Effects of sodium bicarbonate infusion on mortality in medical-surgical ICU patients with metabolic acidosis-A single-center propensity score matched analysis.

(1)

http://www.medintensiva.org/

ORIGINAL ARTICLE

Effects of sodium bicarbonate infusion on mortality in medical---surgical ICU patients with metabolic

acidosis----A single-center propensity score matched analysis

J. Waskowski

^a^,^b^,∗,¹

, B. Hess

^a^,^c^,¹

, L. Cioccari

^a

, I. Irincheeva

^d

, C.A. Pfortmueller

^a^,¹

, J.C. Schefold

^a^,¹

aDepartmentofIntensiveCareMedicine,Inselspital,BernUniversityHospital,UniversityofBern,Bern,Switzerland

bDepartmentofAnaesthesiologyandPainMedicine,Inselspital,BernUniversityHospital,UniversityofBern,Bern,Switzerland

cCenterofIntensiveCareMedicine,LuzernerKantonsspital,Lucerne,Switzerland

dCTUBern,UniversityofBern,Switzerland

Received23February2021;accepted23April2021

KEYWORDS Sodiumbicarbonate;

Metabolicacidosis;

Mortality;

Criticalillness;

ICU

Abstract

Objective:Metabolicacidosisisassociatedwithhighmortality.Despitetheoreticalbeneﬁtsof sodium-bicarbonate(SB),currentevidenceremainscontroversial.WeinvestigatedSB-related effectsonoutcomesinICUpatientswithmetabolicacidosis.

Design:Retrospectiveanalysis.

Setting:Academicmedicalcenter.

Patientsorparticipants: 971ICUpatientswithmetabolicacidosisdeﬁnedasarterialpH<7.3 andCO2<45mmHgtreatedbetween2012and2016.Apropensityscore(PS)wasestimatedusing logisticregression.PatientswerematchedinpairsusingthePS.

Interventions:441patientsweretreatedwithSB8.4%(SB-group)andn=530patientswerenot (controlgroup).

Mainvariablesofinterest: Primaryoutcomewasall-causemortalityatICU-discharge.Average TreatmentEffect (ATE), AverageTreatment effectinTreated(ATT), andestimatedrelative survivaleffectsat20dayswerecomputed.

Results:Inthefullcohort,weobservedconsiderabledifferencesinpH,baseexcess,additional acidosis-relatedindices,andICUmortality(controls31%vs.SB-group56%,p<.001)atbaseline betweenthetwogroups.AfterPS-matching(n=174ineachgroup),nosigniﬁcantdifferencein ICUmortalitywasobserved(controls32%vs.SB-group41%;p=.07).Oddsratios(OR)forATE andATTshowed noassociationwithICUmortality(ORATE:1.08,95%-CI0.99---1.17;p=.08;

ORATT1.09;95%-CI0.99---1.2;p=.09). Hazardratiosat20-days(multivariableHR,matched samplen=348:1.16,95%-CI0.86---1.56,p=.33)showedsimilarsurvivalinthetwostudygroups.

∗Correspondingauthor.

E-mailaddress:jan.waskowski@insel.ch(J.Waskowski).

1 Theseauthorscontributedequallytothiswork.

https://doi.org/10.1016/j.medin.2021.04.010

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

Pleasecitethisarticleas:J.Waskowski,B.Hess,L.Cioccarietal.,Effectsofsodiumbicarbonateinfusiononmortality inmedical---surgicalICUpatientswithmetabolicacidosis----Asingle-centerpropensityscorematchedanalysis,Medicina

(2)

Conclusions: WedidnotobserveeffectsofSBinfusiononall-causemortalityincriticallyill patientswithmetabolicacidosis.

PALABRASCLAVE Bicarbonatodesodio;

Acidosismetabólica;

Mortalidad;

Enfermedadcrítica;

UCI

Efectosdelainfusióndebicarbonatodesodiosobrelamortalidadenpacientesde UCImédico-quirúrgicaconacidosismetabólica:unanálisisdepuntuaciónde propensiónenunsoloCentro

Resumen

Objetivo: Laacidosismetabólicaseasociaconunaaltamortalidad.Apesardelosbeneﬁcios teóricosdelbicarbonatodesodio(BS),laevidenciaactualsiguesiendocontrovertida.Investig- amoslosefectosrelacionadosconelBSsobrelosresultadosenpacientesdelaUCIconacidosis metabólica.

Dise˜no: Análisisretrospectivo.

Ámbito: Centromédicoacadémico.

Pacientesoparticipantes:Se incluyeron 971 pacientesde la Unidadde CuidadosIntensivos (UCI)conacidosismetabólica(pH<7,3,CO2<45mmHg)tratadosentre2012y2016.Secalculó unapuntuacióndepropensión(PS)medianteregresiónlogística.Lospacientesseemparejaron utilizandoelPS.

Variablesdeinterésprincipales: Intervenciones; 441 pacientesfueron tratados con BS 8,4%

(grupoBS)yn=530pacientesno(grupocontrol).

Resultados: ElresultadoprimariofuelamortalidadportodaslascausasalaltadelaUCI.Se calcularonelefectopromediodeltratamiento(ATE),elefectopromediodeltratamientoen lostratados(ATT)ylosefectosdesupervivenciarelativaestimadosalos20días.Enlacohorte completaseobservarondiferenciasconsiderablesenelpH,elexcesodebasesylamortalidad enlaUCI(control31%vs.grupoBS56%,p<0,001)aliniciodelestudioentrelosgrupos.Después delemparejamientodePS(n=174encadagrupo),noseobservarondiferenciassigniﬁcativas enlamortalidadenlaUCI(control32%vs.grupoBS41%;p=0,07).Losoddsratios(OR)para ATEyATTnomostraronasociaciónconlamortalidadenlaUCI(ORATE:1,08,IC95%;0,99-1,17;

p=0,08;ORATT1,09;IC95%;0,99-1,2;p=0,09).Loscocientesderiesgoalos20días(HR multivariable,muestraemparejadan=348:1,16,IC95%;0,86-1,56,p=0,33)mostraronuna supervivenciacomparable.

Conclusiones: Noobservamosefectosdela infusióndeBSsobrela mortalidadportodaslas causasenpacientesconacidosismetabólica.

Introduction

Metabolic acidosis (arterial pH<7.3) is observed in about 8% of intensive care unit (ICU) admissions¹ and is associated with particularly high mortality (up to 57%).^1,2 The etiology of severe metabolic acidosis typically includes tissue hypoxemia, shock, diabetic ketoacidosis, hepatic and renal failure, and intoxications.² Metabolic acidosis has numerous deleterious consequences on various physiological systems, including enzyme-/protein func- tionality, tissue metabolism, and increased production of nitric oxide leading to vasodilatation,³ which may further amplify organ dysfunction. Furthermore, acidosis is associated with reduced adrenoceptor numbers on cellular surfaces,³ depression of myocardial func- tion including contractility and relaxation,^3,4 cardiac arrhythmia,³andashiftoftheoxyhemoglobindissociation curve.³

Previous data show that low serumbicarbonate levels predictmortalityatICUadmissionine.g.cardiogenicshock patients.^5,6Intravenoussodiumbicarbonatecanbeapplied torapidlycorrect metabolicacidosis andreports indicate thatabouttwothirdsofNorthAmericancriticalcarespe- cialists prescribe sodium bicarbonate for this purpose.⁷ However, it remains uncertain whether metabolic acido- sisshouldbecorrectedusingsodiumbicarbonate(orother buffers).^8,9 Importantly, there are concerns that sodium bicarbonatetherapycouldleadtosodiumand/orﬂuidover- load,increasedlactateand/or carbondioxideproduction, and/or decreased ionized serum calcium.¹⁰ Thus, sodium bicarbonate might theoretically worsen the outcome of affectedpatients.¹¹

Current evidence on whether sodium bicarbonate impacts on mortality in the critically ill is controver- siallydiscussed,withsparsedatafrommostlyretrospective studies^12---14available.

2

(3)

As most of theavailable studies areperformed in aci- dotic patients withsepsis^13,15 or hyperlactatemia,^12,16 we thusembarkedtoperformaretrospectiveanalysiswiththe primaryobjectivetoinvestigatewhetherinfusionofsodium- bicarbonate impactson all-causemortalityin critically ill patientswithall-causemetabolicacidosisincludingbutnot restrictedtosepticandlacticacidosis.

Patients and methods

Patients

A monocentric propensity-scorematched analysis of elec- tronicpatientchartswasperformed.Patientsweretreated at the Department of Intensive CareMedicine, University ofBern,Switzerland,betweenJanuary2012andDecember 2016.Patientswereexcludedwhenanyofthefollowingcri- teriawasmetI)refusaltoprovidegeneralconsentforuseof electronichealthcaredataforpatientstreatedafterJan- uary1st2015(standardizedgeneralconsent[GC]procedure establishedintheDepartmentofIntensiveCareMedicine), and II) patients with incomplete basic datasets in regard toage,gender, AcutePhysiologyandChronicHealthEval- uation (APACHE)II-Score at ICU-admission, discharge data and/or ICU mortality data. Routinely recorded data from electronic patientcharts was assessed(Centricity Critical Care;GeneralElectrics,Helsinki,Finland).

The study wasapprovedby thelocalEthicsCommittee onHumanResearch(KantonaleEthikkommission,KEK,Bern, Nr.2018-01829),whowaivedtheneedforindividualwritten informedconsentduetotheGCproceduredescribedabove.

Patientsweregroupedfortreatmentwithsodiumbicarbon- ate8.4%duringtheICUstayvs.withoutsodium-bicarbonate intosodiumbicarbonate(SB)andcontrolgroups(noSB).We chosethesodiumbicarbonate8.4%concentrationasthisis theformulationofchoiceinourICUtotreatsevereacidemia ifnecessary.Forthepreliminarydataselection,weusedthe availablebasicdataset(age,gender,andAPACHEIIscoresat ICUadmission)topaircontrolpatientstosodiumbicarbon- atetreatedpatientsina3:1ratiotogainameaningfulsubset ofICUpatientswithcomparablebaselinediseaseseverity.

Primaryandsecondaryoutcomes

Primary outcome was all-cause mortality (crude and adjusted) at ICU-dischargeandall-causemortalityat hos- pitaldischargeinpatientswith(non-respiratory)metabolic acidosis(deﬁnedaspH≤7.3,pCO2<45mmHg)treatedwith versus without sodium bicarbonate infusion. Secondary objectiveswereexcessmortality(mortalitybeyondICUmor- tality)at30days,atoneyearfollowing ICUdischargeand excess mortality after morethan one year until the last- possiblefollow-up(hospitaldischargeuntilSeptember2019) tobetterdifferentiateashorttermeffectfromahypothet- icallongtermeffect.

Variablescollected/studydata

Datawasderivedfromelectronicpatientcharts.Sourcedata weredoublecheckedbytwoindividualICUphysicians(JW

andBH) beforeextraction.Inconsistencies werediscussed andaconsensuswassoughtbetweenthetwoexaminers.The following routinelycollected datawere available: patient age, diagnostic and comorbidity groups (APACHE IV diagnostic groups), body mass index (BMI), APACHE II scores, timeof ICU admission/discharge, time of hospitaladmis- sion/discharge, cumulative dose of sodium-bicarbonate received,laboratorydataatadmissionincludingarterialpH, arterial carbon dioxide (paCO2), base excess (BE), bicarbonate (HCO³⁻), lactate, sodium, potassium, hemoglobin (Hb),creatinine,timeonandcumulativedoseofvasopres- sors/inotropes (noradrenaline, adrenaline, dobutamine), totalvolumeofpackedredbloodcells(RBC)andfreshfrozen plasma (FFP), need for renal replacement therapy (RRT), timeonmechanicalventilation(in-toextubation),lengthof stay(LOS)inICU/hospital,vitalstatusatICU-andhospital discharge(alive/dead),timeofdeath.

Statisticalanalysis

We explored the marginal univariate distributions of the treatmentvariablesinthe‘‘crude’’dataset(Suppl.Fig.1).

Duetoasymmetryinthecontinuousvariableswereportthe p-valuesof Kolmogorov-Smirnovtest for continuous varia- blesandchi-squaredtestfor categoricalvariables(forthe nullhypothesisofequalfrequenciesbetweentwogroups).

Thep-valuesaretobeinterpretedasmeasureofdissimilar- ity(p-valuescloseto0correspondtohighdissimilarity).

Apropensityscore(PS)ê(x)wasestimatedusinglogistic regressioninwhichtheSBtreatmentstatuswasregressed onpH, paCO2 and HCO3−. Pairs of treated and untreated patients were matched on the propensity score using a caliperof width0.3of thestandarddeviation ofthelogit ofthe propensity score (usingthe Rpackage ‘‘Matching’’

by Diamond and Sekhon, 2013). We considered mortality atICUdischargeaswellasdeathoutcomesathospitaldis- charge,after30days,atoneyearanddeathaftermorethan one year (after hospital discharge until maximum follow- up(thelatterwithoutICU-mortality).Mortalityoccurrences wereregressedonthesetof covariateswithlogitlogistic regression.Thecovariatesforthefullweightedsamplewere SB treatment status, propensity score, sex, age, BMI and APACHEIIscores.Inthematchedsample,giventhattreated patientsandcontrolswerepairedwithrespecttopropen- sityscores(aswellaspH,paCO2andHCO³⁻),thecovariates setwasrestrictedtoSBtreatmentstatus,sex,age,BMIand APACHEII.

The Average Treatment effect in Treated (ATT) is the expectedeffectofthetreatmentforindividualsinthetreat- mentgroup,¹⁷ whiletheaverage treatmenteffect(ATE)is the expectedeffect of the treatment across all individu- alsinthepopulation.RandomizedClinicalTrials(RCTs)are usuallyorganizedinsuchwaythatATEequalsATT (unless explicitly specified exceptions). Difference between ATE andATTwouldindicatethatthetreatmentassignmentwas notrandom or not randomenough in the study consider- ingcovariatesofinterest.WecomputedtheATEOddsRatio (OR)andATTOR¹⁸withstandarderrorsbyAbadieandImbens (2006)onamatchedsample¹⁹;aswellastheATEORonthe fullsample withinverse weightingby ATEweights²⁰ given by1/(ê(x))forthetreatedsubjectsandby1/(1−ê(x))for

(4)

untreated;finally we provide an ATT ORon afull sample withinverseweightingbyATTweights²¹ givenby1forthe treatedsubjectsandby ê(x)/(1−ê(x))foruntreated.

Weestimatedsurvivalcurvesandrelativesurvivaleffects with Cox regression models by regressing survival on SB treatment status,sex,age, BMIandAPACHEII. Bothanal- yseswere donefor the maximumfollow-up aswellasfor a20daysfollow-upasforthemajorityofpatients(97.5%) theICUstaywasbelow20days.Inthe‘‘crude’’unmatched dataset, we addedthe propensity scoretothe covariates set. We used a robust variance estimator to account for theclusteringwithinthematchedsetandweightinginthe

‘‘crude’’samples.

Results

Atotalof18,754admissionsofICUpatientswerescreenedin thestudyinterval.Afterremovalofduplicatesandpatients withincompletedatasets, 16,966 datasetsremained.Of these,749patientsreceivedsodiumbicarbonate(SBgroup) duringICUstay.Weperformeda1:3pairingofpatientswith regardtoage,genderandAPACHEIIscoreatICUadmission tonarrowfurtherdatasearchonpatientswithcomparable disease severity. Forrespective n=2165 patients, further data collectionwasperformed. While exploringthe data, weremovedn=69veryextremeoutlierswithaberrantval- ues(eventuallydataentrymistakes)inICUlengthofstayand veryhighvasopressorsupportinadditiontopatientswithout acidosis(pH>7.3,n=1125).441(SB)and530(noSB)patients remainedintheﬁnal‘‘crude’’dataset.AConsortﬂowchart isgiven(Fig.1).

Patient characteristics of this ‘‘crude’’ data set are givenin Table1.Mediandose ofsodiumbicarbonate 8.4%

givenintheSB-groupwas100ml(interquartilerange[IQR]

100---242ml).Patients receivedSB after amedian time of 5.14h followingICUadmission(IQR14.64h).The SB-group had lower APACHE II scores than the noSB-group (median 30; IQR 11 vs. 32; IQR 19; p<0.001). Further, statisti- callysigniﬁcantdifferenceswereobservedbetweengroups regarding pH (SB median 7.17; IQR 0.15] vs. noSB 7.23;

IQR0.09;p<0.01),arterialHCO3−(SBmedian12.4mmol/l;

IQR 8.11] vs. noSB 17.7mmol/l; IQR 5.28; p<0.001) and additionalacidosis-relatedindices(Table1).Patientsinthe SBgroupdifferedsigniﬁcantlyfromnonSBpatientsinregard toadmissiondiagnosissuchassepsis,metabolicdisorders, intoxications,needforvasoactivesandbloodproducts,and needforRRT(allp<0.05).ICUmortalitywashigherinthe sodium-bicarbonate group (SB 46% vs noSB 31%; p<.001) (Table1). Mediantimetodeathin non-survivorswas7.66 days(IQR279.3days)inthenoSB-Groupvs.3.32daysinthe SB-group(IQR37.53days)(p<.001).

TheORsfortheATTwithregardto(all-cause)ICUmor- talityand(all-cause)in-hospitalexcessmortality(mortality beyondICUmortality)inthe‘‘crude’’datasetadjustedfor SBtreatmentstatus,sex,age,BMIandAPACHEscoreswere 1.56(95%conﬁdenceinterval[CI]1.17---2.07;p=.002)resp.

0.67(95%CI0.43---1.03;p=.07).TheATTORforexcessmor- talityat30days,betweenday31and364andoveroneyear are0.52(95%CI0.32---0.84;p=.007),1.31(95%CI0.74---2.31;

p=.357) and0.72 (95%CI0.47---1.09; p=.121) respectively (SupplementTable1).

Propensityscoreanalysis

WeconstructedthePSmodelthatﬁttedto971observations andexplained34%ofvariabilityintreatmentprescription.

Whiledropping615observations(describedinSuppl.Table 2),weachievedasoundmatching betweenthetreatment andcontrolgroups(Suppl.Fig.2.).

Usingpropensityscores,weobtainedadatasetwith348 patientsequallysplitbetweenthecontrolandsodiumbicar- bonate groups. Patient characteristics are given (Fig. 1, Table 2). Median cumulative dose of sodium bicarbonate was100ml(IQR100,200)appliedafteramedianof8.77h followingICUadmission(IQR2.56,23.51).ICU-mortality,in- hospitalexcessmortalityandexcessmortalityafterhospital dischargeaswellasmediantime-to-deathdidnotdiffersig- niﬁcantly(pforall>.05)(Table2).MedianhospitalLOSwas longerintheSBgroupthaninthenoSBgroup(9days;IQR 17.2daysvs.6days;IQR16days;p=.01)

Inthematched dataset,weobservednoassociationof sodiumbicarbonate withICU-mortalityor deathoutcomes neitherintheATEORnorintheATTOR(Table3).

As for the majority of patients (97.5%), ICU length of staywasbelow20days,weestimatedcrudeKaplan---Meier survival curvesup to20 days (i.e. ifdeath didnot occur after20 daysfromadmission,we reportedit asnodeath eventduring 20 days).The Kaplan---Meier estimateof the

‘‘crude’’datashowedthat thepatientswhoreceivedthe sodiumbicarbonatetreatmenthadconsiderablylowersur- vivalprobability(log-ranktest:p=.005;Fig.2),whileusing thematcheddata,therewasnosignificantdifference(stratifiedlog-ranktestp=.53).ApplyingtheATEandATTweights inthe‘‘crude’’sample(n=971)didnotresultinadifference inprobabilityofsurvival(stratifiedlog-ranktestp=.91and .98,respectively)(Fig.2).EstimatingKaplan---Meiersurvival curvesfor the whole follow-up time showed nodifferent results(Suppl.Fig.3).

Cox regression modeling did not reveal a signiﬁcant differencein survivalat 20days (Hazardratio [HR] (multivariable,matchedsample,n=348:1.16;95%CI0.86---1.56;

p=.33)(Table4).

Discussion

Inthismonocentricpropensity scorematchedanalysis,we observednoinﬂuenceofsodiumbicarbonateinfusiononICU- mortalityincriticallyillpatientswithmetabolicacidosis.

Incriticallyillpatients,itappearsthatsodiumbicarbon- ateinfusion ismostly prescribedassymptomaticmeasure aiming to inﬂuence effects of severe metabolic acidosis (e.g. on the cardiovascular system) until a more causal therapy (e.g. improvement of tissue oxygenation, source control strategies, and/or antibiotics) becomes effective.

Currently,thereis uncertainty whetherSB infusion would impactonpatientsurvivalbeneﬁciallyorwhetheritwould e.g.induce sodiumand/or ﬂuid overload,increase paCO2

production,andwouldpotentially bedetrimental.¹⁰ Previ- ousstudiessupportedthisassumptionandshowedthatSB administrationinmetabolicacidosismayactuallyincrease mortality.¹⁶ In contrast, the BICAR-ICU-trial¹² and a retrospective study of Zhang et al.¹³ showed no impact on mortality when sodium bicarbonate was administered to

(5)

Figure1 Flowchartofpatientselection.

(6)

Table1 Patientdemographics,ICUtreatment,andfollow-updataof‘‘crude’’patientpopulation.

Variable n(all)=971 n(noSB)=530 n(SB)=441 navail.obs. p

Gender Female:318(33%) Female:164(31%) Female:154(35%) 971 .21

BMI(kg/m²) 26.12[6] 26.12[5.95] 26.23[6.26] 971 .77

Age(years) 67[17.5] 67[17] 66[18] 971 .12

APACHEII 31[10.5] 32[19] 30[11] 971 <.001

DGCardiovasc. 342(35%) 191(36%) 151(34%) 971 .61

DGRespiratory 176(18%) 114(22%) 62(14%) 971 <.001

DGGastroint. 107(11%) 49(9%) 58(13%) 971 .07

DGNeurological 80(8%) 63(12%) 17(4%) 971 <.001

DGTrauma&TBI 42(4%) 31(6%) 11(2%) 971 .02

DGMetabolic&Intox 62(6%) 16(3%) 46(10%) 971 <.001

DGHaemat.disease 17(2%) 9(2%) 8(2%) 971 1

DGRenal 16(2%) 8(2%) 8(2%) 971 .91

DGSepsis 99(10%) 40(8%) 59(13%) 971 <.001

apH 7.21[0.13] 7.23[0.09] 7.17[0.15] 971 <.001

aHCO3(mmol/L) 15.5[6.4] 17.7[5.28] 12.4[8.11] 971 <.001

aLactate(mmol/L) 1[1] 0.9[0.6] 1.2[2.4] 971 <.001

apCO2(mmHg) 28.7[8.8] 30.55[7.28] 26.1[8] 971 <.001

BE −10.9[−8] −8.15[−6.3] −14.7[7.7] 971 <.001

Hc 0.27[0.07] 0.28[0.08] 0.26[0.07] 934 <.001

Na(mmol/L) 134[6] 134[6] 134[6] 957 .8

K(mmolL/) 3.6[0.7] 3.6[0.7] 3.6[0.8] 957 .21

CreaS(mmol/L) 110[94.5] 101[88.5] 120[109] 920 <.001

Adrenaline(y/n) 440(45%) 193(36%) 247(56%) 971 <.001

AdrenalineCD(g) 3980.35[9690.6] 1661.96[5100] 6310.7[10,871.1] 440 <.001

Noradrenaline(y/n) 676(70%) 340(64%) 336(76%) 971 <.001

NoradrenalineCD(g) 3091.82[10,659.1] 2133.73[6632.2] 4614.32[15,641.4] 676 <.001

Ventilation(y/n) 640(66%) 372(70%) 268(61%) 971 <.001

Bloodproducts(y/n) 512(53%) 234(44%) 278(63%) 971 <.001

RRT(y/n) 222(23%) 88(17%) 134(30%) 971 <.001

LengthofICUstay(days) 2.57[4.62] 2.57[4.26] 2.58[4.91] 971 .59

Lengthofhospitalstay(days) 8[15] 7[15] 8[16] 937 .14

ICUmortality 366(38%) 164(31%) 202(46%) 971 <.001

In-hospitalexcessmortality 103(11%) 63(12%) 40(9%) 971 .19

30dexcessmortality 84(9%) 54(10%) 30(7%) 971 .08

Excessmortalityd31---365 69(7%) 40(8%) 29(7%) 971 .64

Excessmortalityatlastfollowup^a 122(13%) 74(14%) 48(11%) 971 0.18

Timetodeath(days) 5.23[92.4] 7.66[279.3] 3.32[37.53] 722 <.001

Medianvalues[interquartileranges]ornumbers(percentages)aregiven.‘p’referstoKolmogorov---Smirnovtestforcontinuousvariables, ortochi-squaredtestforcategoricalvariables(forthenullhypothesisofequalcategoricalfrequenciesbetweentwogroups).Laboratory dataareatadmissiontoICU.NoSB,nonsodium-bicarbonaterecipients;SB,sodium-bicarbonaterecipients;avail.obs.,availableobserva- tions;APACHEII,AcutePhysiologyandChronicHealthEvaluation-II-Score;DG,DiagnosisGroupisreferringtotheAPACHEIV-classiﬁcation ofdiagnosisadadmissionwithaseparatesepsisgroup(allgroupsincludeoperativeandnon-operativediagnoses);Cardiovasc.,Cardiovas- cular;Gastroint.,Gastrointestinal;Haemat.,Hematological;BE,baseexcess;Hc,hematocrit;Na,sodium;K,potassium;CreaS,serum creatinine;TBI,traumaticbraininjury;NA,noradrenalin;CD,cumulativedose;RRT,renalreplacementtherapy;ICU,intensivecare unit;d,day;y,yes;n,no.

Apvalue<0.05wasconsideredstatisticallysigniﬁcant.

a Lastfollowup:September2019.Pleasenotethatexcessmortalityisgivenforin-hospitalmortalityandmortalityatd30,d31---365 andatlastfollowup(mortalityisexcessmortalityplusICUmortality).

criticallyillpatientswithmetabolicacidosis.Inlinewiththe aforementioned studies, we did notobserve an increased mortalityaftersodiumbicarbonatetreatment. Incontrast totheBICAR-study¹²andthestudyofZhang,¹³however,we likelystudiedaratherbroadpopulationofICUpatientswith metabolic acidosis aswe included not only patients with acidosisandhyperlactatemia¹²and/orsepticpatients.¹³

In addition,and in accordance withprevious data,^12,13 ourstudydidnotobservemortalitybeneﬁtsinpatientswith sodium bicarbonate infusion. However, although specula- tive, e.g. negative SB-inducedeffects could theoretically becounterweightedbybeneﬁcialeffectssuchasimproved cardiaccontractilityand/orvascularresponsivenesstocat- echolamine therapy,³ without effects on the mortality

(7)

Table2 Patientdemographics,ICUtreatment,andfollow-updataofpropensityscorematchedpatientpopulation(n=348).

Variable n(noSB)=174 n(SB)=174 p

Gender Female:55(32%) Female:50(29%) .64

BMI(kg/m²) 25.5[6.25] 26.83[7.91] .2

Age(years) 68[16] 68[17] .94

APACHEII 32[8] 28[11] <.001

DGCardiovascular 68(39%) 55(32%) .18

DGRespiratory 18(10%) 33(19%) .03

DGGastrointestinal 27(16%) 26(15%) 1

DGNeurological 15(9%) 10(6%) .41

DGTraumaincl.TBI 9(5%) 5(3%) .41

DGMetabolic&Intox 10(6%) 13(7%) .67

DGHemicdisease 3(2%) 3(2%) 1

DGRenal 4(2%) 2(1%) .68

DGSepsis 15(9%) 18(10%) .71

apH 7.23[0.08] 7.23[0.09] 1

aHCO3---(mmol/L) 15.1[3.4] 15.25[3.7] 1

aLactate(mmol/L) 1[0.8] 1[1.0] .45

apCO2(mmHg) 28[6.1] 28.1[6] .99

BE −11[−4.1] −11.25[−4] .17

Hc 0.27[0.07] 0.26[0.07] .98

Na(mmol/L) 134[4.2] 134[6] .98

K(mmol/L) 3.6[0.7] 3.6[0.8] .98

CreaS(mmol/L) 118[84.5] 125[131] <.001

Adrenaline(y/n) 76(44%) 85(49%) .39

AdrenalineCD(g) 2675.07[11,215.1] 5956.06[11,098.4] .13

Noradrenaline(y/n) 109(63%) 135(78%) <.001

NoradrenalineCD(g) 2471.82[6512.9] 3330.52[13,183.4] .09

Ventilation(y/n) 126(72%) 111(64%) .11

BloodProducts(y/n) 91(52%) 111(64%) .04

RRT(y/n) 44(25%) 43(25%) 1

LengthofICUstay(days) 2.65[4.2] 2.87[5.1] .54

Lengthofhospitalstay(days) 6[16] 9[17.2] .01

ICUmortality 55(32%) 72(41%) .07

InHospitalexcessmortality 22(13%) 18(10%) .61

30dexcessmortality 20(11%) 11(6%) .13

Excessmortalityd31---365 12(7%) 15(9%) .69

Excessmortalityatlastfollowup^a 26(15%) 23(13%) .76

Timetodeath(days) 5.22[274.9] 6.25[47.3] .91

Medianvalues[interquartileranges]orcounts(percentages)aregiven.‘p’referstoKolmogorov---Smirnovtestforcontinuousvariables, ortochi-squaredtestforcategoricalvariables(forthenullhypothesisofequalcategoricalfrequenciesbetweentwogroups).Laboratory dataareatadmissiontoICU.DG,DiagnosisGroupisreferringtotheAPACHEIV-classiﬁcationofdiagnosisadadmissionwithasepa- ratesepsisgroup(allgroupsincludeoperativeandnon-operativediagnoses);Cardiovasc.,Cardiovascular;Gastroint.,Gastrointestinal;

Haemat.,Hematological;NoSB,nonsodiumbicarbonaterecipients;SB,sodium-bicarbonaterecipients;APACHEII,AcutePhysiologyand ChronicHealthEvaluationIIScore;Hc,hematocrit;Na,sodium;K,potassium;CreaS,serumcreatinine;TBI,traumaticbraininjury;CD, cumulativedose;RRT,renalreplacementtherapy;ICU,intensivecareunit.

a Lastfollowup:September2019.Pleasenotethatexcessmortalityisgivenforin-hospitalmortalityandmortalityatd30,d31---365 andatlastfollowup(mortalityisexcessmortalityplusICUmortality).

endpoints.Further,effects(whetherbeneﬁcialornot)could theoreticallybeshort-ratherthanlong-lived.

Importantly, about 50% or more patients whoreceived sodium bicarbonate did not have a ‘‘severe’’ metabolic acidosiswithpH≤7.2¹(Tables1and3).However,whencom- paredtootherinvestigations,ourresults(medianpHof7.17 in ‘‘crude’’ and 7.23 in matched groups) may be consid- eredrathercomparabletootherretrospectivesstudies(e.g.

Kim etal.: meanpH 7.244±0.168)¹⁶; Zhanget al.:mini- mummeanpH7.16±0.1)¹³).Moreover,themedianarterial pCO2 inthematchedgroupswas28mmHg.Therefore,the observedpHatadmissionmayberespiratorycompensated andtruepHinfactconsiderablylower.

Further,onemightassumethatSBpatientswouldhave hada higherrisk of death. However,in oursubset of ICU patients,SBtreatedpatientshadaslightly,butstatistically

(8)

Table3 Oddsratios(OR)fortreatmenteffectsformortalityinthematcheddataset.

ATE(matched sample),n=348

p-Value ATT(matched sample),n=348

p-Value

ICUmortality 1.08(0.991,1.17) .08 1.09(0.987,1.2) .09

InHospitalexcessmortality 0.982(0.924,1.04) .55 0.984(0.92,1.05) .62

30dexcessmortality 0.96(0.91,1.01) .13 0.958(0.903,1.02) .16

Excessmortalityd31---365 1.02(0.968,1.08) .44 1(0.94,1.06) .98

Excessmortalityatlastfollow-up^a 0.994(0.935,1.06) .82 0.993(0.928,1.06) .83 Reportedaretheoddsratios(OR)forthetreatmenteffectsformortalityatvariousstagesinpresenceofsodiumbicarbonatetreatment.

Estimatedbylogitlogisticregressions,regressingmortalityoutcomeonSBtreatmentstatus,sex,age,BMIandAPACHE.

a LastfollowupwasinSeptember2019.ATE,AverageTreatmentEffect;ATT,AverageTreatmenteffectinTreated.Pleasenotethat excessmortalityisgivenforin-hospitalmortalityandmortalityatd30,d31---365andatlastfollowup(mortalityisexcessmortalityplus ICUmortality).

Figure2 Kaplan---Meiersurvivalcurves(20days)obtainedusingdifferentpropensityscoremethods.Inthetop-leftpanelcrude Kaplan---Meiersurvivalcurvesfortreatedanduntreatedsubjectsinthe‘‘crude’’sample(n=971;log-ranktest:p=0.005).Inthe top-rightpanelKaplan---Meiersurvivalcurvesfortreatedanduntreatedsubjectsinthepropensityscorematchedsample(n=348 stratiﬁedlog-ranktestp=0.53).Inthebottom-leftandbottom-rightpanelssurvivalcurvesinthesampleweightedusingtheATE weights(n=971)andthesampleweightedusingtheATTweights(n=971)arereported(adjustedlog-ranktestwithp-values0.91 and0.98).ATE,AverageTreatmentEffect;ATT,AverageTreatmenteffectinTreated.

signiﬁcant, lower baseline APACHE II score, whereas ICU mortality tended tobe increased. This stands in contrast to previous data^13,14 but maytheoretically underline that metabolicacidosis isadditionally associatedwithahigher mortality.

Anumberofimportantadditionallimitationsofouranal- ysis deserve discussion. First, limitations arise from the retrospective,single-center,andexplorativedesignofthis studyandallrespectiveinherentlimitationsapplythatare drivenby studydesign.Inthisretrospectivestudy,evenif

(9)

Table4 HazardratiosestimatedwithCoxregressions.

Predictorfor20 day-survival

HR

(univariable,

‘‘crude’’

samplen=971) HR

(multivariable,

‘‘crude’’

samplen=971) HR

(multivariable, matched samplen=348)

HR(multivariable, fullsamplewithATE weights,n=971)

HR(multivariable, fullsamplewithATT weights,n=971) Sodium

bicarbonate treatment

1.30 (1.08---1.55, p=.005)

1.08 (0.87---1.34, p=.50)

1.16 (0.86---1.56, p=.33)

1.07(0.93---1.22, p=.35)

1.05(0.87---1.27, p=.58)

Propensity score

1.17 (1.10---1.23, p<.001)

1.15 (1.07---1.23, p<.001)

--- 1.14(1.09---1.19,

p<.001)

1.18(1.09---1.28, p<.001)

APACHEII 1.03

(1.02---1.05, p<.001)

1.03 (1.02---1.05, p<.001)

1.03 (1.00---1.05, p=.05)

1.02(1.01---1.03, p<.001)

1.02(1.01---1.04, p=.002)

Hazardratios(HRwith95%conﬁdenceintervals)ofpredictorvariablesaregiven.ATE,AverageTreatmentEffect;ATT,AverageTreatment effectinTreated.

mortalityanalyseswereadjustedfortypicalpotentialcon- founders, our data may theoretically be subject to some degree of unmeasured confounding. Additionally, by def- inition, the matched sample analysis required extensive exclusionofpatientsoftheoverallsample,whichmayper seintroduce(e.g.selection)bias.Moreover,thismaypoint toheterogeneity ofthe totalcohortand maybeonerea- son why respectivepatients were particularlychallenging to match. By deliberately curtailing our ICU data set to a subset of patients and by removing some outliers that aimedforacomparablesubsetofICUpatients,itmaythe- oretically be possible that some degree of selection bias was introduced. Second, we only considered intravenous sodium bicarbonate formulations of 8.4% for analysis and mighthavemissedotherpreparations,includinge.g.chronic oralbicarbonateuseinfewpatientswithchronickidneydis- ease. Third,mortalitymaynotbean optimalendpoint to examine effects of an early singleintervention in a critically illpopulation,whichmayalsounderline ourinterest inearlyoutcomesofourcohort(i.e.20daysofICUstay).It appearsthatendpointsrelatedtotherapy-inducedeffects (e.g.increaseinpH)mightbeparticularinterestinginsubse- quentinvestigations.Fourth,ourretrospectivestudydesign made itimpossible toidentifythe medicalreasonfor the (independenttreating)physiciantoprescribesodiumbicar- bonateandforthetimingoftheprescription.Wecanthus notexcludewithcertaintythatSBwasalsogivenforaeti- ologiesofmetabolic acidosiswitha betterprognosis(e.g.

diabeticketoacidosis,intoxications) or otherreasons than formetabolicacidosisonly,e.g.incasesofadditionalrhab- domyolysis. Although the current analysis may be one of the largest investigations available, it appeared that the sample size was too limited to conclude back on effects of SBdose and/or exact timing aswell astoconcludeon subgroupsofcriticallyillpatients(e.g.acutekidneyinjury [AKI] or sepsis patients, patient’s post-surgical interventions). Thus, the presented analysis may theoretically be

‘‘underpowered’’.Fifth, wedidnotevaluateotherimpor- tant covariates such asinterventions during the ICU stay (e.g.emergencyhemodialysis)aswellasrespectiveurgency northeimpactofﬂuidintakeotherthansodiumbicarbonate

includingfluidbalance.Sixth,thepresentedmortalitydata reflectsall-causemortalityandweareunabletoconclude backonspecificacidosisand/ortreatment-relatedadverse effects noron differentcauses of death. Seventh,as our retrospectiveanalysismadeitimpossibletoconcludeback onacidbaseanalysisaftersodiumbicarbonateinfusionwe cannotdescribeanyinfluenceofsodiumbicarbonateonthe pH and the bicarbonate level after treatment. Moreover, we refrained from estimating dose-effects of the sodium bicarbonategiven.

Conclusion

In this propensity score matched analysis, intravenous administration of sodium bicarbonate did not appear to effecton mortalityin ICU patients withmetabolic acidosis.Additionalprospectivecontrolledclinicalinvestigations seem required to further determine potential effects of sodiumbicarbonateinfusion insubgroups ofpatients with metabolic acidosis, including potential high-risk groups, suchascriticallyillpatientswithAKI.

Authors’ contributions

JW: Designed the study, performed data collection and assessment,wrotetheﬁrstdraft,coordinatedtheinputof allauthors,supervisedthestudyandrevisedthemanuscript forimportantintellectualcontent.

BH: Designedthestudy,performed data collectionand assessment,wrotetheﬁrstdraft,coordinatedtheinputof allauthors,andrevisedthemanuscriptforimportantintel- lectualcontent.

LC: Contributed to data interpretation, revised the manuscriptforimportantintellectualcontent.

II: Performed all statistical analyses, revised the manuscriptforimportantintellectualcontent.

CAP: Co-designed thestudy, contributedtodatainter- pretation,supervisedthestudy,andrevisedthemanuscript forimportantintellectualcontent.

(10)

JCS:Co-designedthestudy,contributedtodatainterpre- tation,supervisedthestudyandrevisedthemanuscriptfor importantintellectualcontent.

Allauthorsapprovedtheﬁnalversionofthemanuscript.

Conﬂict of interests

Fulldepartmentaldisclosure:CAP,LC,andJCSreportgrants fromOrionPharma,AbbottNutritionInternational,B.Braun MedicalAG,CSEMAG,EdwardsLifesciencesServicesGmbH, KentaBiotechLtd,MaquetCriticalCareAB,OmnicareClin- icalResearchAG, Nestle,Pierre Fabre PharmaAG,Pﬁzer, BardMedicaS.A.,AbbottAG,AnandicMedicalSystems,Pan GasAGHealthcare,Bracco,HamiltonMedicalAG,Fresenius Kabi,GetingeGroupMaquetAG,DrägerAG,TeleﬂexMedi- calGmbH, GlaxoSmithKline,MerckSharpandDohmeAG, Eli Lillyand Company, Baxter,Astellas, Astra Zeneca,CSL Behring,Novartis,Covidien,Hemotune,Phagenesis,Philips Medical,ProlongPharmaceuticalsandNycomedoutsidethe submittedwork.Themoneywentintodepartmentalfunds.

Nopersonalﬁnancialgainapplied.

II is affiliated with CTU Bern, University of Bern, which has a staff policy of not accepting honoraria or consultancy fees. However, CTU Bern is involved in design, conduct, or analysis of clinical studies funded by not-for-profit and for-profit organizations. In particu- lar,pharmaceuticalandmedicaldevicecompaniesprovide directfundingtosomeofthesestudies.Foranup-to-date listofCTUBern’sconflictsofinterestseehttp://www.ctu.

unibe.ch/research/declarationof interest/indexeng.html The additionalauthors declarethattheyhave nocom- petinginterests.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.medin.2021.04.010.

References

1.JungB,RimmeleT,LeGoffC,ChanquesG,CorneP,Jonquet O, et al. Severe metabolic or mixed acidemia on intensive careunitadmission:incidence,prognosisandadministrationof buffertherapy.Aprospective,multiple-centerstudy.CritCare.

2011;15:R238.

2.Kraut JA, Madias NE. Lactic acidosis. N Engl J Med.

2014;371:2309---19.

3.KimmounA, Novy E,Auchet T,Ducrocq N, LevyB. Hemody- namicconsequencesofseverelacticacidosisinshockstates:

frombenchtobedside.CritCare.2015;19:175.

4.SchefoldJC,FilippatosG,HasenfussG,AnkerSD,vonHaehling S.Heartfailureandkidneydysfunction:epidemiology,mecha- nismsandmanagement.NatRevNephrol.2016;12:610---23.

5.Wigger O, Bloechlinger S, Berger D, Haner J, Zanchin T, Windecker S, et al. Baseline serum bicarbonate levels inde- pendentlypredictshort-termmortalityincriticallyillpatients withischaemiccardiogenicshock.EurHeartJAcuteCardiovasc Care.2016,2048872616683526.

6.MitraB,RomanC,ChartersKE,O’ReillyG,GantnerD,Cameron PA. Lactate, bicarbonate and anion gap for evaluation of patientspresentingwithsepsistotheemergencydepartment:

aprospectivecohortstudy.EmergMedAust.2020;32:20---4.

7.KrautJA,KurtzI.Useofbaseinthetreatmentofacutesevere organicacidosisbynephrologistsandcriticalcarephysicians:

resultsofanonlinesurvey.ClinExpNephrol.2006;10:111---7.

8.Jaber S, Jung B. Time to treat metabolic acidosis in ICU withsodiumbicarbonate?Maybe.AnaesthCritCarePainMed.

2018;37:499---500.

9.FujiiT,UdyA,LicariE,RomeroL,BellomoR.Sodiumbicarbon- atetherapyforcriticallyillpatientswithmetabolicacidosis:a scopingandasystematicreview.JCritCare.2019;51:184---91.

10.ForsytheSM,SchmidtGA.Sodiumbicarbonateforthetreatment oflacticacidosis.Chest.2000;117:260---7.

11.RhodesA,EvansLE,AlhazzaniW,LevyMM,AntonelliM,Ferrer R,etal.Survivingsepsiscampaign:internationalguidelinesfor managementofsepsisandsepticshock:2016.IntensiveCare Med.2017;43:304---77.

12.JaberS,PaugamC,FutierE,Lefrant JY,LasockiS,Lescot T, et al. Sodium bicarbonate therapy for patients withsevere metabolicacidaemiaintheintensivecareunit(BICAR-ICU):a multicentre,open-label,randomisedcontrolled,phase3trial.

Lancet.2018;392:31---40.

13.ZhangZ,ZhuC,MoL,HongY.Effectivenessofsodiumbicar- bonateinfusiononmortalityinsepticpatientswithmetabolic acidosis.IntensiveCareMed.2018;44:1888---95.

14.ZhangZ,MoL,HoKM,HongY.Associationbetweentheuseof sodiumbicarbonateandmortalityinacutekidneyinjuryusing marginalstructuralCoxmodel.CritCareMed.2019;47:1402---8.

15.El-SolhAA,AbouJaoudeP,PorhomayonJ.Bicarbonatetherapy inthetreatmentofsepticshock:asecondlook.InternEmerg Med.2010;5:341---7.

16.KimHJ,SonYK,AnWS.Effect ofsodiumbicarbonateadmin- istration on mortality in patients with lactic acidosis: a retrospectiveanalysis.PLoSOne.2013;8:e65283.

17.HartmanE,GrieveR,RamsahaiR,SekhonJS.Fromsampleaver- agetreatmenteffecttopopulationaveragetreatmenteffecton thetreated:combiningexperimentalwithobservationalstud- iestoestimatepopulationtreatmenteffects.JRStatSoc:Ser A(StatSoc).2015;178:757---78.

18.Mao H, Li L, Greene T. Propensity score weighting analysis and treatment effect discovery. Stat Methods Med Res.

2019;28:2439---54.

19.Abadie A, Imbens GW. Large sample properties of matching estimators for averagetreatmenteffects. Econometrica.

2006;74:235---67.

20.JoffeMM,HaveTRT,FeldmanHI,KimmelSE.Modelselection confoundercontrol,andmarginalstructuralmodels:reviewand newapplications.AmStat.2004;58:272---9.

21.Morgan S, Todd J. A diagnostic routinefor thedetection of consequentialheterogeneityofcausaleffects.SociolMethodol.

2008;38:231---81.