Int J Lab Hem. 2017;1–10. wileyonlinelibrary.com/journal/ijlh  | 1© 2017 John Wiley & Sons Ltd

Received:28March2017  |  Accepted:15July2017DOI: 10.1111/ijlh.12728

O R I G I N A L A R T I C L E

Automated cell counts on CSF samples: A multicenter performance evaluation of the GloCyte system

E. A. Hod1,2 | C. Brugnara3 | M. Pilichowska4 | L. M. Sandhaus5 | H. S. Luu6 |  S. K. Forest1,2 | J. C. Netterwald1 | G. M. Reynafarje2 | A. Kratz1,2

1DepartmentofPathologyandCellBiology,ColumbiaUniversityCollegeofPhysiciansandSurgeons,NewYork,NY,USA2NewYork-PresbyterianHospital,NewYork,NY,USA3DepartmentofLaboratoryMedicine,BostonChildren’sHospital,andHarvardMedicalSchool,Boston,MA,USA4TuftsMedicalCenter,Boston,MA,USA5UniversityHospitals,ClevelandMedicalCenter,Cleveland,OH,USA6Children’sMedicalCenterDallas,Dallas,TX,USA

CorrespondenceAlexanderKratz,NewYork-PresbyterianHospital,CoreLaboratory,NewYork,NY,USA.Email:ak2651@cumc.columbia.edu

Funding informationThisstudywasfundedbyAdvancedInstruments

AbstractObjectives:Automated cell countershave replacedmanual enumerationof cells inbloodandmostbodyfluids.However,duetotheunreliabilityofautomatedmethodsatverylowcellcounts,mostlaboratoriescontinuetoperformlabor-intensivemanualcountsonmanyorallcerebrospinalfluid(CSF)samples.ThismulticenterclinicaltrialinvestigatediftheGloCyteSystem(AdvancedInstruments,Norwood,MA),arecentlyFDA-approvedautomatedcellcounter,whichconcentratesandenumeratesredbloodcells(RBCs)andtotalnucleatedcells(TNCs),issufficientlyaccurateandpreciseatverylowcellcountstoreplaceallmanualCSFcounts.Methods:TheGloCyteSystemconcentratesCSFandstainsRBCswithfluorochrome-labeled antibodies and TNCs with nucleic acid dyes. RBCs and TNCs are thencounted by digital image analysis. Residual adult and pediatric CSF samples ob-tainedforclinicalanalysisatfivedifferentmedicalcenterswereusedforthestudy.CellcountswereperformedbythemanualhemocytometermethodandwiththeGloCyteSystem following the sameprotocol at all sites. The limits of theblank,detection,andquantitation,aswellasprecisionandaccuracyoftheGloCyte,weredetermined.Results:TheGloCytedetectedasfewas1TNC/μLand1RBC/μL,andreliablycountedaslowas3TNCs/μLand2RBCs/μL.Thetotalcoefficientofvariationwaslessthan20%.Comparisonwithcellcountsobtainedwithahemocytometershowedgoodcor-relation(>97%)betweentheGloCyteandthehemocytometer,includingatverylowcellcounts.Conclusions:TheGloCyteinstrumentisaprecise,accurate,andstablesystemtoob-tain redcell andnucleatedcell counts inCSFsamples. It allows for theautomatedenumerationofevenverylowcellnumbers,whichiscrucialforCSFanalysis.TheseresultssuggestthatGloCyteisanacceptablealternativetothemanualmethodforallCSFsamples,includingthosewithnormalcellcounts.

K E Y W O R D S

automatedcellcounters,cerebrospinalfluidtesting,laboratorytesting,redcellcounts,whitecellcounts

2  |     HOD et al.

1  | INTRODUCTION

Until the invention of automated cell counters, hematology labora-tories usedmanual counting chambers to enumerate cells in bloodand body fluids.1 Such manual methods have inherent limitations,especiallyfor lowcellcounts;forexample,atacellcountof5cells/μL, the coefficient of variation (CV) of aNeubauer hemocytometer(HausserScientific,Horsham,PA)is45%.2,3WiththedevelopmentoftheCoultercounterandotherautomatedplatforms,mostlaboratoriesquicklyadoptedautomatedcellcountersfortheenumerationofbloodcells.1 Cell counts in body fluids remainedmanual for severalmoredecades,andonlyinrecentyearshavemanufacturersaddedreliablebodyfluidmodestotheirinstruments.4-12

Today,mostautomatedcellcounterscanprovidereasonablyac-curatecellcountsofalmostallbodyfluids.However,cellularanalysisofcerebrospinalfluid(CSF)samplescontinuestopresentuniquechal-lenges, because reference ranges for total nucleated cells (TNCs) inCSFare0-5cells/μLand“normal”CSFsamplesshouldnotcontainanyredbloodcells (RBCs).2Todistinguishbetweena“normal”CSFwithlessthan5TNCs/μLand0RBCs/μLandan“abnormal”CSFwithele-vatedTNCsand/orRBCs,automatedcellcountersmustbeextremelyaccurateandpreciseatverylowcellcounts.

Several instrumentplatformson themarketofferquantitationofRBCs and TNCs in CSF samples. TheADVIA120i (SiemensMedicalSolutions,Malvern,PA)hasbeenshowntoallowreliableTNCenumer-ationinCSFatverylowcellcounts.However,thepresenceofhighRBCcountscaninfluencetheaccuracyoftheTNCcount.13-15Intheevalu-ationoftheMindrayBC-6800(MindrayNorthAmerica,Mahwah,NJ)bodyfluidmodeincerebrospinalfluid,Buoroetalconcludedthattheinstrumentprovidedaneffectivealternative tomanual cell counts.16 However, they recommendedamicroscopic reviewof all cell countsbetween 4.0 and 7.0cells/μL. The Sysmex XT-4000i and XE-5000cell counters (SysmexAmerica,Lincolnshire, IL)had limitedprecisionat<20TNCs/μL.10,17ThenewerSysmexXN-1000hasalowerlimitofquantitationof5TNCs/μL,andtheminimalreportablenumberofRBCsis1000RBCs/μL.18Recently,Flemingandcolleagueshavedescribedahigh-sensitivityanalysis (hsA) researchmodeon theXN-1000sys-tem.19Thisapplicationhasalowerlimitofquantitationof10RBCs/μLand2TNCs/μL.19,20Buoroandco-workershaveinvestigatedtheuseoftheSysmexUF-1000iBodyFluidModeandfoundgoodcorrelationwithmanualTNCenumeration,withamodestoverestimationofcountsbelow30TNCs/μL.21TheuseoftheIrisiQ200BodyFluidModuleforCSFanalysis (Beckman-Coulter, Indianapolis, IN)was investigatedbyGoubardandcolleaguesaswell asWalkerandco-workers.22,23Bothgroupsfoundgoodcorrelationbetweenthemanualmethodandthere-sultsoftheiQ200.TheCVswereslightlyhigherontheIristhanonthehemocytometer.Finally,Glasserandcolleaguescomparedthe iQ200andtheLH750(Beckman-Coulter)tomanualcountsandfoundunac-ceptableratesoferroratlowercellcounts23(Table1).

Becausetheaccuracyandprecisionat lowCSFcellcounts issoclinically important,andbecausemostautomatedcellcountersmaynothavesufficientaccuracyandprecisionat lowcounts,many lab-oratories still regardmanual chamber counts as the “gold standard”

to obtain cell counts on CSF samples, especially on clear (non-bloody)specimens.20,24TheGloCyteAutomatedCellCounterSystem(AdvancedInstruments,Norwood,MA)isanewplatformthatconcen-tratesandenumeratesTNCsandRBCsusingfluorescentmicroscopyanddigitalimageanalysis.25,26Thepurposeofthismulticenterstudywastodetermine ifGloCyteRBCandTNCcountsweresufficientlyaccuratetoreplacemanualCSFcellcounts.

2  | MATERIALS AND METHODS

2.1 | The GloCyte system

The GloCyte is an automated cell counter system. It concentratesTNCsandRBCsinCSFsamplesandenumeratesthem;30μLofsam-pleand30μLofreagentcontainingfluorochrome-labeledantibodiesagainsthumanRBCsaredispensedintoa0.5-mLtube.Another30μLofsampleand30μLofreagentcontainingnucleicaciddyewithde-tergenttostainTNCsaredispensedintoasecond0.5-mLtube.Aftermixing,thetwosamplesaretransferredintoseparatecartridges.TheTNC stained sample is transferred immediately after mixing, whiletheRBCstainedsampleistransferredaftera3-minuteincubationinthetube.Vacuumsuctionisappliedtothecartridges,removingallliq-uidandpullingallcellsontoamembrane.Asemiconductorlaserandadigitalimageanalysissystemthencaptureandenumeratethefluores-centcellimages.Nomanualcellcountingisnecessary.Theinstrumentdoesnotprovideadditionalqualitativeparametersforthecellcount.Noinstrumentcalibrationisnecessary.

The samples are contained in adisposable cartridgeanddonotcomeincontactwiththeinstrument,eliminatingthepossibilityofcar-ryoverandallowingsafehandlingofpotentiallyinfectiousspecimens.

TABLE  1 PerformanceofpresentlyavailableautomatedCSFcellcountingmethods

InstrumentPerformance Issues in counting cells in CSF References

Advia120i PresenceofhighRBCcountscaninterferewiththeaccuracyoftheWBCcount

15,16

MindrayBC-6800 Microscopicreviewofallcellcountsbetween4.0and7.0recommended

17

SysmexXT-4000iandXE-5000

LimitedPrecisionatWBCcountsof<20cells/μL

11,18

SysmexXN-1000 MinimalreportablenumberofRBCsis1000cells/μL

19

SysmexXN-1000high-sensitivitymode

LowerLimitofquantitationofRBCsis10cells/μL;notFDAclearedintheUS

20,21

SysmexUF-1000i ModestoverestimationoftheWBCcountsbelow30cells/μL

7

iQ200 Unacceptableratesoferroratcountsofless50cells/μL

24

LH750 Unacceptableratesoferroratcountsofless200cells/μL

24

     |  3HOD et al.

Wedidnotexperienceanyinstrumentmalfunctionduringthestudythatledtosampleloss.

2.2 | Manual method

Dependingoneachclinicalsite’shemocytometermaterialpreference,manualcountswereperformedinduplicateoneitherglassNeubauerhemocytometers (Hausser Scientific, Horsham, PA) or disposableLevy-Neubauerhemocytometers (INCYTOC-Chip,Seoul,Korea)byadding10μLtoeachchamberandcountingall9squares.27 The mean ofthetwomanualcountswasusedinthestudy.

2.3 | Study protocol

The studywas performed at five sites: Boston Children’s Hospital,Boston, Massachusetts; Columbia University Medical Center, NewYork, New York; Tufts Medical Center, Boston, Massachusetts;UniversityHospitals,ClevelandMedicalCenter,Cleveland,Ohio;andtheUniversityofTexasSouthwesternMedicalCenter,Dallas,Texas.The studywasapprovedby the InstitutionalReviewBoardofeachsite.

ResidualCSFsamplesthathadbeenobtainedforclinicalpurposesand sent for analysis to the clinical laboratory at each participatingsitewereusedforthestudy.Sampleswereincludedinthestudyafterallclinicallynecessarytestinghadbeenperformed.OnlyCSFsamplesthatwere 4hours or less after collection and that still had at least300 μLofsampleremainingqualifiedforstudyinclusion.Thesampleswere de-identified and coded. The coded samples were then splitandanalyzedusingboththeGloCytetestmethodandthegoldstan-dardhemocytometermethod.Eachsamplewasanalyzedfourtimes:twiceusingthehemocytometermethodfollowedbytworunsontheGloCyte.Everysiteanalyzedsamplesfromitsownpatientpopulation.

2.4 | Performance parameters

2.4.1 | Limit analysis

WefollowedCLSIEP17-A2forthedeterminationoftheLOB,LOD,andLOQ.28LoBtestingwasperformedintheAdvancedInstrumentslaboratorywithpurchasedcell-freehumanCSF.LoDtestingwasper-formed at TuftsMedical Center due to availability of CSF sampleswithlowcellcounts.LoQwascalculatedfromLoBandLoD.

• Limit of the Blank (LOB):TheLOBisthehighestnumberofcellsex-pectedtobemeasuredinablanksample.28-30TheLOBwasdefinedas the 95th percentile ofmeasurementsmade on blank samples(samples with no cells detected with the hemocytometer); 240countswereperformedforeachcelltype,withfiveblanksamples,usingtwoGloCytedevices.Countsweremadeover3daysforeachinstrument,withtworeagent lotsforeachblanksampleandfourreplicatesperreagentlotperday.

• Limit of Detection (LOD):TheLOD is the smallestnumberof cellsthatcanreliablybedetectedasdifferentfromzero;itwasdefined

asthevaluewhereatleast95%ofthemeasuredcountsfellabovetheLOB.LODtestingwasperformedusingCSFspecimenswithlowcellcounts(1-2cells/μL).240countswereperformedforeachcelltypeusingtwoGloCytedevices.SixspecimensweretestedforTNCand6weretestedforRBC.Eachspecimenwastested10timeswithonereagentlotandonecartridgelotand10timeswithasecondreagentlotandasecondcartridgelot.

• Limit of Quantitation (LOQ):TheLOQis the lowestcellcount inasamplethatcanbothbereliablydetectedandalsomeetsetguide-linesforprecisionandbias; itwasdefinedasthe lowestcountatwhichtheGloCyteprovidesquantitativemeasurementswithatotalerrorof20%orless.ThiscutoffwasusedbecausesurveysbytheCollegeofAmericanPathologists (CAP) indicateCVsof20%-40%formanualchambercounts.4,31TheLOQwasdeterminedutilizingthesamesamplesusedtodeterminetheLOD,sixsampleswithlowcell counts (LODsamples) for each cell type.The total error (TE)methodwasused in accordancewithCLSIEP17-A2.28 For eachsamplebyinstrumentandreagentlot,thereplicatedmeasurementswere used to estimate the total error (TE) and the percent totalerror (%TE).The“truth”wasassumedtobethemanualreferencecount foreachsample.Therootmeansquare (RMS)approachtocalculatingTEwasemployedutilizingthe%TEoftheLODcounts:

Biaswasdefinedas thedifferencebetween theautomatedcountandthegoldstandardhemocytometermethod.BasedonthedatafromtheassayrunsdetailedfortheLODexperiments,ascatterplotwasgen-eratedwith%TEontheverticalaxisandthemanualreferencecountonthehorizontalaxis.Alinearregressionlinewasfittothe%TEdata.TheLOQwasthendeterminedasthemanualcountforwhichtheaccuracymettheprespecifiedlevelof20%TE.32

2.4.2 | Receiver- operating curve analysis

Receiver-operatingcurveanalysis(ROC)analysiswasconductedusingamanualcountof5cells/μLasthecutoffbetweennormal/abnormal.Anysampleswithamanualcountoffiveorlesswereassignedzero;allotherswereassignedone.AtallGloCyteTNClevels,sensitivityand1-specificitywerecalculatedandplotted,resultingintheROCcurvepresented.

2.4.3 | Precision

• Intrarun Precision: Intrarun precision studies were performed atthreesites,using14samplesrepresentingtheclinically importantrangefortotalnucleatedcells(TNCs)and16samplesrepresentingtheclinicallyrelevantrangeforredbloodcells(RBCs),aswellas6TNCand6RBCGloCytecontrols.Eachsampleandcontrolwererun10timesbyasingleoperatorforRBCand/orTNCcounts.

• Inter-run Precision: Reproducibility was evaluated at three clin-ical sites over 20days. The sites were selected based on theavailability of resources at the sites to perform testing. Testing

TE=

√

bias2+stddev

2and%TE=

TE

manual ref count×100%

4  |     HOD et al.

wasperformed twicedaily, using the same sets of controls, for20days.Eachcontrolsetwastestedinduplicate,independentlybytwooperators,ateachsite;160controlsweretestedforeachlevelofthetwocontrolsateachsite.Thetotalnumberofcontrolstestedateachsitewas640.Standarddeviationsand%CVwerecalculatedforeachfactorofinterest(residual,run,day,site,andoperator).Aprespecifiedacceptancecriterionforthisassaywassetat20%.33

2.4.4 | Method comparison (Accuracy)

Duetothelowprevalenceofabnormalsamples,asamplingplanwasdevelopedtoensurecollectionofadequatenumbersofbothnormalandabnormalsamples.EachclinicaltrialsiteinitiallytestedbothTNCand RBC for each specimen. Once a specified number of samplesweretestedforagivenresultrange,eitherTNCorRBCcountsweretested, as needed, so that a valid statistical analysis could be per-formed.Forexample,bytheendofthestudy,sampleswithhighTNCandlowRBCcountsweretestedforTNCcountsonly,asasufficientnumberof lowRBCcountshadalreadybeencollected.TherewerethereforedifferencesinthenumberofspecimenstestedforTNCandRBCcounts.

Accuracy studies were performed with 321 samples for TNCcounts(203adultspecimensand118pediatricspecimens)and422samples for RBC counts (243 adult specimens and 179 pediatricspecimens). Each sample was counted twice manually and twiceusingtheGloCyteSystem.Formethodcomparison,themeanofthetwomanualcounts,bothperformedbythesameoperator,wasusedasthegoldstandard.OnlythefirstGloCytecountwasusedfordataanalysis. Inadditionto theclinical samples,manipulatedCSFsam-ples(31samplesforTNCand41samplesforRBC)werecreatedbydilutinghumanTNCandRBCintopooledblankhumanCSF.Thesemanipulatedsampleswerenecessary toprovidedataat theupperend of the reportable range. Separate pediatric samples (pediatricvenous blood and pediatric CSF) and adult samples (adult venousbloodandadultCSF)werecreatedforseparatecollectionofpedi-atricandadultdata.Twomanual countswereperformedwith theNeubauer hemocytometer, and two countswere performed usingtheGloCyte.

CUSUMstatisticswereused to test theassumptionof linearityforthesamples.AP-valueoflessthan.05wouldhaveindicatedthattherewasasignificantdeviationfromlinearitybetweenthehemocy-tometerandtheGloCytemethod;asallP-valuesinthisanalysisweregreater than .05, therewasnoevidenceagainst theassumed linearrelationship and Passing-Bablok analysiswas appropriate to deter-minebiasbetweenresultsfromthehemocytometerandtheGloCyteSystem.

2.4.5 | Linearity/Reportable range

LinearitywasdeterminedusingmanipulatedCSFsampleswithcon-centrationsspanningthelinearrangeoftheinstrument.Atotalof14

concentration levels ranging from0 to 8000TNCs/μL and 15 con-centration levels ranging from 0 to 800000RBCs/μL were testedonthreeGloCyteinstrumentsbythreeoperators.Sampleswererunintriplicateateachlevel,resultingin126TNCcountsand135RBCcounts.

2.5 | Statistical analysis

Rversion3.3.0(2016-05-03)wasusedforLimitofBlankHistograms,andSPSS(IBM,Armonk,NY)wasusedformixed-modelanalysisforprecision studies. Mixed-model analysis was confirmed with SAS® statisticalsoftwareversion9.4(SASInstitute,Cary,NC)andMinitab(MinitabInc.,StateCollege,PA).SASsoftwarewasusedforallotheranalysis.

3  | RESULTS

3.1 | Limits of blank, detection, quantitation

The LOB for the GloCyte instrument was 0.73cells/μL for RBCsand 0.47cells/μL for TNCs. Thus,when performing cell counts onsampleswithoutcells,theGloCyteSystemprovidesacountof lessthan 1cell/μL. The LOD for theGloCyte Systemwas 0.8RBCs/μLand1.2TNCs/μL,indicatingthattheinstrumentisabletodistinguishsampleswithapproximately1cell/μLfromsampleswithlessthan1cell/μL. Finally, the LOQ for the newdevicewas 2.0RBCs/μL and2.6TNCs/μL(Figure1), indicatingthat it isabletoreliablyandpre-cisely enumerate cells in samples with at least 2RBCs/μL and/or3TNCs/μL(Table2).

3.2 | Precision

For the inter-run precision, the total CV for each control level(N=480),includingwithin-run,between-run,between-day,between-site,andbetween-operatorvariation,was<12%.Thelargestcontribu-tortothetotalCVwasthewithin-runCVof10.1%(Table3).Fortheintrarunprecision,thetotalCVof14TNCand16RBCpatientsam-pleswas<18%(Table4).AllsampleshadCVsbelow20%,whichisthestandardacceptancecriterionforthistypeofassay.4,31SampleswithlowcellcountsalsoshowedlowCVs,indicatinggoodrepeatabilityatlowcellcounts.

3.3 | Accuracy

For adult TNC counts, differences under 4.0% were observed in95% of counts up to 10000cells/μL (Figure 2). For pediatric TNCcounts,differencesupto6.8%wereobserved in95%ofcountsupto 8000cells/μL. The differences between hemocytometer countsandGloCyteenumerationsaremuchsmalleratlowercellcounts,withbothadultandpediatricsamplesshowingdifferencesgroupedcloselyaroundzerowhenmanualcountsare<100cells/μL.Theseverylowcountsincludethedecisionpointsusedbyclinicianswheninterpret-ing CSF results and are therefore key for a correct diagnosis and

     |  5HOD et al.

treatmentplan.ForRBCs,differencesbetweentheGloCyteSystemandchambercountswerelargerthanforTNCs.However,thepercentdifferencewasstillsmall(5.6%),asthesemeasurementsrangeintothe900000s. Like forTNCs,differencesbetweenmanual andGloCytecountswereclosetozeroatlowcounts.

Passing-BablokanalysisshowednosignificantbiasofGloCyteTNCcountsvsthemanualmethod(Figure2andTable5).Thedatashowedequivalencebetweenthetwomethodsupto9900TNCs/μL for samples from adults and 7672TNCs/μL for pediatric sam-ples (Figure 3). Manual adult RBC counts were interchangeablewith GloCyte counts up to 901250RBCs/μL. Therewas a smallproportionalbiasinpediatricRBCcounts,undercountingRBCsbyapproximately9%upto817500RBCs/μL.Separateanalysisofpe-diatric samples fromgeneralhospitals (whichanalyzedbothadultandpediatricsamples)showednobias,indicatingthattheGloCytemethodcouldbeusedinterchangeablywiththemanualmethodinthese samples. Samples collected at dedicatedpediatric hospitalsshowed a 13% bias on theGloCyte Systemvs the hemocytome-ter.Thisindicatedthatthedifferencemaybeduetodifferentpro-cedures forhandlingsamples inpediatrichospitalsasopposed togeneralhospitals.Thebiaswasnotpresentatmanualcountsbelow100RBCs/μL,whichincludestheclinicaldecisionthresholdsusedbyclinicians.AsthepresenceofanyRBCsinCSFisconsideredab-normal,biasinhighercountsshouldnotbeanissue,andforlowercounts,GloCytecountswereequivalenttomanualcounting.Bland-Altman plots also showed high accuracy and no significant bias(Figure3).

3.4 | Clinical applications

A cutoff or 5 cells was used for ROC analysis and for sensitivity,specificity, positive predictive value (PPV), and negative predictivevalue(NPV)determination(Table6).AtallGloCyteTNClevels,sen-sitivityand1-specificitywerecalculatedandplotted,resultingintheROC curve presented in Figure4. The area under the curve (AUC)was 0.985, indicating strong agreementwith themanual diagnosis.Sensitivity, specificity, positive predictive value (PPV), and negativepredictivevalue(NPV)wereallgreaterthan0.92,whichalsoindicatesstrongagreement (Table6).Table4 shows the intrarunprecisionattheclinicallyrelevantTNClevelsof5to20cells/μL,andthelowestlevelsofRBCstested.

Cell type Reagent lot LOB: Count/μLLOD: Manual Count/μLa

LOQ: based on total error <20%, assuming manual count (gold standard) as truth

RBC 1 0.33cells/μL 0.6cells/μL 2.0cells/μL

2 0.73cells/μL 0.8cells/μL

Combined 0.53cells/μL 0.8cells/μL

TNC 1 0.33cells/μL 0.7cells/μL 2.6cells/μL

2 0.47cells/μL 1.2cells/μL

Combined 0.40cells/μL 1.2cells/μL

LOB,Limitoftheblank;LOD,limitofdetection;LOQ,limitofquantitation.a100%oftheGloCytecountsweregreaterthantheLOB.

TABLE  2 Limitoftheblank,limitofdetection,andlimitofquantitationoftheGloCytesystem

F IGURE  1 ScatterplotsofLimitofDetection(LOD)Experiments.LimitofQuantitation(LOQ)wasdeterminedasthemanualcountforwhichaccuracymetthelevelof20%totalerror

Tota

l Err

or %

0

20

40

60

80

100

120

140

160

180

200

220

RBC manual reference count (cells/uL)

LOQ Determination for RBC based on LOD samples

LOQ Determination for TNC based on LOD samples

Tota

l Err

or %

0

20

40

60

80

100

TNC manual reference count (cells/uL)

0.00 0.33 0.67 1.00 1.33 1.67 2.00 2.33

0.00 0.33 0.67 1.00 1.33 1.67 2.00 2.33 2.66 3.00

(A)

(B)

6  |     HOD et al.

4  | DISCUSSION

AccurateCSFcellcountsarevitalforthediagnosisofseveraldiseases,includingsubarachnoidhemorrhage,meningealinfection,demyelinat-ingdisease,andcentralnervoussystemmalignancies.Thesesamplesrepresentasignificantpercentageofbodyfluidcellcountsperformedinhematologylaboratories.Untilrecently,manymanufacturersofau-tomatedcellcounterswereunabletoachievesufficientprecisionandaccuracyat lowcell counts tovalidate their instruments for clinicaluseforclear (nonbloody)CSFsamples.Flemingandco-workersde-scribedahigh-sensitiveanalysis(hsA)researchmodeontheSysmexXN-1000(Sysmex,Kobe,Japan)specificallyforcountingcellsinflu-idsthatcontain lowcellcounts.19Thelower limitofquantitationofthismethodis10RBCs/μLand2TNCs/μL;thus,thismethodcannotquantitateRBCsaroundthe“normal”CSFrangeofzeroRBCs.UntiltheFDA510Kclearanceof theGloCyteSystem,noautomatedcellcounterwasvalidatedandFDA510Kclearedforverylowcellcounts;thus, laboratorieshadtousemanualchambercountsformanyCSFsamples. Sandhaus and colleagues performed CSF cell counts withtheGloCyteandthehemocytometeronsamplesfrompatientswithacuteleukemia,lymphoma,malignantneoplasmsinvolvingthebrain,meningitis,andencephalitis;resultscorrelatedverywellbetweenthemanualandtheautomatedmethod.25

Thestrengthsofthisstudyaretheparticipationofmultiplemed-icalcenters,makingitpossibletoobtainlargenumbersofsamplesandspanningtheentiretyoftheclinicallysignificantmeasurementranges, including extremely low cell counts,which are crucial forCSF samples.The standardized studyprotocol ensured consistentdatacollectionatallsites.Thismulticenterstudyatfivemajormedi-calcentersshowsthattheGloCyteAutomatedCellCounterSystem

TABLE  4  Intrarunprecisionattheclinicallyrelevantthresholdsof5,10,and20TNCs/μLandat5and10RBCs/μL

Cell type Sample N Mean Median

Standard Deviation %CV

TNC 128 10 7.80 8.00 0.92 11.78

129 10 5.50 5.50 0.85 15.45

199 10 7.90 8.00 0.99 12.59

300 10 4.10 4.00 0.74 18.00

303 10 17.90 18.00 1.10 6.15

306 10 4.90 5.00 0.57 11.58

310 10 11.70 12.00 0.95 8.11

L1103 10 9.80 10.00 0.92 9.38

L1103 10 10.60 11.00 0.84 7.96

L1105 10 10.30 10.50 0.95 9.21

RBC 126 10 5.00 5.00 0.82 16.33

195 10 7.10 7.00 0.99 14.01

307 10 6.30 6.00 0.67 10.71

L1103 10 10.00 10.00 1.05 10.54

L1105 10 10.30 10.50 0.95 9.21

TABLE 3 Overallinterassayprecisionsummarystatistics

With

in ru

nBe

twee

n ru

nBe

twee

n da

yBe

twee

n si

teBe

twee

n op

erat

orTo

tal

Cell

type

Cont

rol

NM

ean

(cel

ls/μL

)SD

%CV

SD%

CVSD

%CV

SD%

CVSD

%CV

SD%

CV

TNC

Low

480

10.56

1.07

10.1

10.

000.

000.

171.58

0.41

3.86

0.25

2.40

1.18

11.2

0

High

480

122.89

7.30

5.94

0.00

0.00

0.00

0.00

3.81

3.10

1.80

1.46

8.43

6.86

RBC

Low

480

11.25

1.03

9.19

0.00

0.00

0.21

1.86

0.43

3.78

0.30

2.68

1.18

10.4

6

High

480

130.

006.

835.25

0.00

0.00

0.87

0.67

2.24

1.72

1.33

1.02

7.36

5.66

     |  7HOD et al.

providesresultsinterchangeablewiththecountsobtainedbytrainedtechnologistswith a hemocytometer.With a CV below 20%, theGloCyte’sprecisionmet thestandardacceptancecriterion for thistypeofassay.20ThecorrelationbetweenGloCyteandhemocytom-eterresultswasexcellent.ThesefindingsindicatethattheGloCyte

Systemcanprovide resultswith thesameclinicalvalueasmanualmethodsevenforextremelylowcellcountsthatarewithinornearthereferenceranges.Theseresultscanbeobtainedwithinapprox-imately5minutes,withouttheneedforstaffwhohavethepatternrecognitionskillsnecessaryforcellidentification.Thisshouldallow

F IGURE  2 Passing-Bablok(PB)Correlationplotsfrommethodcomparisonforall(A)adultTNC,(B)pediatricTNC,(C)adultRBC,and(D)pediatricRBCcounts

0

0

200 000

200 000

400 000

400 000

600 000

600 000

800 000

800 000

Passing-Bablok RBC: Adult

Manual count cells/µL

Glo

Cyt

e co

unt c

ells

/µL

0

0

200 000

200 000

400 000

400 000

600 000

600 000

800 000

800 000

Passing-Bablok RBC: Pediatric

Passing-Bablok TNC: Adult Passing-Bablok TNC: Pediatric

Manual count cells/µL

Glo

Cyt

e co

unt c

ells

/µL

Manual count cells/µL

Glo

Cyt

e co

unt c

ells

/µL

Manual count cells/µL

Glo

Cyt

e co

unt c

ells

/µL

(A) (B)

(C) (D)

Passing- Bablok analysis

Cell type

Age Group

CUSUM test for linearity (P- value) Parameter Estimate

95% Confidence Interval

TNC Adult .524 Slope 1.000 (1.00,1.00)

Intercept 0.000 (0.00,0.00)

Pediatric .415 Slope 0.963 (0.91,1.00)

Intercept 0.037 (0.00,0.18)

RBC Adult .107 Slope 1.000 (0.99,1.01)

Intercept 0.000 (0.00,0.01)

Pediatric .788 Slope 0.910 (0.89,0.93)

Intercept 0.000 (-0.05,0.06)

TABLE  5 ResultsofPassing-Bablokanalysis

8  |     HOD et al.

0 20 40 60 80 100Manual RBC

–40

–20

0

20

Man

ual R

BC

- G

loC

yte

RB

C 1.96SD = 13.5

–1.96SD = –12.4

Mean = 0.5

Pediatric: RBC

0 200 000 400 000 600 000 800 000Manual RBC

–200 000

–100 000

0

100 000

Man

ual R

BC

- G

loC

yte

RB

C 1.96SD = 66764.4

–1.96SD = –66307.6

Mean = 228.4

Pediatric: RBC

0 20 40 60 80 100Manual TNC

–40

–20

0

20

40

Man

ual T

NC

- G

loC

yte

TNC

1.96SD = 11.9

–1.96SD = –12.7

Mean = –0.4

Pediatric: TNC

0 2000 4000 6000 8000Manual TNC

–1000

0

1000

2000

Man

ual T

NC

- G

loC

yte

TNC

1.96SD = 1096.3

–1.96SD = –876.1

Mean = 110.1

Pediatric: TNC

0 20 40 60 80 100Manual RBC

–200

–100

0

Man

ual R

BC

- G

loC

yte

RB

C

1.96SD = 44

–1.96SD = –50.4

Mean = –3.2

Adult: RBC

0 200 000 400 000 600 000 800 000Manual RBC

–100 000

0

100 000

200 000

300 000

Man

ual R

BC

- G

loC

yte

RB

C

1.96SD = 84475.2

–1.96SD = –76655.5

Mean = 3909.8

Adult: RBC

0 20 40 60 80 100Manual TNC

–100

–50

0

Man

ual T

NC

- G

loC

yte

TNC 1.96SD = 18.5

–1.96SD = –20.9

Mean = –1.2

Adult: TNC

0 2000 4000 6000 8000 10 000Manual TNC

–1000

0

1000

Man

ual T

NC

- G

loC

yte

TNC

1.96SD = 902.5

–1.96SD = –1022.3

Mean = –59.9

Adult: TNC

(A) (B)

(C) (D)

(E) (F)

(G) (H)

F IGURE  3 Bland-AltmanPlotsforPediatricandAdultSamples.Todemonstrateaccuracyandlackofsignificantbiasatlowcellcounts,separategraphsareprovidedforcountsbetweenzeroand100cells/microliter(A,C,E,G)andforallcellcounts(B,D,F,H).Resultsarealsoshownforpediatric(A,B,C,D)andadult(E,F,G,H)specimensseparately,aslabeled

     |  9HOD et al.

fortheabilitytoobtainCSFcellcountsonallshifts,includingnightsandholidays.

Ithasrecentlybeensuggestedthathemocytometersmaysoonbeobsolete intheclinical laboratory.20TheavailabilityoftheGloCyteSystem,with the ability to provide accurate andprecise counts atextremely low cell numbers,maymake that prediction a reality inthenearfuture.ThisstudyraisesthehopethattheGloCyteSystem,incombinationwith thebody fluidmodesofotherautomatedcellcountersanddigitalimageanalysis,willallowlaboratoriestodiscon-tinue allmanual cell countingmethods.However, until automatedcell counters can reliably classify all cells that are present in CSFsamples, includingneoplasticcells,therewillstillbeaneedformi-croscopicexamination,eitherbymanualmicroscopyordigitalimageanalysis.

ACKNOWLEDGEMENTS

TheauthorsthankDr.GeorgeHongandJulieMacKenzie(AdvancedInstruments, Norwood, MA) for their input and comments on thepaper,andWarrenHinkle (BioStatSolutions, Inc,FrederickMD)forprovidingthestatisticalanalysis.

REFERENCES

1. Green R, Wachsmann-Hogiu S. Development, history, and futureof automated cell counters. In: KratzA, Brugnara C, eds.Clinics in Laboratory Medicine 35.Philadelphia,PA:Elsevier;2015:1-10.

2. KarcherDS,McPhersonRA.Cerebrospinal,synovial,serousbodyflu-ids, and alternative specimens. In:McPhersonRA,PincusMR, eds.Henry’s Clinical Diagnosis and Management by Laboratory Methods,22ndedn.Philadelphia,PA:Elsevier;2011:480-506.

3. Danise P,MaconiM, RovettiA, et al. Cell counting of body fluids:comparison between three automated haematology analysers andthemanualmicroscopemethod.Int J Lab Hematol.2013;35:608-613.

4. Sandhaus LM. Body fluid cell counts by automated methods. In:Brugnara C, KratzA eds.Automated Hematology Analyzers: State of the Art. Clinics in Laboratory Medicine. Philadelphia, PA: Elsevier;2015:93-103.

5. BoerK,DeufelT,ReinhoeferM.Evaluationof theXE-5000for theautomatedanalysisofbloodcellsincerebrospinalfluid.Clin Biochem. 2009;42:684-691.

6. BuoroS,ApassitiEspositoS,AlessioM,CrippaA,OttomanoC,LippiG.Automatedcerebrospinalfluidcellcountsusingthenewbodyfluidmodeofsysmexuf-1000i.J Clin Lab Anal.2015;30:381-391.

7. ParisA,NhanT,CornetE,PerolJP,MaletM,TroussardX.Performanceevaluation of the body fluid mode on the platform Sysmex XE-5000 series automated hematology analyzer. Int J Lab Hematol. 2010;32:539-547.

8. YangD,ZhouY,ChenB.Performanceevaluationandresultcompari-sonoftheautomatedhematologyanalyzersAbbottCD3700,SysmexXE2100andCoulterLH750forcellcountsinserousfluids.Clin Chim Acta.2013;419:113-118.

9. AulesaC,MainarI,PrietoM,CobosN,GalimanyR.UseoftheAdvia120hematologyanalyzerinthedifferentialcytologicanalysisofbio-logical fluids (cerebrospinal,peritoneal,pleural,pericardial, synovial,andothers).Lab Hematol.2003;9:214-224.

10. LiA,GronlundE,BrattsandG.AutomatedwhitebloodcellcountsincerebrospinalfluidusingthebodyfluidmodeontheplatformSysmexXE-5000.Scand J Clin Lab Invest.2014;74:673-680.

11. BuoroS,AppassitiEspositoS,VavassoriM,etal.ReflextestingrulesforcellcountanddifferentiationofnucleatedelementsinpleuralandasciticfluidsonSysmexXE-5000.J Lab Autom.2016;21:297-304.

12. BuoroS,MeccaT,AzzaraG,etal.CellPopulationDataandreflextesting rules of cell analysis in pleural and ascitic fluids usingbodyfluidmodeonSysmexXN-9000.Clin Chim Acta.2016;452: 92-98.

13. BremellD,MattssonN,WallinF,etal.Automatedcerebrospinalfluidcellcount–newreferencerangesandevaluationofitsclinicaluseincentralnervoussysteminfections.Clin Biochem.2014;47:25-30.

14. TanadaH,IkemotoT,MasutaniR,TanakaH,TakuboT.Evaluationofthe automated hematology analyzerADVIA(R) 120 for cerebrospi-nal fluid analysis and usage of unique hemolysis reagent. Int J Lab Hematol.2014;36:83-91.

15. MahieuS,VertessenF,VanderPlankenM..EvaluationofADVIA120CSFassay(Bayer)vs.chambercountingofcerebrospinalfluidspeci-mens.Clin Lab Haematol.2004;26:195-199.

16. Buoro S, SeghezziM,MeccaT,VavassoriM, CrippaA, LaGioiaA.EvaluationofMindrayBC-6800bodyfluidmodeforautomatedcere-brospinalfluidcellcounting.Clin Chem Lab Med.2016;54:1799-1810.

17. Kleine TO, Nebe CT, Lower C, GeilenkeuserWJ, Dorn-Beineke A.Cellanalysisincerebrospinalfluid(CSF)usingSysmex(R)hematology

TABLE  6 Sensitivity,specificity,positivepredictivevalue,andnegativepredictivevalueoftheGloCyteSystem

Frequency of GloCyte counts

Cell type ≤5 >5 TP TN FN FP Sensitivity Specificity PPV NPV

TNC 176 176 163 164 12 13 0.931 0.927 0.926 0.932

Amanualcountof5wasusedasthecutoffbetweennormalandabnormal.NPV,NegativePredictiveValue;PPV,PositivePredictiveValue.

F IGURE  4 Receiver-operatingcharacteristic(ROC)curvefortotalnucleatedcells(TNC)

0.00.0

0.2

0.2

0.4

0.4Sen

sitiv

ity

1 - Specificity0.6

0.6

0.8

0.8

AUC = 0.9852

1.0

1.0

10  |     HOD et al.

analyzersXT-4000iandXE-5000:evaluationwithCSFcontrolsoftheJointGermanSocietyforClinicalChemistryandLaboratoryMedicine(DGKL).Cytometry A.2012;81:255-264.

18. Fleming C, Brouwer R, Lindemans J, de Jonge R.Validation of thebodyfluidmoduleonthenewSysmexXN-1000forcountingbloodcellsincerebrospinalfluidandotherbodyfluids.Clin Chem Lab Med. 2012;50:1791-1798.

19. FlemingC,RusscherH,BrouwerR,LindemansJ,deJongeR.EvaluationofSysmexXN-1000High-SensitiveAnalysis(hsA)ResearchModeforCountingandDifferentiatingCells inCerebrospinalFluid.Am J Clin Pathol.2016;145:299-307.

20. Sandhaus LM. Is theHemocytometerObsolete forBodyFluidCellCounting?Am J Clin Pathol.2016;145:294-295.

21. Buoro S, Apassiti Esposito S, Alessio M, Crippa A, OttomanoC, Lippi G. Automated Cerebrospinal Fluid Cell Counts UsingtheNewBody FluidMode of SysmexUF-1000i. J Clin Lab Anal. 2016;30:381-391.

22. Walker TJ, Nelson LD, Dunphy BW, Anderson DM, Kickler TS.Comparative evaluation of the Iris iQ200 body fluid module withmanualhemacytometercount.Am J Clin Pathol.2009;131:333-338.

23. GlasserL,MurphyCA,MachanJT.Theclinicalreliabilityofautomatedcerebrospinal fluidcell countson theBeckman-CoulterLH750andIrisiQ200.Am J Clin Pathol.2009;131:58-63.

24. DeMott WR, Skikne BS. Cerebrospinal Fluid Analysis. In: JakobsDS, DeMottWR, Oxley DK, eds. Jacobs & DeMott Laboratory Test Handbook.Hudson,Ohio:Lexi-Comp;2001:416-418.

25. SandhausLM,DillmanCA,HinkleWP,MacKenzieJM,HongG.Anewautomatedtechnologyforcerebrospinalfluidcellcounts:comparisonof accuracy and clinical impact of glocyte, sysmex xn, andmanualmethods.Am J Clin Pathol.2017;147:507-514.

26. CSF Cell Counter: Advanced Instruments,Norwood,MA;2017[Availablefrom:http://www.aicompanies.com/product/glocyte/-features.

27. (CLSI) CaLSI. Body Fluid Analysis for Cellular Composition; Approved Guideline; CLSI Document H56-A.Wayne,PA;2006.

28. CLSI. Evaluation of Detection Capability for Clinical Laboratory Measurement Procedures: Approved Guideline, 2nd edition; EP 17-A2. Wayne,PA:CLSI;2012.

29. BournerG,DelaSalleB,GeorgeT,etal.ICSHguidelinesforthever-ificationandperformanceofautomatedcellcountersforbodyfluids.Int J Lab Hematol.2014;36:598-612.

30. Evaluation of Detection Capability for Clinical Laboratory Measurement Procedures: Approved Guideline, 2nd edition. CLSI document EP17-A2,(2012).

31. College of American Pathologists. Hemocytometer Fluid Count, Participant Summary.Northfield,IL:CollegeofAmericanPathologists;2015.

32. MacdonaldR.Qualityassessmentofquantitativeanalyticalresultsinlaboratorymedicinebyrootmeansquareofmeasurementdeviation.J Lab Med.2006;30:111-117.

33. Evaluation of Precision of Quantitative Measurement Procedures (EP05-A3),(2014).

How to cite this article:HodEA,BrugnaraC,PilichowskaM,etal.AutomatedcellcountsonCSFsamples:AmulticenterperformanceevaluationoftheGloCytesystem.Int J Lab Hem. 2017;00:1–10. https://doi.org/10.1111/ijlh.12728