselineC-reactiveproteinlevelasapredictorofmortalityinbacteraemia patients:apopulation-basedcohortstudy.ClinMicrobiolInfect2011;17:627– 32.

GradelKO,JensenTG,KolmosHJ,PedersenC,VinholtPJ,LassenAT.DoesC-reactive proteinindependentlypredictmortalityinadultcommunity-acquired bacter-emiapatientswithknownsepsisseverity?.APMIS2013;121(9):835–42.

GradelKO,PóvoaP,VinholtPJ,MagnussenB,PedersenC,JensenTG,etal.Real-life data patterns ofC-reactive protein and albumin leveltrajectoriesaround bacteraemia.BiomarkMed2018a;12(11):1251–9.

GradelKO,VinholtPJ,MagnussenB,PedersenC,JensenTG,KolmosHJ,etal. Hypoalbuminaemia asa markerof trans-capillaryleakagein community-acquiredbacteraemiapatients.EpidemiolInfect2018b;146(5):648–55.

GrossPA,BarrettTL,DellingerEP,KrausePJ,MartoneWJ,McGowanJr.JE,etal. Qualitystandardforthetreatmentofbacteremia.InfectiousDiseasesSocietyof America.ClinInfectDis1994;18(3):428–30.

HalmEA,FineMJ,MarrieTJ,ColeyCM,KapoorWN,ObroskyDS,etal.Timeto clinicalstabilityinpatientshospitalizedwithcommunity-acquiredpneumonia: implicationsforpracticeguidelines.JAMA1998;279(18):1452–7.

HosmerDW,LemeshowS.Assessingthefitofthemodel.In:HosmerDW,Lemeshow S,editors.Appliedlogisticregression1.2nded.NewYork,USA:JohnWiley& Sons,Inc.;2000.p.143–202.

Kaukonen KM, Bailey M, Pilcher D, Cooper DJ, Bellomo R. The systemic inflammatoryresponsesyndromecriteriaandtheirdifferentialassociation withmortality.JCritCare2018;46:29–36.

KellumJA,KongL,FinkMP,WeissfeldLA,YealyDM,PinskyMR,etal.Understanding theinflammatorycytokineresponseinpneumoniaandsepsis:resultsofthe GeneticandInflammatoryMarkersofSepsis(GenIMS)Study.ArchInternMed 2007;167(15):1655–63.

Knudtzen FC, Nielsen SL,GradelKO, LassenAT,Kolmos HJ,Jensen TG,etal. Characteristicsofpatientswithcommunity-acquiredbacteremiawhohavelow levelsofC-reactiveprotein(20mg/L).JInfect2014;68(2):149–55.

KumarA.Antimicrobialdelayandoutcomeinseveresepsis.CritCareMed2014;42 (12):e802.

LeeJH,KimJ,KimK,JoYH,RheeJ,KimTY,etal.AlbuminandC-reactiveproteinhave prognosticsignificanceinpatientswithcommunity-acquiredpneumonia.JCrit Care2011;26(3):287–94.

LeeJL,OhES,LeeRW,FinucaneTE.Serumalbuminandprealbuminincalorically restricted,nondiseasedindividuals:asystematicreview.AmJMed2015;128(9) 1023e1-22.

LevittDG,LevittMD.Humanserumalbuminhomeostasis:anewlookatthe roles of synthesis,catabolism, renal andgastrointestinalexcretion, and the clinical value of serum albumin measurements. Int J Gen Med 2016;9:229–55.

LisboaT,SeligmanR,DiazE,RodriguezA,TeixeiraPJ,RelloJ.C-reactiveprotein correlateswithbacterialloadandappropriateantibiotictherapyinsuspected ventilator-associatedpneumonia.CritCareMed2008;36(1):166–71.

MagnussenB, GradelKO,JensenTG,KolmosHJ,PedersenC,VinholtPJ,etal. Association between hypoalbuminaemia and mortality in patients with community-acquiredbacteraemiaisprimarilyrelatedtoacutedisorders.PLoS ONE2016;11(9):e0160466.

McMillan DC. The systemic inflammation-based Glasgow Prognostic Score: a decadeofexperienceinpatientswithcancer.CancerTreatRev2013;39(5):534– 40.

MorenoMS,NietmannH,MatiasCM,LoboSM.C-reactiveprotein:atoolinthe follow-upofnosocomialpneumonia.JInfect2010;61(3):205–11.

PóvoaP.C-reactiveprotein:avaluablemarkerofsepsis.IntensiveCareMed2002;28 (3):235–43.

PóvoaP,CoelhoL,AlmeidaE,FernandesA,MealhaR,MoreiraP,etal.C-reactive protein asamarkerofventilator-associatedpneumoniaresolution:apilot study.EurRespirJ2005a;25(5):804–12.

PóvoaP,CoelhoL,AlmeidaE,FernandesA,MealhaR,MoreiraP,etal.Pilotstudy evaluatingC-reactiveproteinlevelsintheassessmentofresponsetotreatment ofseverebloodstreaminfection.ClinInfectDis2005b;40(12):1855–7.

PóvoaP,Teixeira-PintoAM,CarneiroAH,PortugueseCommunity-AcquiredSepsis StudyGroupS.C-reactiveprotein,anearlymarkerofcommunity-acquired sepsis resolution:amulti-centerprospectiveobservationalstudy.CritCare 2011;15(4):R169.

PóvoaP,Martin-LoechesI,ArtigasA.BiomarkerkineticsinVAP.ClinPulmMed 2015;22:185–91.

PóvoaP,Martin-LoechesI,RamirezP,BosLD,EsperattiM,SilvestreJ,etal.Biomarker kineticsinthepredictionofVAPdiagnosis:resultsfromtheBioVAPstudy.Ann IntensiveCare2016;6(1):32.

PóvoaP,Martin-LoechesI,RamirezP,BosLD,EsperattiM,SilvestreJ,etal.Biomarkers kinetics inthe assessmentofventilator-associatedpneumonia responseto antibiotics–resultsfromtheBioVAPstudy.JCritCare2017;41:91–7.

RabelloL,PóvoaP,LapaESJR,AzevedoLCP,daSilvaRamosFJ,LisboaT,etal.Patterns of C-reactive protein ratio predicts outcomes in healthcare-associated pneumoniaincriticallyillpatientswithcancer.JCritCare2017;42:231–7.

RanzaniOT,ZampieriFG,ForteDN,AzevedoLC,ParkM.C-reactiveprotein/albumin ratiopredicts90-daymortalityofsepticpatients.PLoSONE2013;8(3):e59321.

SchmidtM,PedersenL,SørensenHT.TheDanishCivilRegistrationSystemasatool inepidemiology.EurJEpidemiol2014;29(8):541–9.

Vigushin DM,Pepys MB,Hawkins PN.Metabolicand scintigraphicstudies of radioiodinatedhumanC-reactiveproteininhealthanddisease.JClinInvest 1993;91(4):1351–7.

VincentJL,DuboisMJ,NavickisRJ,WilkesMM.Hypoalbuminemiainacuteillness:is there arationalefor intervention? Ameta-analysisofcohortstudiesand controlledtrials.AnnSurg2003;237(3):319–34.

XuHJ,MaY,DengF,JuWB,SunXY,WangH.TheprognosticvalueofC-reactive protein/albuminratioinhumanmalignancies:anupdatedmeta-analysis.Onco TargetsTher2017;10:3059–70.

YendeS,D’AngeloG,KellumJA,WeissfeldL,FineJ,WelchRD,etal.Inflammatory markersathospitaldischargepredictsubsequentmortalityafterpneumonia andsepsis.AmJRespirCritCareMed2008;177(11):1242–7.

YendeS,IwashynaTJ,AngusDC.Interplaybetweensepsisandchronichealth. TrendsMolMed2014;20(4):234–8.

YendeS,KellumJA,TalisaVB,PeckPalmerOM,ChangCH,FilbinMR,etal.Long-term hostimmuneresponsetrajectoriesamonghospitalizedpatientswithsepsis. JAMANetwOpen2019;2(8):e198686.