Available online at www.sciencedirect.com

(2008) 177–179

Clinical Biochemistry 41

Use of 2 years of patient data to estimate intra-laboratory totalimprecision of HbA1c measured by multiple HPLC analyzers

D.V. Tran a, George S. Cembrowski a,⁎, T.N. Higgins b

a University of Alberta Hospital Edmonton, Alberta, Canadab Dynacare Kasper Medical Laboratories Edmonton, Alberta, Canada

Received 30 March 2007; received in revised form 26 October 2007; accepted 29 October 2007Available online 13 November 2007

Abstract

Background: Analytic imprecision is used to assess the acceptability of HbA1c methods performed on a single analyzer.Whenmultiple analyzersare used interchangeably in a laboratory, the analytic imprecision is usually increased and can obscure the detection of a genuine HbA1c trend or resultin an artefactual patient trend. We have estimated the imprecision of HbA1c testing of patient specimens by three HbA1c analyzers independent ofreference sample analysis.

Methods: Over 2 years, approximately 150,000HbA1cmeasurements were obtained from any one of three different Bio-RadVARIANT II HPLCanalyzers operated in a large reference laboratory. We tabulated the HbA1c measurements of paired intra-patient blood samples drawn within 30 daysof each other. We calculated the standard deviations of duplicates (SDD) of the intra-patient HbA1c pairs grouped by the following time intervals: 0–3 days, 4–6 days, 7–9 days, 28–30 days. The SDDswere then regressed against time with extrapolation to zero time representing the random analyticerror.

Results: At a mean HbA1c of 7.16%, the total analytic imprecision (coefficient of variation [CV]) is 3.6%.Conclusions: This variation is remarkably low, given that the HbA1c measurements were obtained over a 2-year period on any one of three

analyzers and the long-term within-analyzer CVwas usually 2.3–3.1% as assessed by reference control analysis. This approach could be extended toall HbA1c analyzers since unlike reference control statistics, the patient-derived random error should allow easy comparison of analytic imprecisionamong different analytical systems.© 2007 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.

Keywords: Glycohemoglobin; Hemoglobin A1c; Accuracy; Precision; Patient data; Quality control

Introduction

HbA1c measurements provide a 2- to 3-month estimate ofmean blood glucose and are the best measure of long-termglucose control. Since HbA1c measurements are used for patienteducation, counseling and ultimately patient motivation, itsmeasurement should be optimally accurate and precise. Basedon hemoglobin A1c (HbA1c) variation in intensively controlledtype I diabetes patients, we have previously proposed that long-term analytic coefficients of variation (CVs) be no more than2.1% [1]. This maximally allowable CV is significantly lower

⁎ Corresponding author.E-mail address: cembr001@cha.ab.ca (G.S. Cembrowski).

0009-9120/$ - see front matter © 2007 The Canadian Society of Clinical Chemistsdoi:10.1016/j.clinbiochem.2007.10.014

than those recommended by the National GlycohemoglobinStandardization Program [2] (NGSP) of 3% for laboratoriesinvolved in clinical trials and 4% for reagent/instrument ma-nufacturers. Skeie et al. [3,4], using patient expectations andphysician expectations, and Philipou and Phillips [5], using amathematical approach based on biological variation of HbA1c,came to the conclusion that a desirable precision goal for HbA1c

analysis was 3%.Both the American and Canadian Diabetes Associations have

set a 7.0% treatment goal for HbA1c. At this HbA1c concentra-tion and an average biological variation for HbA1c of 1.7%, thecritical difference (the minimum difference required to signify agenuine change in the HbA1c concentration [two-tailed test,Pb0.05]) is 0.5%, 0.7%, 0.85% and 1.0% at HbA1c analytic CVsof 2.0%, 3.0%, 4.0% and 5.0%, respectively.

. Published by Elsevier Inc. All rights reserved.

Table 1The number of paired HbA1c values used to calculate the SDD for each 3-dayinterval (inter-instrument variation)

Sample sizes—all variants

Midpoint of timeinterval, days

All points Within +3.5 SD Within +3.0 SD

1.5 361 323 3195 240 235 2328 337 334 33311 220 216 21414 392 382 38117 244 238 23720 368 362 35923 319 313 31026 354 352 34529 737 723 720N 3572 3478 3450

178 D.V. Tran et al. / Clinical Biochemistry 41 (2008) 177–179

Larger laboratories, both for the purposes of efficiency andredundancy, will operate multiple identical analytic systems.Use of replicate (but somewhat analytically dissimilar) systemscan introduce between-analyzer analytic error that increases thetotal analytic error that is encountered by the patient specimen.This total analytic error involves not only periods when themultiple analyzers run the same reagent lot but also periodswhen different systems may be using different reagent lots.While this total intra-laboratory/inter-instrument imprecision isnot overtly addressed by the NGSP, their Web site providessummaries of the College of American Pathologists (CAP)proficiency test results. These CAP inter-laboratory statisticsmight establish an upper limit for short-term inter-instrumentimprecision.

We describe a novel, independent approach for deriving thelong-term intra-laboratory/inter-instrument imprecision of

Fig. 1. Estimates of inter-instrument imprecision. SDD is plotted against time. The re100%) by excluding data outside 3.5 SD (labeled 99.95%) and by excluding data o

HbA1c in which summaries of intra-individual HbA1c variationsare plotted against the time between sampling. Linear regressionanalysis allows extrapolation of the variation to time zero wherethe intra-patient physiologic variation in HbA1c will be zero.The variation at time zero thus will correspond to the averageanalytic variation.

Methods

HbA1c measurements for 2 years (November 7, 2002, throughOctober 27, 2004) performed on patients with type 1 and type 2diabetes at Dynacare Kasper Medical Laboratories, a referrallaboratory, were selected. HbA1c measurements (150,048) wereperformed on any one of three cation exchange HPLC Bio-RadVARIANT II analyzers (Bio-Rad, Hercules CA). The analyzersare operated on a 24-h basis, 7 days a week. Quality controlsamples are analyzed every 100 to 200 specimens with columnreplacement usually every 500 samples. For a representativeinstrument, the 6-month CVs are 2.5% and 2.7% at HbA1c valuesof 5.7% and 9.6%, respectively.

Programmed query-basedVisual BasicMicrosoft Access wasused to select all of the intra-patient pairs of HbA1c samplesdrawn within 30 days of each other. All of the paired patientHbA1c were grouped into the following 3-day intervals betweensampling: 0 to 3 days, 4 to 6 days, 7 to 9 days, 28 to 30 days. Theaverage variations in these groups of paired HbA1c were cal-culated from the formula for the standard deviation of duplicates(SDD), SDD=√(Σ(xi1−xi2)2 / 2n) [6]. The SDD values weregraphed against the midpoints of the time interval and thenregressed against time with extrapolation to zero time represent-ing the long-term random error. The SDD values were re-calculated after elimination of any paired HbA1c differencesexceeding either ±3.5 SD or ±3.0 SD and again regressedagainst time.

gression lines are derived from using all of the paired observations (line labeledutside 3.0 SD (labeled 99.7%).

179D.V. Tran et al. / Clinical Biochemistry 41 (2008) 177–179

Results

A total of 3572 sequential intra-patient HbA1c pairs werecollected with the HbA1c separated by 0 to 30 days. Table 1shows the numbers of paired HbA1c data obtained over each 3-day interval for the inter-instrument analysis.

Fig. 1 illustrates the derivation of the inter-instrument va-riation. The inter-instrument analytic CV was calculated fromthe formula CV=100×(SDDt=0 /population mean). By exclud-ing 94 data pairs outside 3.5 SD (labeled 99.95%), the SDDt=0

decreases from 0.446 to 0.262; when 122 data pairs exceeding3.0 SD are excluded (labeled 99.7%), SDDt=0 decreases to aneven lower 0.259. With the population mean of 7.16% andSDDt=0=0.262 and 0.259, the 2-year inter-instrument analyticCV is 3.7% and 3.6%, respectively.

Discussion

Imprecision evaluations are usually based on the repeatedanalysis of reference samples. These estimates are dependent onthe characteristics of the reference sample including its matrixand stability. A reference sample that is submitted to the la-boratory in the same pre-analytical stream as the actual patientsample should demonstrate more variation than that submitted atthe analytical work station. The use of multiple analyzers shouldintroduce even more variation.

The imprecision usually stated by a manufacturer for theirHbA1c method reflects the only the variation at a single workstation and may represent the variation in no more than two orthree reagent lots.

For the three VARIANT II analyzers used in this study, thecumulative impression for a single analyzer, expressed as CV%,for a single lot of reference control for a 12-month period (themaximum length a single lot of control used in the time frame ofthe study) varied from 2.30–2.71% to 2.80–3.11% at HbA1c

concentrations of 5.62% and 9.60%, respectively.The current analysis considers the variation observed in

specimens separated over periods of up to 30 days over 2 yearsthat patient specimens were analyzed and reported. Very fewdata pairs were excluded (122 out of a total of 3572 [3.4%]); afew of these outlying data pairs might have been due to mi-sidentified specimens. The 3.5 SD exclusion criteria used iscommonly recognized to exclude outlier data points. In mostdata sets studied, this 3.5 SD criteria equated to a difference inthe HbA1c value of more than 1.5%. When patient records wereexamined, it was found that when the change in HbA1c levelexceeded 2%, there was a change in therapy usually from oralhypoglycemic medication to insulin therapy. It was felt in-appropriate to include these patients in the study as this does notrepresent true biological variation. Three-day groupings wereused to provide sufficient data points within a group and tomaximize the number of groupings within the specified timeframe. The 30-day time frame was chosen as HbA1c utilizationstudies showed a peak in reordering of HbA1c 28 days after theinitial order. A far more compelling reason for the selection of30 days is that the majority of the HbA1c value is weighted tothe last 30-day glycemic control rather than 120 days. This

weighting and bias is due to the body's natural destruction andreplacement of red blood cells. The analytic variation is due tothe variation in multiple reagent lots and in the three instruments,multiple calibrators and multiple calibrations as well as long-and short-term environmental differences including variations inambient temperature.

The cumulative patient-derived imprecision over three ana-lyzers compares very favorably with the reference control basedimprecision values.

The inter-instrument and inter-reagent random error averagesout to be 3.6% over 2 years. This is a low variation; the upperlimit CVof most NGSP-certified method in 2002 was 5%. Usingthe same patient-derived metrics, in a similar study performed atthe University of Alberta Hospital using a single Bio-RadVARIANT II over approximately 2 years, we obtained a randomerror of 2.8% [7]. We previously showed a patient-derivedrandom error of 2.6% over 6 months for that system [8].

On the basis of these, imprecision studies of HbA1c mea-surements obtained on the Bio-Rad VARIANT II system arerecommended for the precise measurement of HbA1c. The use ofthis approach by laboratories using other HbA1c methods tomeasure the imprecision of the HbA1c method would be of interest.An analytical CVof 3% as assessed by reference sample testing isprobably not adequate if multiple work stations are used to produceHbA1c results for a single patient population. A cursory review ofthe College of American Pathologists (CAP) inter-laboratorysurvey data indicates that the CV of most immunoassay exceeds3%, probably indicating that the intra-laboratory HbA1c measure-ment by multiple but identical immunoassay systems will providetoo high frequencies of defective data. To mitigate the effect ofincreased imprecision due to the use of multiple analyzers,manufacturers and laboratories should strive for decreasing theintra-instrument CV to the 2% recommended by Kolatkar et al. [1].

References

[1] Kolatkar NS, Cembrowski CS, Callahan PL, Etzwiler DD. Intensivediabetes management requires very precise testing of glycohemoglobin.Clin Chem 1994;40:1608–10.

[2] Little RR. Glycated haemoglobin standardization-National Glycohemoglo-bin Standardization Program (NGSP) perspective. Clin Chem Lab Med2003;41:1191–8.

[3] Skeie S, Thue G, Sandberg S. Interpretation of hemoglobin A1c(HbA1c)values among diabetic patients: implications for quality specifications forHbA1c. Clin Chem 2001;47:1212–7.

[4] Skeie S, Perich C, Ricos C, et al. Post analytical external quality assessmentof blood glucose and hemoglobin A1c: an international survey. Clin Chem2005;51:1145–53.

[5] Phillipou G, Phillips PJ. Intraindividual variation of glycohemoglobin:implications for interpretation and quality goals. Clin Chem 2003;39:2305–8.

[6] Cembrowski CS, Carey RN. Laboratory Quality Management. Chicago, IL:American Society of Clinical Pathologists; 1989. p. 14.

[7] Cembrowski GS, TranDV, Hofer TL, Lee T. Use of patient data to derivelong-term random error in laboratory assays: Application to glycohemoglo-bin testing. Poster presentation at the meeting of the Institute for QualityLaboratory Medicine, Atlanta, April 2005. http://www.phppo.cdc.gov/dls/IQLM/2005.

[8] Tran DV, Hofer TL, Lee T, Cembrowski GS. Unique approach to derivationof random error in laboratory assays; application to glycohemoglobintesting demonstrates poor clinical performance for immunoassay. DiabetesTechnol Ther 2003;5:975–8.