h26-a performance goals for the internal quality control ... · h26-a vol. 16 no. 12 replaces h26-p...

H26-AVol. 16 No. 12Replaces H26-P

December 1996 Vol. 9 No. 9

Performance Goals for the Internal Quality Control ofMultichannel Hematology Analyzers; Approved Standard

This document addresses performance goals for analytical accuracy and precision for multichannelhematology analyzers; the relationship of these goals to quality control systems and medicaldecisions; and recommendations for minimum calibrator performance and the detection ofmeasurement errors.

ABC

NCCLS...Serving the World's Medical Science Community Through Voluntary Consensus

NCCLS is an international, interdisciplinary, nonprofit, scope, approach, and utility, and a line-by-line review of itsstandards-developing and educational organization that technical and editorial content.promotes the development and use of voluntary consensusstandards and guidelines within the healthcare community. It Tentative A tentative standard or guideline is made availableis recognized worldwide for the application of its unique for review and comment only when a recommended methodconsensus process in the development of standards and has a well-defined need for a field evaluation or when aguidelines for patient testing and related healthcare issues. recommended protocol requires that specific data be collected. NCCLS is based on the principle that consensus is an effective It should be reviewed to ensure its utility.and cost-effective way to improve patient testing andhealthcare services. Approved An approved standard or guideline has achieved

In addition to developing and promoting the use of voluntary reviewed to assess the utility of the final document, to ensureconsensus standards and guidelines, NCCLS provides an open attainment of consensus (i.e., that comments on earlierand unbiased forum to address critical issues affecting the versions have been satisfactorily addressed), and to identifyquality of patient testing and health care. the need for additional consensus documents.

PUBLICATIONS NCCLS standards and guidelines represent a consensus opinion

An NCCLS document is published as a standard, guideline, or materially affected, competent, and interested parties obtainedcommittee report. by following NCCLS’s established consensus procedures.

Standard A document developed through the consensus less stringent than applicable regulations. Consequently,process that clearly identifies specific, essential requirements conformance to this voluntary consensus document does notfor materials, methods, or practices for use in an unmodified relieve the user of responsibility for compliance with applicableform. A standard may, in addition, contain discretionary regulations.elements, which are clearly identified.

Guideline A document developed through the consensusprocess describing criteria for a general operating practice, The comments of users are essential to the consensusprocedure, or material for voluntary use. A guideline may be process. Anyone may submit a comment, and all commentsused as written or modified by the user to fit specific needs. are addressed, according to the consensus process, by the

Report A document that has not been subjected to con-sensus including those that result in a change to the document whenreview and is released by the Board of Directors. published at the next consensus level and those that do not

CONSENSUS PROCESS appendix to the document. Readers are strongly encouraged

The NCCLS voluntary consensus process is a protocol document. Address comments to the NCCLS Executiveestablishing formal criteria for: Offices, 940 West Valley Road, Suite 1400, Wayne, PA

! The authorization of a project

! The development and open review of documents

! The revision of documents in response to comments by volunteer for participation in NCCLS projects. Please contactusers the NCCLS Executive Offices for additional information on

! The acceptance of a document as a consensus standard orguideline.

Most NCCLS documents are subject to two levels ofconsensus–"proposed" and "approved." Depending on theneed for field evaluation or data collection, documents mayalso be made available for review at an intermediate (i.e.,"tentative") consensus level.

Proposed An NCCLS consensus document undergoes the firststage of review by the healthcare community as a proposedstandard or guideline. The document should receive a wide andthorough technical review, including an overall review of its

consensus within the healthcare community. It should be

on good practices and reflect the substantial agreement by

Provisions in NCCLS standards and guidelines may be more or

COMMENTS

NCCLS committee that wrote the document. All comments,

result in a change, are responded to by the committee in an

to comment in any form and at any time on any NCCLS

19087, USA.

VOLUNTEER PARTICIPATION

Healthcare professionals in all specialities are urged to

committee participation.

December 1996 H26-A

NCCLS VOL. 16 NO. 12 i

THE NCCLS consensus process, which is the mechanism for moving a document through two ormore levels of review by the patient testing community, is an ongoing process. Users shouldexpect revised editions of any given document. Because rapid changes in technology may affectthe procedures, bench and reference methods, and evaluation protocols used in testing, usersshould replace outdated editions with the current editions of NCCLS documents. Current editionsare listed in the NCCLS Catalog, which is distributed to member organizations, or to nonmemberson request. If your organization is not a member and would like to become one, or to request acopy of the NCCLS Catalog, contact the NCCLS Executive Offices. Telephone: 610.688.0100;Fax: 610.688.0700.


Abstract

Performance Goals for the Internal Quality Control of Multichannel Hematology Analyzers; ApprovedStandard (NCCLS document H26-A) provides recommendations for performance goals for theinternal quality control of multichannel hematology analyzers on the basis of the use of physical andchemical standards, accepted reference methods, subcommittee recommendations on what iscurrently achievable, and the concept of medical usefulness. Critical performance characteristics ofquality control systems (i.e., the probabilities of error detection and false rejection) also areconsidered. A well-designed internal quality control program must achieve the level of errordetection specified in this standard; yet it should not be so sensitive as to falsely reject valid results.

[NCCLS. Performance Goals for the Internal Quality Control of Multichannel Hematology Analyzers;Approved Standard. NCCLS document H26-A (ISBN 1-56238-312-4). NCCLS, 940 West ValleyRoad, Suite 1400, Wayne, Pennsylvania 19087, 1996.]

H26-AISBN 1-56238-312-4

December 1996 ISSN 0273-3099


Volume 16 Number 12

A. Richardson Jones, M.D.Joanne Cornbleet, M.D.Berend Houwen, M.D., Ph.D.Luc Van Hove, M.D., Ph.D.John A. Koepke, M.D.Elkin Simson, M.D., Ch.B., M.Med.William R. Swaim, M.D.Onno W. van Assendelft, M.D., Ph.D.

ABC

December 1996 H26-A

NCCLS VOL. 16 NO. 12 iii

This publication is protected by copyright. No part of it may be reproduced, stored in a retrievalsystem, or transmitted in any form or by any means (electronic, mechanical, photocopying,recording, or otherwise) without written permission from NCCLS, except as stated below.

NCCLS hereby grants permission to reproduce limited portions of this publication for use inlaboratory procedure manuals at a single site, for interlibrary loan, or for use in educational programsprovided that multiple copies of such reproduction shall include the following notice, be distributedwithout charge, and, in no event, contain more than 20% of the document's text.

Reproduced with permission, from NCCLS publication H26-A, Performance Goals for theInternal Quality Control of Multichannel Hematology Analyzers; Approved Standard. Copies of the current edition may be obtained from NCCLS, 940 West Valley Road, Suite 1400,Wayne, Pennsylvania 19087 USA.

Permission to reproduce or otherwise use the text of this document to an extent that exceeds theexemptions granted here or under the Copyright Law must be obtained from NCCLS by writtenrequest. To request such permission, address inquiries to the Executive Director, NCCLS, 940 WestValley Road, Suite 1400, Wayne, Pennsylvania 19087 USA.

Copyright ©1996. The National Committee for Clinical Laboratory Standards.

Suggested Citation

NCCLS. Performance Goals for the Internal Quality Control of Multichannel Hematology Analyzers;Approved Standard. NCCLS document H26-A (ISBN 1-56238-312-4). NCCLS, 940 West ValleyRoad, Suite 1400, Wayne, Pennsylvania 19087, 1996.

Proposed StandardOctober 1989

Approved StandardDecember 1996

ISBN 1-56238-312-4ISSN 0273-3099

December 1996 H26-A

NCCLS VOL. 16 NO. 12 iv

Committee Membership

Area Committee on Hematology

Eugene L. Gottfried, M.D. San Francisco General HospitalChairholder San Francisco, California

Subcommittee on the Blood Count

A. Richardson Jones, M.D. Coulter CorporationChairholder Miami, Florida

Joanne Cornbleet, M.D. Stanford University Medical CenterStanford, California

Berend Houwen, M.D., Ph.D. Loma Linda UniversityLoma Linda, California

Luc Van Hove, M.D., Ph.D. Abbott LaboratoriesSanta Clara, California

John A. Koepke, M.D. Duke University Medical CenterDurham, North Carolina

Elkin Simson, M.D., Ch.B., M.Med. Mount Sinai Medical CenterNew York, New York

William R. Swaim, M.D. VA (Minneapolis) Medical CenterMinneapolis, Minnesota

Onno W. van Assendelft, M.D., Ph.D. Centers for Disease Control and PreventionAtlanta, Georgia

Advisors

Marti K. Bailey, M.T.(ASCP) Milton S. Hershey Medical CenterHershey, Pennsylvania

Stuart A. Bentley, M.D. University of North Carolina Medical SchoolChapel Hill, North Carolina

J. David Bessman, M.D. University of Texas Medical BranchGalveston, Texas

William Canfield Bayer CorporationTarrytown, New York

Samuel E. Chappell National Institute of Standards & TechnologyGaithersburg, Maryland

Dr. John M. England Watford General HospitalWatford, England

December 1996 H26-A

NCCLS VOL. 16 NO. 12 v

Virgil F. Fairbanks, M.D. Mayo ClinicRochester, Minnesota

James E. Gill, Ph.D. Abbott LaboratoriesSanta Clara, California

James N. Lowder, M.D. Becton Dickinson Immunocytometry SystemsSan Jose, California

Louise Magruder FDA Center for Devices/Radiological HealthRockville, Maryland

Christine E. McLaren, Ph.D. Morehead State UniversityMorehead, Minnesota

Albert Rabinovitch, M.D., Ph.D. Becton Dickinson VACUTAINER SystemFranklin Lakes, New Jersey

Dean E. Twedt Coulter CorporationMiami, Florida

David E. Nevalainen, Ph.D. Abbott LaboratoriesBoard Liaison Abbott Park, Illinois

Julie A. Alexander, M.T.(ASCP), M.A. NCCLSStaff Liaison Wayne, Pennsylvania

Patrice E. Polgar NCCLSEditor Wayne, Pennsylvania

December 1996 H26-A

NCCLS VOL. 16 NO. 12 vi

ACTIVE MEMBERSHIP (as of 1 October 1996)

Sustaining Members

American Association for Clinical ChemistryBayer CorporationBeckman Instruments, Inc.Becton Dickinson and CompanyBoehringer Mannheim Diagnostics, Inc.College of American PathologistsCoulter CorporationDade International Inc.Johnson & Johnson Clinical Diagnostics Abbott LaboratoriesOrtho Diagnostic Systems Inc. ABC Consulting Group, Ltd.

Professional Members

American Academy of Allergy Asthma & ImmunologyAmerican Academy of Family PhysiciansAmerican Association of BioanalystsAmerican Association of Blood BanksAmerican Association for Clinical ChemistryAmerican Association for Respiratory CareAmerican Chemical SocietyAmerican Medical TechnologistsAmerican Public Health AssociationAmerican Society for Clinical Laboratory ScienceAmerican Society of HematologyAmerican Society forMicrobiologyAmerican Society of Parasitologists, Inc.American Type Culture Collection, Inc.Australasian Association of Clinical BiochemistsCanadian Society of Laboratory TechnologistsClinical Laboratory Management AssociationCollege of American PathologistsCollege of Medical Laboratory Technologists of Ontario Commission on Office Laboratory AccreditationCorps professionnel des technologistes médicaux du QuébecInstitut für Stand. und Dok. im Med. Lab. (INSTAND)

International Federation of National Institute of Standards Clinical Chemistry and TechnologyInternational Society for Ohio Department of Health Analytical Cytology Oklahoma State Department ofItalian Society of Clinical Health Biochemistry Ontario Ministry of Health Japan Association of Medical South African Institute for Technologists Medical ResearchJapanese Committee for Clinical Swedish Institute for Infectious Laboratory Standards Disease ControlJoint Commission on Accreditation of Healthcare OrganizationsNational Academy of Clinical BiochemistryNational Society for Histotechnology, Inc.Ontario Medical Association Laboratory Proficiency Testing ProgramSociedade Brasileira de Analises Clinicas

Government Members

Armed Forces Institute of CT Pathology Beckman Instruments, Inc.Association of State and Becton Dickinson and Company Territorial Public Health Becton Dickinson Consumer Laboratory Directors Products BC Centre for Disease Control Becton DickinsonCenter for Preventive Medicine Immunocytometry Systems (France) Becton Dickinson MicrobiologyCenters for Disease Control and Systems Prevention Becton Dickinson Primary CareChina National Centre for the Diagnostics Clinical Laboratory Becton Dickinson VACUTAINERCommonwealth of Pennsylvania Systems Bureau of Laboratories Behring Diagnostics Inc.Connecticut Department of Behring Diagnostics Inc. - San Public Health & Addiction Jose, CA Services bioMérieux Vitek, Inc.Department of Veterans Affairs Biometrology ConsultantsDeutsches Institut für Normung Bio-Rad Laboratories, Inc. (DIN) Biosite DiagnosticsFDA Center for Devices and Boehringer Mannheim Radiological Health Diagnostics, Inc.FDA Division of Anti-Infective Boehringer Mannheim GmbH Drug Products Bristol-Myers Squibb CompanyHealth Care Financing CASCO Standards Administration ChemTrakINMETRO CholestechInstituto Scientifico HS. Raffaele Ciba Corning Diagnostics Corp, (Italy) A Chiron CompanyIowa State Hygienic Laboratory Ciba Corning Diagnostics Corp,Massachusetts Department of A Chiron Company - Public Health Laboratories Electrophoretic Products Michigan Department of Public Ciba Corning Diagnostics Corp, Health

Industry Members

Advanced Care Products Division (Div. Ortho Diagnostic Systems Inc.)aejesBayer Corporation - Elkhart, INBayer Corporation - Middletown, VABayer Corporation - Tarrytown,NYBayer Corporation - West Haven,

December 1996 H26-A

NCCLS VOL. 16 NO. 12 vii

A Chiron Company - International Remote Imaging Roche Laboratories (Div. International Operations Systems (IRIS) Hoffmann-La Roche Inc.)Ciba Corning Diagnostics Corp, Johnson & Johnson Clinical The R.W. Johnson A Chiron Company - Irvine, CA Diagnostics Pharmaceutical ResearchCiba Corning Diagnostics Corp, LifeScan, Inc. (Sub. Ortho Institute (Div. Ortho Diagnostic A Chiron Company - Reagent Diagnostic Systems Inc.) Systems Inc.) Systems Lilly Research Laboratories Schering CorporationClinical Lab Engineering Madych Associates, Inc. Schleicher & Schuell, Inc.COBE Laboratories, Inc. Mallinckrodt Sensor Systems Second OpinionCosmetic Ingredient Review Medical Device Consultants, Inc. SenDx Medical, Inc.Coulter Corporation Medical Laboratory Automation Sherwood Medical CompanyCytometrics, Inc. Inc. Showa Yakuhin Kako Company,CYTYC Corporation MediSense, Inc. Ltd.Dade International - Deerfield, IL Medix Biochemica Sienna BiotechDade International - Glasgow, Merck & Company, Inc. Sigma Chemical CompanyDE Metra Biosystems SmithKline Beecham CorporationDade International - Miami, FL Micro Media Systems Inc. (Div. SmithKline Diagnostics, Inc.Dade International - Sacramento, Medical Specialties Inc.) (Sub. Beckman Instruments, CA Nellcor Puritan Bennett Inc.)DAKO A/S Neometrics, Inc. Streck Laboratories, Inc.Diagnostic Products Corporation Nissui Pharmaceutical Co., Ltd. Sysmex CorporationDiametrics Medical, Inc. Norfolk Associates, Inc. TOA Medical ElectronicsDifco Laboratories, Inc. North American Biologicals, Inc. TOSOH Medics, Inc.Enterprise Analysis Corporation Olympus Corporation Unipath Co (Oxoid Division)Epoch Pharmaceuticals Optical Sensors, Inc. The Upjohn CompanyEppendorf, Netherler Hinz GmbH Organon Teknika Corporation Vysis, Inc.Donna M. Falcone Consultants Orion Diagnostica, Inc. Wallac OyFujisawa Pharmaceutical Co. Ortho Diagnostic Systems Inc. Warner-Lambert CompanyLtd. Otsuka America Pharmaceutical, The West CompanyGen-Probe Inc. Wheaton PharmaTechGlaxo, Inc. Pfizer Inc Wyeth-AyerstH&S Consultants Procter & Gamble Xyletech Systems, Inc.Health Systems Concepts, Inc. Pharmaceuticals, Inc. ZenecaHelena Laboratories The Product Development GroupHigman Healthcare Radiometer America, Inc.Hoechst Marion Roussel, Inc. David G. Rhoads Associates,Hybritech, Incorporated Inc.Hycor Biomedical Inc. Rhône-Poulenc Roreri-STAT Corporation Roche Diagnostic SystemsInteg, Inc. (Div. Hoffmann-La Roche International Biomedical Inc.) Consultants

Trade Associations

Association of Medical Diagnostic ManufacturersHealth Industry Manufacturers Association

December 1996 H26-A

NCCLS VOL. 16 NO. 12 viii

Associate Active Members

Affinity Health System (WI)Allegheny University of the Health Sciences (PA)Allergy Testing Laboratory (TX)Alton Ochsner Medical Foundation (LA)American Oncologic Hospital (PA)Associated Regional & University Pathologists (UT)Astra Research Center Boston (MA)Baptist Medical Center - Montclair (AL)Battelle (OH)BC Children’s Hospital (Canada)Bethesda Hospital (OH)Bristol Regional Medical Center (TN)Brooks Air Force Base (TX)Broward General Medical Center (FL)Canterbury Health Laboratories (New Zealand)CENTREX Clinical Laboratories (NY)Chester County Hospital (PA)Childrens Hospital Los Angeles (CA)Children's Hospital Medical Center (Akron, OH)Children's Hospital Medical Center (Cincinnati, OH)Children's Hospital - New Orleans (LA)City of Hope National Medical Center (CA)City Hospital (WV)The Cleveland Clinic Foundation (OH)Coler Memorial Hospital (NY)Commonwealth of KentuckyCompuNet Clinical Laboratories (OH)Dean Medical Center (WI)Dhahran Health Center (Saudi Arabia)Diagnostic Systems Laboratories, Inc. (TX)Dianon Systems, Inc. (CT)Duke University Medical Center (NC)Dwight David Eisenhower Army Medical Center (Ft. Gordon, GA)Easton Hospital (PA)East Texas Medical CenterEllis Fischel Cancer Center (MO)Elmhurst Memorial Hospital (IL)Elyria Memorial Hospital (OH)

Evanston Hospital (IL) Montgomery Regional MedicalFederal Medical Center (MN) Center (AL)Fort Leonard Wood Army Montreal Children’s Hospital Community Hospital (MO) (Canada)Grady Memorial Hospital (GA) Mount Sinai Hospital (NY)Great Smokies Diagnostic Mount Sinai Hospital (Toronto, Laboratory (NC) ON, Canada)Harris Methodist Fort Worth National Genetics Institute (CA)(TX) National Institutes of HealthHartford Hospital (CT) (MD)Heritage Hospital (MI) National Naval Medical Center Hopital Saint Pierre (Belgium) (MD)Hunter Area Pathology Service Naval Hospital Cherry Point (NC) (Australia) New Jersey Department ofIncstar Corporation (MN) HealthInstitute for Transfusion The New York Blood Center Medicine (PA) New York State Department ofIowa Methodist Medical Center HealthJapan Association Clinical New York State Library Reagents Ind. (Tokyo, Japan) North Carolina Laboratory ofKaiser Permanente (CA) Public HealthKenora-Rainy River Regional North Carolina School of Laboratory Program (Dryden, Veterinary MedicineON, Canada) North Central Bronx Hospital Laboratorio Clinico Borinquen (NY)(PR) North Shore University Hospital Laboratory Corporation of (NY) America (NC) Northwestern Memorial HospitalLahey Hitchcock Medical Center (IL) (MA) Ocean County MedicalLancaster General Hospital (PA) Laboratories (NJ)Lawrence Memorial Hospital Our Lady of Lourdes Hospital (MA) (NJ)Loma Linda University Medical Our Lady of the Resurrection Center (CA) Medical Center (IL)Maine Medical Center Palo Alto Medical Foundation Malcolm Grow USAF Medical (CA) Center (MD) PAPP Clinic P.A. (GA)Martin Army Community Pathogenesis Corp. (WA) Hospital (Ft. Benning, GA) P athology AssociatesMartin Memorial Medical Center Laboratories (CA) (FL) Permanente Medical Group (CA)Maryview Medical Center (VA) Polly Ryon Memorial Hospital McKennan Hospital (SD) (TX)M.D. Anderson Hospital & Polyclinic Medical Center (PA) Tumor Institute (TX) Puckett Laboratories (MS)MDS Laboratories (Etobicoke, Queens Hospital Center (NY) ON, Canada) The Queen’s Medical Center (HI)The Medical Center of Ocean Ravenswood Hospital Medical County (NJ) Center (IL)Medical College of Virginia Rhode Island Department of Hospital Health LaboratoriesMelbourne Pathology (Australia) Riverside Clinical Laboratories Memorial Medical Center (IL) (VA)Mercy & Baptist Medical Center Riverside-San Bernardino County (LA) Indian Health (CA)Mercy Hospital (MN) St. Anthony’s Hospital (FL)Methodist Hospital of Indiana St. John Hospital and MedicalMethodist Hospitals of Memphis Center (MI) (TN) St. John's Hospital (IL)Mobile Infirmary Association St. Luke’s-Roosevelt Hospital(AL) Center (NY)

December 1996 H26-A

NCCLS VOL. 16 NO. 12 ix

St. Mary of the Plains Hospital University of Alberta Hospitals University of Virginia Medical (TX) (Canada) CenterSt. Mary’s Regional Medical University of California, San VA (Albuquerque) Medical Center (NV) Francisco Center (NM)St. Paul Medical Center (TX) University of Cincinnati Medical VA (Indianapolis) Medical CenterSt. Paul Ramsey Medical Center Center (OH) (IN) (MN) University Community Hospital VA (Jackson) Medical Center San Francisco General Hospital (FL) (MS) (CA) University of Florida VA (Miami) Medical Center (FL)Shadyside Hospital (PA) University of Hawaii at Manoa VA (Milwaukee) Medical Center Shanghai Center for the Clinical University Hospital (Gent) (WI) Laboratory (China) (Belgium) VA (Perry Point) Medical CenterShore Memorial Hospital (NJ) University Hospital (London, (MD)Sinai Hospital of Detroit (MI) ON, Canada) Veterans General HospitalSmithKline Beecham Clinical University Hospital (IN) (Republic of China) Laboratories (GA) University Hospital of Warde Medical Laboratory (MI)SmithKline Beecham Clinical Cleveland (OH) Wilford Hall USAF Medical Laboratories (TX) The University Hospitals (OK) Center (TX)SmithKline Beecham Clinical University of Medicine & William Beaumont Hospital (MI) Laboratories (WA) Dentistry, NJ University Wisconsin State Laboratory of Specialty Laboratories, Inc. (CA) Hospital HygieneStanford Health Services (CA) University of Michigan York Hospital (PA)SUNY @ Stony Brook (NY) University of Nebraska Medical Zale Lipshy University HospitalTravis Air Force Base (CA) Center (TX)Tripler Army Medical Center (HI) University of Utah MedicalUNC Hospitals (NC) Center

OFFICERS BOARD OF DIRECTORS

A. Samuel Koenig, III, M.D., Carl H. Blank, Dr.P.H. Robert F. Moran, Ph.D., President Wyoming Department of FCCM, FAICFamily Medical Care Health Chiron Diagnostics Corporation

William F. Koch, Ph.D., Carl A. Burtis, Ph.D. David E. Nevalainen, Ph.D. President Elect Oak Ridge National Laboratory Abbott LaboratoriesNational Institute of Standards and Technology Sharon S. Ehrmeyer, Ph.D. Donald M. Powers, Ph.D.

F. Alan Andersen, Ph.D., Diagnostics Secretary Helen M. Free, D.Sc.Cosmetic Ingredient Review Bayer Corporation Eric J. Sampson, Ph.D.

Donna M. Meyer, Ph.D., Elizabeth D. Jacobson, Ph.D. and Prevention Treasurer FDA Center for Devices andSt. Joseph Hospital Radiological Health Marianne C. Watters,

Charles F. Galanaugh, Past Kenneth D. McClatchey, M.D., Parkland Memorial Hospital President D.D.S.Becton Dickinson and Company Loyola University Medical Ann M. Willey, Ph.D.

John V. Bergen, Ph.D., Health Executive Director

University of Wisconsin Johnson & Johnson Clinical

Center New York State Department of

Centers for Disease Control

M.T.(ASCP)

December 1996 H26-A

NCCLS VOL. 16 NO. 12 x

ContentsPage

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Committee Membership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

Active Membership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

3 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

4 Performance Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44.1 Analytical Error: Inaccuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44.2 Analytical Error: Calibrator Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54.3 Analytical Error: Calibrator Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54.4 Analytical Error: Nonlinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64.5 Analytical Error: Imprecision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64.6 Analytical Error: Interferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74.7 Analytical Error: Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

5 Relating Performance Goals to Medical Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . 85.1 Effect of Assay Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95.2 Effect of Assay Imprecision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Additional Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Summary of Comments and Subcommittee Responses . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Related NCCLS Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

December 1996 H26-A

NCCLS VOL. 16 NO. 12 xi

Foreword

The primary objective of quality control in the clinical laboratory is to ensure that the analyticalvalues are sufficiently reliable to be used in the care of patients. Once quality goals are establishedfor an assay, their results can be evaluated in terms of clinical usefulness. This approach can havedistinct advantages in the quality control of multichannel hematology analyzers.

This standard examines the effects of the following variables on performance goals:

! Calibration of the instruments ! Imprecision of the analytical measurements ! Analyte variations within an individual! Inherent differences in values among persons.

The first two of these sources of variation define opportunities for improvement in analyzer design. The second two define irreducible biological variables.

In this document, the Subcommittee on the Blood Count provides goals for standards ofperformance that are useful for diagnosis, patient monitoring and control of therapeutic regimens. These performance goals set the stage for a review of quality control options that are intended toprovide a framework for ensuring that patients' assays are made as precisely and accurately as theanalyzer allows. To this end, the production of a companion document is planned that will addressthe principles and methods of quality control that will help users and makers of automatedhematology analyzers achieve these goals. This approach should stimulate improvements inanalyzer design and interpretation of assay results. One goal is that manufacturers will recognizethe need to coordinate the performance of different analytical methods so that assay results fromdifferent analyzers will have a reasonable degree of interchangeability. A further goal in this contextis for designers to strive to minimize some of the existing disparities between analyzer and referenceassays.

Universal Precautions

Because it is often impossible to know which might be infectious, all patient blood specimens are tobe treated with universal precautions. Guidelines for specimen handling are available from the U.S.Centers for Disease Control and Prevention [MMWR 1987; 36 (Suppl 2S) 2S–18S]. NCCLSdocument M29-T2, Protection of Laboratory Workers from Infectious Disease Transmitted by Blood,Body Fluids and Tissue—Second Edition; Tentative Guideline, deals specifically with this issue.

Key Words

Accuracy, analytical bias, calibration, calibrator, imprecision, linearity, quality control, sensitivity,specificity, value.

December 1996 H26-A

NCCLS VOL. 16 NO. 12 1

Performance Goals for the Internal Quality Control of Multichannel Hematology Analyzers; Approved Standard

1 Introduction

The elements of the complete blood cell count(CBC) included in this standard are themeasurement of hemoglobin concentration(Hb), hematocrit (Hct), erythrocyte count(RBC), mean cell volume (MCV), leukocyte count (WBC), and platelet count (Plt). Goalsfor the derived red cell indices, mean cellhemoglobin (MCH), and mean cell hemoglobinconcentration (MCHC) are not included.

The widespread acceptance of automated follows: whole blood analyzers and concomitantimprovements in calibrators and control Accuracy: A measure of agreement betweenmaterials has had a major effect on the the estimate of a value and a "true" value;efficiency of laboratory operation. Also, a quantifiable in terms of departure frommarked improvement in intralaboratory and accuracy; expressed as systematic error orinterlaboratory precision and accuracy has bias.occurred. For values in the adult referencerange, within-laboratory coefficients of ! Accuracy, of an analytic process:variation (CV) of less than 1.0% for RBC, Hb Expressed as the difference betweenand MCV are readily achieved on the newest the average result obtainable by ageneration of analyzers, while WBC counts method under specified conditions andshow CVs less than 2% and Plt counts CVs the result accepted as true orless than 3%. However, maintaining accur- standard; expressed in the same units1-5

acy by preventing or predicting drift during as the result, or as a percentage of theroutine operation remains a problem with standard result (relative accuracy). some types of analyzers.

NOTE: The lower the difference, the2 Scope

This document presents performance goals foranalytical accuracy and precision ofmultichannel hematology analyzers capable ofbeing calibrated. These clinical performancegoals relate the majority of routinely producedanalytical values to reference populations andto action limits established by the laboratory.

Goals for standards for specimen handling,equipment operation, electronic checking,preventive maintenance, and reagent qualityare not included in this document, but theireffect on performance is discussed whererelevant.

This standard will be useful to laboratorydirectors, supervisors, quality control officers,and others who have responsibilities forongoing quality control in hematology

laboratories, particularly in the light of currentregulatory pressures. It is a starting point fornational and international discussion of theissues surrounding the design of qualitycontrol systems for multichannel hematologyanalyzers and as an aid to manufacturers whoseek to improve the performance of theirproducts.

3 Definitions

Within this document, terms are defined as*

higher the accuracy (the lower theinaccuracy). Conventionally, thisdifference includes only processinaccuracy (process bias or systematicerror) because the contribution ofprocess imprecision (random error) isminimized by the averaging of multipledeterminations.

! Accuracy, of a result: Expressed asthe difference between a result andthe "true" value.

References include ICSH, Rules and Operating*

Procedures, 1991; ISO, International Vocabulary of Basicand General Terms in Metrology, 2nd Edition, 1993; NCCLS, NRSCL8-P2, Nomenclature and Definitions forUse in NRSCL and Other NCCLS Documents—2ndEdition; Proposed Guideline.

MCH(pg) 'Hb(g/L)

RBC(x10 12/L)

MCHC(g/L) 'Hb(g/L)

PCV(L/L)

December 1996 H26-A


NOTE: This difference includes may be used to establish the relationship ofcontributions not only from process an analytical method's response to theinaccuracy but also from process character-istic measured for all methods orimprecision, especially when one restricted to some. The calibrator must bedetermination per specimen is the rule. traceable to a national or internationalIt is expressed in the same units as reference preparation or reference materialthe result. when these are available. Calibrators with

! Accuracy, of a control result: Same as establish a quantity/response curve over afor a result. range of interest.

NOTE: In this case, because the (Quality) control material: A substance used inaccepted mean and standard deviation routine practice for checking the concurrentof the population are known, the bias performance of an analytical process.can be expressed alternatively as a Z-score, the signed difference between NOTE: It should be similar to, and is analyzedthe result and known mean divided by along with, patients’ specimens. Controlthe known standard deviation. The Z- materials may or may not have known analytescore is unitless and is a uniform concentrations (i.e., assigned values) withinexpression for all analytes. specified limits (e.g., target value +/!

Bias (synonym for "systematic error"): A not be used for calibration purposes.quantitative measure of inaccuracy orsystematic departure from accuracy under Derived red cell indices: Quantities that mayspecified conditions of analysis. A signed be calculated from the measurement ofdifference (+, !) between two values. hemoglobin (Hb) concentration, packed (red)

NOTE: In general, the difference between the concentration (see the definitions for PCV,true or expected value obtained using an MCH, MCHC, MCV).accepted method and the observed value fromthe method being tested is typically based on ! MCH; mean cell (corpuscular)replicate measurements. Bias is expressed in hemoglobin: the average amount ofthe units of the measurement or as a hemoglobin within the red blood cell inpercentage. a given blood sample.

Calibration: The determination of a biasconversion factor of an analytical processunder specified conditions, in order to obtainaccurate measurement results. The accuracyover the operating range must be establishedby the appropriate use of reference methods,reference materials, or calibrators, or any ! MCHC; mean cell (corpuscular)combination of these. hemoglobin concentration: the average

Calibrator: A (reference) material (e.g., red blood cells of a given bloodsolution) or device of known sample.quantitative/qualitative characteristics (e.g.,concentration, activity, intensity, reactivity)used to calibrate, graduate, or adjust ameasurement procedure or to compare theresponse obtained with the response of a testspecimen/sample.

NOTE: The quantities of the analytes of volume of the red blood cell in a giveninterest in the calibrator are known within blood samplelimits ascertained during its preparation and

different quantities of analyte may be used to

standard deviation). Control materials must

cell volume and erythrocyte (RBC)

hemoglobin concentration within the

! MCV; Mean cell volume: the average

MCV(fL) 'PCV(L/L)

RBC (x10 12/L)

December 1996 H26-A


Linearity: The ability of an analytical processto provide a measurement proportional to the Reference material: A material or substanceanalyte being measured over a defined with one or more property values that areconcentration or counting range. sufficiently homogeneous and well established

NOTE: Linearity refers to overall system the assessment of a measurement method, orresponse (i.e., the final analytical answer for assigning values to materials.rather than the instrument output). Whenanalytical results are plotted against NOTE: The term “reference material(s)” shouldconcentration or counts, the degree to which be used generically. They form a class ofthe plotted curve conforms to a straight line is materials to which "reference" can be made.a measure of system linearity. Criteria for They include certified reference materialslinearity should be based on appropriate slope, (CRM), standard reference materials (SRM),intercept, standard deviation of y about x, as calibrators and standards (see also definitionswell as the linear range. for CRM, calibrators, and standards). A

Packed (red) cell volume, (PCV): The measure and well-defined systems for measurement,of the ratio of the volume occupied by the red makes possible the transfer of the value of ablood cells to the volume of whole blood, measured quantity (physical, chemical,expressed as a fraction (L/L). biological, or technological) between two

NOTE: The term "hematocrit" has been, and mixed gas, a liquid or solid of biological origin,often is, used for this quantity. or a manufactured object.

Precision: Agreement between replicate Reference material, certified (CRM): Ameasurements. It has no numerical value, but reference material, accompanied by ait is expressed in terms of imprecision. certificate, with one or more property values

NOTE: Generally, the degree of imprecision is establishes traceability to an accuratereflected by the standard deviation, which is realization of the unit in which the propertythe measure of random error. For repeated values are expressed, and for which eachmeasurement of any given analyte, the certified value is accompanied by anrandom errors are generally assumed to be uncertainty at a stated level of confidence. distributed normally around the observedmean. Reference population: A group of N persons in

Quality control (internal): The set ofprocedures undertaken in a laboratory for the Sensitivity:continual assessment of work carried out inthe laboratory and the evaluation of tests to ! Sensitivity, analytical: The change indecide whether the results are reliable enough the response of a measuringto be released to the requesting physician. instrument divided by the

NOTE: The procedures should include tests on The sensitivity may depend on thecontrol material, results of which may be value of the stimulus. plotted on a control chart showing upperand/or lower control limits and may includestatistical analysis of patient data. The main NOTE: The term analytical sensitivityobjective is to ensure day-to-day consistency has also been used synonymous forof measurements or observations, if possible detection limit, i.e., the smallestin agreement with an agreed-on indicator of quantity of analyte that can betruth, such as a control material with assigned reproducibly distinguished fromvalues. background noise in a given assay

Quantity (measurable): Attribute of aphenomenon, body, or substance that can bedistinguished qualitatively and determinedquantitatively.

to be used for the calibration of an apparatus,

reference material, together with appropriate

places. A reference material may be a pure or

that are certified by a procedure that

a described state of health or disease.

corresponding change in the stimulus.

December 1996 H26-A


system. It is usually defined at the material measures, measuring0.95 (95%) confidence level (+/! 2 instruments, or reference materials. SD).

! Sensitivity, clinical: The proportion ofpatients with a well-defined clinicaldisorder whose test values exceed adefined decision limit (i.e., a positiveresult and identification of the patientswho have the disorder).

NOTE: The clinical disorder must bedefined by criteria independent of thetest under consideration.

Set point: The level (i.e., analyteconcentration) at which an instrument iscalibrated.

Specificity:

! Specificity, analytical: Freedom frommeasurement interference, i.e., theability of an analytical method todetermine only the component itpurports to measure.

! Specificity, clinical: The proportion ofpatients who do not have a well-defined clinical disorder and whosetest values do not exceed a defineddecision limit (i.e., a negative resultand identification of patients who donot have the disease).

Standard (measurement): (1) Materialmeasure, measuring instrument, referencematerial, or measuring system intended todefine, realize, conserve, or reproduce a unit,or one or more values of a quantity, to serveas a reference; or (2) (preferred) anauthoritative document that sets forth criteriafor performance and characteristics.

! Standard, primary: A standard that isdesignated, or widely acknowledgedas having the highest metrologicalqualities and its value is acceptedwithout reference to other standardsof the same quantity.

! Standard, secondary: A standard, thevalue of which is assigned bycomparison with a primary standard ofthe same quantity.

! Standard, working: A standard that isused routinely to calibrate or check

4 Performance Goals Performance goals for accuracy and precisionof multichannel hematology analyzersrepresent a compromise between what wouldbe ideal for clinical practice and what isrealistically attainable with moderninstrumentation. Bias and imprecision are themajor analytical causes of assay error. Someaspects of the perform-ance of an analyzermay vary as a function of analyteconcentration. The effect of analyteconcentration on imprecision and bias arediscussed in relation to critical decision levels,such as reference intervals and clinical actionlimits.

4.1 Analytical Error: Inaccuracy

Accuracy is a measure of agreement betweenthe estimate of a value and its "true" value. Inaccuracy is a statement of disagreementbetween these parameters. Numericalstatements about degrees of inaccuracyshould take into account the imprecision of itsestimate and the imprecision of the value withwhich it is compared. This helps to determinewhether a disagreement between the methodsis nonsystematic and random (imprecision) orsystematic, unidirectional, and nonrandom(bias).

Bias or inaccuracy can be evaluated relative tothe impact it would have on clinical decisions. Biased assays pose major problems byshifting patients' values relative to fixeddecision levels. For analytes with well-defined upper and lower decision limits, shiftsin test values decrease or increase clinicalspecificity and sensitivity. These changesdepend on the direction of the shift and thealtered positions of the values from patientswith disease compared with the referencepopulations. This is illustrated by examplesgiven in Section 5.

Bias can take one form or a combination ofthree forms.

C It can be proportional to analyte level;multiplication by a single factor canbring all values of the propositusassays into agreement with referenceassays. This condition is a linearslope error.

SEM'SD

n

December 1996 H26-A


C It can be a constant additive error, through accumulation of random error. independent of analyte level; the However, improper use, deterioration as a addition or subtraction of a constant result of improper storage, shipping, or usewill bring the assay values into beyond the specified shelf life can causeagreement with reference assays. calibrator-induced bias. Demonstration of theThis condition is a linear offset error. use of red cell indices for assessing the

C It can reflect a complex, but well- argument that the principles of the methodsdefined, relationship to analyte level. yielding negligible bias for those componentsThis is a special case of nonlinear would also produce a similar standard ofresponse. One example is the non- accuracy for other analytes, has been shown.linear effect of coincident cells withinthe count-sensing zone.

The first form of bias can be caused byincorrectly-assigned calibrator values, but, inaddition, the accuracy of clinical assays canbe adversely affected by improper use of thecalibrator. See Section 4.3 for a furtherdiscussion of this aspect of bias. Additionalinformation on bias appears in NCCLSdocument EP9-A, Method Comparison andBias Estimation Using Patient Samples;Approved Guideline.

Total bias is the greatest possible deviationfrom truth of a single assay. It is a statementof the theoretical degree of error that wouldoccur in the extremely improbable event thatall the components of bias, listed above,operated at full magnitude in the samedirection at the same time. It is calculated asthe linear sum of the limits for set point error,maximum allowable drift error of 2SD, and thesingle-tailed 95% confidence limit of a singleassay rather than the square root of thecombination of the individual variances. Thisis because, for this purpose, it is assumedthat the individual error sources areindependently controlled. The concept oftotal bias is of more value to analyzerdesigners than to analyzer users.

4.2 Analytical Error: Calibrator Bias

Error in the assigned values of the calibrator is of the calibrator has a standard error of propagated to clinical assays. Therefore, the (SD ÷ /10) or (SD ÷ 3.16). Togoal of the process of assigning calibrator com-bine the analyzer SEM with the 95% CL values should be to achieve zero bias. The of the calibrator, the SEM is raised to 95% CLprocess should allow measurement of the by multiplying it by the appropriate t factorerrors contributed by each of its stages . taken from the Appendix. Select the t factorThe propagation of these errors through the for n ! 1 (one less than the number ofassigned value provides a statement of the iterations actually used). Thus, t for 10calibrator confidence limits. Errors should be iterations will be row 9 in the table or 2.2622. controlled in a manner that prevents the The formula for bias limits then becomes:

calibrator from contributing bias except

accuracy of hematology calibrators, using the

8

4.3 Analytical Error: Calibrator Use

Manufacturers have made progress inminimizing the human element in calibrationby automating setpoint adjustment. Thecalibra-tion set point in an instrument withthis feature is free from human error and itsaccuracy depends only on the number ofmeasurements contributing to the meanrecovered value of the calibrator. Feweriterations will result in higher set-pointvariation and a higher likelihood of bias. Fora single assay of an unknown bloodspecimen, the random error (imprecision) ofthe instrument must be combined with theestimate of bias range to give the probableanalytical error range.

The 95% bias range of an analyzer should bemeasured and recorded each time calibrationis performed. It is calculated by combining the 95% confidence limits (CL) of theassigned calibrator value with the standarderror of the mean (SEM) for the number oftimes the calibrator is assayed. SEM isdefined as the range over which the means ofn iterations is expected to vary. It isdescribed by the following equation:

For example, the mean of 10 replicate assays

analyzer analyzer

6,7

95% Bias ' (CL 2analyzer % CL 2

calibrator)

December 1996 H26-A


It is convenient to express assay bias as results that exceed the upper linearity limit. percent because this error is propagated Results that fall above this limit should beproportionately throughout the reportable available to the laboratory, but not for the(linear) range of the instrument. Thus, if the patient's record, and they should be flagged,95% bias limits for Hb set point of 125.0 g/L either through laboratory action limit rules oris ± 2.0 g/L, all assays within the reportable by signals from the analyzer. In most cases,range will have a percentage bias range of quantitative dilution of the specimen with the(2.0 ÷ 125.0 @ 100) or 1.6%. analyzer diluent reagent will bring the result

4.4 Analytical Error: Nonlinearity

Nonlinearity of an assay has an effect similarto that of assay bias, but the magnitude of Analyzer imprecision is important in regard tothe error depends on the value of the both the processes of calibration and qualitymeasurement. Typically, nonlinearity control, and the clinical interpretation ofproduces greater changes with measurements on patients' specimens. Themeasurements beyond the limits of the principal stages at which random error plays areportable range. Values close to the part are shown in Table 2. The calibrationcalibration set point are least affected. process involves transferring assigned valuesBecause clinical decision limits and reference from the calibrator to the analyzer. The betterrange limits are usually located away from the the precision of the analyzer, the more exactset point, nonlinearity can cause changes in is the estimate of set-point error for a givenassay sensitivity and specificity. The goals number of iterations of calibrator assays. for nonlinearity, shown in Table 1, are Also, the more precise an analyzer, the easierimplicitly defined by the assay bias goals it is to detect instrument drift away from thewhen these goals are stated for a reportable set point. Therefore, it requires fewer assaysrange of test values. of control material to decide whether it is

Table 1. Recommended Minimum Ranges for might not perceive the imprecision of Analytic Linearity measurements when interpreting an isolated

Analyte Units Low High Limit Limit

WBC 10 /L 0.1 100.09

RBC 10 /L 1.50 7.012

Hb g/L 20.0 200.0

Hct L/L 0.33 0.54

MCV fL 50.0 130.0

Plt 10 /L 5.0 1,5009

Manufacturers should include claims oflinearity limits in their product labeling. Thisallows the user to estimate the magnitude ofthis error by interpolation between the statedlinearity limit and the reported result. Whenever nonlinearity is algorithmicallycorrected, e.g., for the correction ofcoincidence error, manufacturers should reveal

this to the user and provide an outline of thecorrective method and a statement of itslimitations.

This standard discourages the suppression of

into the linear range of the analyzer.

4.5 Analytical Error: Imprecision

performing within specified limits. A clinician

test result because test values include acombination of both analytical and biologicalvariation. When one of these components ofvariation is much larger than the other, itbecomes a limiting factor. For instance,changes in analytical precision less than 25%of the within-person biological variation mightnot be perceptible without performingreplicate tests. Comparison of these9,10

components of variation can thus be used todefine bounds for analytical precision.

The analytical imprecision componentcontains two subsets:

C The combination of the multiplicity ofvariable physical events that takeplace during an analysis. All theseevents are essentially independent andnot gen-erally related to analyteconcentration. They combineaccording to the square root of thesum of their individual variances.

CV 'N

N@ 100

December 1996 H26-A


C The irreducible error attached to Components to the Confidence sampling a finite number of randomly Limits (CL) of Calibrationdistributed cells or particles that areindependent (not interacting). This isthe Poisson error where under-11

reporting or over-reporting of a countis due to the fact that the actualnumber of detected cells is randomlyvariable as a function of the squareroot of the number of cells counted.

Higher counts have proportionately lessPoisson error than lower counts. Itscontribution to the overall imprecision of theassay, expressed as CV, can be stated asfollows:

Where N is the number of cells counted (notthe number reported).

It is tempting to reduce the Poisson effect onthe error of the calibration set point by usingcalibrators with high values. However, thehoped-for benefit from high cell concentrationmight not be realized because of thecounterbalancing effect of increased viscosityand coincidence error.

By using the mean of iterative (for example10) calibrator assays, the 95% confidencelimit (or 2.2622 @ SD) of the set point isreduced to one-third the standard deviation ofa single assay. See Section 4.3.

Table 2. The Contribution of Random Error

Stage Random errorcomponents

Assign reference Dilution imprecisionvalues*

Reference analyzer imprecision (Poisson)

Calibrate automated analyzer* Stability of

Automated analyzer imprecision

calibrator material

Assign calibrator imprecisionvalues*

Automated analyzer

Stability of calibrator material

Use of calibrator

Error of assigned values

User's analyzer imprecision

Stability of calibrator material

*Components associated with assigning calibrator values.

4.6 Analytical Error: Interferences

The third cause of analytical error is theresponse of the analyzer to the presence of interfering substances that mimic an analytewithout being a true part of it. This leads toreduced analytical specificity. An example isthe influence of lipemic or leukocytic turbidityon hemoglobin measurement. This isdescribed in NCCLS document, H15-A2,Reference and Selected Procedures for theQuantitative Determination of Hemoglobin inBlood–Second Edition; Approved Standard,and is one of the reasons why somecommercial calibrators are system-specific andnot usable with all types of automatedanalyzers. It is the responsibility of themanufacturers of both the analyzer and thecalibrator to document such phenomena andcaution users as to their effects on analyticalaccuracy.

December 1996 H26-A


The phenomenon of interference is important that the goal for the permissible range of in the process of assigning values to variation of results due to drift should be ±calibrators. During this process, accurate twice the instrument SD. This limit istransfer of reference values to the automated selected as conforming to the usual actionanalyzer depends on both systems reacting in limit of the laboratory internal quality controlthe same way to the analyte. If this is not so, system. Calibration bias limits (see above) arethat is, if the analytical specificities of the not affected by drift because drift, bysystems differ, significant bias can be definition, occurs after calibration has takenintroduced. Detection of interference place and is assumed not to occur duringdepends on the principle that the imprecision calibration.of differences of paired reference assaysversus the imprecision of paired analyzer Drift can affect the analyzer response suchassays should, lacking interference, be nearly that its error is proportionate throughout theequal to the combined imprecisions of the reportable range (slope change), or it can takereference and analyzer assays. In the the form of an offset (constant error) thatpresence of interference, the imprecision of adds or subtracts a fixed quantity. the combined differences is greater. Sometimes both factors are at work. The

The manufacturer of a calibrator must choose detection, must be a key factor in the designbetween incorporating an "interference allow- of internal quality control systems.ance" in the assigned values or selectingreference specimens that are free of Drift detection should encompass theinterference. In the former case, the calibrator following factors:carries a bias for all specimens equal to themean frequency of the interference in the C The sensitivity of drift detectionreference population. In the latter case, should be within limits that permit theclinical specimens that contain the clinical use of results that precede itsinterference will have an error; others will not. discovery when no other errors are

The basis for the choice between thesedecisions depends on the frequency and C The laboratory workflow rules mustmagnitude of the interference in the reference allow results to be quarantined afterpopulation versus its frequency and drift detection, pending an analysis ofmagnitude in the clinical population and the causes.sensitivity of the analyzer to the interferingsubstance. C There should be provision for the

4.7 Analytical Error: Drift

Drift is an unintended, unanticipated changeof analyzer response. The change may begradual, reaching a detectable magnitude onlyover a period, or it may be step-wise. Ineither case it can cause bias of assays duringan analysis run so that an unknownproportion of results that preceded itsdetection might be in error. Drift occurring inan instrument that is normally drift-free can bean early symptom of a serious malfunction. Instruments that are prone to drift in ordinaryuse, to a degree that has a clinicalsignificance, should have this propensityspecified by the manufacturer. Currently,there is no information available frommanufacturers to permit a generalized,predictive estimate of how much drift, itsprobable direction, or how often it mightoccur in any given analyzer. It is proposed

possibility of drift, and the need for its

present.

release of selected results based on acareful review of medical need.

5 Relating Performance Goals toMedical Decisions

It is useful to relate performance goals, asapplied to medical decisions, to the terms"specificity" and "sensitivity." According toGalen and Gambino, "The sensitivity of a12**

test is the incidence of true-positive resultsobtained in patients who are known to havethe disease". It is often expressed as its12

complement. That is, the more untrue (false-negative) results, the less the sensitivity. "The specificity of a test is the incidence oftrue-negative results obtained in patients whoare known to be free of the disease". It is12

See also Section 3 of this document.**

December 1996 H26-A


often expressed as its complement. That is, or reference interval is taken to be the center the more untrue (false-positive) results, the 95% of the distribution. Assay results thatless the specificity. fall above the upper reference limit or below

The term "patients who are known to have normal" or, in Fisher's terminology,the disease" is redefined for this standard as "formally significant.” patients with true assay values that eitherviolate the reference limit or the clinical action Laboratories should establish their ownlimits. The term "patients who are free of the reference limits that are appropriate for theirdisease" is redefined as patients with true local populations. NCCLS document C28-A,assay values that are within reference limits or How to Define, Determine, and Utilizedo not violate the clinical action limits. Reference Intervals in the Clinical Laboratory;

Both bias and imprecision can affect establishing reference limits.specificity and sensitivity. Following areexamples of both situations. Figure 1 is a Gaussian model of the

5.1 Effect of Assay Bias

5.1.1 Effect on Reference Limits

The distribution of analyte values in a healthypopulation may only approximately follow anormal or Gaussian curve. The normal range

the lower reference limit are considered "non-13

Approved Guideline, provides methods for

distribution of hemoglobin assay results in ahealthy population (heavy line). The lighter***

lines model the effects of calibration 3.75 g/L(3%) and 6.23 g/L (5%) below an unbiasedset point. Figure 2 is an enlargement of thelower reference limit of this dispersion.

The data on which this and other figures in

subcommittee. This information is intended to illustrateonly the concepts discussed in this section.

***

Section 5 are based wereprovided by a member of the

December 1996 H26-A


Assuming the unbiased set point to be 125.0 This example shows that biased referenceg/L and linear propagation of error through the limits can lead to diagnostic error. Assayreportable range, Table 3 shows the effect of results that fall near the reference limits cantwo magnitudes of bias at the lower reference be falsely labeled "normal" (FN) or falselylimit of 105.0 g/L and the proportion of false- labeled "abnormal" (FP), depending on thepositive results that would be engendered by direction and magnitude of the bias. Bias thatthese errors. A false-positive result is one affects both reference limits and clinicalthat is incorrectly placed outside the reference assays gives the appearance of mutuallimit. The effect of bias on the upper consistency, but distinctions between normalreference limit of 185.0 g/L is the inverse of and abnormal and the recognition of casesthis situation. that transgress clinical action limits might not

Table 3. Effect of Calibration Bias on Hb institution where like bias is not present. Accuracy at the Lower Reference Limit of 105.0 g/L If reference limits are established by a biased

Bias % Error g/L False Positive %

3.0 !3.2 52

5.0 !5.3 88

The percentage proportion of false values iscalculated as follows:

Subjects at unbiased limit = 2.5%Subjects at 3% biased limit = 3.8%Incremental change = (3.8 ! 2.5) = 1.3Increment % = (1.3 ÷ 2.5) @ 100 = 52%

be consistent with results obtained in another

analyzer and the bias is subsequentlycorrected, the proper recognition of specimensthat violate the reference limits is jeopardizedin a similar way.

5.1.2 Effect on Action Limits

Generally, action limits used in the clinicalinterpretation of test results depend on thereference ranges established for the test. If the analyzer response changes, the testvalues do not correspond to the clinician'sreference points and this can result inincorrect interpretation. The specific actionlimit used for each clinical decision depends

December 1996 H26-A


on the clinical problem and the circumstances patient populations, estimates of changes ofinvolved for that specific patient, but these analytical sensitivity and specificity as aaction limits can also be related to the function of bias should be made from the data(normal) reference intervals of the pertinent base of each institution. analyte(s) and are not based on theassumption that these are not biased. The effect of calibration bias on the rate of

The relationship between the magnitude of leukocyte counting is demonstrated byanalytical bias and the clinical sensitivity and modeling a common clinical situation asspecificity of the assay depends on where the shown in Figure 4. Let it be assumed that adecision values lie on the curve of the WBC value of 3.0 x 10 /L is the clinicaldistribution of test values and the slope of the decision level for a chemotherapy regimen. Adistribution curve at that position. The WBC value above the decision level permitsdistribution of hemoglobin test values in a continuation of therapy. A WBC value belowmajor primary care institution differs that level suggests that therapy should besignificantly from the distribution of test discontinued. For the purposes of analysis, itvalues in a healthy population. This is is assumed that bias induced at the calibrationillustrated in Figure 3. The institutional set point (8.0 x 10 /L) is linearly propagatedpopulation is skew-ed in the direction of the to the decision level.most frequently encountered types ofpathology (anemias) and practicallyuninfluenced by rarely encountered diseases(polycythemia). From this example, it is clearthat the dispersion of clinical values cannot betreated as Gaussian with a simple downward-shifted mean. Because of the variability of

false-positive and/or false-negative results in

9

9

December 1996 H26-A


A false-positive result is one that makes the They are given here to enhance theWBC level fall falsely below 3.0 x 10 /L, understanding of the effect of imprecision and9

leading to an unnecessary cessation of inaccuracy on the reliability of diagnostictherapy. A false-negative result is one that criteria. These proposed values are for guid-makes the WBC result rise falsely above 3.0 x ance only. The unique needs and experience10 /L, leading to continuation of therapy in of each institution will exert a modifying9

the face of true depressed leukopoiesis. influence.Figure 4 shows the relationship between biasat the calibration set point and the percentage Different criteria are needed for children orof false results at the clinical decision level. neonates. For example, normal MCV inExpressed in this manner, it appears that children ages 3 to 10 may be as low as 70 fLnegative calibration bias shouldnot exceed and as high as 120 fL in neonates. 0.10 x 10 /L and positivecalibration bias Populations resident above 1,000 m may9

should not exceed 0.20 x 10 /L. require upward adjustment of Hb and RBC9

5.1.3 Tabulation of Action Limits to the hemodilution effect of pregnancy where

Table 4 provides a subcommittee consensusof action limits or quantitative flags. These Examination of a stained blood film may bedecision values are also supported by Kost. useful in cases that violate these limits.

action values. Consideration should be given

appropriate.

14

December 1996 H26-A


Table 4. Suggested Action Limits

Analyte Action Limits Clinical Relevance: Abnormal Results May Reflect(Units) the Following Conditions

WBC hypoplasia, cobalamin, folate, iron deficiency.(10 /L)9

< 3 Sepsis, chemotherapy, radiotherapy, agranulocytosis, marrow

>12 Acute stress (including surgery), infection, malignancy,lymphoma, leukemia.

RBC(10 /L)12

Dehydration, polycythemia, shock, chronic hypoxia.Hb >180% >160&(g/L)

Hct >0.54%L/L >0.48&

RBC <4.4% <3.9&(10 /L)12

Anemia from blood loss, cobalamin, folate, iron deficiency,malignancy, chronic inflammation, chronic liver disease, renaldisease, marrow hypoplasia, chemotherapy, radiotherapy,hemolysis, hemoglobinopathy, thalassemia.

Hb <120% (g/L) <110&

Hct <0.39%L/L <0.30&

MCV <80 inflammation, hemoglobinopathy, thalassemia, sideroblastic(fL) anemia.

Microcytosis from iron deficiency, chronic blood loss, chronic

>100 Macrocytosis from chronic liver disease, cobalamin or folatedeficiency, sprue, smoking, hemolysis.

Plt(10 /L)9

<50 Risk of bleeding. Idiopathic, chemotherapy, radiotherapy.

>800 Risk of thrombosis. Polycythemia, post splenectomy,thrombocythemia.

%, men.&, women.

5.2 Effect of Assay Imprecision

5.2.1 Effect of Analytical Imprecision

Clinical interpretation of assay results isinfluenced by two forms of imprecision. Oneis analytical imprecision as discussed inSection 4.5. Apart from its influence on thecalibration set point, this form of imprecisiondetermines the degree to which the results ofrepeated assays of the same specimen mayrandomly disagree. The clinician should beaware of the possible range of results

(confidence limits) that can be given by asingle assay and the clinician should relatethis range to diagnostic requirements.

Analytical imprecision affects two types ofdecisions:

! Is the value above or below thereference interval limit?

! Does the value violate the clinicaldecision action limit?

>6.2% >5.2&

December 1996 H26-A


It is customary to assume that the reported The laboratory should publish its observedvalue falls on the mode of all possible values 95% confidence limits for all analytes so thatfor that assay. In fact, there is a diminishing clinicians can relate the imprecision of aprobability that the value will fall to one or the reported result to the diagnostic problem. other side of the mode. In the event that a Manufacturers' claims of analyticalclinical decision requires a more exact imprecision should not be used for thisstatement of analyte concentration than is purpose unless they have been confirmed bygiven by a single assay, the mean of multiple test results obtained under actual conditionsassays can be used. Figure 5 uses of use. hemoglobin assay as an example of the wayin which replicate assays can reduce 5.2.2 Effect of Biological Variabilityinterpretive error due to analytical imprecision. This example assumes a "true" or mean Short-term variation of analyte concentrationvalue of 125.0 g/dL and the 95% confidence within an individual person (biologicallimits of a single assay to be ± 3.0 g/L. variability or diurnal variation) was introducedBecause the confidence limits shrink as a in Section 4.5. It can provide a usefulfunction of the square root of the number of yardstick against which to set goals forassays contributing to the mean, little benefit analytical imprecision. Ideally, analyticalis to be gained by performing more than four imprecision should not exceed 25% ofreplications. However, the mean of duplicate biological variability. A ratio of diurnal toassays improves the confidence limits of the analytical variation that does not meet thisresult to ± 2.0 g/L (3.0 ÷ /2). criterion can blur the distinction between

these two causes of change when makingrepeated assays on an individual person.

SDdiurn ' SD 2total & SD 2

analytic

December 1996 H26-A


Table 5 describes the diurnal variation of CBC change of MCV in a patient being monitoredanalytes in healthy subjects from data for response to cobalamin therapy isprovided by three major institutions. measured in days. The time required for the****

Diurnal SD (SD ) is calculated by detection of a significant WBC change in adiurn

subtracting the analytical variation (SD ) patient being monitored for response toanalytic

from the total 24-hour variation (SD ) as antibiotic therapy is measured in hours. total

follows: Considering these extenuating conditions,

Diurnal CV was calculated as (SD ÷ account when evaluating violation ofdiurn

Population Mean) @ 100. reference intervals and action limits.

The column headed “Ratio” provides a figure- The effect of diurnal variation on theof-merit that relates diurnal to analytical measurement of reference intervals has beenvariation. Values below 0.25 indicate that the raised. It will contribute uncertainty (butcriterion for analytical precision has been met. not bias) in the tails of distributionManufacturers should consider within-person histograms. This is most likely for WBC andvariability to be immutable and should relate least likely for MCV.their design goals for analytical imprecision toit. For CBC analytes, the simplistic goal of ____________________analytical SD being 0.25 of the diurnal SDmight not be appropriate. The physiologicalrate of change of the analyte, as well as theclinical implications of measurement error,must be taken into account. For example, thetime required for the detection of a significant

only Plt measurement exhibits unsatisfactoryperformance. Plt measurement at criticalvalues, such as 50 x 10 /L, should, ideally,9

have an analytical SD of the order of 3.0 x10 /L.9 15

Within-person variability should be taken into

16

****Data provided by University of California,San Francisco General Hospital; Veterans Affairs MedicalCenter, Minneapolis; and the University of MiamiHospitals and Clinics.

Table 5. Population Mean Assay Values, Coefficients of Variation of Diurnal Changes, and theRatios of Within-Person Variation, S , to Analytical Imprecision, S .diurn anal

Analyte Units Mean CV% S ÷SPopulation Diurnal Ratio

anal diurn

WBC 10 /L 7.0 14 0.129

RBC 10 /L 4.8 3.5 0.2912

Hgb g/L 142.0 3.0 0.31

Hct L/L 0.425 3.7 0.4

MCV fL 89.2 0.5 1.9

Plt 10 /L 257 5 0.539

December 1996 H26-A


Appendix

Values of t for n Iterations at P (95% Confidence).05

n t n t

2 4.3027 17 2.1098

3 3.1824 18 2.1009

4 2.7764 19 2.0930

5 2.5706 20 2.0860

6 2.4469 21 2.0796

7 2.3646 22 2.0739

8 2.3060 23 2.0687

9 2.2622 24 2.0639

10 2.2281 25 2.0595

11 2.2010 26 2.0555

12 2.1788 27 2.0518

13 2.1604 28 2.0484

14 2.1448 29 2.0452

15 2.1315 30 2.0432

16 2.1199 31 2.0395

December 1996 H26-A


References

1. Wenz B, Ramirez MA, Burns ER. The 9. Statland BE, Winkel P, Harris SC, etH1 hematology analyzer. Arch Pathol al. Evaluation of biologic sources ofLab Med 1987; 111:521–524. variation of leukocyte counts and

2. Billet HH, Simson E, Main P, Bailey C, very precise automated analyzers. AmGuerra P. The MAXM hematology J Clin Pathol 1978;69:49–54.auto-analyzer. An alternative? Am JClin Pathol 1994;102:36–44. 10. Cotlove E, Harris EK, Williams GZ.

3. Sheridan BL, Lollo M, Howe S, variation in long-term studies of serumBergeron N. Evaluation of the Roch constituents in normal subjects. ClinCobas Argos 5Diff automated Chem 1970;16:1028–1032.hematology analyzer with comparisonto a Coulter STKS. Clin Lab Haematol 11. Poisson SD. Recherches sur la1994; 16:177–130. probabilité des jugements. Paris:

4. Hallawell R, O'Malley C, Hussein S, etal. An evaluation of the Sysmex NE 12. Galen RS, Gambino SP. Beyond8000 hematology analyzer. Am J Clin normality: the predictive value andPathol 1991;98:594–601. efficiency of medical diagnosis. New

5. Bentley SA, Johnson A, Bishop CA. Aparallel evaluation of four automated 13. Fisher RA. Statistical Methods forhematology analyzers. Am J Clin Research Workers. 6th ed. Edinburgh:Pathol 1993;100:626–632. Oliver and Boyd, 1936:46.

6. Richardson Jones A. Assignment of 14. Kost GJ. Critical limits for urgentassay values to Coulter controls and clinician notification at US medicalcalibrators. Clin Lab Haematol centers. JAMA 1990:2631990;12(suppl 1):23–30. (5):704–707.

7. ICSH Expert Panel on Cytometry. The 15. Richardson Jones A, Twedt D, Swaimassignment of values to fresh blood W, Gottfried E. Diurnal change ofused for calibrating automated blood blood count analytes in normalcell counters. Clin Lab Haematol subjects. Am J Clin Pathol (in press).1988;10:203–212.

8. Bull BS, Richardson Jones A, Twedt H, Junge B. Seasonal variation ofD, Gibson M. A method for the blood components important forindependent assessment of the diagnosis. Klin Wochenschraccuracy of hematology whole blood 1980;58(15):769–778.calibrators. Am J Clin Pathol1992;98:623–629.

other hematologic quantities using

Biologic and analytic components of

1837.

York: Wiley, 1975:10–11.

16. Rocker L, Feddersen HM, Hoffmeister

December 1996 H26-A


Additional Bibliography

Code of Federal Regulations. 21 CFR 809.10. Code of Federal regulations. 21 CFRLabeling for in vitro Diagnostic Products. US 493.1213. Standard: Establishment andGovernment Printing Office, Washington DC, Verification of Method Performance1992. Specifications. US Government Printing

Office, Washington, DC, 1992.

December 1996 H26-A


Summary of Comments and Subcommittee Responses

H26-P: Performance Goals for the Internal Quality Control of Multichannel Hematology Analyzers;Proposed Standard

General Comments

1. The issue of repeat testing of a specimen over an extended time period should be addressed. For example, a doctor asks us to repeat a CBC that we ran yesterday and the specimen isnow 24 hours older. What we need to know are the precision goals that relate to biologicaldegradation.

! The subcommittee recommends that manufacturers provide information on this issue. Different analyzer designs differ as to their sensitivity to degradation of the various analytesin stored specimens, which makes it difficult to give generalized statements.

2. It would be nice if the committee would prepare a standard method for linearity verificationfor multichannel hematology analyzers.

! In the approved-level standard, the committee avoided method descriptions. Another NCCLSdocument is being developed by the subcommittee that will be a companion document toH26-A, which will include such procedures.

3. The intended goals of this document are admirable, though many of us concur that thedocument is overly theoretical. I recommend more emphasis on the more pragmatic aspectsof this topic and a more succinct summary of the theoretical basis. The establishedperformance goals are outstanding and long overdue.

! The subcommittee appreciates the comment and hopes that the approved-level documentgives greater clarity.

4. The discussion of statistical concepts ought to be included as an appendix, rather than a part of the body of this work. I strongly urge the inclusion of the exact definition andaspects of medical usefulness in quality control.

! The statistical concepts are considered, by the subcommittee, to be inherent to the text. Their relocation would reduce the continuity of the discussion of principles. Section 5,"Relating Performance Goals to Medical Decisions" addresses this comment. A moredetailed discussion of quality control aspects will be given in a forthcoming companiondocument (NCCLS’s H38 project on moving averages of red cell indices), which is beingdeveloped by the subcommittee.

5. I recommend numbering the tables for easier reference.

! The tables are numbered in the approved-level document.

Section 3.0

6. The terms defined are not used consistently in the document and, apparently, sometimes synonyms are used, or the terms are not defined.

! Every effort has been made to ensure that terms are used consistently throughout theapproved-level document.

7. Calibration—Replace “a bias conversion factor” with “bias parameters” to account for the more general calibration model where there is more than just one (slope or proportional bias)parameter.

December 1996 H26-A


! The numerous comments dealing with the need to improve the section on definitions(Section 3) and to use terms consistently have been responded to by the subcommittee byrewriting Section 3.

8. Although “calibrator” is defined and used extensively throughout the document, the term “standard” is never defined or used. Because a primary standard is available for hemoglobin,this term should be included and distinguished from a calibrator.

! The subcommittee agrees; definitions for standard (measurement) and primary, secondary,and working standard are included in H26-A.

9. The definition for internal quality control, as written, is not clear.

! The definition has been revised to address the comment.

10. “Analytical bias - The numerical difference between the limiting mean.” What is a limiting mean?

! The definition for bias has been revised.

11. Why isn’t there a definition for analytical imprecision. There is one for analytical bias.

! See the definition for “precision” included in H26-A. 12. The terms “allowable error” and “error” are undefined. Should this be “analytical allowable

error”?

! See Section 4 for detailed information on analytical errors.

13. The term “assay bias” is undefined.

! See the definition for bias included in H26-A.

14. The terms “assay specificity” and “assay imprecision” are not defined.

! Section 3 has been revised to include definitions for “specificity” and “imprecision” (see“precision”).

15. The term “instrument precision” is undefined.

! A definition for precision is now included in Section 3 and a detailed explanation of theconcept now appears in Section 4.5.

Section 4.1.3

16. In Section 4.1.3, it is unclear what is meant by “easier to detect” and “determinestatistically.” If these refer to the number of controls needed to detect a statically out-of-control condition, the size of the standard deviation is irrelevant. The probabilities are thesame because the limits and out-of-control patterns are relative to the SD. If something elseis meant, it would be informative to explain what.

! This comment is pertinent to internal quality control and will be dealt with in the H38companion document that is being developed by the subcommittee.

Section 4.1.5

17. The whole section on biological variability and the effect of analytical bias and imprecision isdiscussed at the end rather than before the setting of analytical performance goals. Thus, it

December 1996 H26-A


seems that the effect is not used in deciding what the goals should be. For example, Figure3 states that a shift of 1 SD will decrease the specificity from 95 to 84%; yet theperformance goals are stated at ±3 SDs. Won’t the loss of specificity be so great as tomake the analytical results useless?

! Biological variation is of significance in revealing those aspects of analytical precision thatmight be improved. Section 5.2.2 of the approved-level document clarifies these issues.

Section 4.2

18. In Section 4.2, Goals for Analytical Bias, there is no reference to hematocrit (hct). Thisparameter is calibrated and measured on our analyzer; therefore, we need goals for this aswell. The hct is calculated on other analyzers, thus, goals are not necessary.

! Revisions to the document have corrected this omission.

19. Section 4.2 is confusing and inconsistent with the definitions.

! This document has been revised to correct this problem.

Section 4.2.1

20. In Section 4.2.1 the second sentence is not clear. Can it be stated in more detail as to howbias is calculated?

! This, and other parts of Section 4, have been revised to address this question.

Section 4.2.2

21. Section 4.2.2 assumes that errors at all levels are linearly proportional. This may not betrue, e.g., at high and low levels of platelets.

! The document has been revised to show that linearly proportional error is limited to thereportable or linear range of the instrument.

22. There is no discussion of calibration at clinically critical levels. This can be more importantthan at the normal, midrange, or even ends of the "normal" range.

! The clinical consequences of calibration-induced bias are considered in depth in various partsof the approved-level standard, particularly in Section 5. It is not within the scope of thedocument to require calibration at levels other than those recommended by the instrumentmanufacturer.

23. “Analytical instrument bias” as defined—low, medium and high range—are arbitrarilydefined.

! It is agreed that low, medium, and high bias limits were arbitrarily defined. Sections 4.1,4.2, and 4.3 in the approved-level document deal with this issue in detail.

24. There seems to be some confusion of terms. This section is entitled “Bias,” yet it proceeds

to talk about variability. This confusion might be because the term “variability” is undefined. Variability typically refers to imprecision. This section discusses the variability of bias,which doesn’t make much sense. Biases do not combine in a root mean square manner, sothis needs to be explained through the correct use of terms that are defined, and perhapsdefining other terms that might be necessary to explain.

December 1996 H26-A


! The approved-level document deals with the effect of confidence limits on bias range. Forcalculating single-tailed maximum bias limits, linear summing should be used. For predictingprobable net bias range, root-mean-square summing of the components of the confidencelimits is more appropriate.

25. As I understand calibration, and using this document’s definition, how can there be short-term calibration error? Each calibration produces a bias (it may be zero). Variation orimprecision now results from other factors, but not from that calibration. The stated valuehas absolutely no uncertainty. Its effect on bias might be uncertain, but that, by no means,indicates that the result is imprecise. Also the equation for SD seems irrelevant. It is nottotal

used anywhere in the document. This whole section on variation of bias is not used in thecalculation of performance goals.

! Revisions to the document address this comment.

Section 4.2.3

26. As stated in Section 4.2.3, the recommended bias goals for calibrator assignment,instrument calibration, and drift detection are calculated as ±3 times the estimated error ofeach process. What happens to the four sources of bias?

! The subcommittee revised this recommendation. Sections 4.1, 4.2, and 4.3 in H26-A dealwith this issue in detail.

Section 4.2.4

27. As stated in Section 4.2.4, the Poisson counting error adds only to precision. Should thisbe imprecision?

! Yes. The effect of Poisson error is discussed at greater length in Section 4.5 of theapproved-level document.

Section 4.2.5

28. Calibrator assignment goals (vial-to-vial maximum difference) would be difficult for the user to monitor, because, in most cases, only two vials are shipped at a time, or, where multiplevials are shipped, only one or two are in use at a time. The document does go on to recommend that manufacturers report "calibration imprecision" on a long-term basis, buteither component is of little practical use. The numbers should be available for "comparisonshopping" between manufacturers.

! Error in the assigned values of the calibrator is a significant source of bias. This is discussedin Sections 4.1 and 4.7 of the approved-level document. The H38 document that is beingdeveloped by the subcommittee as a companion document to H26-A will deal in depth withthis and related matters.

29. As stated in Section 4.2, the values for the goals are expressed as coefficients of variation;presumably these are goals for analytical bias. Bias is defined as the difference betweenmean and true value. Why are the goals expressed as CVs and not as limits of allowabledifferences? If this is not comparing different things, then would explaining why not beinformative?

! The approved-level document addresses this comment. It is often convenient to use CV incases where error, such as calibration set point bias, is propagated in a linear manner. Expressing this goal as the limit of allowable error in units would require the value of the setpoint to be stipulated.

December 1996 H26-A


Section 4.2.6

30. The numbers given in the chart are not tight enough for our use. The example given allowsa ± 0.4 difference on a WBC of 8.0, whereas the manufacturer of the calibrator we useallows only a 1.25% or 0.1 difference. I would not want as much "play" in an instrument'sset point as the document allows.

! The subcommittee agrees and has revised this section accordingly. See Section 5.

Section 4.2.7

31. I assume the term "drift detection" refers to a standard QC program of controls every shift,but it is not defined. The numbers are closer to 3 SD than 2 SD. If these are derived from"short-term precision data" from the manufacturers, could they not be customized for eachinstrument, and in many cases, tightened?

! The subcommittee has minimized the implication that instrument drift is an inherentcharacteristic of automated hematology analyzers and has added further discussion on thetopic. The goals have been tightened and are now limited to mid-range values.

32. Two important topics (for us) relevant to short-term (shift-to-shift) drift detection were notdiscussed: moving averages and within-shift patient/precision verification.

! These matters will be discussed in detail in the H38 companion document that is beingdeveloped by the subcommittee.

Section 4.2.8

33. The numbers stated in Section 4.2.8 can be tightened only as the individual components aretightened.

! The subcommittee agrees. Note, however, that standards must attempt to embrace asbroad a spectrum of instruments as possible without degrading the practice of medicine.

34. We are assuming that Section 4.2.8 suggests goals (%) regarding instrument to instrumentvariation.

! The standard assumes that automated hematology analyzers, calibrated with the samematerial will have similar bias. References dealing with comparisons among differentmanufacturers are cited in Section 1.

35. Section 4.2.8 says that the goals were set at three times the short-term precision. Shouldthis be “imprecision”? Because, by the definition in this document, precision is measured inCVs and SDs, then three times would still make them CVs or SDs. Are not the values in thetable SDs? So instrument drift is comparing a change (difference?) to an SD? Could this beexplained more carefully.

! The subcommittee agrees that this statement needs both clarification and justification. Thedocument has been revised accordingly.

36. Section 4.2.8 says that errors are combined linearly. Yet, they are the sum of the threegoals that are stated as SDs and, in Section 4.2.2, there was an equation for SD thattotal

combined them in a square root manner. Are the goals for the table in SD units?

! The subcommittee agrees that this statement requires clarification and justification. Therefore, Section 5 in the approved-level document addresses the clinical consequences ofimprecision and bias and uses consistent terminology.

December 1996 H26-A


Section 4.2.9

37. I believe that the stated lower limit (recommended) of hemoglobin at 2 g/dL is unrealisticallylow and a more practical figure of 2.5 or 3 g/dL should be considered.

! The subcommittee agrees that an assay result of 2.0 g/dL is clinically unlikely. This value isa design goal. Design goals should always exceed the maximum expected clinical range.

Section 4.3

38. The "goals for precision" are much more liberal than the "analytic imprecision numbers(Section 5.2.2), and it is not clear which set of numbers would be used in practice.

! The revisions in H26-A in Sections 4.5 and 5.2 address this question.

39. After having three tables with goals stated on SDs in the section on analytical bias, thedocument now addresses precision goals. This is confusing.

! The approved-level document now uses consistent terminology (either SD or confidencelimits) in the numerous references to imprecision. In the few cases where CV is moreappropriate, its use is explained.

Section 6.0

40. The section on quality control should be more elaborate. This section could be more usefulif it explained how out-of-control rules would help maintain certain minimum true-positiveand true-negative rates. I think the section is weak in making a connection between theperformance goals recommended in Section 4.2 and the out-of-control rules recommended.

! To correct this situation, the subcommittee has undertaken the preparation of a companiondocument (i.e., NCCLS’s H38 project on moving averages of red cell indices) that makesrecommendations for calibration and quality control that are consistent with performancegoals.

Section 6.2

41. Section 6.2 states that the quality control system for an instrument with good accuracy andprecision could be less elaborate than (one) with less ideal performance. The accuracy andprecision is irrelevant to whether the quality control system is more or less elaborate, if oneis speaking of statistical quality control. The issue of elaborateness depends on the stabilityof the system and what type of change one wants to detect. Because this section statesthat the performance goals are the same for all systems, then only the former issue is ofconcern. However, it is not addressed in the document.

! See the response to Comment 40.

Section 6.3

42. Though the recommendations to manufacturers are laudable, I don't believe compliancewould be readily forthcoming. Perhaps development of these number sets could come fromagencies with access to data, i.e., College of American Pathologists.

! The subcommittee agrees that manufacturers might find difficulty in complying with theoriginal Section 6.3. This will be addressed in greater depth—within the framework ofvarious regulations with which compliance is required—in the companion document that isbeing developed by the subcommittee.

December 1996 H26-A


Related NCCLS Documents

C28-A How to Define, Determine, and Utilize Reference Intervals in the Clinical Laboratory;Approved Guideline (1995). C28-A provides a protocol for the determination ofreference ranges for defined populations as an aid to the interpretation of laboratorydata.

EP9-A Method Comparison and Bias Estimation Using Patient Samples; Approved Guideline(1993). EP9-T discusses procedures for determining the relative bias between twomethods or devices; the design of a method-comparison experiment using splitpatient samples; and analysis of the data.

H7-A2 Procedure for Determining Packed Cell Volume by the MicrohematocritMethod—Second Edition; Approved Standard (1993). H7-A2 discusses the standardmicrohematocrit method for determining packed cell volume, recommendedmaterials, and information on potential sources of error.

H15-A2 Reference Procedure for the Quantitative Determination of Hemoglobin inBlood—Second Edition; Approved Standard (1994). H15-A2 describes theinternationally accepted reference method for hemoglobin against which all methodsshould be evaluated. It includes specifications for the accurate measurement ofblood hemoglobin content using the hemiglobincyanide method, the construction ofstandard curves and calibration graphs or tables, and the special characteristics ofHiCN solutions suitable for use as secondary calibration material, including thecalculation of HiCN content from spectrophotometric measurements.

H20-A Reference Leukocyte Differential Count (Proportional) and Evaluation of InstrumentalMethods; Approved Standard (1992). H20-A discusses automated differentialcounters and establishes a reference method based on the visual (or manual)differential count for leukocyte differential counting to which an automated ormanual test method can be compared.

H44-P Reticulocyte Counting by Flow Cytometry; Proposed Guideline (1993). H44-P offersguidelines to help laboratorians in reticulocyte counting by flow cytometry. Itincludes factors that affect the accuracy and precision of reticulocyte counting, anda recommended reference procedure.

M29-T2 Protection of Laboratory Workers from Infectious Disease Transmitted by Blood,Body Fluids, and Tissue—Second Edition; Tentative Guideline (1991). M29-T2provides guidance on the risk of transmission of hepatitis B virus and the humanimmunodeficiency virus in the laboratory. Specific precautions for preventing thetransmission of bloodborne infection during clinical and anatomical laboratoryprocedures are addressed.

NRSCL8-P2 Nomenclature and Definitions for Use in NRSCL and Other NCCLSDocuments—Second Edition; Proposed Guideline (1993). NRSCL8-P2 providesproposed definitions for use in NCCLS standards and guidelines, and for submittingcandidate reference methods and materials to the National Reference System for theClinical Laboratory.