Printer Friendly

Variation with time in components of variance for measurements of therapeutic drugs.

Analyses for therapeutic drugs are commonly made with automated, high-throughput, multichannel instruments that require minimal operator intervention. Performance monitoring of these assays is accomplished by means of internal and external (proficiency testing) quality-control procedures. Proficiency test schemes have demonstrated that the within-laboratory sources of variation are more important than between-laboratory sources (1-3), and the College of American Pathologists laboratory improvement program showed that the within-laboratory variance doubled for samples measured 4 months apart compared with measurements made at the same time (2). The present study was designed to track this decrease in within-laboratory precision over time to provide insights into the possible sources of imprecision in routine clinical measurements of therapeutic drugs in serum.

A lyophilized proficiency test sample was prepared by adding midtherapeutic concentrations of 14 drugs to 5.9 L of human serum (Scipac Ltd.). Drug concentrations were as follows: phenytoin, 15.2 mg/L; phenobarbital, 30.3 mg/L; primidone, 7.1 mg/L; carbamazepine, 7.7 mg/L; carbamazepine 10,11-epoxide, 1.9 mg/L; ethosuximide, 68.1 mg/L; valproate, 76.8 mg/L; clonazepam, 42.0 [micro]g/L; lamotrigine, 3.1 mg/L; theophylline, 15.0 mg/L; caffeine, 7.8 mg/L; digoxin, 1.2 [micro]g/L; gentamicin, 2.7 mg/L; and lithium 0.76, mmol/L. The CV of dispensing of test sample aliquots by weight was 0.08%. We distributed 5 differently coded aliquots of the proficiency test sample, on 4 occasions, for analysis by members of the United Kingdom National External Quality Assessment scheme for drug assays. The scheme has an international membership of mostly hospital- or clinic-based sites, with 60% of their participants from the United Kingdom, 30% from Europe, and 10% outside Europe. A pair of samples was sent for analysis 1 month, and 3 single samples were sent at intervals to permit comparisons between pairs of measurements on the same sample, measured 0 to 6 months apart. The 5 samples formed part of the routine circulation of materials, and their identity was blinded to scheme participants. Laboratories reported the measured drug concentrations for their available range of drug assays and, for each measurement, the analytical technique used.

Drug measurements for all techniques combined were screened individually for the 5 sample distributions to reject those >3 SD from the sample mean (4). The percentage of outliers for each technique was as follows: HPLC, 4.1%; Abbott TDx, 1.9%; Abbott AxSYTM, 0.6%; Roche fluorescence polarization immunoassay (FPIA), 1.9%; Roche kinetic interaction of microparticles (KIMS) immunoassay, 1.5%; Roche Tina-quant, 1.8%; Beckman turbidimetric assay, 2.6%; Bayer chemiluminescent assay, 0.9%;cloned enzyme donor immunoassay (CEDIA), 4.3%; Olympus, 3.5%; Vitros, 2.8% (see footnotes to Table 1 for manufacturers of assays used). Data were taken for analysis when a laboratory had reported nonrejected data for a drug for all 5 samples assayed by the same technique. Data for each analytical technique were analyzed independently, thereby excluding the known variation attributable to differences in accuracy among techniques (3). Data for a technique were analyzed when data for a drug were available from >10 scheme participants. Results are presented for 6 of the study analytes, for which data from >5 techniques were available for comparison. The 6 selected drugs and the number of laboratories using the different techniques are listed in Table 1. Data for the other drug/technique combinations are listed in Table 1 of the Data Supplement that accompanies the online version of this Technical Brief at For each drug-technique combination, ANOVA between samples and laboratories was used to extract within-and between-laboratory components of variance. References to variance in the text refer to variances, whereas Table 1 and Fig. 1 present the data in terms of CVs.

Stability of the drugs in the proficiency test sample over time was monitored by 2-way ANOVA between samples and laboratories of the measurements reported. Betweensample differences (P < 0.05) were detected for 3 of the 14 analytes, and these cases were investigated further by linear regression. For valproate, for which the maximum difference between samples was 1.7%, the variation was not time-related, there being no significant (P > 0.4) linear regression with time. For carbamazepine and gentamicin, a significant (P <0.05) time-dependent decrease in drug concentration was detected, equivalent to 2.0% over 6 months for carbamazepine and 5.3% for gentamicin. The measurement data were corrected before analysis in the latter 2 cases to compensate for the observed decrease in concentration.

For each drug measured by each technique, 3 components of variance were estimated from the variances calculated between pairs of samples analyzed at different time intervals. The 2 samples circulated in the same distribution were assumed to have been analyzed by laboratories in the same assay. The within-laboratory variance measured between this pair of samples is thus a measure of within-assay variance. The within-laboratory, between-assay variance was estimated as the mean difference between the within-and between-laboratory components of variance at month 0 and the mean for months 2 to 6. The third variance component, between-laboratory variance, was estimated by the mean value of the between-laboratory variance between pairs of samples for months 2 to 6.

The between-laboratory variance measured between pairs of samples circulated in separate distributions decreased initially with increasing time intervals between samples, to stabilize after a period of 2 months. Within-laboratory variance showed an inverse relationship. Time plots for the 6 drugs measured by >5 techniques are presented in Fig. 1; plots for the other drug-technique combinations are shown in Fig. 1 of the online Data Supplement. The time-dependent changes are attributable to between-assay variability within a laboratory, which will increase as the time interval between drug analyses increases. Comparison between techniques failed to show consistent differences. Indeed, within the limits of the precision of these correlated data, the time course of the change in variance estimates was remarkably constant for all drug-technique combinations. On average, across all 14 analytes included in the study, 71% of the change in variance occurred in the first month, a lesser decrement or increment occurred in the second month, and there was no change thereafter. The estimates of the 3 components of variance are presented in Table 1 and in Table 1 of the online Data Supplement as CVs.


All 3 components of variance differed (P < 0.05, 3-way ANOVA between the drug, technique, and variance components, followed by the Student-Newman-Keuls post hoc test), with the mean within-laboratory, between-assay variance > within-assay variance > between-laboratory variance, being 60%, 24%, and 16%, respectively, of the total variance. Immunoassays for digoxin were less precise than for other analytes (P < 0.05, Student-Newman-Keuls test). No significant differences in variance components were detected between techniques (P > 0.05) in within-drug comparisons.

The present data confirm that the largest variance component in measurements of therapeutic drugs in serum is within-laboratory, between-assay variance. In our study, on average, it was 60% of the total variance. The novel finding of the study is that between-assay variance resulted from within-laboratory factors that occurred, for the most part, within a 1-month time window. Importantly, the time course of accumulation of these measurement differences was similar for HPLC and for a range of commercial immunoassay techniques. This suggests that their sources are most likely general in nature and are not technique specific. The latter finding, and the short duration over which these accumulations occurred, is inconsistent with the source of the variation being changes in lot numbers of reagents, instrumentation drift, or calibration changes that occur over longer periods. It seems likely, therefore, that this, the largest component causing imprecision in therapeutic drug monitoring results, will be open to manipulation by individual laboratories and not to reagent or analyzer manufacturers. To minimize this measurement variation, laboratories should look to local influences that occur within a daily or weekly time frame. They may include analyte sequence effects from nondrug analyses undertaken on the same instrument as well as carryover effects.


(1.) Tsanaclis LM, Wilson JF. Intra-and interlaboratory sources of imprecision in drug measurements by different techniques. Clin Chem 1993;39:851-5.

(2.) Steele BW, Wang E, Palomaki G, Klee GG, Elin RJ, Witte DL. Sources of variability: a College of Amercian Pathologists therapeutic drug monitoring survey study. Arch Pathol Lab Med 2001;125:183-90.

(3.) Wilson JF, Watson ID, Williams J, Toseland PA, Thomson AH, Sweeney G, et al. Primary standardization of assays for anticonvulsant drugs: comparison of accuracy and precision. Clin Chem 2002;48:1963-9.

(4.) Healy MJR. Outliers in clinical chemistry quality-control schemes. Clin Chem 1979;25:675-7.

Previously published online at D01: 10.1373/clinchem.2005.056499

John F. Wilson [1] * and Kathleen Barnett [2] ([1] Department of Pharmacology, Therapeutics and Toxicology, Wales College of Medicine, Cardiff University, Cardiff, Wales, United Kingdom; [2] Cardiff Bioanalytical Services Ltd., Cardiff, Wales, United Kingdom; * address correspondence to this author at: Cardiff Bioanalytical Services Ltd., 16 Mount Stuart Square, Cardiff CF10 5DP, Wales, United Kingdom; fax 44-29-2048-9003, e-mail
Table 1. Components of variance for different analytical techniques and
number of laboratories using each technique for measurement of
therapeutic drugs in samples distributed by the United Kingdom National
External Quality Assessment Scheme. (a)

 CV, %
Technique (b) Phenytoin Phenobarbital Carbamazepine

HPLC 2.7; 2.8; 1.6 1.6; 3.4; 2.4 3.2; 6.0; 3.1
 n 21 21 24
Abbott TDx 1.4; 3.3; 1.4 3.8; 2.8; 2.2 2.4; 3.0; 1.5
 n 18 17 14
Abbott AxSYM 2.3; 2.7; 1.3 3.9; 0.9; 2.3 2.5; 3.5; 2.9
 n 32 30 32
Roche FPIA 2.0; 3.0; 2.5 1.7; 2.2; 2.7 2.2; 4.2; 2.0
 n 43 28 43
Roche KIMS
Roche Tina-quant
Beckman turbidimetric 1.5; 3.1; 1.7 1.7; 4.0; 1.4 3.9; 6.4; 1.3
 n 19 11 16
Bayer chemiluminescent
CEDIA 2.1; 5.4; 1.9 1.8; 4.5; 2.5 2.1; 4.0; 2.0
 n 31 26 37
Olympus 2.0; 4.3; 3.0 2.7; 5.1; 1.7 2.4; 4.0; 2.3
 n 23 14 26
Vitros 1.8; 4.3; 0.1 1.4; 4.7; 2.2 3.0; 3.4; 2.1
 n 20 17 17
Mean of all techniques 2.0; 3.6; 1.7 2.5; 3.3; 2.2 2.5; 4.5; 2.2

 CV, %
Technique (b) Valproate Theophylline Digoxin

Abbott TDx 1.6; 3.7; 2.0 1.4; 2.5; 1.9
 n 17 14
Abbott AxSYM 2.4; 3.4; 2.1 1.9; 2.5; 2.3 5.6; 5.4; 2.5
 n 36 26 25
Roche FPIA 1.4; 2.8; 2.2 2.3; 2.8; 2.4
 n 38 40
Roche KIMS 5.6; 7.1; 1.1
 n 25
Roche Tina-quant 4.1; 7.0; 2.9
 n 27
Beckman turbidimetric 1.3; 6.4; 1.8 2.0; 3.9; 1.0 7.2; 7.6; 5.1
 n 13 21 21
Bayer chemiluminescent 3.3; 4.6; 1.7
 n 18
CEDIA 1.6; 3.7; 1.6 2.5; 4.1; 1.9
 n 25 28
Olympus 2.5; 3.1; 2.3 1.9; 4.4; 3.1 3.4; 8.6; 3.3
 n 18 28 35
Vitros 6.5; 5.9; 4.8
 n 18
Mean of all techniques 1.8; 3.9; 2.0 2.1; 3.4; 2.1 5.1; 6.6; 3.1

(a) CVs (%) are presented as within-assay variance; within-laboratory,
between-assay variance; and between-laboratory variance, with n being
the number of laboratories reporting.

(b) Manufacturers: TDx and AxSYM (Abbott Laboratories, Diagnostics
Division); Roche FPIA, KIMS, and Tina-quant (Hoffmann-La Roche Ltd.,
Diagnostics Division); Beckman turbidimetric assay (Beckman Coulter
Inc.); Bayer chemiluminescent assay (Bayer HealthCare, Diagnostics
Division); CEDIA (Microgenics Corp.); Olympus (Olympus Diagnostica
GmbH); Vitros (Ortho-Clinical Diagnostics).
COPYRIGHT 2005 American Association for Clinical Chemistry, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2005 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Technical Briefs
Author:Wilson, John F.; Barnett, Kathleen
Publication:Clinical Chemistry
Date:Dec 1, 2005
Previous Article:Serum tartrate-resistant acid phosphatase 5b or amino-terminal propeptide of type I procollagen for monitoring bisphosphonate therapy in...
Next Article:Detection of MC1R polymorphisms with protease-mediated allele-specific extension as an alternative to direct sequencing.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |