Letter
to
the
Editor
Responsetocorrespondenceconcerning:“Early hydroxychloroquinebutnotchloroquineuse reducesICUadmissioninCOVID-19patients”
We would like to thank all our colleagues for taking the opportunity to start a scientific discussion on the matter of hydroxychloroquineinCOVID-19.Wefeelthatitisimportantin thecurrentclimate,wherepoliticsandeconomicsinterferewith thescientificdomain,tohaveacalmandscientificdiscussionon thistopic.Belowyouwillfindourresponsetotheletters.
Authors’reply:
Thefirstletter“Comparisonofhospitaltreatmentstrategyor oftreatmentactuallyreceivedinCOVID-19?”asksusaboutour methodology.Weunderstandthenatureofthequestionquitewell, sincewehavediscussedthisissuebeforedecidingwhichanalysis strategytochoose.Clearly,thisisanobservationalstudythatmight be proneto potential bias. To avoid this, we used a two-step procedure. Thefirststepencompassestheinclusionof patients. Thesecondstepinvolvesthecomparisonoftreatedanduntreated patientsaftertheirhospitalization.
Theinclusionofpatientsforourstudywasbasedonhospital treatmentstrategy.PatientswereadmittedtoaHCQ,CQor non-treatmenthospitalrandomly;thehospitalpolicyregardingtheuse of HCQ/CQ was not communicated to the patient beforehand, neitherwasitpartofaselectionprocedureforthestudy.Thus,the factthatapatientwasadmittedtoaHCQ,CQornon-treatment hospitalwasarandomeventandbasedonarandomassignment, based ongeographical loctiononly. Thatis thereason whywe believethatourstudydesignandpatientinclusionapproachesthe intention-to-treatstrategyusedinrandomized-controlledtrials.
TheassignmenttoreceiveHCQorCQtreatmentoccursafter hospitalizationandisinfluencedbypatientcharacteristics,suchas co-morbidity. This means that the actual assignment of a treatmentmayleadtodifferencesbetweentreatedpatientsand untreatedpatients.Thus,tomakeaccurateinferences,itiscrucial tocontrolforconfoundingbyindicationandforthedifferences betweentreatedanduntreatedpatients.Sincetheassignmentof treatment occurs afterhospitalization, it is independentof the actualhospitaltreatmentstrategy.Therefore,webelievethatitis justifiedin this second step topool the data and compare all treatedanduntreatedpatientsdirectly.Subsequently,weusedthe propensity score (PS) matching as a balancing score between treatedanduntreatedpatients,andtocontrolforconfoundingby indication.PSmatchingisaproventooltoadjustatreatmenteffect for measuredconfounders in non-randomized studies, such as ours.AteachvalueofthePS, thedistributionoftheconsidered covariatesisthesameinthetreatedandtheuntreatedorcontrol group.Thebalancethatarandomizedexperimentisexpectedto create by design, is in this way established through statistical
matching.Inourstudy,afterPSmatchingnosignificantstatistical differences were foundbetween thebaseline characteristicsof treatedanduntreatedpatients,includingthedifferencesbetween hospitalstrategies(asisshowninTable1).Furthermore,thetype of hospital treatment was not a confounder in the definite weightedregressionmodel.
In conclusion, we believethat with ourtwo-step procedure (hospital treatment strategy at inclusion and actual treatment receivedintheanalysis),ourstudylimitstheinfluenceofpotential bias and enables us to make correct inferences based on the availabledata.
Thesecondletter,byAsselbergsetalraisesseveralquestionson themethodologyaswell.
The authors stated: “in the presence of ICU restriction for medicalreasonsorpatientpreference,analysesontheoutcome ‘transfertoICU’arelimitedtopatientswithoutICUrestriction,a particularlyselectedgroup.Thismightimpactthegeneralizability oftheresultsconsiderably”.
SincepatientswithanICUrestrictioncannotreachtheendpoint “Transfer to ICU”, we started by computing the outcome as CompositeAdverseEndpoint;takingthetwoendpoints(Mortality andTransfertoICU)together.Thiseliminatesthebiasofcompeting riskbetweenthetwoendpoints.Wedidnotconsider“Discharge” asacompetingrisk,sinceitwasdefinedasdischargeforcureonly, dischargewiththeexpectationofthepatienttodieforexamplein ahospicefacilitywasconsidered“Death”.Toadjustforpotential biasbyICUrestriction,weperformedstratifiedanalysisforthis “composite adverse endpoint” to reflect underlying potential differences between the two groups with regard to adverse incidences andrisk factorprevalence.In addition,multivariable competing risk analysis was conducted; including for ICU restriction.Thehazardratio(HR)changedfrom0.49without,to 0.47withICUrestriction.Thiscorrespondswitha4%change,which islessthanthe10%threshold,whichiscommonlyused,ifitwerea confounder.WiththeresultsofPSmatching,competing multivar-iableriskanalysesandstratifiedanalyses,wedonotbelievethat ICU restriction influenced the effect of HCQ/CQ that we have reportedonICUadmission.Subsequently,wedecidedtoreportthe twooutcomesseparately,afterperformingadditionalcompeting riskanalysiscorrections.Here,wedidnotincludepatientswithan ICUrestriction,thereforeourdataindeedonlydescribeapossible protectiveeffectofHCQinpatientswithoutanICUrestriction.As wedescribedinourdiscussion,prospectiveinterventionstudies areneededtoconfirmourresults.However,eveninrandomized controlled trials generalizability can still be a problem to extrapolate fromtheresults, although a causal effecthas been shown.
Furthermore, the authors mention that the ICU-population changedinthecourseofthepandemic.Wedidnotincludepatients ontheICUinourstudy,forthepatientsonthewardwecorrected
https://doi.org/10.1016/j.ijid.2020.12.008
1201-9712/©2020TheAuthors.PublishedbyElsevierLtdonbehalfofInternationalSocietyforInfectiousDiseases.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).
InternationalJournalofInfectiousDiseases103(2021)478–479
ContentslistsavailableatScienceDirect
International
Journal
of
Infectious
Diseases
for age (which was not different betweenthe HCQ,CQ or no-treatmentgroups).
Itissuggestedthatregionaldifferencesininfectionratesmight havecauseddifferentpatientpopulations.Thisisunlikely,since14 Dutchhospitalsparticipated(fromGoestoGroningen,bothrural hospitalsascentersinlargecities).Fortunately,wedidnotreach “codeblack”intheNetherlands,makinglargedifferencesinICU admissionratesunlikely.Obviously,allobservationalresearchisat riskforbias,includingselectionbiasbyleavingpatientsoutofthe analysisthatweretransferredbetweenhospitals.
Foranansweronthequestiononthemethodologyofinclusion by treatment center or treatment exposure, please see our responsetotheletterbyPetersetal.Patientsthatweredirectly admittedtotheICUorpatientsthatdiedwithinthefirst24hafter admission,wereexcludedfromtheanalysis.Inaddition,wehave usedskimmingasaroutineprocedureinouranalyses.Separate sensitivity analyses were performed to investigate whether immortal bias or informative censoring could influence our results.First,therewerenoeventsbeforethestartofthetherapy. Second, time-dependent analysis shows no difference in our reported results (HRHCQ = 0.47; 95% = 0.27–0.81). Third, in a logisticregressionanalysiswefoundthesameeffectofHCQonICU admission (ORHCQ = 0.40; 0.21–0.77). Thus, neither potential immortal bias nor informative censoring had a significant influenceonthereportedresults.
Finally, wewould like tothankthe authorsof theletter by Burger et al, ‘More gastro-intestinal adverse eventsin non-ICU hospitalized COVID-19patientstreatedwithchloroquineversus hydroxychloroquine’for sharingthe resultsof theirinteresting study.TheauthorsshowthatHCQhasabettersafetyprofilethan CQ,withlessgastro-intestinaladverseevents,andsuggestthatthis maypartiallyexplainthedifferenceintheriskoftransfertotheICU thatwefound.
We deliberately choose to refrain from collecting data on adverseeffects,sincesymptomsofCOVID-19arepossiblydifficult todistinguishfromtreatmentside-effects,especiallyintheelderly population,sotheseresultsaddvaluableinformationontheuseof bothdrugs.
Intheirstudy, 1.3%ofthepatientsdiscontinuedtreatmentinthe HCQgroupduetogastro-intestinaladverseevents,ascomparedto 17%intheCQgroup(atotalof2patientsonHCQversus13patients onCQ).
Wecompletelyagreewiththeauthors’suggestionthatsincethe useofCQisassociatedwithahigherriskofdiscontinuationdueto toxicity, more patients in the HCQ are able to complete their therapy,possiblymakingitmoreeffective.
Conflictofinterest
Westateonbehalfofalltheauthorsthattherearenoknown competingfinancialinterests,orpersonalrelationshipsthatcould haveappearedtoinfluencetheworkreportedinthispaper. Fundingsource
TherewerenoFundingSourcesorStudySponsorsinvolvedin thisstudy.s
Ethicalapproval
TheMedicalEthicsReviewCommitteewaivedethicalApproval forthisstudy.
A.J.J.Lammers*
Isala,Zwolle,TheNetherlands R.M.Brohet DepartmentofEpidemiologyandStatistics,IsalaAcademy,Zwolle, TheNetherlands R.E.P.Theunissen C.Koster Isala,Zwolle,TheNetherlands R.Rood DiakonessenHospital,Utrecht,TheNetherlands D.W.M.Verhagen MedischCentrumJanvanGoyen,Amsterdam,TheNetherlands K.Brinkman OLVG,Amsterdam,TheNetherlands R.J.Hassing Rijnstate,Arnhem,TheNetherlands A.Dofferhoff CanisiusWilhelminaHospital,Nijmegen,TheNetherlands R.elMoussaoui MaasstadHospitalRotterdam,TheNetherlands G.Hermanides RodeKruisHospital,Beverwijk,TheNetherlands J.Ellerbroek ReinierdeGraafGasthuis,Delft,TheNetherlands N.Bokhizzou BovenIJHospital,Amsterdam,TheNetherlands H.Visser BeatrixHospitalGorinchem,TheNetherlands M.vandenBerge AdmiraaldeRuiterHospital,Goes,TheNetherlands H.Bax ErasmusMCRotterdam,TheNetherlands D.F.Postma UniversityMedicalCenterGroningen,Groningen,TheNetherlands P.H.P.Groeneveld Isala,Zwolle,TheNetherlands *Correspondingauthorat:IsalaHospital,DepartmentofInternal MedicineandInfectiousDiseases,DrvanHeesweg2,8025AB Zwolle,TheNetherlands. E-mailaddress:a.j.j.lammers@isala.nl(A.Lammers). Received3December2020
LettertotheEditor InternationalJournalofInfectiousDiseases103(2021)478–479